Financial Condition Reporting for South African Short Term Insurers

Draft Report Financial Condition Reporting for South African Short Term Insurers Calibration Project December 2005 Prepared for: Financial Services Board Prepared by: Deloitte and Insight ABC 1

Table of Contents 1 Glossary of Terms...4 2 Overview...5 2.1 Background, scope and overall framework... 5 2.2 Global comparisons... 9 2.3 Reliances and limitations... 10 2.4 Structure of the report... 11 3 Insurance Capital Charge... 12 3.1 Data Received... 12 3.2 Claims paid data store... 17 3.3 Ultimate Loss Ratio Investigation... 18 3.4 Framework... 20 3.5 Underlying risks and Risk Factors... 24 3.6 Notional companies... 25 3.7 Underwriting risk... 27 3.8 Reserving risk... 32 3.9 Earnings patterns... 33 3.10 Claims payment patterns... 35 3.11 Dynamic Financial Analysis... 38 3.12 Risk Measures... 41 3.13 Tables/Surfaces of Gross Risk Capital... 43 3.14 Diversification & Correlation... 46 3.15 Investment allowance... 48 4 Investment Capital Charge...50 4.1 Overview... 50 4.2 Investment capital requirements in other territories and in Life assurance... 50 4.3 Estimation of capital factors for SA short term insurance... 53 4.4 Calculating the investment risk capital charge... 59 5 Other charges... 61 5.1 Operational risk... 61 6 Total Capital Required...63 6.1 Covariance effects and Grossing-up... 63 7 Liability Estimation...66 7.1 Claims reserves... 66 7.2 IBNR Investigation... 68 7.3 Premium reserves... 69 2

8 Results of the calibration...70 8.1 Overall results of the final calibration of the model... 70 8.2 Applicability of the proposed default model... 77 8.3 Applying the industry calibration in practice... 80 9 Recommendations...83 9.1 Recommendations... 83 10 Acknowledgements...85 11 References...86 12 Appendices...88 12.1 Appendix A: ULR s vs Gross Earned Premium for Different Classes of Business... 89 12.2 Appendix B: Means and Standard Deviations of ULR s... 94 12.3 Appendix C: Reserving Ratios per Class of Business... 98 12.4 Appendix D: CV of Reserving Ratio vs Gross Reserves... 102 12.5 Appendix E: Gross Stand-Alone Capital Simulation Results... 107 12.6 Appendix F: Graphs of Gross Stand-Alone Capital... 115 12.7 Appendix G: Description of The Smith Model... 119 12.8 Appendix H: Allocation of FSB Asset categories to modelled asset classes... 124 12.9 Appendix I: Parameters for the calculation of the diversification/correlation factor... 127 12.10 Appendix J: Example of individual company feedback... 1 3

1 Glossary of Terms ASSA: Actuarial Society of South Africa AURR: Additional Unexpired Risk Reserve BE: Best Estimate DFA: Dynamic Financial Analysis EP: Earned Premium FCR: Financial Condition Report / Reporting FSB: Financial Services Board GIBNR: Gross IBNR GOCR: Gross OCR GWP: Gross Written Premium GUPR: Gross UPR IBNR: Incurred But Not Reported MCR: Minimum Capital Requirement NIBNR: Net IBNR NOCR: Net OCR NWP: Net Written Premium NUPR: Net UPR OCR: Outstanding Claims Reserves PGN: Professional Guidance Note (of ASSA) PM: Prescribed Method or Prescribed Margin RBC: Risk Based Capital TSM: The Smith Model ULR: Ultimate Loss Ratio UPR: Unearned Premium Reserve URR: Unexpired Risk Reserve VAR: Value at Risk WP: Written Premium 4

2 Overview 2.1 Background, scope and overall framework Actuarial & Insurance Solutions at Deloitte, and Insight ABC, were appointed by the Financial Services Board (FSB) in April 2005 to calibrate Financial Condition Reporting (FCR) requirements for the Short Term Insurance Industry in South Africa. Broadly, our aim was to construct a formula, on the basis of data from Star Returns, and Dynamic Financial Analysis, which would be an appropriate basis for a solvency requirement for the industry, for those companies that choose not to use an approved Internal Model. This had to be done in accordance with work performed by the FSB and by the Financial Condition Reporting Sub-Committee of the Short Term Insurance Committee of the Actuarial Society of South Africa. The formula has to take into account international developments, but at the same time be suitable for application in South Africa given available data in the STAR returns. Throughout our project, there were two major constraints, namely that STAR returns did not contain the data required or that data was not reliable. Secondly, the application of a central formula to the Short Term Industry as a whole will inevitably lead to situations where the formula does not fit individual companies with specific circumstances. Given the above, we sent individual results to all registered companies in South Africa, requesting feedback. This process was successful in the sense that it made companies aware of the project, highlighted areas where the modelling or data had to be modified, and indicated specific companies or sectors of the market where the application of an industry framework may not give optimal results for the individual companies involved. In line with the work of the Financial Condition Reporting Sub-Committee, we therefore support a regulatory framework that would allow companies to apply to the FSB for approval to determine a different level of capital, in accordance with a set of guidelines to be specified. For those companies that do not construct a complete risk-based internal model, the option of a certified model allows them to adapt elements of the regulatory framework to take into account their specific circumstances without having to set up a complex internal model. Further, we understand that the FSB would be open to approaches by companies with particular circumstances that may not be taken into account in the industry framework, or where a certified model or internal model specifies capital requirements that the company would struggle to meet. These companies would in practice be allowed to motivate why a special dispensation should apply to them, and the FSB would be free to consider the circumstances of the company and decide whether to grant permission for the company not to meet the capital requirements determined under any or all of the three models. It is envisaged that the FSB would, however, for benchmarking purposes check the industry calibration for every company registered in South Africa, whether the company in the end applies the industry framework, a certified model or an internal model. 5

The framework of industry calibration vs certified model vs internal model, can be represented graphically in the following way: Increasing Complexity Industry Calibration Certified Model Internal Model Increasing Appropriateness for Individual Company Increasing Cost Increasing usefulness in risk management Given the above, the characteristics of each of these different models can be summarised as follows: Industry Calibration Certified Model Internal Model Necessarily approximate Must be prudent for all companies May not be appropriate for circumstances of individual companies More precise for liabilities & individual circumstances of companies Involves judgment Hence requires professional certification Maximum precision for liabilities and assets Also requires professional certification Leads to greatest understanding of risks Provided models are transparent & realistic 6

The above overall framework is preferable to the existing capital requirement (effectively 25% of Net Written Premium), which: does not take into account the real risks faced by companies (i.e. it does not take into account the size of the insurer 1, the class of business written, the combination of classes of business written (i.e. correlation and diversification), expenses, and so on) requires a level of capital which is prudent for some companies but not prudent for others The only advantage of the current model is its simplicity. We hope that the industry calibration would be easy to apply in practice (i.e. driven by spreadsheets, possibly contained in the STAR returns) even though the mathematics operating in the background may be complex. The industry calibration should balance the following opposing factors: a desire for greater complexity to allow as accurately as possible for the individual circumstances of companies a desire to keep the model as simple as possible to apply and use We believe that a spreadsheet-based model contained in statutory returns would achieve this. Such a model would allow companies also to test new levels of capital required should they consider expansion, merger or other management actions. We now describe the regulatory framework in more detail. The following graphical representation of the new solvency requirements, taken from a presentation by Ms Hantie van Heerden of the FSB in August 2005, applies regardless of whether a company uses the industry calibration, a certified model or an internal model: Free assets Fair value of assets Fair value of admissable assets Excess assets Minimum capital requirement M inimum o f R10m. Prescribed basis. Internal model method. Liabilities Consists of best estimate plus additional prescribed margins. Prescribed method or internal model method. Risk management Financial condition report 1 Of course, the required Rand amount of capital increases as the net written premium increases, but as shown below, a capital requirement which is a constant percentage of net written premium does not accurately reflect the considerably higher risks faced by smaller companies, and lower risks faced by larger companies. 7

This model indicates that the new framework would establish the following principles for a Financial Condition Report: 1. Assets should be valued at fair value 2. Some assets will continue to be inadmissible for solvency calculation purposes (e.g. art) 3. A certain proportion of assets will be regarded as covering, or allocated to, insurance liabilities, or reserves. 4. For this purpose, insurance liabilities will consist of: a. Claims reserves, which in turn consists of: i. Incurred-but-not-reported (IBNR claims); and ii. Outstanding reported claims; and b. Premium reserves, which consist of: i. The Unexpired Premium Reserve (UPR); and ii. Where appropriate and needed, the Unexpired Risk Reserve (URR); and 5. The claims and premium reserves should be determined to be best estimates of the appropriate reserves. 6. A prescribed margin added to this takes the insurance liabilities up to the 75 th percentile 2. 7. Once this has been done, the minimum capital requirement is determined in such a way that the total capital minus the prescribed margins will reflect a certain level of sufficiency: 98%, 99% or 99.5%. In other words, the total required capital (which would be sufficient at the 98 th, 99 th or 99.5 th percentile) minus the prescribed margins would represent the minimum capital requirement, or MCR. 8. For smaller companies, the MCR would be subject to a minimum of R10m. 9. Excess assets are the admissible assets in excess of insurance liabilities (at a 75% level of sufficiency), and the MCR must be covered by excess assets under this framework. In assessing capital adequacy, this framework was developed by considering a one year time horizon. In other words, we did not consider the probability of insolvency over 5 years, but looked at the probability that a company would be insolvent within a one year period if it held a certain level of capital. In the rest of this report, we describe the methodology followed and assumptions made to calibrate an industry capital requirement that would meet the above objective. Our calibration aimed to provide an MCR that would take into account insurance risks and asset risks, and the determination of insurance liabilities (or reserves). Throughout our analysis, we fitted the model such that the total capital requirement is set at three levels of sufficiency: 98%, 99% and 99.5%. The reason is that we wanted to measure the impact of different levels of sufficiency on the capital requirement, in order to allow the FSB to judge the 2 Throughout this report, terminology such as 50 th percentile, or sufficiency at the 75% percentile level or 99% sufficient will be used interchangeably. All of these phrases express the same concept, namely that we can construct a probability distribution of outcomes for a company (i.e. a distribution reflecting the number of times out of many possible scenarios where a company s total capital would not be sufficient to meets its liabilities), and then measure the capital requirement at a point which would be sufficient to protect a company against insolvency half the time (set at the 50 th percentile), or against the 3 rd worst loss out of 4 (75 th percentile), or against the worst loss out of 200 (99.5 th percentile) 8

conservatism or otherwise of its final requirement. In the recommendations of this report, we also include a recommendation on the level of sufficiency that we believe should be adopted. 2.2 Global comparisons This section highlights some similarities and differences between our approach and that of other countries, particularly the UK and Australia. It must be noted that due to the multi-faceted nature of each country s regulatory solvency framework it is very difficult to make comparisons. Globally there has been a strong trend for regulators to move towards a risk-based capital approach for short-term insurance (and other financial services). Our proposal and the FSB s requirements are in line with this regulatory trend. Among others, the following countries have regulatory regimes applying an RBC approach: Australia United Kingdom USA Canada Germany Holland Switzerland Some broad similarities are the following: A value at risk (VAR) approach in determining capital requirements with a one year time horizon and a 99.5% level of sufficiency/confidence. A total balance sheet approach where risk in both assets and liabilities are considered Sufficiency in technical provisions set at a 75% level The framework addresses underwriting, reserving, credit and market risks with operational risk being difficult to quantify The allowance for internal models to be used It must be noted that there are still differences in the above from country to country. Some countries (most notably Switzerland) have adopted a Tail VAR approach as a measure of risk. It is recognised that Tail VAR may be mathematically more appealing since it is a consistent measure of risk. Despite this most countries have adopted a VAR approach to capital requirements. This may be due to the relative complexity of communicating tail VAR to stakeholders as compared to VAR. Our scope, as specified by the FSB, was to consider a VAR approach. In this regard we have produced results at the 99% and 98% levels of sufficiency as well as the 99.5% level. We have split our analysis into the eight classes of business set out in the Star returns. In the UK the comparable classes were further sub-divided into classes for proportional and non-proportional reinsurance. Unfortunately, the SA data available did not allow us to perform such a detailed analysis by reinsurance type. In Australia classes of business are grouped into three types that are felt to be broadly homogeneous and they also make allowance for a differential factor to be applied to reinsurance (without differentiating between proportional and non-proportional reinsurance). 9

The allowance for extreme events is an example of an area handled differently by different regulatory models. For example, in Australia they have explicitly modelled extreme events and made allowance for them. In the Star return data we did not have the ability to separate extreme events from attritional losses and our calibration thus implicitly models these different types of losses together. We have taken the opportunity of allowing our insurance risk charges to vary by the size of the account. This is an outgrowth of the investigations we performed where we noticed a distinct trend for the results to reduce in volatility as company size increased. We have taken into account risk dependencies as per the International Actuarial Association s (IAA) global framework. We have built-in a company-specific allowance for diversification and correlation and our base charges by class of business allow for this. This may also make comparison between our stand-alone charges and those in other regimes potentially misleading. Our diversification and correlation factor approach is similar to a study on reserving in Australia where adjustments to the sum of stand-alone reserves are made with reference to a predetermined model. This model explicitly links certain factors (like number of lines of business and concentration of business within one line) to a recommended diversification and correlation factor. We have been more direct in allowing the company in question to calculate a factor more specifically attuned to their circumstances. Due to the complexity of the resulting formula this can only be achieved through providing a pre-programmed spreadsheet for companies to use for this calculation. Another principle from the IAA global framework is the allowance for risk management measures. We have allowed for these in the form of reinsurance and believe that coinsurance should be treated similarly. Unfortunately not all risk management measures are straightforward to quantify and calibrate on an industry-wide basis, in particular those of cell captives. Bearing this in mind, the application of the framework needs to be flexible enough to allow for such situations. The allowance for covariance effects between the insurance capital charge and the investment capital charge as well as the grossing-up of these charges is similar in nature to that allowed for in the South African Life Insurance CAR calculations. 2.3 Reliances and limitations This report is produced on instruction of and for the purposes of the Financial Services Board. It is based on data and information provided to us, which, although we checked and cleansed such information as far as possible, we cannot guarantee the accuracy of. The approach taken in this report is for the purposes of an industry calibration, and should be interpreted in the context of an industrywide regulatory framework. We make no guarantees on the effectiveness of this framework applied to a particular company or a group of companies and we accept no responsibility for the solvency of one or more companies in the industry where this framework is applied, as solvency can be impacted by many other factors that cannot be taken into account in an industry capital requirement, such as inappropriate management action, etc. We do not accept responsibility for the application of the concepts explained here to a specific company and recommend that professional advice be obtained if a company wishes to determine an appropriate level of risk-based capital. The scope of this report is limited to an industry calibration for Financial Condition Reporting purposes. Whilst we make comments on the application of certified models and internal models, the details of such models and how they will be applied in practice are still to be determined and beyond the scope of this report. Further, at several stages of the project we were faced with inadequate data that prevented us from refining our calibration of the model. We therefore make some comments on areas where it may be appropriate to collect more data in STAR returns. It is beyond the scope of this report to make complete recommendations on the way in which STAR returns should be modified, and these comments and recommendations should be seen as preliminary. 10

This report represents our recommendations to the FSB and does not necessarily represent the final format and way in which the capital requirements will be implemented in practice. It is beyond the scope of this report to consider other factors that may impact on the format in which the FSB finally decides to implement an industry calibration, such as Black Economic Empowerment within the industry. As discussed in s2.1 above, the FSB may well allow a company to meet capital requirements under a special dispensation. 2.4 Structure of the report We first describe how the insurance risk capital charge was determined. This includes, among other matters, discussion of: data used and data cleansing the investigation of ultimate loss ratios the analysis of underwriting risk and reserving risk the dynamic financial analysis applied allowance for diversification and correlation We then describe the determination of the asset risk capital charge, which includes: the allocation of asset categories to liabilities the determination of capital factors using The Smith Model, including a description of The Smith Model The next section of the report describes how the insurance risk capital charge and investment risk capital charge are combined to determine the total capital required. We then describe the calibration of industry reserving levels, which includes claims reserves and premium reserves. We also discuss the determination of the prescribed margin, to take the estimate of liabilities up to the 75% level of sufficiency. Once this has been done, we are in a position to describe the Minimum Capital Requirement (MCR) for each company in the industry. We show how the suggested MCR affect the different sectors of the market, and how it changes if we set capital at different levels of sufficiency. Finally, we make recommendations. 11

3 Insurance Capital Charge 3.1 Data Received The main source of information for this exercise was the historic Star returns of all companies in the short-term insurance industry in South Africa. Where available, the FSB provided us with STAR returns dating back to 1990. The first phase of this project consisted of extracting the appropriate data from the Star returns and creating the appropriate data stores for the investigations needed (such as those for underwriting risk and reserving risk). In total 1075 star returns were included in the analysis. Initial calibrations of the model were performed using subsets of the data in order the speed up calculations. Initially analyses were performed on 33 companies, thus covering about 95% of the market by written premiums. The final calibration of the model was performed including all 118 companies present in our sample at one time or another. Approximately 35 companies had not submitted their 2004 returns when the initial batch of data was received from the FSB. The FSB then forwarded the returns as they became available. At this stage the 2004 returns were not available for 27 companies. In most cases this was because the respective company had ceased to exist, a restructuring such as a merger had taken place or the company had changed its name. Consequently after allowing for all these factors, returns for only two active companies were outstanding. Data for years before1999 was generally less reliable. Most of the files were in Excel format, with a number of returns for years prior to 1999 being in Lotus format. These were converted to Excel format. The table below indicates the number of returns received for each year: Year Number of STAR returns 1990 52 1991 50 1992 53 1993 55 1994 55 1995 61 1996 62 1997 68 1998 75 1999 84 2000 92 2001 90 2002 95 2003 93 2004 90 Total 1075 12

3.1.1 Duplicates After removing blank and corrupted files, allowing for changes in company names as well as restructurings duplicate file were found for 5 companies. In most cases these simply represented duplicates and one of the files was excluded. However in some cases one of the files represented a data update received from the company. These were identified and the updated versions were included in the analyses. 3.1.2 Formats The format of the STAR return changed in 1999. A mapping of the old format onto the new format was constructed and subsequently allowed for when extracting data from the STAR returns. The line of business classification also changed, with the Miscellaneous class being further subdivided into Miscellaneous, Liability and Engineering. The consistency of the return formats was checked and a small number of files were found in which the layout and/or location of certain sections was different from that of the standard STAR return format. These were corrected and included in the analyses. It should be noted that the STAR returns showed considerable deviations from the standard format from section 13 onwards. However these sections did not contain information crucial to our analysis and as a result this posed no problem. 3.1.3 Consistency of dates Considerable time was spent cleaning the date fields in the returns. This arose because dates were either left blank or entered in incorrect formats. For example: 1999, 25th of January as opposed to 25-Jan-1999. This problem related mainly to the old STAR return format as the new format has separate cells for the day, month and year values. Financial year end changes also required individual consideration to avoid inappropriate financial years being recorded. 3.1.4 Consistency of company names As mentioned earlier, some companies underwent name changes and restructurings. This had to be considered when extracting our data to ensure that the relevant STAR return data be included under the appropriate companies. A great deal of time was also spent on cleaning the company names as recorded in the STAR returns. This arose because a company s name would be spelled differently for each year s submitted returns. The following are examples of the spelling variations present in the data: ABC XYZ A.B.C. XYZ INS CO. LTD A.B.C. XYZ INSURANCE CO LTD ABC Xyz Insurance Company Limited A B C XYZ Ins Company Limited In addition, a few STAR returns were found to have blank company name fields. These names were inserted after identifying which companies the data related to. 13

3.1.5 Invalid/erroneous entries A phenomenon which resulted in a significant amount of cleaning and subjective consideration was the presence of blank cells. It would seem that many companies simply leave cells blank where in fact they should be recording zeros. This presented a problem because it was also found that companies often neglect to complete certain sections of the STAR return. Consequently, it becomes difficult to distinguish between incomplete data and zeros. In these cases it became necessary to investigate the matter individually. Invalid characters were occasionally found in the data. For example, some companies entered the text character - instead of a zero. The character was also found which indicated the use of nonstandard characters in the cells. 3.1.6 Points relating to the various analyses performed: 3.1.6.1 Triangle extraction This exercise highlighted to a great extent where data was unreliable or incomplete. Blank diagonals were often noticed, indicating that companies did not complete a particular development triangle in a particular year. Interpolation and/or extrapolation were applied in certain cases to estimate individual blank cells to avoid losing a significant amount of otherwise valuable data. This exercise of trying to salvage data was of particular importance since the data was generally sparse to begin with. 3.1.6.2 ULR analysis While calculating and analysing claims run-off patterns, we found the calculated development factors were very useful in identifying tainted data. Claims were considered in relation to premiums, thus providing a better view of the reasonability of the magnitudes of the recorded claim amounts. Extreme ULR s (e.g. 3000%, -80%) indicated potentially incorrectly recorded earned premiums. Where these were identified the points were excluded from our analyses. Where it was fairly obvious that a data capturing error had been made (e.g. a zero left off the end of a figure), the data was corrected and included in further analyses. Where claims triangles had one or two blanks but otherwise contained reasonable data, linear interpolation was used to estimate the blanks from the surrounding data. This was done so that we were still able to calculate development factors for a company from its data and include the resulting ULR estimates in the subsequent statistical analyses. Where no earned premium data was available, we could not calculate ULR s for the relevant companies. However, where earned premium values were not recorded for a year but a reasonable amount of data existed around the missing value we estimated it by linear interpolation. In doing so we avoided losing potentially valuable data. 14

The following table indicates the number of data points used per class for the ULR analysis: Class Number of data points included in analysis Accident 358 Engineering 98 Guarantee 217 Liability 117 Miscellaneous 167 Motor 435 Property 522 Transport 349 3.1.6.3 Reserving Risk analyses As with the ULR analyses, further insight into the overall reasonability of the data was obtained by considering the claims development triangles in relation to the OCR and IBNR development triangles. The OCR and IBNR development areas in the STAR returns were often filled out poorly. Cells are often left blank as opposed to explicitly being recorded as being zero. Often it is not clear whether the data is not available or actually zero. Judgement had to be applied on a company-by-company basis to determine which of the above applies. The following table indicates the number of data points used per class for the Reserving Risk analysis: Class Number of data points included in analysis Accident 332 Engineering 74 Guarantee 249 Liability 93 Miscellaneous 158 Motor 490 Property 441 Transport 257 3.1.6.4 IBNR Analysis We extracted the IBNR triangles from the STAR returns for this analysis. In some cases insufficient data existed for the purposes of calculating chain ladder development factors, yet the data was sufficient to calculate at least two years IBNR figures. In these cases industry average development factors (weighted by gross earned premium) were calculated from the reliable data and used to estimate the relevant companies IBNR figures. 15

The following table indicates the number of data points used per class for the IBNR analysis: Class Number of data points included in analysis Accident 192 Engineering 67 Guarantee 87 Liability 66 Miscellaneous 77 Motor 242 Property 238 Transport 153 3.1.7 Other points: Values in separate areas of STAR returns did not always reconcile. More automatic cross-checks in place could reduce the amount of invalid data input conditional formatting could be used. STAR returns are supposed to be locked with a password. In many cases it became evident that this was somehow sidestepped or overridden especially where structural changes had been made to the layout of the spreadsheet. 3.1.8 Extraction tool To check the consistency and validity of the data, we created a spreadsheet tool that runs through the STAR returns and extracts a specified field. This tool helped us to identify a number of data issues such as: Duplicate, blank and corrupted files Inconsistent financial year ends Inconsistencies or changes in company names and structures (e.g. mergers) The matters identified were presented to the FSB and most were resolved. An issue which received particular attention was the consistent naming of companies. It was necessary to construct a mapping of the companies names to obtain a list of companies which allowed for spelling errors and the fact that some companies legally changed names or ownership. This was done with the help of the FSB. A similar mapping had to be done on the financial year ends to allow for companies who either entered invalid dates or underwent a change of financial year end. The spreadsheet described above also allows the user to compile a custom database of information contained in the STAR returns and has been made available to the FSB for their own future use. Where possible we included the most recent Star returns (2004). Some companies submitted their 2004 Star returns during our investigation. 16

3.1.9 Grouping tool An additional tool that we created during our data extraction and analysis was one that enables the user to create various custom combinations of STAR returns. For example if two companies merged in 2003 and the user wishes to analyse their consolidated returns for years before 2003, he/she could join the STAR returns for the required periods required using this spreadsheet tool. The spreadsheet allows the user to define custom groupings. For example all re-insurers could be grouped and analysed together. The results produced by this spreadsheet are in the latest STAR returns format. The user also has the flexibility of being able to specify exactly which areas of the STAR returns he/she would like to have amalgamated. This spreadsheet has also been made available to the FSB for internal use. Custom combinations of STAR returns 3.2 Claims paid data store The two primary data investigations centred around the analysis of claims paid triangles (for the purposes of determining suitable ultimate loss ratios) and reserve run-off triangles (for the purpose of determining reserving uncertainty). This section describes the extraction of this data and the creation of this data store for each class of business. We applied the spreadsheet extraction tool described above which loops through the available STAR returns for every company and extracts the relevant data. Payment and reserve triangles are then constructed from this data. This spreadsheet allows the user to extract the following data for each development year: Claims payment run-offs Outstanding claims reserves IBNR reserves Relevant earned premium 17

Each line in the data corresponds to a particular historic accident year for a particular company. It is then possible to see how claims and reserves have developed for that particular accident year. The program also automatically creates a cumulative paid claims triangle from the incremental claims paid triangle. This enables the analysis of the claims run-off via traditional methods such as the Chain- Ladder method and Bornheutter-Ferguson method. Development triangle extraction tool 3.3 Ultimate Loss Ratio Investigation This section describes our use of the claims paid data store for the purposes of analysing ultimate loss ratios. Incurred loss ratios are included in the STAR returns. They include estimates of reserves for outstanding claims at the end of each period that are subject to estimation uncertainty. Our investigations attempted to separate this reserving uncertainty from the underwriting uncertainty by looking at how claims paid had historically run-off in each historic accident year. For each of the accident years that was fully run-off we had an estimate of the ultimate loss ratio and for those years that were not fully run-off we made an estimate of the ultimate loss ratio by applying grossing-up factors based on Chain-Ladder and/or Bornheutter-Ferguson ( BHF ) methods, where appropriate. In a sense, some of the ultimate loss ratios represent partially manufactured data. The primary reason for this is the benefits of having observations undistorted by reserving estimates. The following steps were applied for each company, for each accident year: The traditional chain ladder method was used to calculate the development factors for each of the development periods. Unfortunately, the information available in the STAR returns only extends for five development years. To estimate the tail factor (the cumulative development beyond development year five) we fitted a company specific curve to the development factors and used this as a basis for extrapolation. 18

For all companies with sufficient data in their triangles we consolidate their results to estimate an industry-wide development pattern. Where the data/triangle for a particular company is sparse we use the industry estimate for each class as discussed above. This then yields a first estimate of the ultimate loss for each accident year based on fitted chainladder factors. A second independent estimate of the ultimate loss was calculated using the BHF method based on observed historical loss ratios and earned premiums specific to the company in question. If the claims are near to being fully run-off we use the first estimate whereas if the claims from the accident year are less than 50% developed we use the BHF estimate. The resulting ULR s give our best estimate of the ultimate claims emerging for each company from each accident year. These manufactured observations form the basis of later investigations into the distribution characteristics of the loss ratio and reserving risk. The following two graphs show cumulative claims development data from the motor class for two example companies: Cumulative claims development 120.0% 100.0% 80.0% 60.0% 40.0% 20.0% 0.0% -20.0% Company A 1 2 3 4 5 6 Development Year 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 19

120.0% Company B Cumulative claims development 100.0% 80.0% 60.0% 40.0% 20.0% 0.0% -20.0% 1 2 3 4 5 6 Development Year 1999 2000 2001 2002 2003 2004 The above two graphs confirm the short-tailed nature of the motor class. For both companies shown claims are close to 100% developed by the second development year. Due to the short tailed nature of the motor line of business, the BHF method is not strictly necessary. However, we believe it to be more appropriate for the other, longer tailed classes. As such we have built in a measure of flexibility regarding the level at which the BHF method would apply. 3.4 Framework We are proposing a building-block approach for the calculation of the required insurance capital to promote transparency and flexibility in the use of the framework. In setting the insurance capital the following aspects are allowed for: The type of business underwritten and the relative risks and rewards of the type of business The amount of business written and concomitant diversification effects The relative amount of underwriting risk versus reserving risk The mitigation of risk through use of reinsurance Expenses Correlation between classes of business Diversification effects of writing different classes of business The diagram below shows how the insurance capital of a short-term insurer is built-up from various components. This diagram is best understood from the bottom-up, since each component of risk feeds upwards into the component above it. 20

Insurance Capital Required Investment income and capital gains Diversification / Correlation Net stand-alone capital ACC ENG GTEE LIAB MISC MOT PROP TRANS Retention Retention Expenses Expenses Gross stand-alone risk capital ACC ENG GTEE LIAB MISC MOT PROP TRANS GWP GWP GUPR GUPR Insurance capital for a short-term insurer The workings of the diagram above are given below, with more detailed explanations of each component. 3.4.1 Gross stand-alone risk capital This is an estimate of the capital required for pure insurance risks (underwriting and reserving) before allowance for reinsurance and expenses. Stand-alone capital is what an insurer needs to hold without reference to any of the other classes of business being underwritten. Our framework differentiates between: Different classes of insurance business Different account sizes within each class of insurance business 21

We calibrated each class of business within the STAR returns separately on the basis that each class has different underlying risk and return characteristics. These classes are: Accident (ACC) Engineering (ENG) Guarantee (GTEE) Liability (LIAB) Miscellaneous (MISC) Motor (MOT) Property (PROP) Transport (TRA) Each class was investigated individually, though the same types of investigations were performed for each class. We are aware that a more detailed split within each class may have yielded a more accurate identification of risk. However, sufficient quantities of data was not available to achieve statistical significance at this level. Some factors that may affect risk within a class are: Policy term Different types of policy Differing target markets To further allow for the specific risks inherent within each class of business we also analysed each class/account by size. An account with a higher gross written premium and/or gross unearned premium is likely to exhibit lower volatility due to having a greater number of risks on the books. This is an example of diversification within a class of business. We allow for diversification between classes of business at a later stage. The data suggests that larger accounts tend to exhibit lower volatility and in some cases the larger accounts also exhibited greater pricing power. This can be seen in Appendix A where we show the scatter-plot for each class of business of the ultimate loss ratio versus the gross earned premium. In general, as gross earned premium increases we see greater stability in the resulting loss ratios. The following two items are used as a measure of the size of the account: Gross written premium (GWP) Gross unearned premium (GUPR) In essence we are building up the capital required in the coming financial year so we need to consider the specifics of each class of business in the coming financial year. As such, the gross written premium used needs to be an estimate of gross written premium in the coming year. Note that such an estimate would have to be made at the time of submission of STAR returns, and hence always with at least a three month lag. For this reason, we believe that the estimate of gross written premium over the following year would not present problems to most companies, and it would be easy for the FSB to check whether a company consistently under- or over-estimates its gross written premium for the purposes of setting capital. The estimate of gross written premium can be made with reference to the past financial year if it felt that the business level will remain broadly constant. Where changes to the level of business are projected and/or budgeted for these should be taken into account. Since the capital requirements will be calculated at the same time as the sending of the statutory returns, insurers will already be partially into the new financial year and will thus have a further indication of business levels. 22

3.4.2 Net stand-alone capital As with gross written premium above, companies will need to consider the specifics of each class of business in the coming financial year. As such, the following estimates need to be made for each class of business: Retention Expenses The retention percentage for the coming financial year can be estimated with reference to the past financial year if the reinsurance strategy for that class has not changed. Where changes in reinsurance strategy are known, they should be taken into account in determining the retention percentage to use. The gross stand-alone risk capital is multiplied by the anticipated retention percentage in each class of business to give a net stand-alone risk capital. It is important to note that the allowance for reinsurance in the calculation is not ideal, in particular for non-proportional reinsurance. However, we did not have detailed data available in STAR returns for a more in-depth calibration taking into account the type of reinsurance taken out by a company. We recommend that more detailed statistics be collected about proportional and non-proportional reinsurance in STAR returns. We discuss the allowance for reinsurance in more detail below, in the section on Underwriting Risk (s3.7). In the event of a worst-case insurance loss (as envisaged by the net stand-alone risk capital calculated at this point) the company will also incur expenses in the normal course of business. These expenses thus need to be allowed for in the capital requirements of each class of business by adding them to the net stand-alone risk capital. Companies can use their current level of expenses as an indication of the likely level of expenses in the coming year. Further, should companies feel that their expenses would rise significantly in times of high insurance losses they should estimate their expenses on this basis. 3.4.3 Allowance for diversification and correlation The total net stand-alone capital above does not take into account the following effects from writing multiple lines of business: Diversification effects due to writing more than one class of business (these will in general reduce the capital required) Correlation effects between the different classes of business (these will in general increase the capital required since classes tend to be positively correlated) If it is expected that the worst-case insurance event will occur at the same time for each class of business then we would hold the sum of the net stand-alone capitals calculated above as insurance capital against such an event. However, the risks in the eight separate classes of business are not perfectly correlated and we allow for this by multiplying the sum of the net stand-alone capitals by a statistically determined factor that allows for the relative mix of business in each class. The derivation of this factor is discussed later. 3.4.4 Allowance for investment returns The final allowance is for investment returns on assets backing the relevant liabilities. The argument is that even in a worst case scenario, a company will still earn some return on its existing assets, and this can be used to reduce the capital requirement. The default approach will be for companies to take into account investment return on all assets backing the premiums liabilities (unearned premium reserve, unexpired risk reserve) and claims liabilities (outstanding reported claims reserve and incurred but not reported claims reserve). We discuss our suggested allowance for this in more detail below. 23

3.5 Underlying risks and Risk Factors This section discusses the main components in the calibration of gross stand-alone risk capital and sets the scene for the various sections that follow. 3.5.1 Underlying risks Within each class of business we are proposing a framework that links the underlying risks to certain observable risk factors. The main risks underlying the insurance capital charge can be labelled as: Underwriting risk Reserving risk Underwriting risk is the risk that premium earned in future periods will be insufficient to cover claims incurred in those periods. Inherently, underwriting risk is forward looking. Reserving risk is the risk that claims incurred in historic periods will be greater than anticipated (and reserved for) in those periods. Inherently, reserving risk is backward looking. Calibrating the uncertainty in both of these risks formed a major part of this exercise and the resulting data analysis is discussed more fully in later sections. We use the results of this analysis to parameterise our simulation model. These parameters then flow into the DFA engine where underwriting results are calculated. Underwriting Risk Reserving Risk DFA Engine Simulation Results Gross Written Premium Gross Unearned Premium Simulation Model 24

We performed 100,000 simulations for each class of business and company size to ascertain a reliable estimate of capital required at each level of sufficiency. These estimates were then linked to company size via the following risk factors, as shown in the graphic illustration above: Gross written premium Gross unearned premium 3.5.2 Risk Factors This section details how we derived the stand-alone capital charge for each line of business. In line with discussions above we have applied different assumptions for accounts of differing size. In general this implies that larger accounts have lower capital requirements relative to written premium (essentially representing a diversification credit). Conversely smaller accounts have higher relative capital requirements. The factors we have chosen to measure the size of the account are: Gross written premium (projected over the next financial year) Gross unearned premium The reason for our choice of these two figures is that they are broadly verifiable and are seen to be less subject to manipulation than other alternative figures. The main alternative candidate for use was outstanding claims reserves. A historic shortfall of regulatory frameworks has been the inability to distinguish between regular companies, young companies growing rapidly and older companies running-off their existing book of business. The two factors above allow for this differentiation to some extent. The reason for this is that the written premium is a forward looking measure and the unearned premium is a retrospective measure. Young, rapidly growing companies will have a lower component of unearned premiums (relative to regular, established companies). Companies in run-off may project lower future premiums but may well have a high component of unearned premiums. A potential weakness using these factors is that a company may be in run-off, have low future premiums and low unearned premiums but still may exhibit significant risk from its outstanding claims reserves. This is provided for, to an extent, in the prescribed margin included in insurance liabilities. 3.6 Notional companies It is not possible (nor necessary) to run models for every single possible size of insurance account (measured by gross written premium and unearned premium). Instead, we have calculated the exact gross stand-alone capital requirements for a broad and representative sample of hypothetical accounts (to be discussed below) with the aim of applying these results in a smooth manner across the range of account sizes. For each class of business we have run DFA models for 99 different account sizes. These allow for: 11 different levels of gross written premium 9 different levels of gross unearned premium Future growth in account sizes from present levels Our approach can therefore be illustrated approximately in the following way: out of a universe of 36 potential or observed combinations of GUPR and GWP, we choose 4 and model these explicitly. Any actual combination of GUPR and GWP that falls within these four points can then be estimated by 25

interpolation, thereby allowing the model to be applicable to any actual combination of GUPR and GWP, without having to model for an infinite number of combinations. Selection of representative notional companies We suggest that in practice companies will interpolate their capital values from the closest available companies using a set method of interpolation. This will be also automated via STAR returns. In deciding on the levels for gross written premium and gross unearned premium to model, our starting point was the current actual range of business (by size) within each class of business for all insurers writing business in that class, as shown in the latest (2004) STAR returns. This was chiefly measured by the largest value within each class of business. We then extended this range to allow for future growth. In extending the range our rule of thumb was to allow for at least twice the current largest value in the industry in 2004. We applied a logarithmic scale (as opposed to a linear scale) to the spread of company sizes. We felt that this scale allowed us to capture the diversity in companies with greater accuracy. For example, the difference in size between a company writing R20m in premiums and a company writing R10m in premiums can be compared to the relationship of a company writing R2bn of premium versus another writing R1bn of premium. Our approach can be likened to that taken in experimental design where calculations are performed at various levels with the aim of linking the result with its appropriate level. These levels of gross written premium and gross unearned premium are used (together with parameters described in the sections that follow) to set the levels of the following variables for each notional company model: Historic gross written premium 26

Outstanding claims reserves Earned premiums Claims paid Underwriting uncertainty Reserving uncertainty These derivations are discussed in more detail in the DFA section. 3.7 Underwriting risk 3.7.1 Ultimate Loss Ratios One of the primary risks faced by insurers is underwriting risk: the risk that premium earned in future periods will be insufficient to cover claims incurred in those periods. We have applied simulated loss ratios to earned premium in our industry model to take this risk into account. The simulated loss ratios are derived from log-normal distributions specifically parameterised for each class of business and for the account size of the notional company concerned. The starting point of this analysis is the results of the ultimate loss ratio investigation described in the Data section. These results paired each ultimate loss ratio with its corresponding gross earned premium. The results of the ultimate loss ratio investigation can be seen in Appendix A. We have fitted means and standard deviations to this data and the results can be found in Appendix B. The primary patterns emerging from the data are as follows: In general, as gross earned premium increases, the ultimate loss ratio tends to stabilise (and thus exhibit lower volatility). In some classes the average ultimate loss ratio decreases slightly as gross earned premiums increase. This may be indicative in some instances of greater pricing power associated with bigger players in the market and in other instances with an ability to price more accurately for the risk. The results fit our expectations, as an account with a higher gross written premium and/or gross unearned premium is likely to exhibit lower volatility due to having a greater number of risks on the books. This is an example of diversification within a class of business. We then fitted a statistical distribution to the ULR s the lognormal distribution. The parameters used for the log-normal distribution are derived via a method of moments transformation applied to the mean and standard deviation applicable to the particular class and notional company size. It is important to note that the ultimate loss ratios used in the above investigations are all net ultimate loss ratios even though they are paired with gross earned premium. Using net ULR allows for the use of reinsurance and in particular for the use of non-proportional reinsurance. This is because all forms of reinsurance are implicitly shown in the volatility of the net ULR data. We would expect that a gross ULR investigation would exhibit far greater volatility than the investigation performed since the industry s use of non-proportional reinsurance tempers adverse experience. In other words, while we did not have sufficient data to explicitly calibrate the model for different reinsurance programmes, we still make an approximate allowance for the use of proportional and nonproportional reinsurance to the extent that its impact on ULR s is contained in the data. As mentioned, this is a less than ideal solution, but the only one available to us given the lack of reinsurance data in STAR returns. Over time, the industry calibration can be fine-tuned to ensure that especially nonproportional reinsurance is allowed for more accurately. For companies with significant nonproportional reinsurance programmes, it may be worthwhile to consider the use of a certified model. 27

One drawback of the above approach is that the results have been calibrated for the average use of reinsurance prevalent in the industry. Should this change materially from historic usage patterns then the results may be biased accordingly. The following graph shows the mean net ULR level for each class of business versus gross earned premium: ULR mean - All classes 1.50 1.40 1.30 1.20 1.10 1.00 Motor Property Transport Accident Guarantee Liability Engineering Misc Mean 0.90 0.80 0.70 0.60 0.50 0.40 0.30-500,000 1,000,000 1,500,000 2,000,000 2,500,000 3,000,000 3,500,000 4,000,000 4,500,000 Gross earned premium (R'000) 28

The following graph shows the standard deviation of net ULR for each class of business versus gross earned premium: ULR standard deviation - All classes Standard deviation 1.50 1.40 1.30 1.20 1.10 1.00 0.90 0.80 0.70 0.60 Motor Property Transport Accident Guarantee Misc Liability Engineering 0.50 0.40 0.30 0.20 0.10 0.00-500,000 1,000,000 1,500,000 2,000,000 2,500,000 3,000,000 3,500,000 4,000,000 4,500,000 Gross earned premium (R'000) 29

In the following graph we show a derived measure of return per unit of risk for each class of business. The risk-return measure is calculated as one minus the loss ratio (i.e. a form of return ) divided by the standard deviation. This is similar in spirit to risk-return measures used in financial economics. As expected, return increases per unit of risk as Gross Earned Premium increases. ULR risk-return - All classes 3.00 2.80 2.60 2.40 2.20 2.00 1.80 Risk-return 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00-0.20-0.40-500,000 1,000,000 1,500,000 2,000,000 2,500,000 3,000,000 3,500,000 4,000,000 4,500,000 Gross earned premium (R'000) Motor Property Transport Accident Guarantee Misc Liability Engineering 30

3.7.2 Underwriting Cycle A cursory analysis of the data confirms the existence of an underwriting cycle. This is seen when looking at the average loss ratios across accident years. Average annual ULR 120% 117% 110% 104% 100% 90% 91% 92% 91% 80% 80% 77% 78% 70% 67% 65% 69% 70% 72% 68% 60% 59% 50% 40% 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 It is possible to strip out the effect of the underwriting cycle using various techniques. One simple approach to use is average loss ratios and factors/multiples of the average for adjustment of the data. A more detailed analysis using ARIMA time series models could also be undertaken. However, it must be noted that by removing the effects of the underwriting cycle we are reducing the variability in the dataset, essentially attributing this to underwriting cycle effects. This removal may not be desirable for FCR purposes. It is very difficult to know where the industry is in the underwriting cycle at any point. Further, we are looking at specifying a single capital formula appropriate throughout the underwriting cycle. The calibration of this formula should thus take the full spectrum of variability in the underwriting cycle into account as we would want capital to be adequate throughout the underwriting cycle. In light of this we have not removed the effects of the underwriting cycle from the data. Further, the calibration to industry data, without the removal of the underwriting cycle, does allow for historical variability caused by the underwriting cycle, and hence there is an implicit allowance for the variability caused by the underwriting cycle included in our calibration. 31

3.8 Reserving risk Another primary insurance risk is reserving risk: the risk that claims incurred in historic periods will be greater than anticipated and reserved for. The purpose of this investigation is to estimate the risk of under-reserving inherent in outstanding claims reserves (OCR). The investigation is based on the claims triangle data extracted in the data section above. In our definition, OCR includes both IBNR and notified but outstanding claims. As in the ULR investigation above we perform our investigations for each class of business since each is likely to have its own inherent volatility. We calculate the following ratio for the purposes of estimating reserving risk: Re serve Re serveratio = Re serve + k + 1 Paid k +1 In other words, we compare the reserve set up at the end of a period with the sum of the reserve set up in the following period and the claims paid in the following period. Over a one year period this ratio exhibits perfect reserving (in hindsight) if it is equal to one. If the ratio is greater than one it theoretically exhibits under-reserving and similarly if it is lower than one it theoretically exhibits overreserving. It is this reserving uncertainty, as measured over a one year time period, that is allowed for in our industry model. The reserving ratio as given above has a highly heterogeneous distribution. This can be seen in the distribution of reserving ratios for each class of business in Appendix C. The following graph is an example of this distribution for the motor class: k Reserving ratio distribution - Motor 14% 12% 10% 8% 6% 4% 2% 0% 0 0.3 0.6 0.9 1.2 1.5 1.8 2.1 2.4 2.7 3 3.3 3.6 3.9 4.2 4.5 4.8 5.1 5.4 5.7 6 6.3 6.6 6.9 7.2 7.5 7.8 8.1 8.4 8.7 9 9.3 9.6 9.9 Reserving risk As in the ULR analysis we investigated the possibility that the reserving uncertainty may be linked to the size of the account. In this instance we measured the size of the account by the gross reserve size attaching to each reserving ratio. The resulting scatter-plots can be seen in Appendix D. The 32

uncertainty inherent in reserving exhibits a similar pattern to the ULR investigation in that larger account sizes appear to be associated with less uncertainty. We have fitted a functional form to the coefficient of variation (CV) of the Reserving Ratio as it varies by account size, the results of which can also be seen in Appendix D. The CV is the ratio that the standard deviation of the Reserving Ratio bears to the mean. The following graph is an example of the coefficient of variation and scatter-plot of reserving ratios for the motor class. Note that the coefficient of variation is not a curve of best fit in the graph below, but a separate ratio determined from the underlying data. Reserving ratio distribution - Motor 25.0 Reserving risk 20.0 Coefficient of variation 15.0 10.0 5.0 0.0-100 200 300 400 500 600 Gross claims reserve (R'000) As with ultimate loss ratios the ratio above is measured on net data. It must also be noted that only net data was available at this level of detail. 3.9 Earnings patterns For the purpose of running the DFA model for each of the notional companies it is necessary to have an estimate of the premium earning pattern for each class of business. The premium earning pattern is the percentage of premium written that is earned in the same financial year in which it is written. This earning pattern is used primarily in two places: It is applied to projected gross written premium to arrive at an estimate of gross earned premium in the year after the valuation date Unearned premium in the most recent year is grossed up by it to estimate written premium in the previous financial year. As explained above, both gross earned premium in the year after the valuation date and written premium in the year prior to the valuation date are used for the calibration. 33

While the earning pattern within a class of business may vary from company to company depending on the nature of the business written, the aim of this investigation is to estimate an industry average for each class. The STAR returns contained information regarding the term of business written for each company where premium is allocated to the following broad categories: Monthly Annual Term (greater than annual) We assigned the following earning patterns to each of these broad categories: Policy Term Earning patterns Monthly 95% Annual 50% Term 16% The average term used was obtained by multiplying these earning patterns by the average amounts of business in each category of policy term (in each class of business), as obtained from the data. These can be seen in the table below: Class of business Monthly Annual Term Accident 38.06% 40.28% 21.66% Engineering 51.67% 44.68% 3.65% Guarantee 60.76% 30.58% 8.65% Liability 46.08% 41.31% 12.60% Miscellaneous 32.33% 42.97% 24.70% Motor 61.55% 32.91% 5.54% Property 47.00% 43.88% 9.11% Transport 40.73% 49.26% 10.01% 34

Distribution of business written 100% 90% 80% 70% 60% 50% 40% 30% 20% Term 10% Annual 0% Monthly Accident Engineering Guarantee Liability Miscellaneous Motor Property Transport The resulting earning patterns are given in the table below: Class of business Earning Pattern Accident 60.23% Engineering 72.46% Guarantee 74.96% Liability 66.92% Miscellaneous 56.58% Motor 76.36% Property 68.51% Transport 65.33% 3.10 Claims payment patterns A claims payment pattern for each class of business was derived indirectly through our investigation into ultimate loss ratios (discussed more fully earlier in the Data section). Where company data was not sparse, we used the chain ladder factors derived to augment our industry analysis for that class of business. As discussed earlier, where the data was sparse we used these industry averages to give estimated ultimate loss ratios from claims paid to date, effectively using the Bornhuetter-Ferguson method. 35

The following tables and graphs show the calibration of cumulative as well as incremental claims by development year for each class of business. Cumulative claims development pattern Years since claim event Class of business 0 1 2 3 4 5 6 Accident 55.02% 86.42% 96.33% 99.01% 99.74% 99.91% 100.00% Engineering 63.02% 95.69% 99.56% 99.94% 99.99% 100.00% 100.00% Guarantee 28.29% 75.07% 91.64% 95.36% 96.34% 96.69% 100.00% Liability 38.32% 67.66% 85.45% 92.94% 96.17% 98.04% 100.00% Miscellaneous 48.52% 85.79% 96.75% 99.22% 99.81% 99.95% 100.00% Motor 78.66% 98.00% 99.73% 99.94% 99.98% 99.99% 100.00% Property 52.36% 91.84% 98.41% 99.49% 99.74% 99.87% 100.00% Transport 47.12% 87.63% 97.49% 99.41% 99.84% 99.95% 100.00% Cumulative claims development 110% 100% 90% 80% 70% 60% 50% 40% 30% Accident Engineering Guarantee Liability Miscellaneous Motor Property Transport 20% 0 1 2 3 4 5 6 Years since claim event The following graphs indicate the incremental development of claims, underlying the above cumulative development. 36

Incremental claims development pattern Years since claim event Class of business 0 1 2 3 4 5 6 Accident 55.02% 31.40% 9.92% 2.68% 0.73% 0.17% 0.09% Engineering 63.02% 32.67% 3.87% 0.38% 0.05% 0.01% 0.00% Guarantee 28.29% 46.78% 16.57% 3.72% 0.98% 0.36% 3.31% Liability 38.32% 29.35% 17.79% 7.49% 3.23% 1.87% 1.96% Miscellaneous 48.52% 37.28% 10.96% 2.47% 0.59% 0.14% 0.05% Motor 78.66% 19.34% 1.73% 0.21% 0.04% 0.01% 0.01% Property 52.36% 39.48% 6.58% 1.07% 0.26% 0.13% 0.13% Transport 47.12% 40.50% 9.86% 1.93% 0.43% 0.11% 0.05% Incremental claims development 80% 70% 60% 50% 40% 30% 20% 10% Accident Engineering Guarantee Liability Miscellaneous Motor Property Transport 0% 0 1 2 3 4 5 6 Years since claim event The factors above were also applied in the calibration of the size of the notional companies. We have already discussed in previous sections how the notional companies are chiefly parameterised by gross written premium and gross unearned premium. Each notional company also needs to have a value for its outstanding claims reserves (outstanding reported and incurred but not reported reserves). In our framework the OCR is a function of the gross unearned premium using the following approach: We divide the gross unearned premium reserves by the one minus the earning pattern to get an estimate of the previous year s historic written premium. Historic premium in prior years is then attained by dividing each year s historic premium by a growth assumption. This gives us five years of historic premium for each notional company size, which is related to the size of that notional company via the gross unearned premium. We then use these historic gross written premiums to get estimates of claims currently outstanding. This is done by multiplying the gross written premium by an expected claims ratio to get the full ultimate loss arising from writing that level of premium. The ultimate loss is then multiplied by one minus the appropriate claims development factor (as an estimate of the percentage of claims currently outstanding). For example, R100m of gross written premium may result in a loss ratio of 60% giving ultimate claims of R60m. If this premium relates to motor business that was written two years ago then the 37

appropriate factor to apply (from the table above) is 2% yielding an estimate of outstanding claims for that year of R1.2m (R60m x 2%). 3.11 Dynamic Financial Analysis This section sets out the working of the DFA engine (also known as the simulation model) referred to earlier. We have used the Prophet DFA library 3 for this exercise and it must be noted that the formulas within this library are far too lengthy to be listed individually. Nevertheless, we have included salient formulas for the purposes of understanding the behaviour of the simulation engine. The DFA engine is concerned with producing results of stand alone capital requirements for each class of business. Since there are 8 classes of business considered, 99 company sizes for each class and 100,000 simulations run for each company size, collectively we have run 79.2 million simulations in calibrating this model. Each notional company is modelled identically except for the inputs described below (relating mainly to the size of the account and relevant risk). It should be noted that all inputs and calculations are performed on a gross basis. The following are the inputs into the system for each line of business, for each company size: Variable Description WP UPR PRM_PTN OCR(k) CLM_PTN(k) written premium for the coming year of new business unearned premium reserve at the end of the year Premium earning pattern (the % of premium earned in the year it is written) outstanding claims reserves plus IBNR for each historic accident year (mean estimate) Claim payment pattern over time: the proportion of unpaid ultimate claims paid from a particular accident year, k k Historic accident year where k = 1 is the most recent historic accident year & k = 0 is the coming accident year ULR distribution OCR distribution Ultimate loss ratio distribution and parameters (these vary according to the size of the earned premium) Coefficient of variation for the outstanding claims reserves (these differ according to the size of the reserve) 3.11.1 Earned premium Earned premium (EP) is calculated for the coming year of business. EP = UPR + WP * PRM_PTN Since the average earning pattern for all classes of business are less than a year, the full UPR is assumed to be earned in the coming year. Claims arise from two sources: Premium earned in the coming accident year (underwriting losses/profits) 3 Prophet is proprietary actuarial software developed by Deloitte and used internationally. 38

Premium earned in previous accident years (reserve run-off) The following two sections deal with each of these in turn. 3.11.2 Underwriting losses/profits The ultimate claims arising from premium earned in the coming accident year is attained by simulating an ultimate loss ratio and applying it to the projected earned premium in the period: FUT_AUC = EP * Simulated(ULR) For each simulation the ULR is simulated from a lognormal distribution appropriately parameterised for the size of the account. This parameterisation is based on data analysis, as described above in the underwriting risk sections. It is important to note that the claims simulated are statistically consistent with both the size of company and the amount of earned premium. The total ultimate claims are paid out in the coming accident year as follows: FUT_PAY = FUT_AUC * CLM_PTN(0) The reserves set up at the start and end of the first projection period are calculated according to the following formulas: FUT_RES_END = FUT_AUC * {1-CLM_PTN(0)} FUT_RES_BGN = 0 The reserve set-up is thus the portion of the actual ultimate claim that remains unpaid according to the development pattern. The change in reserve is calculated as follows: FUT_CHG_RES = FUT_RES_END - FUT_RES_BGN The two primary items from this section that affect the income statement (and hence profitability) are the claim payment made in each accident year (FUT_PAY) and the change in reserves (FUT_CHG_RES). 3.11.3 Reserve Run-off This section deals with the run-off of reserves set-up for historic accident periods. The ultimate losses as a result of outstanding reserves are simulated as follows: HIST_RRO(k) = OCR(k) * Simulate (OCR_CV) for k = 1,,6 The mean estimate of outstanding claims (given by OCR) is adjusted by a factor simulated using a lognormal distribution where the coefficient of variation (CV) of the Reserving Ratio is the primary parameter. The run-off of outstanding claims is then paid over the remaining development period in a manner similar to the claims in future periods: HIST_PAY(k) = HIST_RRO(k) * CLM_PTN(k) 39

HIST _ PAY = HIST _ PAY ( k) k The reserves set at the start and end of each period are then calculated according to the following formulas: HIST_RRO_END(k) = HIST_RRO(k) * {1- CLM_PTN(k)} HIST_RRO_BGN(k) = OCR(k) HIST_CHG_RES(k) = HIST_RRO_END(k) HIST_RRO_BGN(k) HIST _ CHG _ RES = HIST _ CHG _ RES( k) k The reserve set-up is thus the portion of the final run-off that remains unpaid according to the development pattern. The two primary items that affect the income statement (and hence profitability) are the claim payment made in each accident year (HIST_PAY) and the change in reserves (HIST_CHG_RES). 3.11.4 Underwriting Profit Since we are considering insurance profits/losses over a one year time horizon, underwriting profit (over the first projection year) is the primary item of interest to us. This is calculated as follows: UWP = EP FUT_PAY FUT_CHG_RES HIST_PAY HIST_CHG_RES 100,000 simulations for each account size give us a statistical distribution of underwriting profit. We examine the distribution of this variable and make special note of the 99.5 th, 99 th and 98 th percentile for each hypothetical company we run in our model. The results are then used to determine a set of capital requirements for each hypothetical company under consideration. 40

3.12 Risk Measures 3.12.1 Choice of risk measure There are a plethora of risk measures available to use for capital adequacy. We have used a value at risk approach (VAR) where we analyse the 99.5%, 99% and 98% sufficiency levels. While certain risk measures have different desirable mathematical properties (such as Tail VAR and others) our brief was to analyse the simulations using a VAR approach. It must be noted that the primary work has been performed in obtaining the simulations and other risk measures can be analysed at a later time if this is deemed appropriate. Further, it appears that the international regulatory practice is moving towards a VAR approach, with a 99.5% sufficiency level being the most widely used. 3.12.2 Number of simulations As discussed in the previous section 79.2 million simulations were run. In this section we detail the investigations performed by us in determining a satisfactory number of simulations to perform. We have concentrated our investigation on analysis of a single class of business and a single notional company size. Performing estimations in the tail of a long-tailed distribution by simulation is made more complex by the extent of simulation error. Simulation error arises since the result estimated is a function of the specific simulated outcomes of the model. It is an established fact that greater numbers of simulations reduces this error. We have aimed for a 1% error tolerance in our simulated analysis and the following heuristic investigation was performed to check if the number of simulations performed achieves the set error tolerance. We use simulations from a log-normal distribution to estimate the VAR (at the various sufficiency levels) as a result of a cumulatively increasing number of simulations. It must be noted that the distribution for capital requirements is not log-normally distributed though it should behave in a similar manner to an appropriately parameterised log-normal distribution. The following graph compares three simulated companies (as the number of simulations increase) against the actual 99.5% sufficiency level for an appropriately parameterised lognormal distribution: 41

The following graph compares three simulated companies (as the number of simulations increase) against the actual 99% sufficiency level for an appropriately parameterised lognormal distribution: 42

The following graph compares three simulated companies (as the number of simulations increase) against the actual 98% sufficiency level for an appropriately parameterised lognormal distribution: As a result of this investigation and the result of the same investigation performed for other suitably parameterised lognormal distributions we are comfortable with the number of simulations (100,000) run for each class of business, for each notional company size. 3.13 Tables/Surfaces of Gross Risk Capital The results of our simulation model are in the form of specified amounts of gross risk capital for each combined level of gross written premium and gross unearned premium. This section details the approach taken to apply these simulation results to specific companies. Our original approach involved fitting a three dimensional surface to the simulation results. Unfortunately, this approach was not successful but it is instructive to discuss the original approach and its shortcomings. 3.13.1 Original approach (surfaces) Fitting a function to the simulation results involves proposing a functional relationship between capital required and the risk factors of written premium and unearned premium. The advantages of fitting a function to the simulations results are: It is easy to apply in practice The fitted function will further smooth simulation error from the simulation results if the underlying functional relationship is correct It provides a transparent form for communicating the simulation results 43

The fitted coefficients can be interpreted and possibly compared between classes An over-riding disadvantage arises if the functional relationship does not adequately fit the full range of the data without becoming overly complex. We attempted to fit surfaces using of the following forms for the gross stand-alone risk capital: Quadratic: 2 2 SC i = β0 + β1 GWP+ β2 ( GWP) + β3 GUPR+ β4 ( GUPR) + β5 ( GWP) ( GUPR) and Power: SC i = β ( β 0 + β 1 GWP 3 2 β 4 + β GWP ) These resulted in intuitively appealing surfaces with the following shape: β 5 The properties/constraints that such a function needs to observe are: It should be non-negative (i.e. results in positive capital at every combination of written premium and unearned premium) It should be non-decreasing in both directions (i.e. it increases as written premium increases and unearned premium increases) It should be increasing at a decreasing rate in both directions (written premium and unearned premium) a result of diversification 44

It should allow adequately for diversification when levels of written premium and unearned premium change together Satisfying the above turns out to be non-trivial for the families of functions set out above. In particular, the final property (the correct interaction between written premium and unearned premium) proved difficult to model adequately. The difficulty is best seen in the following graph that measures the relative error of fitting a functional form to the simulation results against the values fitted. The two items of importance are the satisfactory fit until a certain level is reached and, thereafter, the subsequent over-fitting that occurs for larger companies. Over-fitting in this situation means that the fitted capital requirements are greater than necessary when compared to the simulation model results, and is shown by negative relative errors. 0.50 0.45 0.40 Relative Error 0.35 0.30 0.25 0.20 0.15 0.10 0.05 - - -0.05 500 1,000 1,500 2,000 2,500-0.10-0.15-0.20-0.25-0.30-0.35-0.40-0.45-0.50 3.13.2 Adjusted approach (tables) The approach of fitting a global surface was not adequate for the entire range of risk factors (written premium and unearned premium). This suggests using a local estimation method where we give more weight to simulations results closer to the point we are estimating. One practical approach is to interpolate the raw simulation results based on the levels of written premium and unearned premium. The specific form of the interpolation we have used in the industry impact study is a bilinear interpolation taking into account the relative distances from both written premium and unearned premium. Thus, instead of being based on an explicit functional form, the gross stand-alone capital results are obtained from pre-specified tables. These tables of raw simulation results can be found in Appendix E for each class of business at the various levels of sufficiency. Corresponding graphs (at the 99.5% level) can be found in Appendix F. Note that the axes for gross written premium and gross unearned premium are on a logarithmic and not linear scale. 45

We envisage that the FSB will implement this approach by way of a spreadsheet provided to industry with the following inputs: class of business level of gross written premium level of gross unearned premium The spreadsheet would then provide the gross stand-alone capital required as an output. This spreadsheet will work in concert with that needed for the correlation and diversification factor calculation, detailed in the following section. 3.14 Diversification & Correlation The gross risk capital obtained from tables (as discussed in the previous section), is adjusted to allow for specific projected retention levels and expenses of the company concerned in estimating their regulatory capital. This yields a net stand-alone capital level for each class of business written. The subject of this section is the combination of these net stand-alone capital levels to obtain the total net capital requirement (before allowance for investment returns). An inappropriately prudent estimate of the total net capital requirement can be obtained from summing the individual net stand-alone capital requirements. However, this does not take into account the following two effects from writing multiple lines of business: Diversification effects due to writing more than one class of business (these will in general reduce the capital required) Correlation effects between the different classes of business (these will in general increase the capital required since classes tend to be positively correlated) When stand-alone capital levels are added, the implicit assumption is that worst-case outcomes will occur at the same time. In reality, when combining classes of business, diversification benefits arise since it is unlikely that worst-case outcomes for each risk will occur at precisely the same time. In other words, if the risks are independent, a worst-case outcome for one risk is likely to occur with a normal outcome for another risk. The resulting total capital required is thus lower than the sum of the stand-alone capitals. One caveat to the above is that some classes of business are correlated. Essentially, where classes of business are positively correlated this means that there will be a tendency for adverse outcomes to occur together. This tends to increase the capital levels compared to those that arise where diversification alone is considered. These effects are taken into account by multiplying the sum of the net stand-alone capitals above by a statistically determined factor that allows for the above effects on a company s relative mix of business. This factor is calculated as the following ratio: factor ( p) = DCC( p) TSC( p) Where: DCC(p) = Diversified and Correlated Capital at the p th percentile TSC(p) = Total Stand-Alone Capital at the p th percentile 46

DCC is the appropriate percentile (99.5%, 99%, 98%) of a log-normal distribution with the following mean (mu or E[DCC]) and variance (sigma squared or V[DCC]) appropriately parameterised for underwriting risk: E [ DCC] = E[ Where: N i= 1 EPi E( ULR i )] N = The number of classes of business i = The specific class of business under consideration EP(i) = The earned premium in class of business i E[ULR(i)] = The expected ultimate loss ratio for class of business i V[ DCC] = Where: N 2 ( EPi ) V[ ULRi ] + 2 i< i= 1 ( EP )( EP ) Cov[ ULR ; ULR ] V[ULR(i)] = The variance of the ultimate loss ratio for class of business i Cov[ULR(i) ; ULR(j)] = The co-variance of the ultimate loss ratios of class i & j j i j i j TSC is the sum of the appropriate percentiles (99.5%, 99%, 98%) of a series of lognormal distributions where we allow for underwriting risk only. The following mean and variance are used in estimating the parameters of each lognormal distribution. E SC ] = EP E( ULR ) [ i i i 2 V [ SC ] = ( EP ) V[ ULR ] i i i The ratio of DCC to TSC is an approximate measure of the appropriate factor to be applied to the net total capital required to allow for diversification and correlation effects. The advantages of using this factor are: It has been calibrated consistently with the rest of the framework It allows for different concentrations of business written in the various classes It allows for both diversification and correlation effects It has satisfied various tests of robustness to changing parameter values The main weakness is that it is a univariate measurement allowing for underwriting risk only. This is mitigated to some extent by the fact that underwriting risk is the main underlying risk. Further, it is difficult to maintain the simplicity of the framework presented if reserving risk is included. The parameters for the above calculation can be found in the appendix. It is not envisaged that companies will perform this portion of the calculation in practice due to its complexity. Instead, a preprogrammed spreadsheet (provided by the FSB) will calculate the specific factor based on the split of earned premium in the various classes of business. The following hypothetical example serves to explain how the above would work in practice. The numbers given are simplified for illustrating the application of factors for correlation and diversification. In practice the results would be impacted by numerous other factors not discussed below. Company A single line of business, no allowance for correlation/diversification 47

Company A has R100m of gross earned premium in the motor class of business. Given Company A s level of premium and unearned premium and other factors, its net capital requirement is R40m. In particular, no diversification/correlation factor is applied to this net capital requirement since only one class of business is written. (Alternately one can consider a factor of one being applied to the net capital requirement thus leaving it unchanged). Company B multiple lines of business, allowance for correlation/diversification In contrast, Company B has R50m of gross earned premium in each of motor and property. Given Company B s levels of premiums, unearned premium and other factors its net capital requirement is R23m for motor and R27m for property. This gives a total net stand-alone capital requirement of R50m. Since Company B writes two classes of business a factor for correlation and diversification is calculated. The inputs into the calculation of this factor are as follows: Class of business Gross Earned Premium (Rm) Accident 0 Engineering 0 Guarantee 0 Liability 0 Miscellaneous 0 Motor 50 Property 50 Transport 0 The output of this calculation is as follows: Diversification & correlation factor 71.64% This yields to the following net capital requirement: Net capital required R35.82m The Net Capital Required is obtained by multiplying the total net stand-alone capital (R23m+27m) by the diversification and correlation factor. In particular, note that the overall capital requirement is lower for Company B than Company A despite the fact that Company B had a higher net stand-alone capital (R50m versus R40m). 3.15 Investment allowance 3.15.1 Asset Allocation An allowance for investment returns on assets backing insurance risks is allowed for. The net capital required (after adjustment for correlation and diversification between the classes of business) is reduced by an estimate of investment returns over the coming financial year. The estimate of investment returns has to take into account the following two factors: 48

The quantum of assets earning an investment return for the purposes of this allowance The asset split for the purposes of this calculation Regarding the quantum of assets we have chosen to allow for investment returns on all assets backing insurance liabilities (premiums and claims reserves). For the asset split we are assuming that a notional hypothecation of assets to liabilities on the balance sheet will be performed in the following broad manner: 1. Current liabilities will be matched with appropriate assets 2. Insurance liabilities will then be matched with the remaining assets 3. The remaining pool of assets are then deemed to be backing the shareholders funds In general, the order of allocating assets to liabilities is as follows: 1. Cash 2. Near cash 3. Fixed Interest 4. Property 5. Equity 6. Other This order of allocation has implications for the investment allowance (by virtue of the returns on the different asset classes) and also on the asset capital charge discussed in the following section. The only exception to this order of allocation applies to the first allocation of assets to current liabilities where it is more appropriate to switch the roles of near cash and cash. 3.15.2 Investment return There are two alternative approaches to follow regarding the returns to use for the investment allowance: Allowing for a risk premium on risky assets Applying a risk free rate of return (measured against an appropriate government security) Both have merits with the latter, in particular, being simpler to apply in practice. Nevertheless, we feel that allowance for the risk premium inherent in the different classes of assets is appropriate since the downside risk is being allowed for in the asset capital charge. Note that we are proposing use of investment returns before allowance for tax since these returns will only be required (for capital purposes) when an underwriting loss has been made in this situation no tax is payable on the investment return. 49

4 Investment Capital Charge 4.1 Overview This section gives a summary of the investigations performed to calibrate an appropriate charge for the investment risk of short term insurers. Investment risk is confined to the market risk aspect, i.e. the risk that market movements cause a loss in value of the assets held to back liabilities and capital requirements to such an extent that solvency is threatened. The aim of the exercise was to arrive at a set of capital adjustment factors for each asset class that should provide protection up to a specified level of confidence but not necessarily against all possible eventualities. The capital adjustment factor is applied to the value of the assets held in that class to arrive at an amount of capital to be held as a charge for investment risk. The aim of the capital adjustment factor is to provide the specified level of protection against loss in market value on only the assets backing the liabilities and other capital elements. The intention is not to protect the free assets that do not cover capital requirements. Therefore the capital adjustment factor only needs to be applied in relation to the assets backing the liabilities and regulatory capital elements. This implies that the assets would have to be allocated to the liabilities and capital elements to decide what assets the capital adjustment factors needs to be applied to. To minimise capital requirements, it is likely that the approach would be to first allocate the asset classes with the lowest capital adjustment factor. A more desirable approach would be to first allocate the assets that match the liabilities by nature, term and currency, but this is likely to produce much the same result given the types of business generally considered here. 4.2 Investment capital requirements in other territories and in Life assurance 4.2.1 Approach followed by other regulators 4.2.1.1 Australia (APRA): GGN110.2 specifies that a probability of default of maximum 0.5% over a 1 year horizon is the minimum requirement. A different combination of time horizon and probability of default may be used as long as it is appropriate for the business and consistent with the minimum requirement. As this method relies on an internal model being used by the insurer no specific capital factors are published. 4.2.1.2 UK (FSA): The minimum capital requirements consist of two items: A risk based measure using a formula approach called the enhanced capital requirement (ECR) and Internal capital assessment called the Individual capital adequacy standard (ICAS) The ECR = Asset risk charge + Insurance risk charge 50

The asset risk charge is obtained by applying factors to the various categories of assets. Details can be obtained on www.fsa.gov.uk 4.2.1.3 Canada (OSFI): In Canada a Minimum Capital Test (MCT) was introduced in 2003 which is a set of requirements based on a risk-based capital framework. A property and casualty insurer s minimum capital requirement is the sum of: Capital for On-Balance sheet assets Margins for Unearned premiums and unpaid claims Catastrophe reserves and additional policy provisions An amount for reinsurance ceded to unregistered reinsurers Capital for off-balance sheet exposures The capital requirements for investment risk is contained in the first bullet and specified as a set of capital factors to be applied to the assets. The capital factors also include an allowance for counterparty risk by distinguishing between different qualities of assets in each class. The capital factors are available at www.osfi-bsif.gc.ca/app/docrepository/1/eng/guidelines/capital/guidelines/mct_guideline_e.pdf 4.2.1.4 SA Life Insurance Capital Adequacy Requirement (CAR): The Capital Adequacy Requirement is calibrated to a 95% confidence level over a five year time horizon. The investment risk is allowed for in the Resilience and Worse investment return items of the Intermediate Ordinary Capital Requirement (IOCAR), which is then further grossed up by an investment factor to form the OCAR. The investment factor used to gross up the IOCAR allows for the fact that the IOCAR itself is backed by assets subject to the adverse scenarios. 51

Under the Resilience Capital adequacy requirement assets and liabilities are subjected to the following adverse scenarios simultaneously (table reproduced from ASSA PGN104, version6): Type of asset Equities FTSE/JSE All Share dividend yield below 4% FTSE/JSE All Share dividend yield 5% or above FTSE/JSE All Share dividend yield at 4% or above, but below 5% Fixed property Fixed income Inflation linked bonds Cash and fluctuating interest rate assets Other assets Foreign currency denominated assets Fall in fair value 30% fall in value 20% fall in value Interpolate between 20% and 30% 15% fall in value Impact of 25% relative increase/decrease in yield to maturity (i.e. when yields are 10%, test resilience to both 7.5% and 12.5% yield environments) Impact of increase/decrease in real yield to maturity by factor of 25% of real yield to maturity (i.e., a real yield of 4% increases to 5%). No change in value 35% fall in value Same as for domestic assets subject to a minimum of 20% fall in value For options and futures the long-term insurer s exposure to the relevant assets (e.g. equities) must be taken into account when calculating this requirement. A long-term insurer might, for example, hedge an equity portfolio by selling futures. Should the composition of the portfolio and the future index sold be identical, a fall in fair value of say 30% would make a resilience Capital Adequacy Requirement unnecessary for the hedged portion of the portfolio. As the composition will normally not be identical the Statutory Actuary would have to consider what offset to allow. In an extreme case, the fair value of the equity portfolio may even fall while the index rises. Notes: The requirement for equity values has been deduced from studying 12 months price movements of the JSE Actuaries All Share Index. The levels of 30% and 20% have been chosen to roughly correspond with a probability of less than 5% that these limits would be exceeded in any 12-month period. The limits for the other asset categories have been chosen to reflect the fact that in general terms the other asset classes are less volatile than equities. The worse investment return scenario assumes that future investment returns would be equal to 0.85x the valuation assumption (test for a 15% relative reduction). This implies that the valuation interest rate used in valuing both assets and liabilities and the assumed growth rates for future dividends and rentals where applicable must all be reduced to 0.85x valuation rate per annum. 4.2.1.5 Capital modelling by the CAR committee of ASSA In modelling the CAR requirements as stated above, the CAR sub-committee of the Life assurance committee of ASSA investigated returns in the South African and some overseas markets over time 52

starting around 1971. Their focus for the CAR requirements above was the 95 th percentile of movements in asset values, but in their investigation of equity returns, they also reported on the 99 th percentile. It shows a 33.4% capital factor for the 99 th percentile for domestic equity. After applying the grossing up factor assuming the capital is itself invested in equity. If the capital is again invested in equity, a capital requirement of 50.15% is obtained. We would expect the adjustment factor for a 99.5% level of confidence to be a somewhat higher than this. The following table shows the results of their analysis of the percentage movement in asset values Fund in Rand value SA UK US Data Start Date Dec-85 Dec-85 Dec-85 90th Percentile -13.7% -4.1% -11.0% 95th Percentile -25.3% -21.0% -19.9% 97.5th Percentile -30.2% -32.9% -35.2% 99th Percentile -33.4% -43.0% -45.1% Remove Foreign Exchange effect 90th Percentile -13.7% -17.0% -15.2% 95th Percentile -25.3% -22.0% -20.5% 97.5th Percentile -30.2% -24.4% -23.5% 99th Percentile -33.4% -28.2% -26.2% Foreign Exchange effect only 90th Percentile -7.6% -11.2% 95th Percentile -16.7% -24.1% 97.5th Percentile -23.9% -29.8% 99th Percentile -27.3% -31.3% * In all cases, the data END date is in March 2004 4.3 Estimation of capital factors for SA short term insurance 4.3.1 Purpose of the capital adjustment factors The investment capital requirement is set to ensure that changes in asset values do not adversely affect an insurer s ability to demonstrate solvency. For example, if an insurer has liabilities plus minimum capital requirement of say 100, backed by assets of 100 then it is technically solvent. However, if the value of the assets changes due to market movements to say 90 (a 10% fall in the 53

market), then the insurer is insolvent 4. However if it had 112 of assets it would still have 100.8 of assets after the fall and would still be solvent. The capital requirements aim to set a level of capital (12 in this example) that would ensure that the insurer is adequately protected against a secondary fall in asset values. This process is performed in the grossing-up of the asset capital charge and the insurance capital charge discussed in detail in the following section. It would not make sense to protect against all possible movements in asset values, as that would be a very onerous requirement. By analysing market behaviour and fitting financial, economic and statistical models to it, we can analyse the probabilities attaching to different possible movements in asset values. We can then set capital requirements at such a level that a desired probability of being able to demonstrate solvency after an adverse event is achieved. The capital requirement can be reworded to something like this: In order to be 99.5% certain that the market value of assets would exceed the value of the liabilities after a fall in market values, a capital amount of x% of the value of the assets needs to be held. The 99.5% can be replaced by whatever the final decision is on the level of confidence to use. 4.3.2 Asset classes considered Due to the lack of data available it is not possible to investigate a wide variety of possible investments separately. We therefore decided to investigate broad groupings of asset classes. The advantage is that this will keep the capital adjustment factors simple and easy to use. We considered the broad groupings of asset classes separately to arrive at capital factors for each of them. The asset classes considered were: Asset class Cash Equity Outstanding term Fixed Interest 1, 2, 5, 7, 10 Property n/a n/a n/a 4.3.3 Method of Estimation The capital factors were estimated using The Smith Model (TSM). TSM is a sophisticated deflator enabled stochastic asset model and economic scenario generator. It produces economic simulations of the future based on well accepted financial theory while also respecting the statistical features observed in the market. TSM can be calibrated to the market prices and yield on any date. The calibration used for this investigation was calibrated to the South African market as at 31 March 2005. Some results were tested for sensitivity by using the South African calibration as at 30 June 2005. For more information on TSM see Appendix G. 10 000 simulations were used to ensure sufficient credibility for the higher percentiles. (An event which has 0.5% probability of occurring is only expected to occur 50 times in 10 000 simulations.) 4.3.4 Cash Cash is defined as a very short term investment (less than one year) that provides return in the form of interest. It is considered to be a very low risk investment with little or no risk of loss of the capital or 4 Insolvency is loosely used here to denote the fact that excess assets have fallen below those required by a regulatory minimum capital requirement 54

market value. Examples of cash investments are deposits, banker s acceptances, commercial paper, treasury bills and certificates of deposit with outstanding term less than one year. A capital factor of 0% is suggested for cash. One would not expect to see any loss in market value from an investment in cash and this is supported by TSM which attaches a 0% probability to a decrease in market value for this asset class. 4.3.5 Equity Equity investment is defined as an investment in the shares or stocks of a company, listed on a recognised exchange. It also includes investments that are substantially similar to equity investment, e.g. investments in equity unit trusts. For the purposes of this investment capital charge, unlisted shares are not included in this category, but are included under other. The reason for this is that the modelling below is based on a listed equity index and reflects the characteristics of that investment. TSM was used to investigate the following problem: If 100 of liabilities is covered by an investment in equity how much of that asset needs to be held to ensure (with the appropriate probability) that the value of the assets would be greater than the value of the liabilities (i.e. 100) after an adverse event. The construction of this problem implies that the capital held to protect it against the fall in market value of equities is also held in equities and also subject to the same adverse event. It is desirable to allow for the fact that the capital itself would also be subject to the adverse event, but it is not necessarily the case that the capital would be invested in the same asset class as the assets for which the capital is set up. We therefore aimed to express the capital adjustment factor as a factor that would give the amount of capital for the assets under consideration, but not for the assets backing the capital as well. Once all the capital amounts for all the asset classes have been determined and accumulated, one would once again go through an allocation exercise where the assets not used to back liabilities is again allocated to the capital amount. These assets then need to be grossed up for the correct adjustment factors, given the assets allocated to the capital. The adverse events considered were the following: Immediate fall in asset value (investigated by using monthly simulated returns in the first year) Fall in asset value in the first year from the date of investigation Fall in asset value in the first 5 years from the date of investigation The rationale for the first option is to ensure sufficient assets at all times. The rationale for the second option is to ensure solvency at the next financial year end. The third option was included for interest sake only and investigates the position for liabilities that have a longer run off period. The first option was tested over 5 different time periods and the average value taken. The capital factor required showed very little variation from year to year, which is as expected. The resulting capital adjustment factors were: Adverse scenario 99.5% 99% 98% 95% 99.5% Immediate fall in asset value 16.6% 14.5% 12.5% 9.5% June05 Fall in asset value in the first year 38% 34% 29.8% 23% 36.7% Fall in asset value in the first 5 years 55.3% 47.9% 39.2% 29% 51% 55

The Immediate fall scenario shows a lower capital factor due to the lower volatility associated with projecting from a known starting point. As the time period of projection increases the volatility increases (the expanding funnel of doubt phenomenon). We recommend using the factors from the fall over first year scenario, for the chosen level of confidence. 56

4.3.6 Fixed Interest A fixed interest investment is defined as an investment of term longer than 1 year that has returns in the form of fixed periodic payments and a return of principal at maturity. These can be instruments issued by governments, local authorities or companies. There is some variation in the behaviour of these investments due to differences in term, payment profile, liquidity and credit risk. However, for the purpose of the investment capital risk we assume that they all have similar characteristics in terms of volatility of market price. Due to the nature of fixed interest investments, one would expect them to have less volatile behaviour than equities. Fixed interest investments are often used to match a fixed or nearly fixed set of cash flows. When the market value of the set of cash flows is calculated as the discounted value of the cash flows and the fixed interest investments match the cash flows exactly, then the value of both the assets and the liabilities will change in the same way to market movements in the interest rate. In that scenario no capital allowance is needed. However that level of exact matching is virtually impossible to achieve and the liabilities held by short term insurers are generally quite inflexible to market movements. In the context of short term insurance investments the risk exists that the market value of the fixed interest investment will decrease without a corresponding change in the value of the liability. The market value of a fixed interest investment can decrease when interest rates change. We have investigated changes in interest rates projected over a short term (monthly changes in the first year) as well as annual changes projected over the next 10 years. As the asset model starts at a fixed point in time where the yield curve is known, the changes in the yields tends to be smaller over the first year than over subsequent years. This is reflected in our results and the instantaneous changes are much smaller than the year-on-year changes. The year-on-year changes decrease slightly over time as the volatility decreases (yields are expected to revert to a long term mean over the very long term). We investigated zero coupon bonds of outstanding term 1, 2, 5, 7 and 10 years to estimate the impact on the market value of such bonds as a result of changes in the yield curve year on year. On a zero coupon bond a loss in market value occurs if the interest rate rises. Once again the amount of asset was investigated that is needed to ensure with the desired probability that the market value after the fall would cover the liability of 100. The average capital amount required over the 12 months investigated for the instantaneous fall and the 10 years investigated for the year on year fall was used to set a capital requirement. It was then again expressed as a percentage asset adjustment factor and gross up factor. The resulting capital adjustment factors were: Adverse scenario Instantaneous fall in MV Year on year fall in MV Outstanding Term 99.5% 99% 95% 99.5% 99% 95% 1 year 1.8% 1.6% 1.05% 6.67% 5.73% 3.7% 2 years 3.14% 2.76% 1.86% 11.27% 9.8% 6.46% 5 years 6.15% 5.45% 3.71% 19.75% 17.4% 11.72% Once again the instantaneous fall scenarios produce quite low results. This is due to the effect of starting at a known starting point and projecting over a very short time period (one month). The yearon-year change in market value is a more appropriate measure to use for a one year view on solvency. Next we investigated the trend in year-on-year changes in interest rates and the resulting capital factors. We added bonds at longer durations (7 years and 10 years) to get a better view of the trend in the capital factors. 57

Outstanding term 1 years 2 years 5 years 7 years 10 years 99.5% 6.67% 11.27% 19.75% 24.6% 27% 99% 5.73% 9.8% 17.4% 21.8% 24% 98% 5% 8.4% 15% 19% 20.9% 95% 3.7% 6.46% 11.75% 14.75% 16.25% 30% Trend in Capital Factors for Zero Coupon Bonds 25% Capital Factor 20% 15% 10% 5% 0% 99.5% 1 2 3 4 5 6 7 8 9 10 Outstanding Term 99% 98% 95% It is quite common to express capital requirements for fixed interest assets as an absolute change or percentage change in the yield for the investment (or a combination of these). This eliminates the need to specify values for all outstanding durations. However, there is clear evidence that the proportionate change in yield required for a specific level of confidence should vary by outstanding term of the bond, especially at the shorter durations. If the capital requirement is specified as a proportionate change it remains robust in changing economic conditions, e.g. periods of very low inflation and yields. The disadvantage of this method is that each company would then have to examine their bond portfolio and calculate the specific capital adjustment factors based on the impact of the yield change for the bonds they hold. Due to the additional complexity this option was not considered to be the optimal solution. However, it would be a simple extension of the work done above to obtain the results for this option if required. 4.3.7 Property It is interesting to note that the Alexander Forbes Property index shows no negative month-on-month movements over the period for which it is available, from 1983 to the present day. However, it would not be reasonable to assume that this continues indefinitely in the future. The version of the Smith Model that we used in this investigation has not yet been calibrated for South African property. It is possible to calibrate the model for property, but the data availability is very poor. The benefits gained by attempting this calibration do not justify the costs. We model an approximation to property by expressing it as a portfolio of equity and fixed interest to reflect the rental income and capital gains associated with property. We propose using a capital factor for property half way between that used for equity and that used for a 10 year bond. 58

4.3.8 Other asset classes It is desirable to have an asset category and capital factor for asset classes not covered by the above analysis. This would cover asset classes like unlisted shares, as well as other assets not easily placed into a category. It is not intended to cover all assets that do not fit 100% into the definitions of the asset classes given above. For each individual asset discretion should be used to ensure that the allocation to asset classes is sensible. Where an asset is clearly very similar to one of the standard classes, e.g. investment in a money market unit trust is very similar to cash, then that asset should be grouped with the standard asset class and not allocated to Other. We recommend that the capital factor for Other assets should be set at the same level as that used for equity. As the other assets can cover a wide range of assets, it is not possible to analyse the category and the capital factor set for it will necessarily be somewhat arbitrary in this case we are also being conservative in our recommendation. 4.3.9 Foreign exchange risk The analysis done by the life insurance CAR committee showed that there was not a statistically significant difference between the volatility of Rand denominated and Sterling (UK) and Dollar (US) denominated funds. They therefore decided not to have different requirements for funds denominated in foreign currencies, but to have a minimum volatility for foreign amounts. Note that for the life insurance capital calculation, the adverse scenarios are also applied to the liability values. So if a liability is denominated in a foreign currency the foreign exchange factor would apply to both the assets and the liabilities. This results in an offsetting effect for foreign assets backed by foreign liabilities. For the short term insurance industry we do not suggest that the liability values be adjusted for the adverse scenario and therefore a more explicit approach has to be followed to allow for foreign exchange risk. We propose that the same capital charges should apply to the foreign assets as to assets denominated in the domestic currency, but that an additional foreign asset capital charge may be appropriate if there is a material mismatch in the currency of assets and liabilities. For example, backing Dollar liabilities with Rand assets or vice versa. 4.4 Calculating the investment risk capital charge The calculation of the capital charge on investments is a multi-step process: 1. First, assets have to be matched or allocated to liabilities, starting with current and other liabilities and ending in the assets remaining being allocated to insurance liabilities (claims and premium reserves). 2. Thereafter, the relevant asset charges (as discussed in this section) are applied to the split of assets backing the insurance liabilities. 3. Finally, this asset capital charge is combined with the insurance capital charge of the previous section to allow for covariance effects. This is explained in the following section and involves another set of asset allocations in this instance to match the asset capital charge and insurance capital charge with appropriate assets. As discussed, in order to calculate the investment risk capital charge, assets have to be allocated to back the liabilities. We suggest that the liabilities are split into the relevant currencies and the liability in each currency considered separately. In the allocation of assets, the aim should be first to match the currency of the liabilities. (This is not a restriction, and in some cases it may be desirable to back liabilities by different currencies even though assets of the same currency are available.) Where the assets backing the liabilities are not in the same currency, a foreign exchange capital requirement 59

needs to be taken into account. Within a currency, the assets should be allocated to the liabilities in the order of least risky (cash) to most risky (other). (Once again this is not a restriction, and any sensible allocation method can be used.) The assets allocated should be the actual assets held by the company and not notional assets as per an investment mandate or other method. When sufficient assets have been allocated to back the full amount of the liabilities, the investment capital charge for each asset class and currency can be calculated by applying the capital factor to the amount of assets. The next step is the combination of the Insurance capital charge and the Investment capital charge. This is discussed in more detail in the section on the Total Capital Required. In Appendix G we include a more detailed description of The Smith Model, and Appendix H shows the allocation of the FSB s asset categories to the modelled asset categories. 60

5 Other charges 5.1 Operational risk We have made no explicit allowance for operational risk. The reasoning behind this is as follows: Where operational risk is not measurable, this is already implicitly included in the calibration, as it is included in the historical data. Where operational risk is measurable, we believe the correct approach in many instances is not to have a higher capital requirement, but rather to address the matters causing operational risk directly. In other words, we do not believe it is appropriate to require all insurers in South Africa to set aside an additional amount of capital to make provision for identifiable operational risks that are only relevant to some insurers. To the extent that operational risk affected historical loss ratios this is already included in the calibration. The fact that there is no explicit allowance for operational risk in the industry calibration does not mean that companies should be less vigilant about operational risk issues. If anything, the more detailed measurement envisaged for certified models and internal models should make it easier for companies and the FSB to identify operational risk, as compared to the current capital framework of 25% of net written premium. The GIRO Working Party paper on Quantifying Operation Risks in Insurance Companies (Tripp et. al.) provides useful insight into the problem of quantifying operational risks, and in particular in a regulatory model. We comment on some of the findings of the authors: The Working Party mentions that there are two methods of allowing for operational risk in a DFA model: implicitly, as we have done, and explicitly. The explicit method would involve the removal of all operational risk losses from the financial history of a company and construct a DFA model that models everything except operational risk. The operational losses should then be modelled separately and added back to the model which allows for operational risk explicitly. The Working Party states that the explicit method is preferred. However, while this may be the preferred approach for an individual company s DFA model, there are significant difficulties with this model in an industry-wide investigation. It is clear that the type of data required would not be available, and the GIRO Working Party paper mentions the collection of data on operational risks in the UK as the first priority. The paper suggests that operational risk has a different application for each company, depending on its own particular circumstances, and that there are many difficulties with quantifying risk relating to soft issues accurately (mainly the people element in operational risk, which, according to the Basle definition consists of risks relating to processes, people and systems). The paper quotes the FSA (Consultation Paper 142), who suggests that companies are encouraged to collect data on operational risks, but that they should consider both quantitative and qualitative factors to understand operational risk. Some of the conclusions of the paper include the following: Actuarial models for operational risk would have to be preceded by an understanding of process management, organisational design including defining roles and responsibilities, occupational psychology and general management. There is a need to collect data The relative importance of operational risk compared with insurance or market risk is unclear. The authors state that: Our illustrations show a relatively small operational risk only 2% of net 61

premiums on average. Further work is needed to quantify the real impact: it could be easily three four or more times the illustrated level, (p.62). Given the above, we believe that the application of an industry-wide operational risk capital requirement is currently not possible in South Africa, mainly as a result of lack of data. We therefore favour the implicit method in this industry calibration, but we also suggest that companies should consider operational risk individually, and that, over time, as more data becomes available, the FSB re-considers whether operational risk should be built into the solvency framework. 62

6 Total Capital Required 6.1 Covariance effects and Grossing-up This section details how the various capital charges are combined. This involves combining the insurance capital charge and the asset capital charge. In combining the capital charges we have allowed for the following: A fall in the value of assets backing the capital requirements The covariance effects (diversification and correlation) between the asset capital charge and the insurance capital charge The former is allowed for by grossing-up both charges by appropriate amounts to allow for a fall in the value of the capital charges. The latter is allowed for using the heuristic rule of summing the squares of the two capital charges and taking the square root (this is less than the sum of the charges and allows for the fact that companies are not likely to experience a worst-case asset event and worstcase insurance event at the same time). The following formulas set out how we envisage all of the above will be achieved in practice: ACC = Asset Capital Charge ICC = Insurance Capital Charge g 1 = Grossing-up factor on asset charge g 2 = Grossing-up factor on insurance charge TCR = Total Capital Required TCR = ACC g 1 2 + ICC g 2 2 The grossing-up factors are calculated via an intermediate calculation described below. This step involves the performance of an asset allocation (after the allocation of assets to current liabilities and reserves) to adjusted values for the asset capital charge and the insurance capital charge. These adjustments are given below and are performed so as not to penalise companies for the composition of elements of their shareholders funds not being used to back their capital requirements: TCR_ADJ = Intermediate total capital required (before grossing-up) ACC_ADJ = Adjusted Asset Capital Charge ICC_ADJ = Adjusted Insurance Capital Charge 2 TCR _ ADJ = ACC + ICC ACC ACC _ ADJ = TCR _ ADJ ACC + ICC 2 63

ICC _ ADJ = TCR _ ADJ ICC ACC + ICC An asset charge (calculated on the same basis as ACC) is calculated for the allocation of assets to ACC_ADJ and ICC_ADJ. These two charges are the weighted average fall in assets that could result for an appropriate level of sufficiency. c 1 = asset charge on ACC_ADJ where 0 < c 1 < 1 c 2 = asset charge on ICC_ADJ where 0 < c 2 < 1 The resulting grossing-up factors are calculated as follows: g 1 = 1- c 1 where 0 < g 1 < 1 g 2 = 1-0.5*c 2 where 0 < g2 < 1 The rationale for the above is that for the asset capital charge, full grossing-up should be allowed for as you will need a grossed-up asset charge in precisely the situation that you need the asset charge itself. The grossing-up of the insurance capital charge only takes half of the appropriate asset charge into account since a worst case insurance event will not always happen at the same time as a worst case asset event. The use of a factor of a half can be seen to be allowing for a 50% correlation between insurance catastrophes and investment market crashes, which is in line with the intended practice in European markets. It is important to note that we envisage that companies will receive a credit towards their total capital requirement equal to their overall prescribed margins (from claim reserves and premium reserves). We expand more on this in the following section. MCR = TCR PM PM = total prescribed margins MCR = minimum capital requirement The following hypothetical example serves to explain how the above would work in practice. The numbers given are simplified for illustrating the application of grossing-up factors. In practice the results would be impacted by numerous other factors not discussed below. Consider a company that before grossing-up its capital charges had the following charges: Charge Amount (R m) Asset Capital Charge (ACC) 30 Insurance Capital Charge (ICC) 100 64

This would result in the following intermediate capital charge (allowing for covariance effects but not grossing-up). Intermediate Capital Charge (TCR_ADJ) R104.4m The above is obtained by summing the squares of the capital charges above (asset capital charge and insurance capital charge) and taking square roots as in the formula for TCR_ADJ above. This leads to the following adjusted capital charges: Adjusted Charge Amount (R m) Asset Capital Charge (ACC_ADJ) 24.09 Insurance Capital Charge (ICC_ADJ) 80.31 These charges are then allocated to the assets remaining after the allocations performed for current liabilities and other liabilities. This asset allocation is performed in a similar manner to those performed for the liabilities. This will result in a set of two calculated asset capital charge percentages for each of the asset capital and insurance capital. The following charge percentages would result from the above allocation: Capital Charge Percentage Percentage Asset Capital Charge Percentage (c1) 0.1 Insurance Capital Charge Percentage (c2) 0.3 Note that the numbers given in the table above will depend on the actual assets held in a particular situation. In this case we have assumed the numbers above for the purpose of this hypothetical illustration. However, the asset capital charge percentage will always be higher than the insurance capital charge percentage since the allocation for the asset charge is performed first. Grossing-up factor Percentage Asset Capital Charge: g 1 ( 1-0.1 ) 0.9 Insurance Capital Charge: g 2 ( 1-0.5 x 0.3 ) 0.85 We are now in a position to calculate TCR from the formula above since we know ACC, ICC, g 1 and g 2. The final value of R122.28m for TCR is obtained as follows: TCR = ( 30 ) + ( 100 ) 2 0.9 2 0. 85 65

7 Liability Estimation 7.1 Claims reserves 7.1.1 Original approach We proposed a default model for the calculation of the best estimate of outstanding claims reserves (outstanding reported claims reserve and incurred but not reported claims reserves). The reason for this original approach was to obtain consistency with the analyses performed on ultimate losses and claims development. The formula applied a percentage to Net Earned Premium for each class of business over the past six years. The formula had two components: a loss ratio applied to net earned premium yielding the net expected claims for that period and an outstanding claims factor applied thereafter (this factor takes into account the duration since that earning period). The loss ratio and outstanding claims factors applied were consistent with those derived for insurance capital (discussed in a previous section). Further, a best estimate of gross claims reserves was required for use in the prescribed margin calculation. This was made by applying the same percentage as above to historic gross earned premiums. 7.1.2 Revised approach A disadvantage of the above approach is that it provides a formula-driven best estimate of outstanding reported claims where it may be felt that companies should be able to perform more accurate estimates for these reported claims since individual case estimates can be made. We thus revised our approach to providing a formula to give a best estimate of IBNR alone (which should then be added to companies case estimates of outstanding reported claims). The estimation of IBNR reserves are more suited to an analytical approach based on past experience. In summary, individual companies will be responsible for determining their own best estimate reserve for outstanding reported claims. Incurred but not reported claims (IBNR) will be reserved for on the prescribed basis described below. The combination of these two best estimate reserves will be scaledup to obtain an overall claims reserve with a 75% sufficiency level. The difference between the best estimate and 75% level is referred to as the prescribed margin. Note that, in the certified model, reserving would be one of the most natural places for a company to deviate from the industry calibration. This may especially be appropriate for niche insurers writing business with different characteristics to standard short term insurance business. 7.1.3 Best estimate IBNR Reserves IBNR reserves (gross and net) are obtained by multiplying earned premium (gross and net) by an appropriate percentage varying by class and development year. For this purpose, earned premium for the last six historic years was considered. The following table contains factors for incurred but not reported claims for each development period: 66

Percentage of claims outstanding at the end of year: Class of business 1 2 3 4 5 6 Accident 9.00% 2.90% 0.94% 0.30% 0.10% 0.03% Engineering 8.67% 2.57% 2.04% 1.99% 1.98% 1.98% Guarantee 24.92% 5.46% 1.39% 0.54% 0.36% 0.33% Liability 17.01% 3.73% 1.32% 0.89% 0.81% 0.80% Miscellaneous 7.30% 1.01% 0.29% 0.20% 0.19% 0.19% Motor 4.33% 0.63% 0.31% 0.28% 0.28% 0.28% Property 6.14% 0.43% 0.12% 0.10% 0.10% 0.10% Transport 9.71% 3.40% 1.69% 1.22% 1.10% 1.06% Traditional chain ladder methods were used to obtain best estimate IBNR reserves based on the IBNR development triangles filled out by insurers in the STAR returns. These reserves were then expressed as a proportion of earned premium and curves were fitted to the resulting values. 7.1.4 Best estimate Outstanding reported claims Reserves As stated above, the outstanding reported claims will be estimated by each company on a best estimate basis. For the purposes of our industry impact analysis we have taken the current OCR levels as best estimates of outstanding reported claims. Depending on the prudence included in current reserving practices this may or may not be accurate. 7.1.5 Prescribed margins The use of prescribed margins aims to yield reserves that have a 75% level of sufficiency. We envisage application of a recommended prescribed margin formula to liabilities calculated on a best estimate basis. It must be noted that all prescribed margins can be offset against the capital requirements calculated above. The best estimate IBNR reserves produced by the factors above are combined with the best estimate OCR reserves and scaled up by the prescribed margin formula to obtain reserves with 75% sufficiency. The prescribed margin formula is consistent with the investigation performed for reserving risk (as discussed in the insurance capital charge) and it thus operates as a function of the gross best estimate claims reserves. The prescribed margin formula has the following functional form for each class of business: PrescibedMargin = a + b(grossclaimsreserves) c 67

The specific parameters used for each class of business are given in the table below. These curves are calculated so as to increase the level of sufficiency of a best estimate reserve to 75%. Class of business a b c Accident -2.22 2.80-0.01 Engineering -1.32 1.61 0.00 Guarantee -7.47 8.22-0.01 Liability -1.29 1.60 0.00 Miscellaneous -4.75 5.57-0.01 Motor -1.86 2.56-0.02 Property -3.65 4.48-0.01 Transport -15.67 16.30 0.00 7.2 IBNR Investigation This section details the investigation performed to analyse the emergence of reported claims by year of development for each class of business. The results have been used in the best estimate IBNR factors. We created a new data store by extracting the IBNR development triangles for each insurer from Section 7 in the STAR returns. Other data used included gross and net earned premium, estimated ultimate losses and estimated ultimate loss ratios this was found from our claims paid data store discussed earlier. The following steps were carried out separately for each class of business and involved an iterative approach with feedback between the various steps: Inflation: All monetary values extracted received an inflation adjustment consistent with similar adjustments made in previous analyses. Data cleaning: Each company s development triangles were considered individually in detail for reasonableness and validity. Further, the corresponding gross and net earned premiums, ultimate losses and ultimate loss ratios were also considered in order to form a view of the overall reliability of a specific insurer s data. Outlying and extreme values were either cleaned or removed from the analysis. For example, in certain cases large negative IBNR provisions held in the first year of development were removed from the data to avoid obtaining negative development factors. In some cases NEP values were extremely low and the IBNR values as a proportion thereof were in the order of a few thousand percent. There were not many of these and they were also removed. Where negative incremental IBNR values were recorded in the STAR returns, it was assumed that these represented downward adjustments to IBNR estimates from earlier years. Hence they were offset against the most recent positive provisions for a given accident year. Where data was extremely sparse (e.g. a company s IBNR development triangle as at 2004 contained only one entry even though at least 6 years of STAR returns were submitted) the data was considered to be inadequate and excluded. In some cases companies recorded IBNR provisions but the NEP s (and often even GEP s) were zero or blank. These data points were also excluded. Development factors: Standard chain ladder methods were used to calculate company-specific development factors between the various development years in the IBNR triangles. Where an insurer did not have data reliable enough to calculate its own development factors, industry average development factors were assigned. The industry average development factors were obtained by calculating weighted average development factors from the companies who did have individual factors, with the results being weighted by the relevant average gross earned premiums. 68

Tail factors were extrapolated by fitting exponential curves of the form f = 1+ be development factors on a company-by-company basis as well as an overall industry tail factor. i ci to the Development of IBNR triangles: The development factors obtained above were then used to calculate the bottom right hand corner of each company s IBNR triangle. This in turn allowed the estimation of the development of claims reporting for each accident period. The IBNR reserves so obtained were expressed as a proportion of the corresponding earned premium. Further cleaning was performed on the figures obtained and an iterative process was used to move between steps two and five. Smoothing of results: A weighted average IBNR reserve percentage varying by year of development was obtained from the above step. The weightings were by gross earned premium, increasing the reliance on data from large companies. Exponential curves of the form b+ ct IBNR = a + e were fitted to these average IBNR percentages in each development year to t smooth out random fluctuations in the data. 7.3 Premium reserves For premium reserves (UPR, URR, etc) it is felt that the degree of prudence contained using current estimation techniques (365ths method for instance) may be broadly equivalent to that required through the application of prescribed margins. The primary reason for this is that an unexpired premium reserve implicitly contains some profit margin. Where a company has specific knowledge that their premiums are inadequate, appropriate prudence would need to be borne in mind when setting an additional unexpired risk reserve. This URR would also form part of the insurance liabilities and would have to be set at a 75% level of sufficiency. Since the calculation of the URR would depend on the context and the specific circumstances of the company, it is not appropriate to prescribe a formula for the URR. For the purposes of our industry impact analysis we have thus assumed that the premium reserves inherently contain a 75% level of sufficiency. Calculation of the prescribed margin for premium reserves is thus used to quantify the credit towards required capital only. The following formula is applied to the net unearned premium reserve to determine the amount of prescribed margins contained therein. UPR_ PM = MAX(0,1 a b( GEP) c ) The parameterisation of the above formula varies by class of business as follows: Class of business a b c Accident 0.90983 0.00000 0.00000 Engineering 0.88405 0.00000 0.00000 Guarantee 0.95079 0.00000 0.00000 Liability 0.86177 0.00000 0.00000 Miscellaneous -25.79795 27.87515-0.00260 Motor 0.86615 4.32524-0.31210 Property -90.13983 91.65954-0.00048 Transport -7.73843 9.02279-0.00400 69

8 Results of the calibration 8.1 Overall results of the final calibration of the model In this section we show the results of the industry calibration described in the previous sections. We start this section by a discussion of the results of the calibration in terms of reserves, and then we discuss the impact of the calibration in terms of the Minimum Capital Requirement. On the basis of the above methodology we have compared our best estimate reserves plus prescribed margin to the current reserves held in the industry. This is shown in the table and graph below: RESERVES (OCR + IBNR) (R'000) Best estimates Prescribed margins Prescribed Reserves Current Captive 193,000 53,320 246,319 183,929 Cell Captive 523,029 112,525 635,554 565,731 Niche 1,050,541 209,751 1,260,292 1,718,092 Not Completing Quarterlies 27,179 7,006 34,185 27,049 Reinsurer 1,422,776 241,520 1,664,296 2,078,296 Run-off 104,869 28,458 133,327 99,604 Typical 5,303,070 827,431 6,130,501 5,350,403 Industry total 8,624,464 1,480,010 10,104,474 10,023,105 7,000,000 6,000,000 Current Current reserves versus Prescribed reserves (R'000) 75% Reserve 5,000,000 4,000,000 3,000,000 2,000,000 1,000,000 - Captive Cell Captive Niche Not Completing Quarterlies Reinsurer Run-off Typical This indicates that the industry calibration may understate reserves for reinsurers and niche insurers. For both of these types of insurance companies, run-off periods are often longer than for other types of insurers, and risks are different. We do not have sufficient data to calibrate reserving requirements 70

specifically for these types of insurers, and this may be an area where a certified model may be usefully applied. The following tables give a further breakdown of the components of the above reserve calculation for different classes of business: IBNR RESERVES COMPARISON (R'000) Best estimate Current Accident 113,677 98,104.33 Engineering 134,467 78,705.55 Guarantee 232,269 606,102.34 Liability 191,315 249,502.96 Miscellaneous 311,940 839,597.68 Motor 678,286 950,872.11 Property 627,814 891,019.57 Transport 125,721 100,224 Industry Total 2,415,488 3,814,129 TOTAL CLAIMS RESERVES COMPARISON (R'000) Best estimate Current Accident 321,425 305,852.05 Engineering 297,190 241,428.93 Guarantee 545,009 918,842.60 Liability 1,155,580 1,213,767.56 Miscellaneous 1,267,628 1,795,284.90 Motor 2,583,686 2,856,271.98 Property 2,089,899 2,353,104.46 Transport 364,049 338,552 Industry Total 8,624,464 10,023,105 The above tables indicate that there may already be margins included in current reserves held, and the 75% level of sufficiency would lead to reserving that compares fairly well with current levels. Initially, we proposed that companies should calculate best estimate reserves themselves, and we will simply specify a calibration for taking reserves up to the 75 th percentile. However, from initial industry feedback, it was clear that there was a need for the calibration to also include formulae for setting IBNR reserves. After further discussion it was also clear that companies would be comfortable to use their own reported outstanding claims reserve, but some would want to rely on the industry calibration of the IBNR reserve. The above calibration was felt to be more accurate than using a simple percentage of written premiums. In the industry impact analysis above we have used the outstanding reported claims reserves specified in the STAR returns as companies best estimates of outstanding reported claims. This may or may not be a valid assumption, for example some companies may have implicit margins built into their estimates of outstanding reported claims. We do, however, recommend that the FSB makes it clear that companies should ensure that the reserves that they submit in their STAR returns comply with the framework requirement of being disclosed on a 75 th percentile basis. The reserving calculation is important to companies in the sense that it may determine the tax liability of companies. However, the tax treatment of the suggested new industry calibration must still be clarified. 71

We now consider the results of the industry calibration in terms of MCR The following table indicates, for different types of insurer 5 : How the overall levels of minimum capital requirement (MCR) under the industry calibration at different levels of sufficiency (98%, 99% and 99.5%) compare against current levels of capital required 6 and against shareholders assets. That the overall levels of shareholders assets in the industry are sufficient to cover capital requirements under the 98% and 99% and 99.5% levels of sufficiency. However, all these levels of capital are considerably higher than the current capital requirement. The following types of insurers would, as a group, have insufficient shareholders assets to meet the MCR under the industry calibration: Cell Captives (at the 98%, 99% and 99.5% levels), niche insurers (at the 99% and 99.5% levels) and reinsurers (at the 99.5% level). Of course, comparisons on an aggregated basis give no indication of the impact on individual companies. This is discussed in more detail below. MINIMUM CAPITAL REQUIRED (R'000) Max(25% NWP ; R10m ) Shareholders' assets (ADJ) 98% 99% 99.5% Captive 120,000 804,082 276,783 386,147 504,384 Cell Captive 607,610 1,251,633 1,814,739 2,286,898 2,777,127 Niche 744,566 4,666,716 3,640,569 4,765,818 6,011,319 Not Completing Quarterlies 40,823 1,451,276 103,592 133,104 162,358 Reinsurer 466,274 1,629,471 1,109,693 1,453,753 1,818,831 Run-off 133,640 2,237,153 166,571 186,430 208,489 Typical 5,656,747 13,559,328 8,079,593 10,236,640 12,520,463 Industry total 7,769,660 25,599,659 15,191,542 19,448,789 24,002,971 The total capital required, i.e. MCR plus prescribed margins (margins on insurance liabilities to take them up to the 75th percentile), can be summarised as follows: TOTAL CAPITAL REQUIRED (R'000) 98% 99% 99.5% Captive 333,413 442,777 561,014 Cell Captive 1,986,906 2,459,064 2,949,293 Niche 3,895,111 5,020,360 6,265,862 Not Completing Quarterlies 112,266 141,778 171,032 Reinsurer 1,385,200 1,729,259 2,094,337 Run-off 195,086 214,944 237,004 Typical 9,135,788 11,292,835 13,576,658 Industry total 17,043,769 21,301,017 25,855,199 5 In identifying a particular insurer as niche, captive, reinsurer, typical and so on, we have used the classification of the FSB. 6 Note that we took the current requirement to be 25% of net written premium with a minimum of R10m. Whilst the R10m minimum is not currently included in legislation it seems this rule will be implemented almost without doubt. Hence, to remove distortions in comparisons, particularly when considering smaller companies, we have done all comparisons including the R10m minimum. 72

The following graph gives an indication of the total capital requirement, MCR and current capital requirement with a R10m minimum, for different sizes of company: It is easier to compare these graphically for different company sizes. The following graph shows the three measurements for the largest one third of the market: 73

In the medium and smaller size segment of the market, the capital requirements at the 99.5% level are onerous, but it is also clear that it is the R10m minimum that affects several small companies: 74

The following graph indicates how setting the capital requirement at 98% and 99% would make the minimum requirement considerably less onerous than that shown in the above tables: We have also determined the impact of the new solvency requirements on those companies that would have to raise additional capital to meet the minimum requirements. This was determined in the following way: The current 10% contingency reserve was not included as a liability, thus releasing its value into the shareholders fund. Where our estimate of reserves on a 75 th percentile basis was lower than the current reserves held by the company, this amount was released and included in shareholders capital against which the MCR was compared. Conversely, where a company s current reserving level is lower than the insurance liabilities determined at the 75 th percentile, shareholders assets reduced accordingly. For every company, the MCR was then determined in accordance with the method described in the rest of this report. Only where the MCR exceeded shareholders assets after adjustment for reserving as described above, the difference between the minimum capital requirement and shareholders assets was determined and recorded. The sum of these differences is contained in the table below. 75

The table below does not contain any allowance for the fact that several companies would be in a position to release capital under the new requirements, and only shows the results for those companies with a shortfall. SHAREHOLDERS' ASSET (ADJ) SHORTFALL TO MINIMUM CAPITAL REQUIREMENT (R'000) 98% 99% 99.5% Captive 5,492 42,151 93,715 Cell Captive 574,307 1,046,465 1,536,322 Niche 1,335,832 1,995,433 2,762,071 Not Completing Quarterlies 7,756 7,756 7,756 Reinsurer 7,122 122,695 269,971 Run-off 83,985 103,843 125,903 Typical 736,897 1,612,240 2,775,940 Industry total 2,751,391 4,930,583 7,571,677 Given the shareholders assets held in the different types of insurers, the above table indicates that the most significant impact of the suggested industry calibration would be on cell captives and niche insurers. The table below measures the shortfall as a percentage of the adjusted shareholders funds for each type of insurer. From this it can be seen that at the 99.5% level, the capital requirement for Typical Insurers (whose shareholders assets fall short of the MCR) becomes a significant proportion of the total shareholders assets currently held by all Typical Insurers in the industry (20.5%). The same is also true for reinsurers (16.6%). Whilst this ratio of additional capital required by those companies that fall short, compared against total shareholders assets currently held, does not represent the additional capital that would be needed for the industry, it still gives a useful indication of the potential impact of the industry calibration as described in the rest of the report. SHAREHOLDERS' ASSET (ADJ) SHORTFALL TO MINIMUM CAPITAL REQUIREMENT (R'000) 98% 99% 99.5% Captive 0.7% 5.2% 11.7% Cell Captive 45.9% 83.6% 122.7% Niche 28.6% 42.8% 59.2% Not Completing Quarterlies 0.5% 0.5% 0.5% Reinsurer 0.4% 7.5% 16.6% Run-off 3.8% 4.6% 5.6% Typical 5.4% 11.9% 20.5% Industry total 10.7% 19.3% 29.6% The table below contains the actual splits between the final insurance and asset capital charges (after grossing-up) for each sufficiency level: 98% 99% 99.5% ICC ACC ICC ACC ICC ACC Captive 99.98% 0.02% 99.98% 0.02% 99.98% 0.02% Cell Captive 99.82% 0.18% 99.84% 0.16% 99.84% 0.16% Niche 95.49% 4.51% 96.04% 3.96% 96.23% 3.77% Not Completing Quarterlies 100.00% 0.00% 100.00% 0.00% 100.00% 0.00% Reinsurer 93.63% 6.37% 94.33% 5.67% 94.54% 5.46% Run-off 90.22% 9.78% 91.38% 8.62% 91.81% 8.19% Typical 91.31% 8.69% 92.00% 8.00% 92.08% 7.92% Total 93.59% 6.41% 94.19% 5.81% 94.33% 5.67% 76

It can be seen that the insurance capital charge has a far greater impact on companies than the asset capital charge. The main reason for this is the conservative investment strategies employed by most short-term insurance companies (predominantly cash-based). Further, the relative significance of the two charges stays roughly constant at the different levels of sufficiency. 8.2 Applicability of the proposed default model This section sets out the areas where the default model may not be best applied and those types of company that have given us feedback to this effect. The model was constructed to be broadly applicable to as many players in the short-term insurance industry as possible. Nevertheless, the model does not apply perfectly to some company types. These can be broken into four main types of company: Captives and cell-captives Re-insurers Niche insurers Companies in run-off 8.2.1 Captives and Cell-Captives Captives and cell captives gave us the most feedback regarding our individual impact analyses. In particular they were concerned that the model did not take into account the special nature of their businesses. The following issues were raised: Star return data was felt to be inadequate for a proper determination of risk The eight different classes of business available in the Star return data do not allow for the varying types of business written by these insurers and the concomitant diversification and correlation effects are thus ignored. The proposed framework would be more suitable to third party business than to first party business. Stop loss reinsurance can be used to limit losses on a particular account and capital requirements should be limited to this potential loss (this principle extends to all types of insurers and should be taken into account in a certified model) In some cases expenses are covered under the reinsurance arrangement. Recapitalization obligations normally form part of the shareholders agreements. While this does not guarantee the availability of capital it is significantly different from the operations of other insurers. Due to the above recapitalization agreements it was proposed that the focus of attention should be on the strength of the cell owner. We agree with this approach and would expect it to play a role in any certified model. There was concern regarding the manner in which non-proportional insurance has been allowed for in the proposed framework. (We trust that this has been addressed in the relevant sections of this report.) 77

Flexible expense structures are used as risk mitigation tools (expense payments decrease as loss ratios increase and vice versa) and this would need to be considered. Such company specific details are naturally important but are beyond the scope of an industry calibration. From our meetings with and comments received from Cell Captives, it is not clear whether the Cell Captive industry is in favour of disclosing more detailed data in the STAR returns, which would effectively be used to motivate the low risks typically taken on by Cell Captives. Often, the only risk carried in practice by the Cell Captive is the credit risk of the shareholder of the cell, and even this risk is limited in practice. Given this, we see no alternative but to recommend that those Cell Captives who cannot meet the capital requirements of the industry calibration should therefore apply a certified model in order to motivate to the FSB why they should be allowed to hold lower levels of capital. It is probably appropriate for the certified model to be applied in different ways to Third Party and First Party cells. Our recommendation would be that the industry calibration be applied in principle to Third Party cells, but that there would of course still be allowance for a company to apply a certified model to the Third Party cells. In practice, such an approach may cause difficulties, but this would depend on the individual circumstances of the Cell Captive. First Party cells however would be more appropriately dealt with via certified models or internal models. Given the large number of specific issues that should be considered for Cell Captives, the lack of data, and the impact of the industry calibration on Cell Captives, we recommend that: There should still be considerable consultation between the FSB and the Cell Captive industry The FSB should encourage the Cell Captive industry to develop certified models to ensure prudent capital levels are held in this industry, taking into account the unique characteristics of cell captive business in a way that has not been possible in the industry calibration 78

8.2.2 Re-insurers There are several reasons that re-insurers may feel that the default capital charges are not applicable to them: Reinsurance business can be seen as more risky/volatile than business written by direct writers Their business has a longer tail due to reporting delays They have indicated that the format of their data for Star return purposes is not appropriate for the way they operate. For reinsurers, we recommend that at least the reserving calculation be done on a certified model basis, given their unique run-off characteristics. It would also be necessary to consult further with the reinsurance industry to determine ways in which the STAR returns have to be modified. Given the small number of reinsurance companies registered in South Africa, we believe it would be impractical to calibrate a model specifically for reinsurers, and we expect that for most of the reinsurers it would make sense to use a certified model in any event. 79

8.2.3 Niche Insurers The industry calibration may or may not be appropriate to a niche insurer depending on the nature of their business. There are a number of niche insurers with very specific defining characteristics which are impossible to take into account in an industry calibration. Hence a niche industry calibration would not be possible, given that each of these insurers face such widely varying risks. Again, we believe the only practical route for niche insurers would be the adoption of a certified model, should the industry calibration result in inappropriate capital requirements. 8.2.4 Companies in run-off These types of company are characterised by having no prospective written premium and very little or no unearned premium reserve. However, they may have significant outstanding claims reserves. Since the insurance capital charge is calibrated via the written premium and unearned premium the result may be artificially low for such companies. The reason behind this is that the level of outstanding claims reserves in our proposed model is set relative to the level of written premium and unearned premium in line with industry averages. As discussed above, we believe that the use of certified or internal models may be indicated for reinsurers, niche insurers, companies in run-off and cell captives. Nevertheless, aspects of the above industry calibration can still be used by these companies where appropriate, and hence the work performed here should have some value to some of these companies. Further work should be done on the way in which the certified model and internal models will apply in practice. 8.3 Applying the industry calibration in practice In this section we make further comments on the way in which the industry calibration should be applied. We would recommend that the implementation of new capital requirements be done in such a way that it does not have a crippling effect on the industry. The ways in which certified models may be applied in practice are beyond the scope of this report, but guidance on this should be provided by the FSB before implementation of the new framework. However, in our view, certified models would typically be applied where inter alia: The industry reserving calculation does not allow adequately for the risks of the company, or does not reflect a unique run-off triangle; and/or Where the insurer has a significant non-proportional reinsurance programme, reducing its risks at the tail end, and hence the insurer should be in a position to motivate a lower capital requirement Given the industry impact of the recommended FCR approach, we also feel that it would not be appropriate for the FSB to adopt the 99.5% level of sufficiency at the outset. We recommend that the industry be given time to build up capital to the required levels. The FSB should take the final decision on this, but we believe an approach whereby the MCR would be set in accordance with the 98% level of sufficiency for the first 4 years, then 99% for another 2 years and finally 99.5% in the 7 th year from the date of implementation would have the following advantages: it would minimise the potential impact of the new requirements on the industry at the outset, while at the same time ensuring that those companies that are under-capitalised increase their capital, and those companies where the industry calibration does not lead to appropriate results can develop internal or certified models 80

it would allow time to collect more data in STAR returns which in turn would enable, if necessary, a more refined fit of the overall model to determine the appropriate level of capital at the 99.5% level. The reasoning behind the above recommendation is the following analysis of net profit in the industry, as compared to the amount of additional capital required by those companies who fall short of the minimum requirement in the industry. While this is a theoretical and imperfect measure, we have divided the MCR of each company who needs additional capital by the net profit (after investment income and tax) generated by each company in 2004. This shows theoretically how long it would take for a company to reach the MCR if it currently has a shortfall and it uses its 2004 levels of profit to meet this shortfall. Clearly, companies should not view this measure as a realistic indication of how long it would take to reach the MCR, as underwriting profits will differ from year to year, and investment income would increase as the company holds higher and higher levels of capital. Nevertheless, the following tables can be interpreted as providing some indication of the potential strain of the new requirements on companies. MCR Shortfall: 98% (measured in years of profit) Years of Percentage of Total Average Net profit Frequency Percentage of Total Total Net Premium (Net Premium) Premium 0 49 55.06% 19,706,834 66.43% 402,180 1 5 60.67% 4,541,426 81.74% 908,285 2 4 65.17% 1,850,284 87.97% 462,571 3 5 70.79% 1,536,462 93.15% 307,292 4 4 75.28% 334,404 94.28% 83,601 5 2 77.53% 239,862 95.09% 119,931 6 3 80.90% 373,910 96.35% 124,637 7 3 84.27% 216,804 97.08% 72,268 8 1 85.39% 185,312 97.70% 185,312 9 1 86.52% 0 97.70% 0 10 1 87.64% 40,396 97.84% 40,396 20 1 88.76% 7,476 97.87% 7,476 Above 4 93.26% 541,998 99.69% 135,500 Losses 6 100.00% 91,177 100.00% 15,196 Total 89 29,666,345 333,330 MCR Shortfall: 99% (measured in years of profit) Years of Percentage of Total Average Net profit Frequency Percentage of Total Total Net Premium (Net Premium) Premium 0 43 48.31% 18,125,783 61.10% 421,530 1 3 51.69% 290,710 62.08% 96,903 2 6 58.43% 5,099,355 79.27% 849,893 3 2 60.67% 768,144 81.86% 384,072 4 5 66.29% 1,551,247 87.09% 310,249 5 5 71.91% 1,226,029 91.22% 245,206 6 1 73.03% 43,919 91.37% 43,919 7 2 75.28% 239,862 92.18% 119,931 8 0 75.28% 0 92.18% 0 9 6 82.02% 683,189 94.48% 113,865 10 0 82.02% 0 94.48% 0 20 6 88.76% 1,004,932 97.87% 167,489 Above 4 93.26% 541,998 99.69% 135,500 Losses 6 100.00% 91,177 100.00% 15,196 Total 89 29,666,345 333,330 81

MCR Shortfall: 99.5% (measured in years of profit) Years of Percentage of Total Average Net profit Frequency Percentage of Total Total Net Premium (Net Premium) Premium 0 37 41.57% 14,775,521 49.81% 399,338 1 6 48.31% 3,350,263 61.10% 558,377 2 4 52.81% 2,189,979 68.48% 547,495 3 5 58.43% 3,200,086 79.27% 640,017 4 1 59.55% 0 79.27% 0 5 3 62.92% 1,029,259 82.74% 343,086 6 4 67.42% 2,020,665 89.55% 505,166 7 3 70.79% 485,593 91.19% 161,864 8 2 73.03% 143,145 91.67% 71,573 9 2 75.28% 140,636 92.14% 70,318 10 1 76.40% 9,903 92.18% 9,903 20 9 86.52% 958,136 95.41% 106,460 Above 6 93.26% 1,271,983 99.69% 211,997 Losses 6 100.00% 91,177 100.00% 15,196 Total 89 29,666,345 333,330 On the basis of the above tables, we can see that 87% of the market (weighted by net premium) would be in a position to build up capital to the MCR at the 99% level of sufficiency within 4 years, if they were to use 2004 level of profits every year. Of this 85%, companies representing 61.10% of the market by size of net premium would immediately meet the 99 th percentile MCR. We therefore recommend that the 98% level of sufficiency should apply for 4 years, and that it would be required that companies reach the 99% level at the end of the 4 years. Note that this analysis does not take into account the fact that most of the companies not meeting the requirement may apply for certified or internal models, and that there may well be a substantially lower percentage of them not meeting the 98% or 99% MCR as a result. Using the same reasoning, we suggest that the 99.5 th percentile capital requirement be implemented after 6 years. A further element of the regulatory framework is the point at which the FSB would intervene. In other words, it would be useful for the FSB to have some point where, if a company meets the MCR, but their capital is just above the MCR, the FSB would want to take some action to avoid the company from falling below the minimum requirement. Given that the industry calibration is performed at a prudent level, we believe that the intervention point should be a fairly small multiple of the MCR. For instance, we suggest that, if a company has capital of more than 1.1 times the MCR, the FSB need not take any regulatory action. However, if the capital falls to between 1 and 1.1 times the MCR, the FSB should take some action, such as requesting a business plan from the company to discuss how it will manage its capital. Only when the capital falls below the MCR the FSB would intervene more directly. The intervention points would have to be specified during the transitional period, and every company falling short of the minimum requirement would have to discuss with the FSB: Whether they will use an internal model or certified model to determine different capital requirements Whether the company would apply for a special dispensation justifying why it would not meet any of the capital requirements determined on the above basis How the company would reach the level of capital agreed with the FSB Appendix I contains an example of company specific feedback, showing details of the application of the model to an individual hypothetical company. 82

9 Recommendations 9.1 Recommendations In this section we do not repeat the detailed description of the DFA model behind the industry calibration. Our main recommendation is that the industry calibration should be done in accordance with the methodology, assumptions and parameterisation contained in the rest of this report. Given this main recommendation, it is worth highlighting a number of specific recommendations. We do not discuss the reasoning behind these recommendations as this is already contained in detail in the rest of the report: STAR returns should be expanded to collect more data on reinsurance, and particularly on nonproportional reinsurance, cell captives and co-insurance. Such reinsurance data may be used to refine the calibration of the model in future. We recommend that no explicit allowance be made for the underwriting cycle. We recommend that no explicit capital requirement be determined to cover operational risk. Specific attention should be given to the development of certified model frameworks for cell captives, reinsurers, niche insurers and companies in run-off. These certified model frameworks should provide guidance on how elements of the industry calibration may be used in practice and the standards that should be applied by companies when modifying elements for their specific requirements. We recommend that the industry calibration would not be applied blindly to companies in the above sectors, and that further consultation is appropriate with these specific sectors of the market. Despite the fact that there may be shortcomings in the industry calibration for companies in specific sectors of the market, we recommend that the FSB use the proposed framework as a benchmark. A framework for internal models has to be established by the FSB. In determining the total capital charge using the square root formula, we recommend that grossing up on the Insurance Capital Charge be limited to 50%, to avoid the overall requirement being too onerous. We recommend that the FSB implements the capital requirement in such a way that it would be at the 98% sufficiency level on implementation, 99% after four years and 99.5% after another two years. Until such time as the implementation of the capital requirement is at the 99.5% level, the FSB should collect additional data in STAR returns to ensure that any necessary refinements can be made at a later stage if deemed appropriate (in particular more detailed allowance for nonproportional reinsurance may be required). The FSB should determine and specify its intervention points. For instance, we suggest that, if a company has capital of more than 1.1 times the MCR, the FSB need not take any regulatory action. However, if the capital falls to between 1 and 1.1 times the MCR, the FSB should take some action, such as requesting a business plan from the company to discuss how it will manage its capital. Only when the capital falls below the MCR the FSB would intervene more directly. The intervention points would have to be specified separately and specifically during the transitional period. 83

Every company falling short of the minimum requirement would have to discuss with the FSB: Whether they will use an internal model or certified model to determine different capital requirements Whether the company would apply for a special dispensation justifying why it would not meet any of the capital requirements determined on the above basis. It is worth emphasising that the FSB has indicated that it would be willing to consider and review applications for special dispensations from companies who feel that their individual circumstances warrant such dispensations. How the company would reach the level of capital agreed with the FSB The industry calibration contains complex calculations and formulae. To make it easier for companies to apply in practice, we recommend that the FSB issues a spreadsheet-based model to companies that would automate the calculation of the MCR, and that this become part of the STAR returns. Finally, this report should be regarded as a step towards the specification of a Financial Condition Reporting framework for South African Short Term Insurers. The FSB now has to decide on how to implement recommendations in this report, and we recommend that there would still be considerable discussion with the industry on the way in which the model would be implemented. Emile Stipp BBusSc LLB FIA FASSA In my capacity as an employee of Deloitte & Touche Hillary Murashiki BSc FIA FASSA In my capacity as an employee of Insight ABC Sam Isaacson BSc(Hons) FIA FASSA In my capacity as an employee of Deloitte & Touche 84

10 Acknowledgements The following individuals were responsible for the design, development and calibration of the proposed framework: Team Actuarial and Insurance Solutions at Deloitte Insight ABC FSB Name Emile Stipp Sam Isaacson Jaco van der Merwe Deshni Subbiah Hillary Murashiki David Zhong Hantie van Heerden André Janse van Vuuren 85

11 References Year Paper Published by Insurance 2004 New Solvency Assessment for Short-Term insurers regulated under the short-term insurance act 53 of 1998 1998 Short-term insurance act SA 2003 Calibration of the General Insurance risk based capital model (prepared for the FSA) 2003 CP190: Enhanced capital requirements and individual capital assessments for non-life insurers 2003 Impact of the new non-life prudential capital regime on the market (prepared for the Association of British Insurers) 2005 Prudential Standard GPS 110: Capital Adequacy for General Insurers 2001 Research and data analysis relevant to the development of standards and guidelines on liability valuation for general insurance (prepared for the Institute of Actuaries of Australia) FSB Watson Wyatt FSA Deloitte APRA 1999 Overview of Dynamic Financial Analysis CAS 1999 Dynamic Financial Analysis models of property-casualty insurers CAS 2000 Dynamic Financial Analysis: Performance and risk measures, using model results Tillinghast-Towers Perrin CAS 2000 Dynamic Financial Analysis: Strategies CAS 2003 Dynamic Financial Analysis: making use of your simulations (presented to the Society of Actuaries, Ireland) 2004 Quantifying Operational Risk in General Insurance Companies (by GIRO Working Party). Authors: Tripp, HM; Bradley, HL; Devitt, R; Orros, GC; Overton, GL; Pryor, LM and Shaw, RZ) Tony Brooke-Taylor, Deloitte Institute of Actuaries, London 2004 White Paper of the Swiss Solvency Test Swiss Federal Office of Private Insurance 2005 Solvency assessment models compared CEA and Mercer Oliver Wyman Investment 1995 Stochastic Asset Models General Insurance Convention 1996 How Actuaries can use Financial Economics Institute of Actuaries 1997 Global Returns on Assets with Fat Tails Investment Convention 1998 Gauge Transforms in Stochastic Models AFIR 1999 Information Structures Investment Convention 86

Year Paper Published by 2000 Incomplete Bond Pricing Models FORC 2000 Fitting Yield Curves with Long Constraints FORC 2000 Das InvestmentModell TSM DAV 2001 Consistent Multinational Assumptions Investment Convention 2001 Asset Model for Fair Values (Timbuk 1) Life Convention 2001 Levy Processes for Investment Models Oxford Levy Seminar 2002 Corporate Bond Models Investment Convention 2003 Option Pricing with Deflators Risk Publications 2003 The Importance of Being Normal ASTIN 2005 PGN104: Valuation of Long-term insurers ASSA 87

12 Appendices 88

12.1 Appendix A: ULR s vs Gross Earned Premium for Different Classes of Business 89

ULR distribution - Accident 2.60 2.40 2.20 ULR 2.00 1.80 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00-50,000 100,000 150,000 200,000 250,000 300,000 350,000 Gross earned premium (R'000) ULR distribution - Engineering 2.40 2.20 2.00 ULR 1.80 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00-50,000 100,000 150,000 200,000 250,000 300,000 350,000 400,000 Gross earned premium (R'000) 90

ULR distribution - Guarantee 3.00 2.80 2.60 ULR 2.40 2.20 2.00 1.80 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00-200,000 400,000 600,000 800,000 1,000,000 1,200,000 Gross earned premium (R'000) ULR distribution - Liability 2.40 2.20 2.00 ULR 1.80 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00-100,000 200,000 300,000 400,000 500,000 600,000 700,000 Gross earned premium (R'000) 91

ULR distribution - Miscellaneous 4.80 4.60 4.40 4.20 4.00 3.80 3.60 3.40 3.20 3.00 2.80 2.60 2.40 2.20 2.00 1.80 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00-200,000 400,000 600,000 800,000 1,000,000 1,200,000 1,400,000 Gross earned premium (R'000) ULR ULR distribution - Motor 3.00 2.80 2.60 ULR 2.40 2.20 2.00 1.80 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00-500,000 1,000,000 1,500,000 2,000,000 2,500,000 3,000,000 3,500,000 4,000,000 Gross earned premium (R'000) 92

ULR distribution - Property 4.20 4.00 3.80 ULR 3.60 3.40 3.20 3.00 2.80 2.60 2.40 2.20 2.00 1.80 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00-500,000 1,000,000 1,500,000 2,000,000 2,500,000 3,000,000 Gross earned premium (R'000) ULR distribution - Transport 2.60 2.40 2.20 ULR 2.00 1.80 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00-50,000 100,000 150,000 200,000 250,000 Gross earned premium (R'000) 93

12.2 Appendix B: Means and Standard Deviations of ULR s ULR fitting - Accident 2.60 2.40 2.20 2.00 ULR mu+sd Fitted mu-sd 1.80 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00-50,000 100,000 150,000 200,000 250,000 300,000 350,000 Gross earned premium (R'000) ULR fitting - Engineering 2.40 2.20 2.00 ULR mu+sd Fitted mu-sd 1.80 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00-50,000 100,000 150,000 200,000 250,000 300,000 350,000 400,000 Gross earned premium (R'000) 94

ULR fitting - Guarantee 3.00 2.80 2.60 2.40 2.20 2.00 1.80 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 ULR mu+sd Fitted mu-sd 0.00-200,000 400,000 600,000 800,000 1,000,000 1,200,000 Gross earned premium (R'000) ULR fitting - Liability 2.40 2.20 2.00 ULR mu+sd Fitted mu-sd 1.80 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00-100,000 200,000 300,000 400,000 500,000 600,000 700,000 Gross earned premium (R'000) 95

ULR fitting - Miscellaneous 6.40 6.20 6.00 5.80 5.60 5.40 5.20 5.00 4.80 4.60 4.40 4.20 4.00 3.80 3.60 3.40 3.20 3.00 2.80 2.60 2.40 2.20 2.00 1.80 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00-200,000 400,000 600,000 800,000 1,000,000 1,200,000 1,400,000 Gross earned premium (R'000) ULR mu+sd Fitted mu-sd ULR fitting - Motor 3.60 3.40 ULR Fitted 3.20 3.00 mu+sd mu-sd 2.80 2.60 2.40 2.20 2.00 1.80 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00-500,000 1,000,000 1,500,000 2,000,000 2,500,000 3,000,000 3,500,000 4,000,000 Gross earned premium (R'000) 96

ULR fitting - Property 4.20 4.00 3.80 ULR Fitted 3.60 mu+sd mu-sd 3.40 3.20 3.00 2.80 2.60 2.40 2.20 2.00 1.80 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00-500,000 1,000,000 1,500,000 2,000,000 2,500,000 3,000,000 Gross earned premium (R'000) ULR fitting - Transport 2.60 2.40 2.20 2.00 ULR mu+sd Fitted mu-sd 1.80 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00-50,000 100,000 150,000 200,000 250,000 Gross earned premium (R'000) 97

12.3 Appendix C: Reserving Ratios per Class of Business Reserving ratio distribution - Accident 10% 9% 8% 7% 6% 5% 4% 3% 2% 1% 0% 0 0.3 0.6 0.9 1.2 1.5 1.8 2.1 2.4 2.7 3 3.3 3.6 3.9 4.2 4.5 4.8 5.1 5.4 5.7 6 6.3 6.6 6.9 7.2 7.5 7.8 8.1 8.4 8.7 9 9.3 9.6 9.9 Reserving risk Reserving ratio distribution - Engineering 12% 10% 8% 6% 4% 2% 0% 0 0.3 0.6 0.9 1.2 1.5 1.8 2.1 2.4 2.7 3 3.3 3.6 3.9 4.2 4.5 4.8 5.1 5.4 5.7 6 6.3 6.6 6.9 7.2 7.5 7.8 8.1 8.4 8.7 9 9.3 9.6 9.9 Reserving risk 98

Reserving ratio distribution - Guarantee 9% 8% 7% 6% 5% 4% 3% 2% 1% 0% 0 0.3 0.6 0.9 1.2 1.5 1.8 2.1 2.4 2.7 3 3.3 3.6 3.9 4.2 4.5 4.8 5.1 5.4 5.7 6 6.3 6.6 6.9 7.2 7.5 7.8 8.1 8.4 8.7 9 9.3 9.6 9.9 Reserving risk Reserving ratio distribution - Liability 14% 12% 10% 8% 6% 4% 2% 0% 0 0.3 0.6 0.9 1.2 1.5 1.8 2.1 2.4 2.7 3 3.3 3.6 3.9 4.2 4.5 4.8 5.1 5.4 5.7 6 6.3 6.6 6.9 7.2 7.5 7.8 8.1 8.4 8.7 9 9.3 9.6 9.9 Reserving risk 99

Reserving ratio distribution - Miscellaneous 14% 12% 10% 8% 6% 4% 2% 0% 0 0.3 0.6 0.9 1.2 1.5 1.8 2.1 2.4 2.7 3 3.3 3.6 3.9 4.2 4.5 4.8 5.1 5.4 5.7 6 6.3 6.6 6.9 7.2 7.5 7.8 8.1 8.4 8.7 9 9.3 9.6 9.9 Reserving risk Reserving ratio distribution - Motor 14% 12% 10% 8% 6% 4% 2% 0% 0 0.3 0.6 0.9 1.2 1.5 1.8 2.1 2.4 2.7 3 3.3 3.6 3.9 4.2 4.5 4.8 5.1 5.4 5.7 6 6.3 6.6 6.9 7.2 7.5 7.8 8.1 8.4 8.7 9 9.3 9.6 9.9 Reserving risk 100

Reserving ratio distribution - Property 10% 9% 8% 7% 6% 5% 4% 3% 2% 1% 0% 0 0.3 0.6 0.9 1.2 1.5 1.8 2.1 2.4 2.7 3 3.3 3.6 3.9 4.2 4.5 4.8 5.1 5.4 5.7 6 6.3 6.6 6.9 7.2 7.5 7.8 8.1 8.4 8.7 9 9.3 9.6 9.9 Reserving risk Reserving ratio distribution - Transport 12% 10% 8% 6% 4% 2% 0% 0 0.3 0.6 0.9 1.2 1.5 1.8 2.1 2.4 2.7 3 3.3 3.6 3.9 4.2 4.5 4.8 5.1 5.4 5.7 6 6.3 6.6 6.9 7.2 7.5 7.8 8.1 8.4 8.7 9 9.3 9.6 9.9 Reserving risk 4 101

12.4 Appendix D: CV of Reserving Ratio vs Gross Reserves Reserving ratio distribution - Accident 20.0 18.0 16.0 Reserving risk Coefficient of variation 14.0 12.0 10.0 8.0 6.0 4.0 2.0 0.0-10,000 20,000 30,000 40,000 50,000 60,000 70,000 80,000 Gross claims reserve (R'000) 102

Reserving ratio distribution - Engineering 10.0 9.0 8.0 Reserving risk Coefficient of variation 7.0 6.0 5.0 4.0 3.0 2.0 1.0 0.0-20,000 40,000 60,000 80,000 100,000 120,000 140,000 160,000 180,000 Gross claims reserve (R'000) Reserving ratio distribution - Guarantee 12.0 Reserving risk 10.0 Coefficient of variation 8.0 6.0 4.0 2.0 0.0-50,000 100,000 150,000 200,000 250,000 Gross claims reserve (R'000) 103

Reserving ratio distribution - Liability 12.0 Reserving risk 10.0 Coefficient of variation 8.0 6.0 4.0 2.0 0.0-20,000 40,000 60,000 80,000 100,000 120,000 Gross claims reserve (R'000) Reserving ratio distribution - Miscellaneous 18.0 16.0 14.0 Reserving risk Coefficient of variation 12.0 10.0 8.0 6.0 4.0 2.0 0.0-50,000 100,000 150,000 200,000 250,000 300,000 350,000 400,000 Gross claims reserve (R'000) 104

Reserving ratio distribution - Motor 25.0 Reserving risk 20.0 Coefficient of variation 15.0 10.0 5.0 0.0-100 200 300 400 500 600 Gross claims reserve (R'000) Reserving ratio distribution - Property 20.0 18.0 16.0 Reserving risk Coefficient of variation 14.0 12.0 10.0 8.0 6.0 4.0 2.0 0.0-100,000 200,000 300,000 400,000 500,000 600,000 700,000 Gross claims reserve (R'000) 105

Reserving ratio distribution - Transport 14.0 12.0 Reserving risk Coefficient of variation 10.0 8.0 6.0 4.0 2.0 0.0-10,000 20,000 30,000 40,000 50,000 60,000 70,000 80,000 90,000 100,000 Gross claims reserve (R'000) 106

12.5 Appendix E: Gross Stand-Alone Capital Simulation Results Gross stand-alone capital at the 99.5% level: GWP level (R'mil) GWP level (R'mil) GWP level (R'mil) GWP level (R'mil) Accident UPR level (R'mil) 0 0.3 1 2 4 11 29 75 193 500 0-0.41 1.15 3.24 8.85 21.22 53.18 123.95 289.17 685.46 1 1.47 2.06 2.94 5.18 10.89 23.12 55.02 125.61 290.67 686.83 2 2.79 3.42 4.24 6.11 11.71 25.39 55.66 126.27 299.24 692.21 4 5.29 5.97 6.70 8.75 13.65 26.68 56.58 127.79 298.09 689.34 8 10.02 10.68 11.17 13.06 18.76 30.18 60.84 132.32 296.87 709.62 16 18.91 19.39 19.85 22.02 27.06 39.26 68.49 137.70 293.23 694.20 32 35.61 36.75 37.64 39.64 43.74 53.09 86.06 147.23 314.06 703.18 63 66.84 68.59 68.84 69.70 71.23 85.28 113.19 185.28 343.07 750.23 126 125.06 123.14 126.28 124.17 128.46 139.13 171.19 233.83 405.56 763.03 251 233.20 222.14 233.19 232.09 236.59 249.78 273.46 333.81 488.81 858.66 501 433.27 440.87 435.24 419.42 424.68 448.63 481.89 517.53 679.56 1,048.83 1000 801.79 786.05 809.99 801.37 785.72 817.22 834.79 916.70 1,051.89 1,359.66 Engineering Guarantee UPR level (R'mil) 0 0.3 1 2 4 11 29 75 193 500 0-0.21 0.67 2.22 6.54 17.72 44.24 103.43 248.91 582.26 1 0.75 1.06 1.71 3.54 8.05 19.30 45.77 104.82 250.20 583.42 2 1.44 1.70 2.21 3.80 8.33 19.32 45.48 106.32 247.55 576.45 4 2.74 3.01 3.34 4.69 8.77 19.23 44.80 105.20 241.10 575.57 8 5.16 5.38 5.58 6.71 10.07 20.16 44.53 105.80 248.05 584.89 16 9.61 9.93 10.03 10.65 13.40 21.96 46.38 108.09 247.69 572.24 32 17.67 17.68 17.43 18.25 20.90 27.18 48.21 106.22 254.14 567.03 63 32.01 31.47 33.08 32.24 35.16 39.60 57.40 109.46 248.92 563.28 126 56.88 57.99 56.61 57.15 57.84 62.42 77.72 120.83 246.81 564.85 251 98.61 97.65 98.41 97.58 100.88 100.60 110.99 141.29 250.39 553.40 501 165.53 160.28 164.03 167.91 163.67 171.81 172.91 198.64 273.09 530.61 1000 265.76 257.16 266.62 267.27 264.46 260.32 274.30 292.64 346.62 526.48 UPR level (R'mil) 0 0.3 1 2 7 19 57 170 506 1500 0-1.05 3.14 9.11 23.16 57.55 125.85 275.67 553.39 1,108.26 1 4.10 5.24 7.37 13.26 26.71 60.52 128.04 277.29 554.48 1,109.00 2 7.58 8.62 10.37 15.80 28.68 59.31 131.77 271.24 559.48 1,145.20 5 13.84 14.39 16.07 20.58 32.75 62.79 132.87 269.96 579.44 1,113.19 10 24.97 25.71 26.67 29.88 41.53 69.50 134.63 278.91 580.29 1,142.13 21 44.50 45.97 44.65 50.34 58.18 84.23 147.34 282.15 580.49 1,164.66 45 78.41 75.22 78.52 81.60 90.29 111.28 168.13 305.55 600.52 1,142.83 96 136.67 137.41 138.16 141.38 147.47 167.12 210.82 345.47 632.89 1,156.26 205 235.62 231.08 233.10 228.47 255.71 261.31 298.47 410.51 656.52 1,240.85 437 401.58 404.22 411.05 405.82 397.21 426.42 450.82 555.57 779.99 1,257.84 935 675.62 687.17 674.44 655.91 695.05 661.49 729.07 776.72 950.85 1,449.66 2000 1,118.76 1,108.28 1,132.29 1,121.56 1,155.08 1,116.25 1,113.18 1,202.07 1,364.79 1,689.82 Liability UPR level (R'mil) 0 0.3 1 2 6 16 45 126 355 1000 0-0.79 2.23 6.39 17.55 42.21 104.52 235.90 563.23 1,320.31 1 2.97 3.93 5.39 9.61 20.68 44.88 106.86 237.77 564.82 1,321.63 2 5.66 6.56 7.90 11.97 21.52 46.57 104.08 241.17 568.25 1,293.20 5 10.69 11.33 12.97 16.66 25.81 49.72 108.13 245.16 560.11 1,320.00 10 20.09 20.12 22.09 24.48 34.47 56.41 112.48 248.76 559.97 1,295.68 21 37.50 37.93 39.75 43.35 50.40 70.74 124.04 262.24 559.97 1,321.76 45 69.59 69.91 71.84 72.86 81.13 96.86 148.14 273.93 587.71 1,318.96 96 128.38 129.57 131.07 130.27 137.35 153.90 201.25 314.83 612.43 1,371.85 205 235.49 234.57 242.08 238.68 248.02 250.36 297.27 404.10 699.29 1,416.55 437 429.40 425.26 428.46 440.45 439.36 452.44 488.35 569.19 849.19 1,510.24 935 778.09 792.11 786.89 782.23 778.95 784.01 821.34 905.26 1,133.84 1,689.85 2000 1,400.32 1,407.00 1,380.50 1,425.05 1,412.17 1,375.43 1,455.65 1,502.31 1,686.36 2,250.78 107

Miscellaneous UPR level (R'mil) 0 0.3 1 2 6 16 45 126 355 1000 0-3.24 9.39 23.72 55.45 122.08 243.11 479.98 896.94 1,610.35 1 11.78 16.21 22.70 35.65 65.34 129.80 248.57 483.80 899.47 1,611.96 2 20.66 23.53 30.16 43.74 71.81 132.77 246.60 487.94 891.18 1,588.80 5 35.72 36.80 42.24 53.39 80.28 138.02 258.53 476.72 920.10 1,568.23 10 60.86 64.09 67.66 77.49 97.94 159.77 270.70 493.85 902.37 1,579.79 21 102.35 107.85 107.98 112.14 133.15 178.44 283.43 518.05 925.77 1,572.75 45 169.98 171.68 181.03 179.98 190.86 241.25 322.15 541.96 934.19 1,631.47 96 278.91 283.80 283.96 287.26 296.37 335.26 407.01 586.73 961.42 1,672.46 205 451.79 462.91 460.27 460.99 449.03 493.99 565.20 728.83 1,065.61 1,752.43 437 720.43 729.08 735.85 715.28 729.22 769.65 792.79 911.84 1,211.08 1,813.50 935 1,124.26 1,104.06 1,129.07 1,122.37 1,142.20 1,187.34 1,188.55 1,289.40 1,493.27 1,931.35 2000 1,697.75 1,732.50 1,734.20 1,690.78 1,680.87 1,750.83 1,800.22 1,798.04 1,914.23 2,355.56 GWP level (R'mil) GWP level (R'mil) GWP level (R'mil) GWP level (R'mil) Motor UPR level (R'mil) 0 0.3 1 2 5 14 37 101 276 750 0-1.52 3.97 10.09 22.86 48.95 95.69 187.14 360.91 633.90 1 6.26 7.61 9.82 15.55 27.40 52.53 98.26 188.99 362.21 634.75 2 11.64 12.68 15.22 19.26 30.14 54.40 103.12 194.28 363.99 648.31 5 21.45 21.99 24.35 27.93 38.19 59.32 104.20 197.88 356.42 640.52 13 39.04 39.23 40.70 45.00 52.20 72.88 114.36 203.17 377.98 678.38 30 70.04 70.29 76.00 73.47 80.07 94.45 131.90 216.27 373.41 637.86 71 123.68 122.28 120.41 123.52 128.08 144.50 176.72 251.52 397.89 694.58 166 214.58 213.32 216.50 212.29 219.87 231.99 255.50 308.81 450.92 694.46 388 364.82 367.81 376.26 366.06 369.69 371.99 389.95 432.83 535.55 783.69 910 604.63 604.28 607.12 610.56 604.29 606.75 634.35 648.23 730.42 924.19 2133 966.68 976.12 962.56 966.32 986.68 975.16 966.98 997.87 1,067.71 1,201.17 5000 1,458.45 1,469.39 1,432.66 1,451.41 1,435.38 1,451.28 1,513.22 1,512.74 1,485.42 1,598.72 Property UPR level (R'mil) 0 0.3 1 2 6 16 45 126 355 1000 0-1.99 5.72 15.42 36.35 76.72 156.53 301.16 543.77 958.43 1 7.74 9.94 13.84 23.17 42.83 81.57 160.04 303.56 545.31 959.39 2 14.48 16.37 19.08 28.12 45.93 83.04 158.30 302.76 545.10 947.08 5 26.55 27.61 31.13 36.22 53.48 92.16 159.17 298.71 543.43 969.36 12 47.67 50.10 51.88 55.71 67.96 102.08 167.52 309.32 561.07 944.19 28 83.88 84.19 87.38 87.69 101.78 128.19 195.59 317.06 561.98 990.52 63 144.72 145.12 147.42 151.21 156.79 180.44 234.52 350.05 580.82 941.87 145 244.73 242.76 244.27 248.57 252.18 271.94 317.18 414.01 619.73 970.40 332 404.56 420.46 407.24 408.06 412.76 426.87 454.48 532.16 710.77 1,039.47 761 649.06 660.50 632.53 661.17 633.61 665.47 685.99 731.75 867.33 1,127.36 1745 994.53 1,000.74 1,006.78 997.90 1,021.65 1,009.73 1,018.27 1,068.17 1,122.17 1,377.90 4000 1,402.88 1,413.74 1,461.64 1,434.28 1,426.67 1,383.76 1,425.76 1,436.96 1,486.92 1,597.29 Transport UPR level (R'mil) 0 0.3 1 2 4 11 29 75 193 500 0-0.73 1.98 5.43 12.66 27.60 55.10 105.50 203.39 371.15 1 2.71 3.63 5.04 8.68 15.59 30.07 57.00 106.91 204.45 371.89 2 4.65 5.59 6.59 9.58 16.62 29.33 56.08 109.92 207.75 377.02 4 7.96 8.46 9.75 12.29 18.39 32.00 56.98 107.44 203.75 361.33 8 13.55 14.01 14.74 17.08 22.70 34.56 60.24 111.30 207.30 377.87 16 22.89 23.51 23.06 25.46 30.19 41.68 67.24 112.30 207.44 367.01 32 38.38 38.58 39.36 39.85 44.53 54.31 75.51 122.69 219.08 377.40 63 63.77 64.52 62.44 63.51 66.79 75.01 97.71 139.26 227.79 388.36 126 104.76 102.80 103.03 104.88 109.36 112.62 132.59 171.96 243.88 403.48 251 169.58 164.66 167.32 170.92 172.14 170.18 189.35 220.08 294.40 446.42 501 269.13 263.14 259.87 274.91 267.68 278.34 281.29 309.77 357.06 482.17 1000 415.30 418.25 415.90 412.54 416.21 426.57 429.22 442.30 481.84 590.19 108

Gross stand-alone capital at the 99% level: GWP level (R'mil) GWP level (R'mil) GWP level (R'mil) GWP level (R'mil) Accident UPR level (R'mil) 0 0.3 1 2 4 11 29 75 193 500 0-0.30 0.84 2.42 6.61 15.71 38.20 90.08 211.52 490.05 1 1.05 1.51 2.14 3.87 8.14 17.11 39.53 91.29 212.61 491.03 2 2.00 2.44 3.12 4.57 8.68 18.29 40.28 91.69 221.36 496.56 4 3.80 4.19 4.85 6.29 9.96 19.42 41.43 92.23 218.64 502.81 8 7.21 7.70 8.11 9.52 13.52 22.05 42.84 95.08 214.65 507.59 16 13.63 14.02 14.68 15.94 19.36 28.21 49.21 100.50 216.26 506.42 32 25.71 26.55 26.98 28.25 31.09 38.75 61.27 108.15 229.72 510.81 63 48.32 48.29 50.06 50.61 51.42 61.66 82.46 128.36 250.65 532.07 126 90.51 89.38 90.85 89.16 94.35 100.06 121.83 167.44 291.68 554.85 251 168.89 160.80 168.87 171.62 171.19 183.83 201.60 242.55 361.99 609.73 501 313.82 329.21 318.20 307.72 313.95 328.03 347.17 366.37 497.26 755.20 1000 580.50 565.60 584.42 579.01 567.94 579.53 609.46 663.40 763.54 953.97 Engineering Guarantee UPR level (R'mil) 0 0.3 1 2 4 11 29 75 193 500 0-0.16 0.52 1.75 5.20 14.08 34.71 82.93 197.70 456.38 1 0.57 0.82 1.33 2.80 6.41 15.34 35.91 84.04 198.72 457.29 2 1.10 1.28 1.70 3.03 6.67 15.43 35.75 83.99 197.80 459.96 4 2.08 2.27 2.57 3.68 6.97 15.54 35.83 83.08 192.67 452.93 8 3.91 4.02 4.25 5.21 7.96 16.25 35.49 83.38 194.91 468.58 16 7.28 7.33 7.64 8.19 10.49 17.09 36.00 84.79 195.19 452.22 32 13.35 12.85 13.31 13.73 15.62 20.86 38.20 84.97 203.09 444.54 63 24.06 23.68 24.41 24.40 26.65 30.17 43.43 86.24 193.18 438.43 126 42.44 43.23 43.12 42.32 43.56 47.54 57.82 93.58 191.92 440.24 251 72.76 70.53 72.34 71.76 71.69 74.57 82.40 105.36 192.09 435.10 501 119.95 119.09 118.94 123.21 121.21 124.56 124.52 145.72 204.42 404.83 1000 186.79 184.17 183.27 186.34 189.93 187.47 187.69 205.58 242.30 395.88 UPR level (R'mil) 0 0.3 1 2 7 19 57 170 506 1500 0-0.74 2.38 6.91 17.86 44.43 98.08 214.89 432.99 833.01 1 2.86 3.68 5.60 10.05 20.60 46.73 99.79 216.15 433.85 833.56 2 5.38 6.06 7.43 11.73 22.00 46.66 101.57 214.72 440.77 862.00 5 9.98 10.64 11.34 15.54 24.94 49.33 106.32 212.32 439.56 834.80 10 18.28 18.91 19.29 22.41 31.65 53.30 107.52 218.44 447.77 866.07 21 33.01 34.37 33.58 37.23 44.83 64.23 114.37 223.98 446.86 880.99 45 58.85 57.46 59.49 61.56 67.20 86.23 130.73 241.14 464.80 878.10 96 103.52 104.17 103.82 107.41 111.69 126.42 164.60 260.00 482.18 889.46 205 179.58 178.54 181.76 174.55 190.73 196.11 229.83 315.47 506.24 920.93 437 306.86 306.17 309.06 306.63 308.75 324.29 347.50 416.54 584.95 960.18 935 515.21 520.41 514.04 501.05 520.85 514.27 541.54 604.07 721.73 1,099.48 2000 846.07 843.59 853.85 842.96 856.41 845.64 846.66 906.91 1,005.09 1,242.80 Liability UPR level (R'mil) 0 0.3 1 2 6 16 45 126 355 1000 0-0.56 1.62 4.76 13.09 32.20 81.38 187.09 449.86 1,045.63 1 2.12 2.78 3.92 7.15 15.43 34.24 83.21 188.58 451.13 1,046.67 2 4.08 4.74 5.84 8.81 15.90 36.06 80.66 189.43 451.03 1,026.69 5 7.80 8.13 9.41 12.20 18.99 38.24 84.55 192.81 443.22 1,058.04 10 14.80 14.87 16.66 18.19 25.67 42.62 87.23 193.29 441.36 1,039.28 21 27.89 28.38 28.96 32.40 37.67 54.45 96.08 203.30 444.04 1,060.51 45 52.17 52.79 53.71 54.26 61.32 74.62 114.84 212.12 465.96 1,055.50 96 96.92 97.94 97.38 99.39 104.46 116.15 152.25 247.14 481.12 1,079.25 205 178.72 173.21 181.23 178.94 187.21 193.71 230.38 309.88 545.18 1,118.21 437 327.06 325.86 328.70 325.96 333.00 340.74 366.41 435.26 661.63 1,180.23 935 593.60 594.42 597.73 596.25 597.08 600.34 620.02 686.72 864.33 1,335.59 2000 1,067.40 1,080.56 1,080.24 1,086.28 1,083.85 1,069.76 1,110.27 1,154.40 1,294.39 1,727.52 109

Miscellaneous UPR level (R'mil) 0 0.3 1 2 6 16 45 126 355 1000 0-2.27 6.67 18.03 42.08 92.27 191.17 369.29 691.20 1,201.26 1 8.18 11.33 16.12 27.10 49.59 98.11 195.46 372.22 693.15 1,202.46 2 14.67 16.72 21.43 31.99 53.36 101.54 192.40 375.63 702.81 1,211.98 5 25.89 26.93 31.43 39.57 61.01 105.51 201.12 376.28 702.79 1,196.37 10 44.98 47.36 49.90 57.06 74.70 122.77 212.62 388.77 701.89 1,194.89 21 76.95 80.83 81.30 85.55 101.59 139.54 222.54 398.81 718.37 1,188.80 45 129.69 131.54 136.37 134.14 147.40 185.01 248.91 421.04 720.98 1,230.39 96 215.18 220.48 217.59 223.04 232.32 261.54 317.26 465.94 735.82 1,251.80 205 350.80 353.34 355.69 356.92 353.91 388.09 436.73 558.94 826.86 1,273.33 437 559.36 562.75 560.82 556.39 570.08 588.78 622.97 714.81 940.63 1,349.27 935 864.41 874.36 866.21 866.71 891.09 893.16 906.46 981.96 1,119.99 1,440.79 2000 1,271.28 1,285.93 1,295.07 1,271.83 1,263.57 1,302.33 1,328.06 1,343.18 1,432.43 1,656.33 GWP level (R'mil) GWP level (R'mil) GWP level (R'mil) GWP level (R'mil) Motor UPR level (R'mil) 0 0.3 1 2 5 14 37 101 276 750 0-1.17 3.08 7.78 17.98 38.27 76.44 150.78 287.16 502.79 1 4.73 5.86 7.62 11.99 21.56 41.07 78.50 152.27 288.20 503.46 2 8.86 9.56 11.56 14.97 23.49 43.32 81.12 155.09 290.31 511.14 5 16.47 17.13 18.57 21.47 29.68 46.11 82.84 156.70 289.29 513.87 13 30.27 30.35 32.09 34.55 40.35 55.95 91.48 159.53 297.87 527.42 30 54.83 54.69 57.29 56.97 61.94 75.45 104.14 170.77 298.50 517.42 71 97.62 96.66 96.86 99.23 102.83 113.05 140.18 199.35 313.64 542.17 166 170.32 168.83 171.03 170.89 175.00 181.25 203.77 245.80 361.89 547.91 388 289.93 288.98 291.43 289.82 294.28 297.28 311.84 347.77 430.72 608.97 910 477.76 474.16 480.56 479.80 482.13 482.94 496.09 515.80 569.08 724.92 2133 750.50 758.90 757.24 762.71 769.13 761.21 745.86 769.63 808.42 906.18 5000 1,086.77 1,106.31 1,081.50 1,100.53 1,088.95 1,076.65 1,106.67 1,086.65 1,109.75 1,173.14 Property UPR level (R'mil) 0 0.3 1 2 6 16 45 126 355 1000 0-1.40 4.14 11.51 27.49 59.20 123.33 237.57 430.95 745.34 1 5.34 6.98 10.01 17.30 32.39 62.95 126.10 239.46 432.16 746.08 2 10.23 11.60 13.88 20.90 34.58 65.68 123.41 236.73 431.83 731.23 5 19.21 19.79 23.03 27.02 39.99 70.65 126.31 237.07 440.13 736.70 12 35.25 36.53 37.29 40.83 52.37 77.65 132.36 247.65 440.93 738.95 28 63.25 63.84 65.04 67.74 76.80 99.04 152.42 251.22 447.08 766.59 63 110.90 110.96 112.72 115.10 120.01 138.26 183.16 274.53 454.83 721.51 145 189.74 187.42 189.62 191.38 198.42 207.93 243.38 325.64 489.55 759.31 332 315.36 325.40 321.84 319.73 317.46 326.17 362.03 419.64 552.27 800.81 761 504.01 510.29 507.48 512.43 505.72 513.80 530.98 571.70 665.51 860.30 1745 757.11 762.95 761.69 760.21 779.33 766.79 784.87 783.66 850.02 994.29 4000 1,010.74 1,013.93 1,044.85 1,019.93 996.97 1,006.81 1,026.96 1,030.07 1,073.17 1,109.27 Transport UPR level (R'mil) 0 0.3 1 2 4 11 29 75 193 500 0-0.56 1.52 4.20 9.93 22.10 43.52 82.72 159.50 271.83 1 2.04 2.79 3.86 6.71 12.23 24.07 45.02 83.83 160.32 272.37 2 3.53 4.28 5.15 7.64 13.12 23.69 44.93 86.03 155.46 280.07 4 6.07 6.51 7.35 9.55 14.34 25.32 45.50 84.70 156.59 266.86 8 10.36 10.73 11.32 13.11 17.43 27.42 47.16 85.83 156.20 274.97 16 17.54 17.91 17.91 19.86 24.11 32.22 51.18 89.12 160.11 266.79 32 29.41 29.84 30.00 30.77 34.13 42.25 57.88 97.02 164.64 279.67 63 48.72 49.51 47.97 48.94 51.23 56.99 73.44 105.03 171.83 284.10 126 79.46 78.76 78.81 80.87 82.89 86.50 97.55 129.00 181.96 295.29 251 127.00 124.04 127.30 125.60 127.44 130.40 140.58 164.77 219.39 318.77 501 197.32 192.16 196.08 202.34 197.92 201.58 208.52 227.57 259.75 338.10 1000 293.94 306.21 293.61 295.55 293.70 295.94 300.32 314.22 324.69 400.91 110

Gross stand-alone capital at the 98% level: GWP level (R'mil) GWP level (R'mil) GWP level (R'mil) GWP level (R'mil) Accident UPR level (R'mil) 0 0.3 1 2 4 11 29 75 193 500 0-0.20 0.58 1.67 4.46 10.77 26.03 58.94 139.38 321.97 1 0.69 1.00 1.48 2.67 5.49 11.73 26.93 59.73 140.10 322.61 2 1.31 1.55 2.03 3.11 5.91 12.17 27.06 60.40 142.71 330.65 4 2.50 2.79 3.16 4.18 6.78 13.26 27.57 61.55 145.11 326.14 8 4.74 4.96 5.18 6.47 8.90 14.76 28.88 62.41 142.45 331.95 16 8.98 9.16 9.57 10.50 13.02 18.61 32.92 66.45 145.67 330.35 32 16.93 17.59 17.51 18.80 19.98 25.77 39.93 71.66 152.16 336.63 63 31.81 31.05 32.51 32.91 34.04 40.80 54.63 83.12 164.37 344.30 126 59.52 59.13 60.81 59.72 61.38 66.52 80.07 107.73 189.41 369.99 251 110.84 106.62 111.82 113.47 111.38 120.03 130.42 158.57 238.05 396.63 501 205.38 213.21 210.89 202.07 207.95 213.17 223.98 241.31 323.90 476.96 1000 378.38 375.02 379.08 368.78 377.91 380.15 393.72 426.29 499.59 622.53 Engineering Guarantee UPR level (R'mil) 0 0.3 1 2 4 11 29 75 193 500 0-0.12 0.39 1.32 3.89 10.48 25.72 62.70 146.83 339.06 1 0.40 0.59 0.98 2.12 4.79 11.42 26.61 63.54 147.59 339.73 2 0.77 0.90 1.21 2.23 4.91 11.63 27.35 62.44 148.41 345.27 4 1.45 1.58 1.83 2.71 5.09 11.53 26.64 62.39 146.10 339.95 8 2.72 2.81 2.98 3.60 5.80 11.99 26.69 62.57 147.25 342.10 16 5.03 5.12 5.29 5.66 7.44 12.57 27.02 63.15 146.54 332.76 32 9.15 8.91 9.14 9.69 10.74 14.88 28.04 62.97 149.69 337.86 63 16.30 16.13 16.75 16.41 18.15 20.87 31.25 63.79 144.22 331.23 126 28.27 28.57 28.41 28.14 28.81 31.78 38.84 66.11 139.81 321.40 251 47.22 45.93 46.47 46.72 47.06 48.91 54.38 72.27 136.18 311.64 501 74.63 73.94 74.59 74.41 76.67 77.02 77.58 92.28 136.63 287.24 1000 107.69 107.40 104.03 106.86 104.36 110.16 103.30 116.24 146.13 258.88 UPR level (R'mil) 0 0.3 1 2 7 19 57 170 506 1500 0-0.49 1.59 4.86 13.44 33.20 74.61 157.99 313.42 577.60 1 1.87 2.45 3.73 7.07 15.50 34.91 75.90 158.92 314.04 577.99 2 3.57 4.03 5.16 8.30 16.25 35.35 76.36 159.95 318.57 595.92 5 6.74 7.06 7.80 10.73 18.18 36.29 78.86 157.77 318.19 589.35 10 12.54 13.17 13.27 15.54 22.87 39.34 80.84 159.92 323.35 593.35 21 22.96 23.54 23.60 26.05 31.48 46.42 85.42 166.84 322.09 611.37 45 41.36 40.93 41.53 44.03 47.78 61.29 96.13 178.45 330.86 597.46 96 73.29 74.53 72.82 75.01 79.81 90.34 119.23 189.27 344.53 608.37 205 127.54 127.36 129.51 127.21 135.17 141.76 165.08 228.01 361.98 627.89 437 217.37 217.95 221.33 215.96 218.48 229.17 244.91 299.91 419.16 658.54 935 361.22 366.35 359.14 353.17 365.01 362.02 380.87 427.52 503.20 748.39 2000 580.48 570.59 585.35 580.30 588.86 582.27 588.16 627.10 701.36 826.12 Liability UPR level (R'mil) 0 0.3 1 2 6 16 45 126 355 1000 0-0.38 1.10 3.31 9.27 23.83 59.66 139.70 337.70 785.15 1 1.42 1.88 2.67 4.98 10.93 25.33 61.00 140.81 338.65 785.93 2 2.76 3.20 3.99 6.12 11.44 26.07 60.12 144.35 340.40 789.67 5 5.33 5.58 6.46 8.32 13.62 28.05 63.22 144.00 336.52 803.94 10 10.22 10.29 11.46 12.75 17.99 31.25 63.53 147.00 331.46 780.41 21 19.43 19.63 20.33 22.21 26.56 38.15 71.10 150.71 336.94 790.71 45 36.61 36.96 37.84 38.59 42.75 51.97 81.91 158.73 348.34 784.62 96 68.39 68.32 68.20 70.12 72.46 82.15 108.69 181.59 356.02 807.98 205 126.54 124.97 126.90 126.59 130.00 137.89 161.39 222.18 397.74 831.68 437 231.75 230.86 235.74 228.34 233.62 239.84 258.50 315.16 481.94 861.99 935 419.62 420.40 421.85 422.42 423.88 426.04 440.81 484.74 607.65 954.59 2000 749.72 753.84 752.45 756.75 747.54 755.52 781.30 817.75 905.09 1,238.93 111

Miscellaneous UPR level (R'mil) 0 0.3 1 2 6 16 45 126 355 1000 0-1.51 4.64 12.73 30.82 68.65 143.07 273.07 511.50 844.97 1 5.44 7.56 11.23 19.13 36.32 72.99 146.28 275.25 512.95 845.82 2 9.98 11.42 14.96 22.31 39.28 75.32 145.03 286.64 514.78 859.51 5 18.01 19.36 22.00 28.15 43.91 78.29 151.33 283.18 517.62 844.57 10 31.93 33.30 35.47 40.87 54.45 88.45 155.53 288.99 518.81 844.81 21 55.62 57.34 57.25 61.98 73.75 102.26 167.57 297.97 530.70 834.54 45 95.11 96.63 98.47 99.38 110.55 134.01 188.28 314.93 541.53 862.52 96 159.40 163.64 159.63 162.99 173.11 192.43 237.41 345.07 547.49 863.10 205 260.88 263.36 263.98 267.20 264.23 288.71 321.86 420.97 603.15 886.60 437 413.85 417.63 410.06 417.23 430.67 433.00 458.33 527.07 675.49 932.16 935 626.94 633.39 618.15 630.53 637.53 655.14 654.28 705.56 799.30 964.61 2000 878.43 882.92 907.64 875.02 877.04 888.26 902.73 916.11 965.50 1,061.92 GWP level (R'mil) GWP level (R'mil) GWP level (R'mil) GWP level (R'mil) Motor UPR level (R'mil) 0 0.3 1 2 5 14 37 101 276 750 0-0.85 2.30 5.84 13.23 28.77 58.19 114.49 216.37 374.58 1 3.44 4.24 5.68 9.00 15.86 30.87 59.75 115.62 217.15 375.08 2 6.48 7.09 8.29 11.10 17.61 32.54 61.25 118.10 218.43 380.82 5 12.13 12.63 13.56 16.07 22.00 34.55 63.45 119.88 222.89 386.62 13 22.48 22.64 23.75 25.59 30.59 42.37 68.59 121.43 225.43 396.00 30 41.03 41.13 42.29 43.22 46.64 56.97 79.81 131.27 225.88 387.47 71 73.49 72.52 73.71 75.42 77.86 84.58 106.37 150.29 236.27 394.24 166 128.51 128.82 129.18 128.75 132.25 136.97 152.15 185.75 268.53 414.30 388 217.78 217.90 218.77 216.09 223.10 220.13 232.17 261.39 320.22 449.15 910 353.28 349.95 357.97 357.72 356.30 356.60 363.10 380.33 413.39 532.26 2133 534.88 542.72 544.81 547.13 538.37 534.46 539.57 549.21 562.92 623.00 5000 710.61 705.49 703.60 705.60 703.78 707.33 717.10 711.60 719.43 739.90 Property UPR level (R'mil) 0 0.3 1 2 6 16 45 126 355 1000 0-0.94 2.88 8.08 20.29 44.21 91.97 181.78 321.83 545.22 1 3.49 4.72 6.97 12.14 23.91 47.00 94.04 183.23 322.74 545.76 2 6.87 7.84 9.58 14.33 25.03 48.78 94.27 181.05 323.07 526.20 5 13.21 13.69 15.60 19.42 28.83 51.89 96.63 178.20 328.54 533.36 12 24.81 26.24 26.70 29.47 37.37 58.20 100.38 186.60 327.72 533.31 28 45.40 46.88 46.52 49.20 55.47 71.92 112.30 189.06 336.38 543.11 63 80.86 79.47 83.08 83.86 88.53 101.91 137.14 204.39 339.38 526.21 145 139.65 138.24 140.68 142.49 147.64 153.37 180.02 241.48 362.39 546.47 332 232.22 237.79 234.39 233.48 234.45 240.26 261.55 310.91 408.87 570.51 761 365.98 372.06 365.61 372.60 370.16 374.19 386.68 417.23 478.78 598.46 1745 527.02 533.55 533.57 533.59 531.28 539.43 540.87 549.68 587.26 647.79 4000 624.57 626.09 635.40 623.13 610.63 625.38 607.30 627.42 652.59 628.80 Transport UPR level (R'mil) 0 0.3 1 2 4 11 29 75 193 500 0-0.40 1.14 3.18 7.64 16.60 32.94 61.28 111.33 178.67 1 1.46 2.00 2.90 5.08 9.41 18.08 34.08 62.10 111.91 179.02 2 2.54 3.00 3.76 5.73 10.04 18.16 33.87 63.09 110.78 183.28 4 4.38 4.72 5.44 6.95 10.63 19.08 34.76 62.95 110.40 176.03 8 7.50 7.71 8.23 9.66 13.00 20.98 35.72 63.26 112.28 176.39 16 12.68 12.84 13.11 14.28 17.32 23.99 38.08 65.83 114.02 177.14 32 21.16 21.56 21.40 22.28 24.31 30.36 42.58 69.91 112.62 185.09 63 34.72 35.37 34.64 35.14 37.30 41.78 52.58 77.30 118.54 190.33 126 55.71 55.54 55.78 56.43 58.52 60.30 68.41 90.18 125.10 189.51 251 86.66 85.17 88.70 86.50 87.36 87.50 96.14 111.19 144.27 199.28 501 128.71 128.51 127.73 129.79 128.23 131.86 136.76 145.68 164.56 202.23 1000 177.03 182.38 179.34 181.40 179.00 179.85 180.34 188.19 189.72 218.42 112

Gross stand-alone capital at the 95% level: GWP level (R'mil) GWP level (R'mil) GWP level (R'mil) GWP level (R'mil) Accident UPR level (R'mil) 0 0.3 1 2 4 11 29 75 193 500 0-0.20 0.58 1.67 4.46 10.77 26.03 58.94 139.38 321.97 1 0.69 1.00 1.48 2.67 5.49 11.73 26.93 59.73 140.10 322.61 2 1.31 1.55 2.03 3.11 5.91 12.17 27.06 60.40 142.71 330.65 4 2.50 2.79 3.16 4.18 6.78 13.26 27.57 61.55 145.11 326.14 8 4.74 4.96 5.18 6.47 8.90 14.76 28.88 62.41 142.45 331.95 16 8.98 9.16 9.57 10.50 13.02 18.61 32.92 66.45 145.67 330.35 32 16.93 17.59 17.51 18.80 19.98 25.77 39.93 71.66 152.16 336.63 63 31.81 31.05 32.51 32.91 34.04 40.80 54.63 83.12 164.37 344.30 126 59.52 59.13 60.81 59.72 61.38 66.52 80.07 107.73 189.41 369.99 251 110.84 106.62 111.82 113.47 111.38 120.03 130.42 158.57 238.05 396.63 501 205.38 213.21 210.89 202.07 207.95 213.17 223.98 241.31 323.90 476.96 1000 378.38 375.02 379.08 368.78 377.91 380.15 393.72 426.29 499.59 622.53 Engineering Guarantee UPR level (R'mil) 0 0.3 1 2 4 11 29 75 193 500 0-0.12 0.39 1.32 3.89 10.48 25.72 62.70 146.83 339.06 1 0.40 0.59 0.98 2.12 4.79 11.42 26.61 63.54 147.59 339.73 2 0.77 0.90 1.21 2.23 4.91 11.63 27.35 62.44 148.41 345.27 4 1.45 1.58 1.83 2.71 5.09 11.53 26.64 62.39 146.10 339.95 8 2.72 2.81 2.98 3.60 5.80 11.99 26.69 62.57 147.25 342.10 16 5.03 5.12 5.29 5.66 7.44 12.57 27.02 63.15 146.54 332.76 32 9.15 8.91 9.14 9.69 10.74 14.88 28.04 62.97 149.69 337.86 63 16.30 16.13 16.75 16.41 18.15 20.87 31.25 63.79 144.22 331.23 126 28.27 28.57 28.41 28.14 28.81 31.78 38.84 66.11 139.81 321.40 251 47.22 45.93 46.47 46.72 47.06 48.91 54.38 72.27 136.18 311.64 501 74.63 73.94 74.59 74.41 76.67 77.02 77.58 92.28 136.63 287.24 1000 107.69 107.40 104.03 106.86 104.36 110.16 103.30 116.24 146.13 258.88 UPR level (R'mil) 0 0.3 1 2 7 19 57 170 506 1500 0-0.49 1.59 4.86 13.44 33.20 74.61 157.99 313.42 577.60 1 1.87 2.45 3.73 7.07 15.50 34.91 75.90 158.92 314.04 577.99 2 3.57 4.03 5.16 8.30 16.25 35.35 76.36 159.95 318.57 595.92 5 6.74 7.06 7.80 10.73 18.18 36.29 78.86 157.77 318.19 589.35 10 12.54 13.17 13.27 15.54 22.87 39.34 80.84 159.92 323.35 593.35 21 22.96 23.54 23.60 26.05 31.48 46.42 85.42 166.84 322.09 611.37 45 41.36 40.93 41.53 44.03 47.78 61.29 96.13 178.45 330.86 597.46 96 73.29 74.53 72.82 75.01 79.81 90.34 119.23 189.27 344.53 608.37 205 127.54 127.36 129.51 127.21 135.17 141.76 165.08 228.01 361.98 627.89 437 217.37 217.95 221.33 215.96 218.48 229.17 244.91 299.91 419.16 658.54 935 361.22 366.35 359.14 353.17 365.01 362.02 380.87 427.52 503.20 748.39 2000 580.48 570.59 585.35 580.30 588.86 582.27 588.16 627.10 701.36 826.12 Liability UPR level (R'mil) 0 0.3 1 2 6 16 45 126 355 1000 0-0.38 1.10 3.31 9.27 23.83 59.66 139.70 337.70 785.15 1 1.42 1.88 2.67 4.98 10.93 25.33 61.00 140.81 338.65 785.93 2 2.76 3.20 3.99 6.12 11.44 26.07 60.12 144.35 340.40 789.67 5 5.33 5.58 6.46 8.32 13.62 28.05 63.22 144.00 336.52 803.94 10 10.22 10.29 11.46 12.75 17.99 31.25 63.53 147.00 331.46 780.41 21 19.43 19.63 20.33 22.21 26.56 38.15 71.10 150.71 336.94 790.71 45 36.61 36.96 37.84 38.59 42.75 51.97 81.91 158.73 348.34 784.62 96 68.39 68.32 68.20 70.12 72.46 82.15 108.69 181.59 356.02 807.98 205 126.54 124.97 126.90 126.59 130.00 137.89 161.39 222.18 397.74 831.68 437 231.75 230.86 235.74 228.34 233.62 239.84 258.50 315.16 481.94 861.99 935 419.62 420.40 421.85 422.42 423.88 426.04 440.81 484.74 607.65 954.59 2000 749.72 753.84 752.45 756.75 747.54 755.52 781.30 817.75 905.09 1,238.93 113

Miscellaneous UPR level (R'mil) 0 0.3 1 2 6 16 45 126 355 1000 0-1.51 4.64 12.73 30.82 68.65 143.07 273.07 511.50 844.97 1 5.44 7.56 11.23 19.13 36.32 72.99 146.28 275.25 512.95 845.82 2 9.98 11.42 14.96 22.31 39.28 75.32 145.03 286.64 514.78 859.51 5 18.01 19.36 22.00 28.15 43.91 78.29 151.33 283.18 517.62 844.57 10 31.93 33.30 35.47 40.87 54.45 88.45 155.53 288.99 518.81 844.81 21 55.62 57.34 57.25 61.98 73.75 102.26 167.57 297.97 530.70 834.54 45 95.11 96.63 98.47 99.38 110.55 134.01 188.28 314.93 541.53 862.52 96 159.40 163.64 159.63 162.99 173.11 192.43 237.41 345.07 547.49 863.10 205 260.88 263.36 263.98 267.20 264.23 288.71 321.86 420.97 603.15 886.60 437 413.85 417.63 410.06 417.23 430.67 433.00 458.33 527.07 675.49 932.16 935 626.94 633.39 618.15 630.53 637.53 655.14 654.28 705.56 799.30 964.61 2000 878.43 882.92 907.64 875.02 877.04 888.26 902.73 916.11 965.50 1,061.92 GWP level (R'mil) GWP level (R'mil) GWP level (R'mil) GWP level (R'mil) Motor UPR level (R'mil) 0 0.3 1 2 5 14 37 101 276 750 0-0.85 2.30 5.84 13.23 28.77 58.19 114.49 216.37 374.58 1 3.44 4.24 5.68 9.00 15.86 30.87 59.75 115.62 217.15 375.08 2 6.48 7.09 8.29 11.10 17.61 32.54 61.25 118.10 218.43 380.82 5 12.13 12.63 13.56 16.07 22.00 34.55 63.45 119.88 222.89 386.62 13 22.48 22.64 23.75 25.59 30.59 42.37 68.59 121.43 225.43 396.00 30 41.03 41.13 42.29 43.22 46.64 56.97 79.81 131.27 225.88 387.47 71 73.49 72.52 73.71 75.42 77.86 84.58 106.37 150.29 236.27 394.24 166 128.51 128.82 129.18 128.75 132.25 136.97 152.15 185.75 268.53 414.30 388 217.78 217.90 218.77 216.09 223.10 220.13 232.17 261.39 320.22 449.15 910 353.28 349.95 357.97 357.72 356.30 356.60 363.10 380.33 413.39 532.26 2133 534.88 542.72 544.81 547.13 538.37 534.46 539.57 549.21 562.92 623.00 5000 710.61 705.49 703.60 705.60 703.78 707.33 717.10 711.60 719.43 739.90 Property UPR level (R'mil) 0 0.3 1 2 6 16 45 126 355 1000 0-0.94 2.88 8.08 20.29 44.21 91.97 181.78 321.83 545.22 1 3.49 4.72 6.97 12.14 23.91 47.00 94.04 183.23 322.74 545.76 2 6.87 7.84 9.58 14.33 25.03 48.78 94.27 181.05 323.07 526.20 5 13.21 13.69 15.60 19.42 28.83 51.89 96.63 178.20 328.54 533.36 12 24.81 26.24 26.70 29.47 37.37 58.20 100.38 186.60 327.72 533.31 28 45.40 46.88 46.52 49.20 55.47 71.92 112.30 189.06 336.38 543.11 63 80.86 79.47 83.08 83.86 88.53 101.91 137.14 204.39 339.38 526.21 145 139.65 138.24 140.68 142.49 147.64 153.37 180.02 241.48 362.39 546.47 332 232.22 237.79 234.39 233.48 234.45 240.26 261.55 310.91 408.87 570.51 761 365.98 372.06 365.61 372.60 370.16 374.19 386.68 417.23 478.78 598.46 1745 527.02 533.55 533.57 533.59 531.28 539.43 540.87 549.68 587.26 647.79 4000 624.57 626.09 635.40 623.13 610.63 625.38 607.30 627.42 652.59 628.80 Transport UPR level (R'mil) 0 0.3 1 2 4 11 29 75 193 500 0-0.40 1.14 3.18 7.64 16.60 32.94 61.28 111.33 178.67 1 1.46 2.00 2.90 5.08 9.41 18.08 34.08 62.10 111.91 179.02 2 2.54 3.00 3.76 5.73 10.04 18.16 33.87 63.09 110.78 183.28 4 4.38 4.72 5.44 6.95 10.63 19.08 34.76 62.95 110.40 176.03 8 7.50 7.71 8.23 9.66 13.00 20.98 35.72 63.26 112.28 176.39 16 12.68 12.84 13.11 14.28 17.32 23.99 38.08 65.83 114.02 177.14 32 21.16 21.56 21.40 22.28 24.31 30.36 42.58 69.91 112.62 185.09 63 34.72 35.37 34.64 35.14 37.30 41.78 52.58 77.30 118.54 190.33 126 55.71 55.54 55.78 56.43 58.52 60.30 68.41 90.18 125.10 189.51 251 86.66 85.17 88.70 86.50 87.36 87.50 96.14 111.19 144.27 199.28 501 128.71 128.51 127.73 129.79 128.23 131.86 136.76 145.68 164.56 202.23 1000 177.03 182.38 179.34 181.40 179.00 179.85 180.34 188.19 189.72 218.42 114

12.6 Appendix F: Graphs of Gross Stand-Alone Capital 99.5% Capital curve - Accident 1,400 1,300 1,200 1,100 1,000 900 800 700 600 500 400 300 200 100-500 193 75 29 11 4 GUPR (R'mil) 2 1 0 0 0 1 2 4 8 16 32 63 126 251 501 1000 GWP (R'mil) 99.5% Capital curve - Engineering 600 500 400 300 200 100-2 0 4 1 8 16 32 63 126 251 501 1000-0 1 500 193 75 GUPR (R'mil) 29 11 4 2 GWP (R'mil) 115

99.5% Capital curve - Guarantee 1,700 1,600 1,500 1,400 1,300 1,200 1,100 1,000 900 800 700 600 500 400 300 200 100-1,500 506 GUPR (R'mil) 170 57 19 0 7 1 2 2 5 10 21 45 96 205 437 935 2000 GWP (R'mil) - 0 1 99.5% Capital curve - Liability 2,300 2,200 2,100 2,000 1,900 1,800 1,700 1,600 1,500 1,400 1,300 1,200 1,100 1,000 900 800 700 600 500 400 300 200 100-1,000 355 GUPR (R'mil) 126 45 2-5 0 1 10 21 45 96 205 437 935 2000 0 1 16 6 2 GWP (R'mil) 116

99.5% Capital curve - Miscellaneous 2,400 2,300 2,200 2,100 2,000 1,900 1,800 1,700 1,600 1,500 1,400 1,300 1,200 1,100 1,000 900 800 700 600 500 400 300 200 100 - GUPR (R'mil) - - 1,000 355 126 45 16 6 2 1 0 1 2 5 10 21 45 96 205 437 935 2,000 GWP (R'mil) 99.5% Capital curve - Motor 1,600 1,500 1,400 1,300 1,200 1,100 1,000 900 800 700 600 500 400 300 200 100-0 1 2 5 13 30 71 166 388 910 2133 5000-0 1 750 276 101 GUPR (R'mil) 37 14 5 2 GWP (R'mil) 117

99.5% Capital curve - Property 1,600 1,500 1,400 1,300 1,200 1,100 1,000 900 800 700 600 500 400 300 200 100-1,000 355 GUPR (R'mil) 126 45 16 6 - - 2 1 0 1 2 5 12 28 63 145 332 761 1,745 4,000 GWP (R'mil) 99.5% Capital curve - Transport 600 500 400 300 200 100-2 4 0 1 8 16 32 63 126 251 501 1000-0 1 500 193 75 GUPR (R'mil) 29 11 4 2 GWP (R'mil) 118

12.7 Appendix G: Description of The Smith Model 12.7.1 General description of the model The Smith Model (TSM) is a deflator enabled multi-variate stochastic asset model. It produces economic simulations of the future based on widely accepted finance theory whilst also respecting the statistical features observed in the market. It can be calibrated to the observed market prices and yield curve at any date for any economy. The Smith Model is proprietary and is produced by the Capital Markets Group (a division of Deloitte). It was developed over the past 10 years and has a dedicated team ensuring it is continually updated. Although the model is proprietary, the general principles on which it is built are published in a variety of articles (see further reading below) and have therefore stood up to scrutiny by its peers. 12.7.2 Model structure and Calibration 12.7.2.1 The model structure It is relatively easy to build a stochastic asset model that can produce economic simulations. However a closer examination of the simulations frequently reveals features that can be exploited using dynamic hedging strategies. This is important for financial planning because optimal strategies that exploit features of a stochastic asset model make genuine value added strategies harder to find. The best approach to exposing the genuine and valuable dynamic strategies is to eliminate inefficiencies from the stochastic model that is being used. Only then can the user be sure that the effects that they are seeing are due to the business dynamics and not the underlying model. TSM is dedicated to eliminating model inefficiencies. This is captured by the following features. Arbitrage-Free Yield Curves Arbitrage-Free Prices Efficient Return Projections - Market Equilibrium Initial Yield Curves Efficient Yield Projections Efficient Exchange Rate Projections TSM is also dedicated to capturing widely accepted statistical features of markets as follows: Mean-Reversion of interest rates Fat-Tailed Distributions 119

12.7.2.2 Calibration (data and parameters) Historical data: For the UK calibration 10 years worth of historical data (weekly where available) is used. This is to ensure that the data used is relevant. The data sources include government publications, stock exchanges, third party data, broker reports and other published market analyses. For the standard South African calibrations, the Smith model uses the past 10 years weekly data of the following: Consumer Price Index Equity Index Dividend yield Real yield, if available Government Bond Index Initial yield curve (typically the model uses a set of 5 government bonds of different durations, e.g. 1 year, 2 year, 5 year, 10 year and longest available) The data is provided by DataStream. If more than one economy is required then the relevant exchange rates are used as well. Note that the model uses past data together with today s data to arrive at its overall results. For example, as it is assumed that the market is arbitrage free the reported future yields will be driven by information in today s yield curves. A specific calibration replicates market prices for a specific date. Market forecasts are used in instances where the statistical parameter estimates are not clear or contradicts with other parameters, rather that just using a pure statistical best estimate. Parameters - Initial conditions Two initial conditions are required to calibrate the model. The equity risk premium is defined as the log return on the total return index in excess of the return on a risk-free deposit account. The risk premiums for TSM are calculated to be consistent with market equilibrium. The infinite spot rate is set based on the yield curve published by the Bond Exchange of South Africa. For the South African calibration of 31 March 2005 an annually compounded rate of 5.5% p.a. was used. 12.7.3 Key features of the model 12.7.3.1 Statistical features The Smith model is based on a mean reverting Ornstein-Uhlenbeck process for the yields and a random walk process for the total return of each asset class modelled. This assumes that the economy is in a stable state and yield values exhibit mean reversion. The fat tailed returns observed in economic markets are modelled using a five-parameter family of distributions that are fitted to the first 5 moments of the distribution. Sophisticated techniques are employed to avoid negative interest rates but still produce low interest rates with reasonable frequency and without creating arbitrage. The model starts off from a pre-developed economic framework. The equations of the model are built from widely accepted and scrutinised finance theory. The covariance matrix is used to 120

parameterise the model. As a recent development the model is now also able to calibrate to prospective covariance information, e.g. the implied volatilities in swaption data. The Smith model is a stationary model; hence these covariance structures are fixed for a specific calibration. It is believed that these matrices hold sufficient data to accurately model the economy. The Smith model does not support regime switching due to the lack of theoretical backing for it. The Smith model uses jump processes to model excessive variability in all modelled variables i.e. jump processes are used to model the fat tails experienced in the markets. 12.7.3.2 Economic features The Smith model assumes that markets are efficient and arbitrage free. It generates asset prices by equating supply and demand for assets, assuming that the market is near equilibrium. This means that near-arbitrage opportunities are avoided too. It can accurately reproduce market prices and produces an arbitrage free yield curve. Exchange rate fluctuations are constrained by the differences in the yield curves between the different countries. 12.7.4 The valuation methods supported by the model The Smith model is a deflator enabled model and therefore supports deflator valuation techniques and real world probabilities. Further, the model provides the returns on all the asset categories and economies requested. The user can determine their discount rate based on their assumed future asset distributions backing their reserves for each simulation generated if they prefer. 12.7.5 The outputs of the model 12.7.5.1 Asset Classes The standard calibration includes the following asset classes for the local economy: Equities Total return Equity growth Dividend income and yield Government bonds of any duration (i.e. the yield curve) Gross redemption yield Income Growth Running yield Total return Cash or money market interest rates Price Inflation Deflators 121

Additional asset classes such as property and corporate bonds and derivatives can be added if required. Additional economies can be provided as required. If additional economies are included in the calibration, all the asset classes listed above are provided for each economy. In addition the exchange rates between the economies are provided. The Smith Model can be calibrated for any economy for which sufficient data is available. 12.7.5.2 Frequency Term of Projections The model is a continuous time model. The user can therefore draw simulations at any time horizon and any frequency. 12.7.6 Ease of use 12.7.6.1 Calibration of the model general The Capital Markets Group in the UK has a dedicated team of people who calibrate the model and provide support. The process they follow is to determine the variance-covariance matrix of all the modelled assets from the historical data gathered. If more than one economy is modelled, then the covariance matrix of the different economies is also constructed. A full risk free yield curve is modelled based on government bonds. An exact calibration is done to the initial yield curve represented by the set of individual government bonds as at date of calibration. The user may indicate which bonds to use especially if they have certain bond exposures in their asset portfolios. The calibration will provide a good fit to historic covariance data. The calibration will also provide a good fit to expected future interest rates as implied by the current term structure of interest rates and the prospective covariance structure, e.g. as implied in swaption volatilities. Sophisticated users have the option of doing the calibration themselves. This is only practical for companies with sufficiently skilled resources. 122

12.7.7 Further reading The following articles discuss The Smith Model or some of the principles on which it is built: Year Article Context 1995 Stochastic Asset Models General Insurance Convention 1996 How Actuaries can use Financial Economics Institute of Actuaries 1997 Global Returns on Assets with Fat Tails Investment Convention 1998 Gauge Transforms in Stochastic Models AFIR 1999 Information Structures Investment Convention 2000 Incomplete Bond Pricing Models FORC 2000 Fitting Yield Curves with Long Constraints FORC 2000 Das InvestmentModell TSM DAV 2001 Consistent Multinational Assumptions Investment Convention 2001 Asset Model for Fair Values (Timbuk 1) Life Convention 2001 Levy Processes for Investment Models Oxford Levy Seminar 2002 Corporate Bond Models Investment Convention 2003 Option Pricing with Deflators Risk Publications 2003 The Importance of Being Normal ASTIN 123

12.8 Appendix H: Allocation of FSB Asset categories to modelled asset classes Description of asset 1. Bank notes and coins, including Krugerrand coins of all denominations, issued or caused to be issued in terms of the South African Reserve Bank Act, 1989 (Act No. 90 of 1989). 2. A credit balance in an account with, or a deposit, including a negotiable deposit and a bill, accepted by, an institution finally registered under the Banks Act, 1990 (Act No. 94 of 1990), or the Mutual Banks Act, 1993 (Act No. 124 of 1993). Modeled asset category Cash 0% Cash 0% 3. Public deposits with the Corporation for Public Deposits established by section2 of the Corporation for Public Deposits Act, 1984 (Act No. 46 of 1984). 4. Securities issued by, and loans made to, the Government of the Republic in terms of section 19 of the Exchequer Act, 1975 (Act No. 66 of 1975). 5. Securities and loans guaranteed by a Minister of the Republic under section 35 of the Exchequer Act, 1975. 6. Securities issued or guaranteed by, and loans made to or guaranteed by, a body, council or institution under the repealed Provincial Government Act, 1961 (Act No. 32 of 1961). 7. Securities issued by, and loans made to, the Local Authorities Loans Fund Board under the Local Authorities Loans Fund Act, 1984 (Act No. 67 of 1984). 8. Securities issued or guaranteed by, and loans made to or guaranteed by, the Rand Water Board under the Rand Water Board Statutes (Private) Act, 1950 (Act No. 17 of 1950). 9. Securities issued or guaranteed by, and loans made to or guaranteed by, Eskom under the Eskom Act, 1987 (Act No. 40 of 1987). 10. Securities issued or guaranteed by, and loans made to or guaranteed by, and deposits with, the Land and Agricultural Bank of South Africa under the Land Bank Act, 1944 (Act No. 13 of 1944). 11. Securities issued or guaranteed, and loans raised or guaranteed, under the Legal Succession to the South African Transport Services Act, 1989 (Act No. 9 of 1989). Fixed interest if > 1 year, cash if <=1 year Fixed interest if > 1 year, cash if <=1 year Fixed interest if > 1 year, cash if <=1 year Fixed interest if > 1 year, cash if <=1 year Fixed interest if > 1 year, cash if <=1 year Fixed interest if > 1 year, cash if <=1 year Fixed interest if > 1 year, cash if <=1 year Fixed interest if > 1 year, cash if <=1 year Fixed interest if > 1 year, cash if <=1 year 5% 0% 0% 5% 5% 5% 5% 5% 5% 124

12. Securities and loans, not elsewhere stated, which are: issued by or made to a body corporate established by a law of the Republic; and approved by the Registrar for the purposes of the Schedule generally by notice in the Gazette subject to the conditions determined by the Registrar and specified in the notice. Fixed interest if > 1 year, cash if <=1 year 5% 13. Immovable property in the Republic Property 50% 14. Motor vehicles, furniture and office equipment, including computer equipment used by the short-term insurer concerned in the course of its business in the Republic. 15. Listed shares and securities issued by a company incorporated in the Republic. 16. Unlisted shares and securities issued by a company incorporated in the Republic. 17. Shares, debentures and depository receipts which are Other 50% Equity 30% Other 50% Equity 30% issued by an institution incorporated outside the Republic; and listed on a licensed stock exchange in the Republic. 18. Linked units Equity 30% in respect of institutions one or more of which is or re incorporated outside the Republic; and which are listed on a licensed stock exchange in the Republic. 19. Loan stock listed on a licensed stock exchange in the Republic issued by a company incorporated in the Republic. 20. Listed securities issued by a government of a country other than the Republic or listed securities and shares issued by an institution incorporated outside the Republic. Fixed interest if > 1 year, cash if <=1 year Foreign, Fixed interest and Equity 30% 20% 21. A credit balance in an account with, or a deposit, including a negotiable certificate of deposit, or a bill, accepted by, an institution incorporated outside the Republic, which would have been a bank in terms of the Banks Act, 1990, if it were incorporated in the Republic. 22. Units which are derived from or linked to one or more assts referred to in (20) and (21) above. 23. Derivatives and margin deposits on the assets referred to in (20) and (21) above Foreign cash 20% Look through to underlying 20% Other 20% 24. Units in a unit trust scheme registered in terms of the Unit Trusts Control Act 1981 (Act No 54 of 1981) Look through to underlying 30% 125

Trusts Control Act, 1981 (Act No. 54 of 1981). 25. Derivatives and the margin deposit in the Republic. Other Traded on a recognised exchange Over the counter instruments 26. Claims secured by mortgages over immovable property in the Republic. 30% 75% Other 50% 27. Other claims, not elsewhere stated, against a long-term insurer in terms of a long-term policy a person in the Republic; and a body corporate and any stock or shares in a body corporate which is not incorporated and registered in the Republic but which, in the opinion of the Registrar, carries on business in the Republic and which has been approved by the Registrar generally by notice in the Gazette and subject to the conditions determined by the Registrar and specified in the notice. 28. Premiums due and payable to the short-term insure in respect of short-term insurance business carried on in the Republic. 29. Claims against companies or body corporates incorporated outside the RSA, not specified elsewhere. Other and Foreign Other 20% 50% 50% Other 100 % Foreign Other 75% 30. Unlisted foreign investments. Foreign Other 75% 126

Draft Report 12.9 Appendix I: Parameters for the calculation of the diversification/correlation factor Correlation Matrix Accident Engineering Guarantee Liability Miscellaneous Motor Property Transport Accident 1 0.121465048 0.183651933-0.047304819-0.118772641 0.076799189 0.267285892 0.351504303 Engineering 0.121465048 1 0.034966036 0.220727973 0.105283921 0.037499426 0.218654276 0.551756731 Guarantee 0.183651933 0.034966036 1 0.185352476-0.045480726 0.18299882 0.06058422 0.229965053 Liability -0.047304819 0.220727973 0.185352476 1 0.517525719 0.215258095 0.517222906 0.298063816 Miscellaneous -0.118772641 0.105283921-0.045480726 0.517525719 1 0.057163811-0.056104331-0.207188902 Motor 0.076799189 0.037499426 0.18299882 0.215258095 0.057163811 1 0.179959272 0.181373381 Property 0.267285892 0.218654276 0.06058422 0.517222906-0.056104331 0.179959272 1 0.15259213 Transport 0.351504303 0.551756731 0.229965053 0.298063816-0.207188902 0.181373381 0.15259213 1 Mean Standard Deviation ULR: Curve Parameters Accident Engineering Guarantee Liability Miscellaneous Motor Property Transport a 0.50254307 0.612636264 0.618170107 0.61434177-26.22393264 0.639600626-90.42079051-8.022356358 b 0 0 0 0 27.87514672 4.325235775 91.65953926 9.022793339 c 0 0 0 0-0.002596084-0.312100426-0.000481849-0.004003602 Accident Engineering Guarantee Liability Miscellaneous Motor Property Transport a -1.220247172-4.911405719-383.0620459-383.5786695-528.5630863-382.8654151-572.0061182-518.4092584 b -1.693099285-0.068637528 383.5423913 383.0539543 529.9521492 383.733916 574.3269511 517.9805142 c 0.041844124 0.215291039-0.000920735-0.000667254-0.000739973-0.001246635-0.00092317-0.000640623 d 4.590146054 6.084347389 4.015095659 3.934541169 5.992095791 6.254777558 5.316390968 4.491149893 Formulae for the calculation of the ULR Mean & Standard deviation: ULR Means: a + b ( GEP ) c when GEP > 0 0 otherwise ULR Standard Deviations: c a+ b( GEP) + 1 d 2 e when GEP > 0 0 otherwise 127

Draft Report 12.10 Appendix J: Example of individual company feedback 1

Draft Report Company Feedback FSB FCR Recalibration: October 2005 Prepared for IMAGINARY COMPANY Prepared by 1

Company Feedback FSB FCR Recalibration Deloitte and Insight ABC Page 2

Company Feedback FSB FCR Recalibration Table of Contents 1. Overview... 3 1.1 Background, scope and overall framework... 3 1.2 Global comparisons... 6 1.3 Reliances and limitations... 8 2. Introduction... 9 3. Potential Capital Requirements...10 4. Insurance Capital Charge...12 5. Asset Capital Charge...16 6. Total Capital Charge and Minimum Capital Requirement...18 7. Reserving...20 8. Conclusion...23 9. Recommendations...24 Page 3

Company Feedback FSB FCR Recalibration 1. Overview 1.1 Background, scope and overall framework Actuarial & Insurance Solutions at Deloitte, and Insight ABC, were appointed by the Financial Services Board (FSB) in April 2005 to calibrate Financial Condition Reporting (FCR) requirements for the Short Term Insurance Industry in South Africa. Broadly, our aim was to construct a formula, on the basis of data from Star Returns, and Dynamic Financial Analysis, which would be an appropriate basis for a solvency requirement for the industry, for those companies that choose not to use an approved Internal Model. This had to be done in accordance with work performed by the FSB and by the Financial Condition Reporting Sub-Committee of the Short Term Insurance Committee of the Actuarial Society of South Africa. The formula has to take into account international developments, but at the same time be suitable for application in South Africa given available data in the STAR returns. Throughout our project, there were two major constraints, namely that STAR returns did not contain the data required or that data was not reliable. Secondly, the application of a central formula to the Short Term Industry as a whole will inevitably lead to situations where the formula does not fit individual companies with specific circumstances. Given the above, we sent individual results to all registered companies in South Africa, requesting feedback. This process was successful in the sense that it made companies aware of the project, highlighted areas where the modelling or data had to be modified, and indicated specific companies or sectors of the market where the application of an industry framework may not give optimal results for the individual companies involved. In line with the work of the Financial Condition Reporting Sub-Committee, we therefore support a regulatory framework that would allow companies to apply to the FSB for approval to determine a different level of capital, in accordance with a set of guidelines to be specified. For those companies that do not construct a complete risk-based internal model, the option of a certified model allows them to adapt elements of the regulatory framework to take into account their specific circumstances without having to set up a complex internal model. Further, we understand that the FSB would be open to approaches by companies with particular circumstances that may not be taken into account in the industry framework, or where a certified model or internal model specifies capital requirements that the company would struggle to meet. These companies would in practice be allowed to motivate why a special dispensation should apply to them, and the FSB would be free to consider the circumstances of the company and decide whether to grant permission for the company not to meet the capital requirements determined under any or all of the three models. It is envisaged that the FSB would, however, for benchmarking purposes check the industry calibration for every company registered in South Africa, whether the company in the end applies the industry framework, a certified model or an internal model. The framework of industry calibration vs certified model vs internal model, can be represented graphically in the following way: Page 4

Company Feedback FSB FCR Recalibration Increasing Complexity Industry Calibration Certified Model Internal Model Increasing Appropriateness for Individual Company Increasing Cost Increasing usefulness in risk management Given the above, the characteristics of each of these different models can be summarised as follows: Industry Calibration Certified Model Internal Model Necessarily approximate Must be prudent for all companies May not be appropriate for circumstances of individual companies More precise for liabilities & individual cricumstances of companies Involves judgment Hence requires professional certification Maximum precision for liabilities and assets Also requires professional certification Leads to greatest understanfing of risks Provided models are transparent & realistic The above overall framework is preferable to the existing capital requirement (effectively 25% of Net Written Premium), which: does not take into account the real risks faced by companies (i.e. it does not take into account the size of the insurer 7, the class of business written, the combination of classes of business written (i.e. correlation and diversification), expenses, and so on) 7 Of course, the required Rand amount of capital increases as the net written premium increases, but as shown below, a capital requirement which is a constant percentage of net written premium does not accurately reflect the considerably higher risks faced by smaller companies, and lower risks faced by larger companies. Page 5

Company Feedback FSB FCR Recalibration requires a level of capital which is prudent for some companies but not prudent for others The only advantage of the current model is its simplicity. We hope that the industry calibration would be easy to apply in practice (i.e. driven by spreadsheets, possibly contained in the STAR returns) even though the mathematics operating in the background may be complex. The industry calibration should balance the following opposing factors: a desire for greater complexity to allow as accurately as possible for the individual circumstances of companies a desire to keep the model as simple as possible to apply and use We believe that a spreadsheet-based model contained in statutory returns would achieve this. Such a model would allow companies also to test new levels of capital required should they consider expansion, merger or other management actions. We now describe the regulatory framework in more detail. The following graphical representation of the new solvency requirements, taken from a presentation by Ms Hantie van Heerden of the FSB in August 2005, applies regardless of whether a company uses the industry calibration, a certified model or an internal model: Free assets Fair value of assets Fair value of admissable assets Excess assets Minimum capital requirement M inimum o f R10m. Prescribed basis. Internal model method. Liabilities Consists of best estimate plus additional prescribed margins. Prescribed method or internal model method. Risk management Financial condition report This model indicates that the new framework would establish the following principles for a Financial Condition Report: Assets should be valued at fair value Some assets will continue to be inadmissible for solvency calculation purposes (e.g. art) A certain proportion of assets will be regarded as covering, or allocated to, insurance liabilities, or reserves. For this purpose, insurance liabilities will consist of: Claims Reserve, which in turn consists of: Page 6

Company Feedback FSB FCR Recalibration Incurred-but-not-reported (IBNR claims); and Outstanding reported claims The premium reserves, which consist of: The Unearned Premium Reserve (UPR); and Where appropriate and needed, the Additional Unexpired Risk Reserve (AURR) The OCR and premium reserves should be determined to be sufficient as best estimates of the appropriate liability The prescribed margin added to this takes the insurance liabilities up to the 75 th percentile 8. Once this has been done, the minimum capital requirement is determined in such a way that the total capital minus the prescribed margins will reflect a certain level of sufficiency: 98%, 99% or 99.5%. In other words, the total required capital (which would be sufficient at the 98 th, 99 th or 99.5 th percentile) minus the prescribed margins, would represent the minimum capital requirement, or MCR. For smaller companies, the MCR would be subject to a minimum of R10m. Excess assets are the admissible assets in excess of insurance liabilities (at a 75% sufficiency level), and the MCR must be covered by excess assets under this framework. In assessing capital adequacy, this framework was developed by considering a one year time horizon. In other words, we did not consider the probability of insolvency over 5 years, but looked at the probability that a company would be insolvent within a one year period if it held a certain level of capital. In our final report we describe the methodology followed and assumptions made to calibrate an industry capital requirement that would meet the above objective. Our calibration aimed to provide an MCR that would take into account insurance risks and asset risks, and the determination of insurance liabilities (or reserves). Throughout our analysis, we fitted the model such that the total capital requirement is set at three levels of sufficiency: 98%, 99% and 99.5%. The reason is that we wanted to measure the impact of different levels of sufficiency on the capital requirement, in order to allow the FSB to judge the conservatism or otherwise of its final requirement. In the recommendations in the final report, we also include a recommendation on the level of sufficiency that we believe should be adopted. 1.2 Global comparisons This section highlights some similarities and differences between our approach and that followed in other countries, particularly the UK and Australia. It must be noted that due to the multi-faceted nature of each country s regulatory solvency framework it is very difficult to make comparisons. Globally there has been a strong trend for regulators to move towards a risk based capital approach for short-term insurance (and other financial services). Our proposal and the FSB s requirements are in line with this regulatory trend. Amongst others, the following countries have regulatory regimes applying an RBC approach: Australia United Kingdom 8 Throughout this report, terminology such as 50 th percentile, or sufficiency at the 75% percentile level or 99% sufficient will be used interchangeably. All of these phrases express the same concept, namely that we can construct a probability distribution of outcomes for a company (i.e. a distribution reflecting the number of times out of many possible scenarios where a company s total capital would not be sufficient to meets its liabilities), and then measure the capital requirement at a point which would be sufficient to protect a company against insolvency half the time (set at the 50 th percentile), or against the 3 rd worst loss out of 4 (75 th percentile), or against the worst loss out of 200 (99.5 th percentile) Page 7

Company Feedback FSB FCR Recalibration USA Canada Germany Holland Switzerland Some broad similarities are the following: A value at risk (VAR) approach in determining capital requirements with a one year time horizon and a 99.5% level of sufficiency/confidence. A total balance sheet approach where risk in both assets and liabilities are considered Sufficiency in technical provisions set at a 75% level The framework addresses underwriting, reserving, credit and market risks with operational risk being difficult to quantify The allowance for internal models to be used It must be noted that there are still differences in the above from country to country. Some countries (most notably Switzerland) have adopted a Tail VAR approach as a measure of risk. It is recognised that Tail VAR may be mathematically more appealing since it is a consistent measure of risk. Despite this most countries have adopted a VAR approach to capital requirements. This may be due to the relative complexity of communicating tail VAR to stakeholders as compared to VAR. Our scope, as specified by the FSB, was to consider a VAR approach. In this regard we have produced results at the 99% and 98% levels of sufficiency as well as the 99.5% level. We have split our analysis into the eight classes of business set out in the Star returns. In the UK the comparable classes were further sub-divided into classes for proportional and non-proportional reinsurance. Unfortunately, the SA data available did not allow us to perform such a detailed analysis by reinsurance type. In Australia classes of business are grouped into three types that are felt to be broadly homogeneous and they also make allowance for a differential factor to be applied to reinsurance (without differentiating between proportional and non-proportional reinsurance). The allowance for extreme events is an example of an area handled differently by different regulatory models. For example, in Australia they have explicitly modelled extreme events and made allowance for them. In the Star return data we did not have the ability to separate extreme events from attritional losses and our calibration thus implicitly models these different types of losses together. We have taken the opportunity of allowing our insurance risk charges to vary by the size of the account. This is an outgrowth of the investigations we performed where we noticed a distinct trend for the results to reduce in volatility as company size increased. We have taken into account risk dependencies as per the International Actuarial Association s (IAA) global framework. We have built-in a company-specific allowance for diversification and correlation and our base charges by class of business allow for this. This may also make comparison between our stand-alone charges and those in other regimes potentially misleading. Our diversification and correlation factor approach is similar to a study on reserving in Australia where adjustments to the sum of stand-alone reserves are made with reference to a predetermined model. This model explicitly links certain factors (like number of lines of business and concentration of business within one line) to a recommended diversification and correlation factor. We have been more direct in allowing the company in question to calculate a factor more specifically attuned to their circumstances. Due to the complexity of the resulting formula this can only be achieved through providing a pre-programmed spreadsheet for companies to use for this calculation. Another principle from the IAA global framework is the allowance for risk management measures. We have allowed for these in the form of reinsurance and believe that coinsurance should be treated similarly. Unfortunately not all risk management measures are straightforward to quantify and calibrate Page 8

Company Feedback FSB FCR Recalibration on an industry-wide basis, in particular those of cell captives. Bearing this in mind, the application of the framework needs to be flexible enough to allow for such situations. The allowance for covariance effects between the insurance capital charge and the investment capital charge as well as the grossing-up of these charges is similar in nature to that allowed for in the South African Life Insurance CAR calculations. 1.3 Reliances and limitations This report is produced on instruction of and for the purposes of the Financial Services Board. It is based on data and information provided to us, which, although we checked and cleansed such information as far as possible, we cannot guarantee the accuracy of. The approach taken in this report is for the purposes of an industry calibration, and should be interpreted in the context of an industrywide regulatory framework. We make no guarantees on the effectiveness of this framework applied to a particular company or a group of companies and we accept no responsibility for the solvency of one or more companies in the industry where this framework is applied, as solvency can be impacted by many other factors that cannot be taken into account in an industry capital requirement, such as inappropriate management action, and the like. We do not accept responsibility for the application of the concepts explained here to a specific company and recommend that professional advice be obtained if a company wishes to determine an appropriate level of risk-based capital. The scope of this report is limited to an industry calibration for Financial Condition Reporting purposes. Whilst we make comments on the application of certified models and internal models, the details of such models and how they will be applied in practice are still to be determined and beyond the scope of this report. Further, we were at several stages of the project faced with inadequate data that prevented us from refining our calibration of the model. We therefore make some comments on areas where it may be appropriate to collect more data in STAR returns. It is beyond the scope of this report to make complete recommendations on the way in which STAR returns should be modified, and these comments and recommendations should be seen as preliminary. This report represents our recommendations to the FSB and does not necessarily represent the final format and way in which the capital requirements will be implemented in practice. It is beyond the scope of this report to consider other factors that may impact on the format in which the FSB finally decides to implement an industry calibration, such as Black economic empowerment within the industry. As discussed in s2.1 above, the FSB may well allow a company to meet capital requirements under a special dispensation. Page 9

Company Feedback FSB FCR Recalibration 2. Introduction This document presents an illustrative analysis in an attempt to quantify the impact of the proposed new solvency requirements on your company. The figures are based on data contained in your latest Star return submission and the framework proposed by us. This document follows on from: The industry workshop held at Deloitte on 11 August 2005 as well as the company specific calculations provided at the workshop. More detailed company specific feedback provided during September 2005 via email in response to requests for more detail The above documents were preliminary and aimed to encourage feedback from the various stakeholders in the short-term insurance industry. We did indeed receive valuable feedback from various stakeholders and have taken this into account in determining our framework and final recommendations. This document contains company specific figures as well as a brief explanation of the framework and will be better understood if read together with our final report to the FSB that will be published shortly. It must be noted that the figures are based on an application of the (default) prescribed model. Where companies feel that the prescribed model used below is not appropriate for their company, they will be able to submit alternative numbers as part of a certified model. The FSB will also consider alternative results that come from a full internal model. These two approaches were discussed in greater detail at the industry workshop and are further discussed in our final report. In short, we envisage that a certified model will involve modifications to the prescribed model that remain within the framework presented. On the other hand, an internal model would constitute some form of a dynamic financial analysis performed at the company specific level. Further, actual application of the framework is likely to produce different results to those presented here since we have applied Star return data in an automated fashion that may differ from the individual attention that will be given when an actual calibration is performed. In particular, the framework is forward-looking and requires estimates of financial variables for the coming year. To avoid subjectivity on our part we have taken all of these estimates at their current values. Note that there will still be several opportunities for the industry to give feedback on the new capital requirements to the FSB but that it would also be useful to receive as much comment from the industry as early in the process as possible. The following figures are examples of information from your Star returns (R 000): Net Written Premium: R61 000 Shareholders Funds (Statutory basis): R117 640 Please let us know if these figures are not accurate. Page 10

Company Feedback FSB FCR Recalibration 3. Potential Capital Requirements This section sets out our estimates of the proposed new solvency requirements for your company based on data contained in your most recent Star return submission. For comparison purposes it is suitable to use shareholders funds adjusted to allow for the release of contingency reserves since it is not envisaged that companies will be required to hold the MCR (as set out below) in addition to the contingency reserves. Further, shareholders funds may also be adjusted to allow for reserving on the prescribed basis. The details of this adjustment are provided in a later section. Shareholders Funds (Adjusted): R116 817 We propose 2 figures against which to compare the proposed new solvency requirements: 25% of Net Written Premium, no minimum capital R15 250 25% of Net Written Premium, R10m minimum capital R15 250 The second figure represents the existing regime together with a related proposal to implement a ten million Rand minimum capital requirement for all companies. The prescribed margins are quantified as that amount necessary to bring reserves from a best estimate to a 75% sufficiency level. This applies to both premium reserves and claim reserves. Prescribed Margins: R2 701 Part of the total capital requirement can be met by prescribed margins with the difference between the total capital required and the prescribed margins being the minimum capital requirement (MCR). Page 11

Company Feedback FSB FCR Recalibration The following potential capital requirements were determined for your company. These can be compared to shareholders funds and the current statutory solvency requirements as set out above: At the 99.5% level of sufficiency: Minimum Capital Requirement R105 117 Total Capital Requirement R107 818 At the 99% level of sufficiency: Minimum Capital Requirement R83 992 Total Capital Requirement R86 694 At the 98% level of sufficiency: Minimum Capital Requirement R65 264 Total Capital Requirement R67 965 We will elaborate on the calculation of the MCR at the 99.5% sufficiency level for your company in the rest of this document. Page 12

Company Feedback FSB FCR Recalibration 4. Insurance Capital Charge The primary building block of our proposal is the gross stand-alone risk capital for each class of business written. This is the capital required for a class of business in isolation, for the risk of claims exceeding premiums and as such resulting in an underwriting loss. In particular it does not take into account reinsurance, commission, expenses or the diversification effects of writing multiple lines of business. This is allowed for later. Further, a subtle point that has caused some confusion is that the term gross stand-alone risk capital is a slight misnomer since the volatility has been calibrated using net data 9. This is important in light of the application of reinsurance in subsequent steps. The capital level for each class of business is calculated according to the level of gross written premium (GWP) and gross unearned premium (GUPR) in each class of business. The gross standalone risk capital per line of business is read from tables based on predetermined levels of GWP and GUPR (and interpolated accordingly). The figures in the above mentioned table are calibrated from data in historic Star returns and to the results of a DFA model and take account of: the varying levels of risk and reward between different classes of business the diversification benefits achieved by companies of varying size within each class of business Application of the above result in the following Gross stand-alone risk capital requirements for your company: Class of business Gross Stand-Alone Risk Capital Accident R32 577 Engineering R0 Guarantee R0 Liability R58 794 Miscellaneous R0 Motor R107 046 Property R119 058 Transport R0 These numbers are then adjusted for reinsurance to give Net stand-alone risk capital requirements for each class of business written. The figures above are multiplied by a retention percentage actual net written premium divided by gross written premium for each class of business. Allowance for reinsurance in this manner implicitly gives credit for non-proportional reinsurance in equal value to the premium paid for such reinsurance, relative to the gross stand-alone risk capital 9 If we had calibrated to gross data then we would expect that the gross stand-alone risk capital would be higher. In fact, gross data at the level of detail required was not available in the Star returns. Page 13

Company Feedback FSB FCR Recalibration above. We realise that this may not be completely accurate for companies with an extensive nonproportional reinsurance programme, however we are constrained by a lack of data in the STAR returns in this regard. This is an area where a certified model or internal model would help a company to calculate more accurate figures. The retention figures used are shown in the following table: Class of business Retention% Accident 50.00% Engineering 0.00% Guarantee 0.00% Liability 40.00% Miscellaneous 0.00% Motor 50.00% Property 35.00% Transport 0.00% Actual net expenses and commission for each class of business are added to the net stand-alone risk capital to yield a net stand-alone capital. In this case the figures used for expenses plus commission are as follows: Class of business Net Expenses and Commission Accident R2 800 Engineering R0 Guarantee R0 Liability R3 360 Miscellaneous R0 Motor R9 188 Property R5 145 Transport R0 Page 14

Company Feedback FSB FCR Recalibration The resulting Net stand-alone capital is given below: Class of business Net Stand-Alone Capital Accident R19 089 Engineering R0 Guarantee R0 Liability R26 877 Miscellaneous R0 Motor R62 710 Property R46 815 Transport R0 Total R155 492 The total Net stand-alone Capital in the table above does not take into account the following two effects from writing multiple lines of business: Diversification effects due to writing more than one class of business (these will in general reduce the capital required) Correlation effects between the different classes of business (these will in general increase the capital required since classes tend to be positively correlated) These effects are taken into account by multiplying the sum of the net stand-alone capitals above by a statistically determined factor that allows for the above effects on your relative mix of business: Total Net stand-alone Capital R155 492 Correlation and Diversification factor 61.45% Insurance Capital charge (allowing for diversification and correlation) (R 000) R95 557 Page 15

Company Feedback FSB FCR Recalibration The last step for the Insurance Capital charge is making allowance for investment returns on assets backing relevant liabilities. In this case the relevant liabilities are the claims reserves 10 (OCR and IBNR) and premiums reserves (UPR, URR, etc ). We have performed an asset allocation exercise to your Star return data (discussed more fully in the next section) and the following mix of assets backing the relevant liabilities resulted: Asset Class Rand Value Cash R19 273 Near cash R0 Fixed Interest R0 Property R0 Equity R0 Other R0 Total R19 273 For the purposes of these illustrative calculations we have applied a flat return of 8% per annum to the value of assets above. The final insurance capital charge is attained by subtracting the allowance for investment income and capital gains from the insurance charge (allowing for diversification and correlation) above. Insurance Charge (before investment return) R95 557 Investment Income and Capital Gains R1 542 Insurance Capital Charge required R94 016 10 Claims reserves used here are adjusted for calculation of reserves on the prescribed basis Page 16

Company Feedback FSB FCR Recalibration 5. Asset Capital Charge We now turn our attention to the charge for investment capital. In this section we rely on the results of a rule-based allocation of assets to liabilities, performed on information in your latest Star return. In general, allocations were made that resulted in the lowest asset capital charge and/or the most appropriate match of assets to liabilities by nature and term. The asset capital charge focuses on the mix of assets backing the relevant liabilities (claims reserves and premium reserves). The charges are calculated to allow for a fall in the relevant asset class at the 99.5% level. The following charges are made in general for a 99.5% level of protection: Asset Class Asset Charge Cash 0% Near cash 0% Fixed Interest 11 11.25% Property 24.63% Equity 38% Other 38% An allocation of assets to other liabilities (mainly current liabilities) is performed before the allocation of assets to the relevant liabilities discussed above. For your company, the mix of assets backing the relevant liabilities was deemed to be as follows: (this asset split also determined the allowance for investment income and capital gains in the previous section) Asset Class Asset Split Cash R19 273 Near cash Fixed Interest Property Equity Other R0 R0 R0 R0 R0 11 The proposed framework allows for varying charges on fixed interest securities of different duration. Unfortunately, data in the Star returns did not allow us to separate out fixed interest holdings at this level of detail. Page 17

Company Feedback FSB FCR Recalibration This results in the following charges against the assets backing the relevant liabilities. The sum of these charges is the overall asset capital charge. Asset Class Cash Near cash Fixed Interest Property Equity Other Total Rand Value R0 R0 R0 R0 R0 R0 R0 Page 18

Company Feedback FSB FCR Recalibration 6. Total Capital Charge and Minimum Capital Requirement This section details how the two capital charges are combined. This involves combining the insurance capital charge and the asset capital charge allowing for the following: A fall in the value of assets backing the capital requirements The covariance effects (diversification and correlation) between the asset capital charge and the insurance capital charge The former is allowed for by grossing-up both charges by appropriate amounts to allow for a fall in the value of the capital charges. The latter is allowed for using the heuristic rule of summing the squares of the two capital charges and taking the square root (this is less than the sum of the charges and allows for the fact that companies are not likely to experience a worst-case asset event and worstcase insurance event at the same time). The following formulas set out how we envisage all of the above will be achieved in practice: ACC = Asset Capital Charge (as calculated above) ICC = Insurance Capital Charge (as calculated above) g 1 = Grossing-up factor on asset charge g 2 = Grossing-up factor on insurance charge TCR = Total Capital Required TCR = ACC g 1 2 + ICC g 2 2 The grossing-up factors are calculated via an intermediate calculation described below. This step involves the performance of an asset allocation (after the allocation of assets to current liabilities and reserves, described above) to adjusted values for the asset capital charge and the insurance capital charge. These adjustments are given below 12 : TCR_ADJ = Intermediate total capital required (before grossing-up) ACC_ADJ = Adjusted Asset Capital Charge ICC_ADJ = Adjusted Insurance Capital Charge 2 TCR _ ADJ = ACC + ICC 2 ACC _ ADJ = TCR _ ADJ ACC ACC + ICC ICC _ ADJ = TCR _ ADJ ICC ACC + ICC 12 The reason for these adjustments are so as not to unnecessarily penalise the composition of companies shareholders funds not being used to back their capital requirements Page 19

Company Feedback FSB FCR Recalibration For your company these adjusted capital charges were: Adjusted Charge Value TCR_ADJ R94 016 ACC_ADJ R0 ICC_ADJ R94 016 An asset charge (calculated on the same basis as the asset capital charge in the previous section) is calculated for the allocation of assets to ACC_ADJ and ICC_ADJ. These two charges are the weighted average fall in assets that could result, and are calculated at an appropriate level of sufficiency. c 1 = asset charge on ACC_ADJ where 0 < c 1 < 1 c 2 = asset charge on ICC_ADJ where 0 < c 2 < 1 The resulting grossing-up factors are calculated as follows: g 1 = 1- c 1 where 0 < g 1 < 1 g 2 = 1-0.5*c 2 where 0 < g 2 < 1 This resulted in the following charges and resulting grossing-up factors for your company: Charge c g Assets (1) 0.00% 100.00% Insurance (2) 25.60% 87.20% The rationale for the above is that for the asset capital charge, full grossing-up should be allowed for as you will need a grossed-up asset charge in precisely the situation that you need the asset charge itself. The grossing-up of the insurance capital charge only takes half of the appropriate asset charge into account since a worst case insurance event will not always happen at the same time as a worst case asset event. The use of a factor of a half can be seen to be allowing for a 50% correlation between insurance catastrophes and investment market crashes, which is in line with the intended practice in European markets. Calculation of the grossing-up factors in this section (g1 and g2) and the capital charges in previous sections is used in the formula for total capital required (TCR) at the beginning of this section. This yields the following: Adjusted Charge Value Total Capital Required R107 818 Total Capital Required (R10m minimum) R107 818 Prescribed Margins R2 701 MCR R105 117 We envisage that companies will receive a credit towards their total capital requirement equal to their overall prescribed margins (prescribed margin from claim reserves and prescribed margin from premium reserves). We expand more on the calculation of this prescribed margin in the following section. Page 20

Company Feedback FSB FCR Recalibration 7. Reserving This section gives more information on the application of the prescribed formula for reserves and for the calculation of prescribed margins. Note that our framework (with respect to reserves) has changed materially since the last feedback document. Where companies feel that the prescribed model is not appropriate given their particular circumstances they may submit alternative estimates from a certified model. Incurred but not reported (IBNR) We have calibrated a formula to industry data in the Star returns to give a prescribed best estimate of IBNR reserves for each class of business. IBNR reserves (gross and net) are obtained by multiplying earned premium (gross and net) by an appropriate percentage varying by class and development year. For this purpose, earned premium for the last six historic years was considered. The following table contains the best estimate IBNR reserve for each class: Class of business IBNR Best Estimate Accident R1 090 Engineering R0 Guarantee R0 Liability R2 719 Miscellaneous R0 Motor R1 417 Property R943 Transport R0 Total R6 168 Note that the best estimate reserve for each class is made up of six components an amount of earned premium for each historic year multiplied by an appropriate factor which varies by development period. Outstanding reported claims Companies will continue to estimate their own outstanding reported claims and will be required to give a best estimate of outstanding reported claims for use in the reserving framework. Page 21

Company Feedback FSB FCR Recalibration We have used the current levels of OCR as a proxy for these best estimates. This may or may not be appropriate depending on the level of prudence contained in the current OCR levels. These OCR levels (obtained from your Star returns) are given below: Class of business OCR Best Estimate Accident R300 Engineering R0 Guarantee R0 Liability R420 Miscellaneous R0 Motor R500 Property R280 Transport R0 Total R1 500 Prescribed Margin: Claims reserves The use of prescribed margins aims to yield reserves that have a 75% level of sufficiency. We envisage application of a recommended prescribed margin formula to claims reserves calculated on a best estimate basis. It must be noted that all prescribed margins can be offset against the capital requirements calculated above. The sum of the best estimates of IBNR and the outstanding reported claims yield the best estimate of claims reserves. We then apply the prescribed margin formula to adjust this best estimate to a 75% level of sufficiency. Application of the proposed formula results in the following: Class of business Prescribed Claims Reserves Accident R1 774 Engineering R0 Guarantee R0 Liability R3 949 Miscellaneous R0 Motor R2 500 Property R1 700 Transport R0 Total R9 923 Page 22

Company Feedback FSB FCR Recalibration Prescribed Margin: Premium reserves For premium reserves (UPR, URR, etc) it is felt that the degree of prudence contained using current estimation techniques (365ths method for instance) may be broadly equivalent to that required through the application of prescribed margins. The primary reason for this is that an unexpired premium reserve implicitly contains some profit margin. Where a company has specific knowledge that their premiums are inadequate, appropriate prudence would need to be borne in mind when setting an additional unexpired risk reserve. This URR would also form part of the insurance liabilities and would have to be set at a 75% level of sufficiency. Since the calculation of the URR would depend on the context and the specific circumstances of the company, it is not appropriate to prescribe a formula for the URR. For the purposes of our industry impact analysis we have thus assumed that the premium reserves inherently contain a 75% level of sufficiency. Calculation of the prescribed margin for premium reserves is thus used to quantify the credit towards required capital only. Application of the prescribed margin formula for premium reserves results in the following results for your company: Class of business Accident Engineering Guarantee Liability Miscellaneous Motor Property Transport Total UPR Prescribed Margins R225 R0 R0 R221 R0 R0 R0 R0 R447 Prescribed Margins The total prescribed margins are those contained in the claims and premium reserves above and amount to the following in your case Reserve Prescribed Margin Claims reserves R2 255 Premium reserves R447 Total prescribed margins R2 701 The total prescribed margins are used as a credit toward the total capital required to yield the minimum capital required (MCR) as described in the previous section. Page 23

Company Feedback FSB FCR Recalibration 8. Conclusion Should your company use the prescribed methods to calculate capital and reserves then it is suitable to compare the MCR to the shareholders funds adjusted for contingency reserves and the prescribed reserving basis described above. If the MCR is greater than the adjusted shareholders funds then additional capital is required. Shareholders funds (adjusted) R116 817 At the 99.5% level of sufficiency: Minimum Capital Requirement (MCR) R105 117 Additional capital required (if any) R0 At the 99% level of sufficiency: Minimum Capital Requirement (MCR) R83 992 Additional capital required (if any) R0 At the 98% level of sufficiency: Minimum Capital Requirement (MCR) R65 264 Additional capital required (if any) R0 Page 24

Company Feedback FSB FCR Recalibration 9. Recommendations In this section we do not repeat the detailed description of the DFA model behind the industry calibration. Our main recommendation is that the industry calibration should be done in accordance with the methodology, assumptions and parameterisation contained in the rest of this report. Given this main recommendation, it is worth highlighting a number of specific recommendations. We do not discuss the reasoning behind these recommendations as this is already contained in detail in the rest of the report: STAR returns should be expanded to collect more data on reinsurance, and particularly on nonproportional reinsurance, cell captives and co-insurance. Such reinsurance data may be used to refine the calibration of the model in future. We recommend that no explicit allowance be made for the underwriting cycle. We recommend that no explicit capital requirement be determined to cover operational risk. Specific attention should be given to the development of certified model frameworks for cell captives, reinsurers, niche insurers and companies in run-off. These certified model frameworks should provide guidance on how elements of the industry calibration may be used in practice and the standards that should be applied by companies when modifying elements for their specific requirements. We recommend that the industry calibration would not be applied blindly to companies in the above sectors, and that further consultation is appropriate with these specific sectors of the market. Despite the fact that there may be shortcomings in the industry calibration for companies in specific sectors of the market, we recommend that the FSB use the proposed framework as a benchmark. A framework for internal models has to be established by the FSB. In determining the total capital charge using the square root formula, we recommend that grossing up on the Insurance Capital Charge be limited to 50%, to avoid the overall requirement being too onerous. We recommend that the FSB implements the capital requirement in such a way that it would be at the 98% sufficiency level on implementation, 99% after four years and 99.5% after another two years. Until such time as the implementation of the capital requirement at the 99.5% level, the FSB should collect additional data in STAR returns to ensure that any necessary refinements can be made at a later stage if deemed appropriate (in particular more detailed allowance for nonproportional reinsurance may be required). The FSB should determined and specify its intervention points. For instance, we suggest that, if a company has capital of more than 1.1 times the MCR, the FSB need not take any regulatory action. However, if the capital falls to between 1 and 1.1 times the MCR, the FSB should take some action, such as requesting a business plan from the company to discuss how it will manage its capital. Only when the capital falls below the MCR the FSB would intervene more directly. The intervention points would have to be specified separately and specifically during the transitional period. Every company falling short of the minimum requirement would have to discuss with the FSB: Whether they will use an internal model or certified model to determine different capital requirements Whether the company would apply for a special dispensation justifying why it would not meet any of the capital requirements determined on the above basis. It is worth emphasising that Page 25

Company Feedback FSB FCR Recalibration the FSB has indicated that it would be willing to consider and review applications for special dispensations from companies who feel that their individual circumstances warrant such dispensations. How the company would reach the level of capital agreed with the FSB The industry calibration contains complex calculations and formulae. To make it easier for companies to apply in practice, we recommend that the FSB issues a spreadsheet-based model to companies that would automate the calculation of the MCR, and that this become part of the STAR returns. Finally, this report should be regarded as a step towards the specification of a Financial Condition Reporting framework for South African Short Term Insurers. The FSB now has to decide on how to take the recommendations in this report forward, and we recommend that there would still be considerable discussion with the industry on the way in which the model would be implemented. Page 26

Company Feedback FSB FCR Recalibration Should you have any queries or anything further you may wish to discuss please do not hesitate to contact us on any of the following numbers: Emile Stipp Tel 011 209 8102 Fax 011 209 8200 Cell 082 336 5170 e-mail estipp@deloitte.co.za Sam Isaacson 011 209 8159 011 209 8200 073 190 1978 saisaacson@deloitte.co.za Hillary Murashiki Tel 011 781 1614 Fax 011 781 1749 Cell 073 797 3244 e-mail othillary@workersacademy.co.za Page 27