Bankruptcy Navigator Index 2.0 Customer Handbook Prepared by: Larry Macdonald, Sr. Product Manager 20-Dec-2012
Table of Contents Introducing the Bankruptcy Navigator Index 2.0... 3 1. Modeling Concepts... 3 1.1 Population of Interest / Minimum Scoring Criteria... 4 1.2 Response Variable... 4 1.3 Available Data / Independent Variables... 5 2. The BNI 2.0 Model... 5 2.1 Segmentation... 5 2.2 Modeling Technique... 6 2.3 Scaling... 7 2.4 Attributes... 7 3. Output... 9 4. Evaluating a Model... 9 5. Using BNI 2.0 with ERS... 12 5.1 Dual Matrix Strategy... 12 5.2 Refreshing Scores for Account Management... 14 Summary... 14 Appendix Reason Codes, Scorecard Indicator, Reject Codes... 16 Reason Codes... 16 Scorecard Indicators... 18 Reject Codes... 18 2012 Equifax Canada Co. All rights reserved. Page 2
Introducing the Bankruptcy Navigator Index 2.0 Risk managers have a difficult task. Their companies are in the business of lending money or granting credit to ordinary Canadians who need credit cards for everyday purchases, loans to buy goods and services, lines of credit to optimize and bring flexibility to credit management, and mortgages so they will have a place to live. Unfortunately, it is not profitable for these companies if their customers don t pay them back. A credit file, supplemented with a credit score, gives a risk manager the ability to assess the likelihood that a customer will meet their financial obligations to make regular payments on the credit that they use. Credit scores are systematic and predictive, enabling the application of consistent business rules. Low risk customers may receive better product offerings, better terms, or higher limits, while high risk customers may be required to provide securitization or may not be offered credit at all. Most Canadian consumers regularly pay their bills and have established good credit histories. These are profitable consumers for the lending institutions in the financial industry. Unfortunately, the consumers who are non-payers cause significant losses for the lenders, resulting in increased interest rates for all. Identifying these consumers and mitigating their losses in a timely manner makes the entire lending practice more efficient for the lender and borrower alike. The Equifax Risk Score (ERS) is our most accurate solution in the prediction of consumer delinquency risk. Consumers with low scores have a high probability of going 90 days past due or worse on their debt obligations over the next 12 months. ERS uses such credit file characteristics as delinquency, utilization and balances, inquiries, public records, and the ages and types of credit products, to assess delinquency risk. Bankruptcy and delinquency clients portray different behaviours. Delinquent consumers are in moderate financial difficulties, and this is often a long-term chronic behaviour. They manage credit poorly, and occasionally miss payments, sometimes with multiple missed payments leading to collection efforts and write-offs. Consumers heading into bankruptcy are in severe financial difficulty. They often continue to make payments any way they can. They make smaller payments on their credit cards as their balances rise. They tap into their line of credit to make their instalment loan payments. They shop for additional credit cards to use, a new line of credit, and consolidation loans. Eventually they reach their limit on their credit cards and lines of credit, and are forced to declare for bankruptcy, but because they have not missed payments, their delinquency score remains high. Nationally, about 70% of consumer accounts are paid as agreed when the consumer files for bankruptcy. The Bankruptcy Navigator Index (BNI 2.0) weighs the information in the credit file that is predictive of bankruptcies. When used with a delinquency score like ERS, BNI 2.0 can help companies reduce losses while maintaining volumes, or increase new accounts at the same loss level. BNI 2.0 identifies those consumers who are in financial difficulty and yet not missing payments. 1. Modeling Concepts BNI 2.0 is a predictive model offered by Equifax that risk managers use to help determine which customers or applicants share credit characteristics consistent with consumers in the past who later filed for bankruptcy. These predictive characteristics can be used to estimate the bankruptcy risk of any other individual consumer. 2012 Equifax Canada Co. All rights reserved. Page 3
In order to build a model, three key components are required: population, outcome, and data. The population is the collection of records used to build the model, and should be representative of the population where the model will be used. The outcome is the value to be modelled, and represents the unknown quantity that the model predicts. The data are the known attributes that are available at the time and used in the calculations. 1.1 Population of Interest / Minimum Scoring Criteria There are two considerations necessary to define a population for a credit model. The first is to decide on the target population where the model is of interest. For BNI 2.0, the ideal population may be any Canadian consumer. Assessing the bankruptcy risk of every prospect or existing customer is the best case. The second consideration is available information. It may be difficult or impossible to make an accurate assessment of risk given the sparcity of data in some cases. Among the common limitations: Credit File Not Found. Sometimes an applicant does not have a credit file. This includes young Canadians who apply for credit for the first time, or new immigrants. Some people may go for many years without building a credit history, preferring to pay cash for everything so as not to owe any money, or they use their family s credit (parents or spouse) for their own needs. A change in their family status, such as death or divorce, forces them to establish their own credit. Additionally, a credit file may not be found in cases where there are mismatches between the information in the credit file and the application. This could be due to typographical errors, information variants such as different name versions like Robert and Bob, revised information that hasn t been updated on the credit file like new address or change of name, or format errors, such as having the input data in the wrong fields, or supplying an address out of the country. Death Notice on File. Although the estate of a deceased individual may be responsible for the financial commitments in some cases, the information in the credit file may not be predictive of the future performance of the account. Inactivity. If a credit file has not been updated for a period of time, the information that it contains may be stale. In many cases, it may be more accurate for an institution to make lending decisions based on other information, such as income statements or wealth (including property equity), by providing collateral or security, or having a co-signer. The combination of the target population and the available information to generate a score is known as the minimum scoring criteria. For BNI 2.0, the minimum scoring criteria is all Canadian consumers with credit activity within the last 24 months. Activity is defined as a trade line updated (based on the date reported) or a hard inquiry 1. Credit files with death notices are excluded from the scoring criteria as well. 1.2 Response Variable Credit scoring is done to help risk managers understand how likely it is that their customers are going to make the required payments on their credit products (loans, credit cards, lines of credit, mortgages, etc.). The model requires 1 Inquiries are posted whenever somebody views or receives information contained in the credit report. Hard inquiries indicate that a consumer has applied for credit and granted permission for someone to see their credit report for the purpose of adjudicating that credit. Soft inquiries are posted whenever a company refreshes information about their customers, but are not motivated by consumer activity. Soft inquiries are only visible to the consumer, and do not affect any credit scores. 2012 Equifax Canada Co. All rights reserved. Page 4
taking a representative sample from the target population from a recent archive period (known as the observation point), and then defining and calculating the response variable; consumer credit files are observed at a more current period (the performance period, 24 months after the observation period for BNI) to determine whether there has been negative behaviour. For BNI 2.0, this is defined as a bankruptcy on file 2 within the performance window. For the development of BNI 2.0, records were chosen from time periods (observation and performance) spaced 24 months apart, April 2000 and April 2002. While BNI 2.0 was developed as a score that predicts a bankruptcy on file within 24 months for a consumer, BNI 2.0 is also very predictive of other similar outcomes, such as a bankruptcy on file within 12 months rather than 24. BNI 2.0 can be used to predict the likelihood of a consumer going bankrupt for consumers within an individual portfolio, or on prospects or new applicants, and is therefore an excellent tool to use both at adjudication / acquisition as well as account management. 1.3 Available Data / Independent Variables A model estimates an unknown quantity by developing the relationship between the known attributes and the required outcome (the performance as defined in the previous section). The known attributes have to be available at the time that the unknown quantity is to be estimated. The statistical term for these attributes is independent variables. For BNI 2.0, the attributes come from the credit file. Equifax Canada Risk Modelling Segments (RMS) consist of over 400 proprietary credit file attributes covering a wide spectrum of credit file characteristics including delinquency, utilization and balances, inquiries, public records, and the make-up of the wallet. These segments include many industry-specific attributes as well as some that are aggregated for all trades or inquiries. 2. The BNI 2.0 Model The BNI 2.0 model returns a three-digit numerical score that corresponds to the bankruptcy risk for the individual consumer with the given credit file information. Consumers with high credit scores are less likely to have bankruptcies than consumers with low scores. This section discusses properties of the score: how it is built and how the results can be interpreted. 2.1 Segmentation The predictability of a model is often greatly enhanced by segmenting the population into a number of subgroups, and creating a different predictive formula in each segment. Different formulas may be needed because there may be differences in the availability of data for certain parts of the population. For example, there is no need to have attributes for public records in all formulas if there are different segments for consumers with and without public records. Another reason is that different business decisions may apply, such as if companies have different strategies for consumers who are new to credit. A third reason for segmentation is that there may be certain subgroups of the population for which there is a different relationship between the modeling attributes and the outcome. 2 Bankruptcy on file includes either a new bankruptcy public record or a trade line reported as included in the bankruptcy filing. 2012 Equifax Canada Co. All rights reserved. Page 5
BNI 2.0 uses a segmentation scheme based on presence of a previous bankruptcy, the number of trades on file, and the age of the oldest trade. A total of six segments are defined. A different formula is to be applied to each, so that the attributes can predict the outcome over each segment. These distinct formulas are called scorecards. Since there is a direct relationship between the segment and scorecard, the two terms are commonly used interchangeably 3. Some credit files have robust data, with a long credit history and a large number of trades, and the future performance of the consumer can be accurately assessed with great confidence. On the other hand, when the open date of the oldest trade is recent and/or the number of trades is few, the consumer doesn t have a robust credit history and there isn t a lot of information that can be used in identifying if these consumers are good credit risks. These files are often called thin files. Consumers with good but short payment histories may be considered low risk for continuing payment and obligations with the credit that they already have, but may not be as low risk for new credit granted. They may be able to handle payments on their existing accounts, but introducing a new credit product may require new payments, and the consumer has yet to provide proof of the ability to handle these additional responsibilities. In addition, new accounts added to a thin file indicate a change in behaviour, and adds uncertainly and risk to an estimate of the probability of bankruptcy. Identifying the segment for these consumers will help a risk manager deal with these cases. 2.2 Modeling Technique Within each segment, a logistic regression model is developed. Logistic regression is a modeling technique designed to model the relationship between a binary 4 outcome and the explanatory variables. For each variable, a weight is assigned, multiplying the weight by the value, or by giving a set number of points for each possible value of the variable. In addition to the logistic regression models, there is a neural network. Neural networks are statistical models that look at combinations of variables, and these combination variables are used to build the model. Neural networks can be highly predictive when the event that is being predicted is a rare event or when the amount of available information is limited. With BNI 2.0, the predictability of the logistic regression models is enhanced by combining the logistic regression result with the result of the neural networks. The two results are blended using a methodology called score fusion, transforming them into one estimate of the likelihood of bankruptcy. The Inquiry Only segment uses a decision tree algorithm to segment the files by bankruptcy risk. The bankruptcy rates observed in the nodes of the decision tree were used as the model probability for this segment. After score fusion, the formulas applied to each credit file have derived a probability of bankruptcy. These formulas have been proven accurate, in the development dataset, by taking all of the records with similar probabilities and calculating the observed bad rates and comparing them to the expected bad rate. 3 There is an additional scorecard developed for the population for which there are no trades or public records. This inquiry only sub-population uses the inquiries in the credit file to generate the score and is handled separately from the other scorecards. 4 Binary outcomes are those that can take two values. These are often denoted by yes and no, or true and false. They are represented in code as 0 and 1. 2012 Equifax Canada Co. All rights reserved. Page 6
2.3 Scaling BNI 2.0 uses a 1 to 999 scale 5 rather than a probability estimate, where a high score indicates a low bankruptcy risk. The value does not correspond directly to a probability; a 950 BNI does not imply a 95% probability of not filing for bankruptcy. The key is that the score should separate and rank order bankruptcy risk. This means that here should be a relationship between the score and the bankruptcy rate. Consumers with low scores have a high bankruptcy rate, while consumers with higher scores should have a lower bankruptcy rate. The relationship should be consistent, and there should be a large difference in bankruptcy rates between consumers with the best and worst BNI scores. Companies should validate scores on their own portfolio in order that their risk managers can customize and optimize their strategy for their own business needs. Over time, changes in lending policies, reporting policies, lending institutions, consumer behaviour, and the economy can change the relationship between scores and the expected bad rate. BNI 2.0 will continue to perform in the future and continue to rank order bankruptcies, but the bad rate at various scores may shift if the data or the economy changes significantly. Risk managers should monitor their portfolios regularly and determine if decisions should be implemented at a different score. 2.4 Attributes The credit file attributes that are included in the final BNI 2.0 model are those credit file characteristics that are found to be predictive of future bankruptcy. These tend to be the same characteristics that risk managers consider when they look at a credit file. They can be classified into a number of categories: Utilization and balances. This is the most important factor in predicting bankruptcies. Consumers with acute financial stress are unable to pay off their bills. In the early stages, they may be able to make minimum payments on credit cards but not pay as much as they did before, and as their balances increase towards the top of their credit limit, and the danger of maxing out their cards becomes a reality. At that point, some consumers tap into their line of credit to pay down their high interest cards and make the payments on their instalment loan obligations. The balance owed is one factor; consumers usually only file for bankruptcy when they owe a lot. Utilization is another related factor. Utilization is the ratio between balance and high credit. Low utilization indicates a difference between the balance and high credit, an amount often called open to buy. A large amount open to buy gives consumers flexibility, a way to pay bills for a short time that can t be covered by income, by tapping into their lines of credit. A small amount of open to buy reduces the incentive to make a payment on a credit card or line of credit since they won t be able to use much of it anyway. And once the credit cards are maxed out and the lines of credit are used up, they may not be able to cover their payments. Bankruptcy may become their only option. 5 Many risk managers are used to a 300 to 900 range for credit scores such as ERS or BEACON. BNI 2.0 deliberately uses a different score range, as it measures a different behaviour. The BNI 2.0 score distribution is highly skewed to the high end, with most consumers scoring above 900, which is higher than the upper limit for delinquency scores. However, caution is to be exercised when BNI dips below 900. Risk manager who see a 775 BNI 2.0 score, for example, may initially think that this is a good score, falling back on experience with other scores. However, more than 90% of the Canadian population scores higher than 775, so this should not be considered a good BNI 2.0 score. By viewing the details within the credit report, the risk manager is likely to find some areas of concern which lead to the low BNI 2.0 score. 2012 Equifax Canada Co. All rights reserved. Page 7
Inquiries. Consumers who are going through financial difficulties, whether through job loss, family or health situations, or general financial woes, often look for additional credit products to provide additional open to buy. They may apply for a loan to pay down the credit card they have maxed out, and try to get a new credit card. The inquiry may be the leading indicator, the first sign of danger that appears on the credit file. Of course not every inquiry is a sign of financial difficulty, and only a number of inquiries, in combination with other warning signals should lead to a significant decline in a credit score. Consumers sometimes shop around when they are looking for certain products, and multiple inquiries over a short period of time can be considered as shopping for one product. Three mortgage inquiries in a week rarely means that a consumer is trying to buy three houses, while three credit card inquiries may mean that they are going to have three new credit cards. Mortgage inquiries, auto finance inquiries, and Telco inquiries are deduped, meaning that multiple inquiries within 30 days count as one inquiry in the calculation of the score. In addition, inquiries within the first 30 days do not count in the score calculation, allowing consumers a chance to shop at different places without there being an advantage to the first lender that pulls a file that receives a score with no inquiries counting. Public records. Those who have a prior history of bankruptcy, or have had collection issues or other derogatory public records may be considered risky. The presence of these events, though relatively rare, has a significant negative impact on a credit score. Payment history. Missing payments on a credit file is always of concern to a lender. In many cases, they indicate a difficulty in paying bills, perhaps due to cash flow problems, or mismanagement of the individual or family budget. Some consumers miss payments due to sloppiness in paying bills rather than an inability to do so, but even this indicates a lack of responsibility and may lead to a higher expectation of bankruptcy. However, the majority of consumers who file for bankruptcy are not chronically delinquent. Overall, an Equifax study has shown that about 70% of the outstanding balance shown on a credit file from the month prior to the date when the bankruptcy was file is on accounts that are paid as agreed. Credit history. A consumer who has managed credit for many years is considered lower risk than someone new to credit. Over a long period of time, many people have significant life events, including moving, changing or losing job, marriage or divorce, having children, and serious illness or injury. Those who have experienced these events in the past and have not filed for bankruptcy have a strong likelihood to continue doing so. They have shown responsibility and consistency over a long period of time and can be considered a good credit risk. On the other hand, young Canadians or new residents, new to credit, who have managed a first credit card with a small credit limit have not demonstrated a history of managing large amounts of debt and haven t proven the ability to make regular payments over several years (such as a mortgage or auto loan). They are less likely to have dealt with significant life events, and if one occurs, they have not demonstrated the ability to deal with it and maintain their financial responsibilities. Also in this category is the number and type of accounts. Consumers with many different accounts may be of higher risk, and the properties of some trades indicate a higher risk than they do in other industries. For example, high utilization in a line of credit may be a risk factor, while high utilization of a loan just means that it is a new account and the consumer hasn t had a chance to pay off much of the balance. 2012 Equifax Canada Co. All rights reserved. Page 8
The presence of many new accounts may also be an indication of higher risk. For one thing, consumers in financial difficulty often try to open new accounts in order to extend their open to buy and continue to pay their bills through the tough times. Additionally, a number of new accounts may indicate that something has changed, and this brings uncertainly to the risk prediction, which in turn means higher risk to the lender. 3. Output Along with the three digit numeric BNI 2.0 score, four reason codes are provided that help explain why a consumer file has the score that it does. Reason codes correspond to the attributes that have values that lower the score, such as high balances or utilization, delinquent accounts, or excessive inquiries. The first three reason codes correspond to the three attributes from the credit file whose values lower the score by the largest amount. The fourth reason code denotes the scorecard, or subpopulation, indicator. The identification of the subpopulation used in the score may be a strong indication of the score. For example, the subpopulations defined by delinquency or public records tend to score lower than the subpopulations without them. In addition, the scorecard indicator highlights files with limited credit information. In many cases, these thin files may be low risk for the accounts that they already have, but would be high risk if they were given a new and large loan, line of credit, or mortgage. If they suddenly had a credit vehicle to fall tens of thousands of dollars or more in debt, they may not have the ability to dig themselves out of that hole, which would increase their bankruptcy risk. Risk managers can use the scorecard indicator to automate different credit policies for consumers belonging to different subpopulations. For credit files that do not meet the minimum scoring criteria, a reject code or message is returned to explain why a score wasn t returned. Companies can use the reject code to guide them in deciding how to make a decision when the credit score is not available. See Appendix for a complete list of reason and reject codes for BNI 2.0. 4. Evaluating a Model What do we mean when we say that a model works? It means that a model can be used in predicting the unknown quantity that it is supposed to predict. There are a number of different ways to determine this. Formal methodologies include creating tables and graphs and calculating statistics. Credit scores, especially with scaling, work if the scores separate and rank order credit risk. A population that is representative of the target population that has values for the required quantity are used to show how well the score predicts the outcome. First, records are segmented by score. This may be done by fixed score ranges, like 20-point score bands, or by score distribution, such as deciles. The keys to looking at data expressed this way: Separation. Files with low scores should have a high percentage of records with the negative occurrence that was measured, while low risk scores should have a low percentage of bad outcomes. The more 2012 Equifax Canada Co. All rights reserved. Page 9
separation there is, the better the score is performing. It is better if the bad rate is 50% in the high risk section and 2% in the low risk, than if the differences vary between 15% and 10%. Rank ordering. As scores improve, the bad rate should improve in an orderly and predictable fashion. BNI 2.0 scores increase when the risk decreases, so the observed bad rate should also decrease as scores increase. Here is an example illustrating how scores separate and rank order risk: All Accounts Negative Performance Bankruptcy Separate and rank order Consumers Cumulative % Consumers Cumulative % Bad Rate Goods to Bad 1 to 830 2,209,760 10.0% 62,778 72.6% 2.8% 34.2 830 to 941 2,209,761 20.0% 14,463 89.3% 0.7% 151.8 941 to 961 2,209,760 30.0% 5,050 95.1% 0.2% 436.6 961 to 967 2,209,761 40.0% 2,090 97.5% 0.1% 1,056.3 967 to 974 2,209,760 50.0% 925 98.6% 0.0% 2,387.9 974 to 982 2,209,760 60.0% 513 99.2% 0.0% 4,306.5 982 to 987 2,209,761 70.0% 303 99.5% 0.0% 7,291.9 987 to 990 2,209,760 80.0% 180 99.7% 0.0% 12,275.4 990 to 992 2,209,761 90.0% 144 99.9% 0.0% 15,344.6 992 to 999 2,209,760 100.0% 77 100.0% 0.0% 28,697.2 Scorables 22,097,604 86,523 0.4% 254.4 In this example, the records are grouped in deciles, 10% in each row of the table. Within the entire population, only 0.4% of the records were classified with negative performance (86,523 out of 22M), having a bankruptcy within the performance window. The lowest scores, 1 to 830, correspond to the 2.2M consumers (10% of the total) with the highest risk. They had a 2.8% bad rate, which is more than 7 times higher than the overall bad rate. The highest scores, 992 to 999, had a very low bad rate of 0.003%. The bad rates rank order through the score ranges, decreasing with every decile. Bad rates are often shown graphically. This is a plot of the column represented in the table above by the red up arrow: 3.0% 2.8% 2.5% 2.0% 1.5% 1.0% 0.5% 0.0% 0.7% 0.2% 0.1% 0.0% 0.0% 0.0% 0.0% 0.0% 0.003% 1 to 830 830 to 941 941 to 961 961 to 967 967 to 974 974 to 982 982 to 987 987 to 990 990 to 992 992 to 999 2012 Equifax Canada Co. All rights reserved. Page 10
Companies will often set a score cut-off, with strategies to take adverse action to consumers with low scores, or will decline applicants with low scores that fall below the cut-off that has been set. Since they want to limit declines or adverse action to a small percentage of their consumers or applicants, it is crucial that most of the consumers who will, in the future, exhibit bad behaviour have scores that are below the cut-off now. Some good customers 6 have low scores for various reasons, and companies do not want to strain many of these relationships and lose good business. In the table above, of the 86,523 consumers with serious delinquency or bankruptcy, 62,778 scored in the bottom 10%. This represented 72.6% of all bads. It is often a key measure of the predictability of a credit score to determine the percentage of bad records identified in the highest risk scores, and the higher the better. This is the column in the table above highlighted by the purple down arrow. Graphically, this is called a lift chart: 100% 95% 98% 99% 99% 100% 100% 100% 100% 90% 89% 80% 70% 73% 60% 50% 40% 30% 20% 10% 0% 0% 1 830 941 961 967 974 982 987 990 992 999 The better a score works, the higher the line will be in the lift chart. One way to measure how well the score works is called the Kolmogorov-Smirnov statistic, or K-S. The K-S looks at the lift chart, and calculates the percentage of good and bad records (represented by the blue line, below) scoring below each available score value. The maximum difference is the K-S 7 : 6 Good means that they will be good customers in the future; they will meet their financial obligations, and not be classified as bad according to the definition that is used to test the score. 7 Other metrics, such as the Gini and AUROC, measure the lift of a model differently, but K-S is often the standard metric in credit risk scoring. 2012 Equifax Canada Co. All rights reserved. Page 11
100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 73% 10% 89% 69.5% KS 20% 95% 98% 99% 99% 100% 100% 100% 100% 90% 30% 40% 50% 0% 1 830 941 961 967 974 982 987 990 992 999 60% 70% 80% Here, 89% of the bads (red line) score at 941 or below, but only 20% of the goods. The 69% difference is the largest difference at any point on the graph, so it is the K-S value. The higher the K-S, the better 8. 5. Using BNI 2.0 with ERS While BNI 2.0 is an excellent score on its own, the power of BNI 2.0 is how it can be used in conjunction with a delinquency score like ERS. Companies using a different risk score such as CRP 9 or BEACON will see a similar benefit. Most companies are already using one of these scores. BNI 2.0 is a complimentary score and is not designed to replace a risk score, but to be used together to present a more robust assessment of risk covering different patterns of behaviour. As stated before, delinquency is most often characterised as a chronic behaviour, with missed payment due to individuals who mismanage their budgets. Bankruptcy is more of an acute financial difficulty where a consumer loses the ability to pay bills. The key driver for delinquency is a pattern of missed payments. The key driver for bankruptcy is outstanding debts. Many consumers file for bankruptcy without going through stages of delinquency and because they don t miss payments, their delinquency score often remains high. BNI 2.0 identifies bankruptcy risk, for consumers who are not missing payments. Identifying these surprise bankruptcies is the key benefit of using BNI 2.0. 5.1 Dual Matrix Strategy This section presents one way to build a strategy using multiple scores. Different strategies for combining credit risk score may also be valid, and can be used at the discretion of the risk manager. Consider accounts paid as agreed in March 2011, and scorable with ERS and BNI 2.0. 8 What is a good K-S? It depends greatly on the use of the model and the available data. Sometimes a very small amount of lift can have tremendous benefit. One of the values of K-S is as a comparison tool between different models on the same data, as an evaluation of the models. Another is to evaluate how well a model performs over time as things change. 9 The Consumer Risk Predictor (CRP) is an Equifax proprietary generic delinquency risk score. It is very similar to ERS and BEACON in that it predicts delinquency, with a similar score range and interpretation as the other scores. 2012 Equifax Canada Co. All rights reserved. Page 12
Number of Accounts (All Companies - All Trades - Canada) BNI 2.0 Range ERS Range 1 to 499 500 to 699 700 to 799 800 to 849 850 to 899 900 to 949 950 to 964 965 to 999 Total Under 500 248,842 35,323 17,264 8,515 8,855 10,536 906 18 330,259 500 to 549 350,386 121,356 68,885 37,344 40,715 55,910 10,016 602 685,214 550 to 574 236,430 113,531 73,145 44,080 51,169 76,431 17,399 1,865 614,050 575 to 599 260,067 149,253 108,738 68,548 84,483 143,651 39,600 10,008 864,348 600 to 624 285,959 183,403 145,156 98,535 128,480 245,878 78,522 22,971 1,188,904 625 to 649 298,664 233,989 197,448 141,766 199,252 446,764 186,492 51,692 1,756,067 650 to 674 296,750 308,138 277,854 210,296 325,025 925,800 590,525 245,835 3,180,223 675 to 699 176,135 272,379 271,893 212,359 325,767 1,028,671 996,463 811,903 4,095,570 700 to 749 64,426 215,830 342,867 343,838 656,733 2,694,716 3,353,154 4,094,830 11,766,394 750 and Up 141 1,806 8,394 17,701 64,427 941,166 3,886,651 31,679,199 36,599,485 Total 2,217,800 1,635,008 1,511,644 1,182,982 1,884,906 6,569,523 9,159,728 36,918,923 61,080,514 3,581,232 accounts (5.9%) with ERS 600+ but BNI 2.0 below 800 In this example, about half of the accounts are showing very little risk. The BNI 2.0 and ERS scores are in the highest categories, with nearly 32M of the 61M accounts in the cell in the bottom right corner of the table. Meanwhile, about 1.8M accounts are in the top left corner of the table with low ERS (under 600) and BNI 2.0 (under 800) scores. For these, BNI 2.0 adds little value, as ERS has already identified them as high risk. The cells highlighted in pink compose 3.6M accounts, or 5.9% of the total. They belong to individuals who have ERS scores of 600 or better, which may be high enough to be considered low risk. These consumers also have BNI 2.0 scores below 800 10. BNI 2.0 considers these consumers to have a heightened risk of filing for bankruptcy. The general characteristic of this segment of the population are consumers making regular payments (positive for the key driver of ERS) but carrying a lot of debt (negative factor for BNI 2.0). By March 2012, one year later, many consumers filed for bankruptcy. There were 280,000 accounts out of the 61M total, on the credit files of those consumers who filed for bankruptcy. Using the March 2011 scores, these accounts are distributed in the following table: Number of Accounts Bankrupt (All Companies - All Trades - Canada) BNI 2.0 Range ERS Range 1 to 499 500 to 699 700 to 799 800 to 849 850 to 899 900 to 949 950 to 964 965 to 999 Total Under 500 18,658 930 278 83 83 27 4 0 20,063 500 to 549 21,123 3,367 1,338 531 435 286 24 5 27,109 550 to 574 13,426 2,960 1,409 641 530 511 45 4 19,526 575 to 599 13,107 3,649 1,765 922 902 743 101 30 21,219 600 to 624 13,333 4,443 2,444 1,170 1,231 1,388 161 6 24,176 625 to 649 13,334 5,553 3,062 1,591 1,848 2,223 390 10 28,011 650 to 674 12,752 8,063 4,519 2,669 2,994 4,600 1,173 154 36,924 675 to 699 7,658 7,832 5,597 3,275 3,569 5,116 1,584 452 35,083 700 to 749 2,581 5,977 7,249 5,389 7,951 14,955 5,667 1,853 51,622 750 and Up 8 37 126 216 735 4,883 5,418 5,439 16,862 Total 115,980 42,811 27,787 16,487 20,278 34,732 14,567 7,953 280,595 104,568 accounts bankrupt (37.3%) with ERS 600+ but BNI 2.0 below 800 10 The choice of 600 and 800 as ERS and BNI cut-offs is purely arbitrary for this example, and care should be taken by the risk manager to find the right values for his book of business. 2012 Equifax Canada Co. All rights reserved. Page 13
The cells highlighted in pink contain over 104,000 accounts, which is 37.3% of the total number of accounts for those consumers who filed for bankruptcy. This is disproportionally high, as only 5.9% of the accounts for the entire population fall into this segment. The bankruptcy rate is 6.3 times higher than the average for the entire population, and yet their delinquency score is above 600, suggesting that these consumers are low credit risk. Using BNI 2.0, the risk manager can identify this small high risk population (only 5.9% of accounts) and take action, and save a significant exposure to bankruptcy loss. 5.2 Refreshing Scores for Account Management Using BNI 2.0 at adjudication has obvious benefits, as risk managers can identify these high risk consumers before extending credit, but regular refreshes of BNI 2.0 scores for account management has a significant impact as well. Most large companies refresh their delinquency scores regularly throughout the year. However, remember the key drivers. Delinquency scores depend greatly on maintaining a consistent payment pattern. For consumers with safe 11 ERS scores at last refresh, they are probably going to remain safe as long as the consumer continues to make all regular payments. If they begin to miss payments, chances are that the risk manager is aware of this, since there is a good chance that there are missed payments to his company. He has an opportunity to take immediate action and possibly begin to make collection efforts at early-stage delinquency. On the other hand, the key drivers for BNI 2.0 are outstanding debts and excessive inquiries. Consumers in acute financial difficulty are carrying larger balances on their credit cards, tapping into their lines of credit, and applying for new credit cards as they max out the ones they have, or looking for consolidation loans. Much of this activity will go on with other companies, and the risk manager may be unaware of this consumer activity when monitoring the account in his company, especially if no payments are missed. Because BNI 2.0 scores change when heightened risk factors occur on events that a risk manager will not see when monitoring his own account, it is crucial to refresh BNI 2.0 scores regularly to identify these behaviours and be proactive in risk management for these accounts. Summary Canadian consumers need credit cards for everyday purchases, loans to buy goods and services, lines of credit to optimize and bring flexibility to credit management, and mortgages so they will have a place to live. To finance these things, banks and other lending or credit granting institutions provide these credit products, but need to ensure that they can do so profitably. They send their data to Equifax, and Equifax builds credit files that contain the credit history of over 24 million active credit holders. The information in these credit files allows risk managers the opportunity to evaluate the likelihood that the consumer or applicant is in a good financial position and will be able to repay the loan, or make regular payments on the revolving credit. Credit scores, like ERS and BNI 2.0, assist this process. In particular, BNI 2.0 provides additional lift to the process, by identifying a different pattern of behaviour, that of acute financial stress. Identifying high bankruptcy risk consumers who are in good standing with their accounts allows the risk manager to account for different types of credit risk, and can optimize strategies to be as profitable and successful as possible. 11 Safe scores are scores that are high enough that the risk manager has no need to take any adverse action. 2012 Equifax Canada Co. All rights reserved. Page 14
2012 Equifax Canada Co. All rights reserved. Page 15
Appendix Reason Codes, Scorecard Indicator, Reject Codes Reason Codes Reason Code Description 03 number of trades with recent high utilization 04 number of retail trades 05 number of national card trades currently past due 06 number of trades one payment past due in last 12 months 07 number of bank revolving trades with high utilization in last 12 months 08 number of personal finance trades 11 number of finance inquiries in last 24 months 12 number of inquiries in last 3 months 13 number of inquiries in last 12 months 14 age of oldest trade 15 number of recently opened trades 16 total balance for open retail trades 19 number of national card trades 21 average age of national card trades 22 number of national card trades with high utilization 24 ratio balance to limit/high credit for open trades 30 number of trades currently past due 37 total high credit for open department store trades 44 number of trades ever one payment past due or worse 46 age of derogatory public records 47 number of collections 48 age of oldest personal finance trade 49 total balance for open personal finance trades 50 total high credit for open personal finance trades 51 total balance for trades with date opened in last 6 months 52 utilization for bank revolving trades with date opened in last 3 months 53 worst current rating for miscellaneous finance trades 54 worst current rating for bank installment trades 55 worst current rating 56 total monthly payments 57 total high credit for revolving trades opened in last 6 months 58 total high credit for open revolving trades 59 total high credit for open retail trades 60 total high credit for open national card trades 61 total high credit for open bank revolving trades 62 total balance for open revolving trades 2012 Equifax Canada Co. All rights reserved. Page 16
63 total balance for open national card trades 64 total balance for open department store trades 65 total balance for open bank installment trades 66 total balance for auto finance/auto loan trades 67 ratio average retail utilization to average revolving utilization 68 number of trades opened in last 24 months 70 number of open trades 71 number of open personal finance trades 72 number of open national card trades 73 number of open credit union trades 74 number of open bank revolving trades 75 number of open bank installment trades 76 number of open auto loan trades 77 number of derogatory public records 78 average utilization for open trades 79 amount currently past due 80 age of oldest sales finance trade 81 age of oldest bank revolving trade 82 age of oldest auto loan trade 83 number of department store trades high utilization in last 3 months 84 ratio open personal finance trades to total open trades 85 number of installment trades with high utilization 86 utilization for bank installment trades 87 number of bank installment trades with high utilization in last 12 months 88 total balance for bank installment trades opened in last 3 months 89 total high credit for bank installment trades opened in last 3 months 90 utilization for bank installment trades opened in last 3 months 91 number of satisfactory retail trades in last 24 months 92 number of retail trades with high utilization in last 6 months 93 number of satisfactory national card trades in last 24 months 94 ratio satisfactory national card trades to total national card trades 95 number of national card trades with high utilization in last 3 months 96 total high credit for national card trades opened in last 6 months 97 number of trades 60 days in last 12 months 98 number of trades 30+ days in last 12 months 99 number of trades 90+ days in last 12 months A0 A1 A2 A3 average utilization for revolving trades utilization for revolving trades number of revolving trades with high utilization in last 6 months number of miscellaneous finance trades ever 30+ days 2012 Equifax Canada Co. All rights reserved. Page 17
A4 number of bank revolving trades bad debt in last 6 months A5 ratio open auto finance/auto loan trades to total open trades A6 utilization for auto finance/auto loan trades A7 number of auto finance/auto loan trades bad debt in last 12 months A8 number of revolving trades with high utilization A9 worst rating for revolving trades B0 number of trades with high utilization B1 number of trades with low utilization B2 number of trades with balance > $0 B3 number of always satisfactory bank revolving trades B4 number of always satisfactory department store trades B5 number of always satisfactory credit union trades B6 number of satisfactory trades with balance > $0 B7 ratio trades opened in last 24 months to total open trades B8 ratio trades opened in last 12 months to total open trades B9 number of inquiries in last 6 months C1 number of trades with high utilization in last 6 months C2 number of retail trades with high utilization in last 3 months C3 number of bank revolving trades bad debt in last 24 months Z0 no trade or public record information Scorecard Indicators Scorecard Indicator Description S0 S1 S2 S3 S4 S5 Existing Bankruptcy Scorecard Young File Scorecard Thin Credit File Scorecard Thick Prime Credit File Scorecard Thick Subprime Credit File Scorecard Inquiries only Scorecard Reject Codes Code Description 091 'D' Subject Deceased 092 'F' File Under Review 093 'I' No Recent Activity 094 'S' File Cannot Be Scored 2012 Equifax Canada Co. All rights reserved. Page 18