Data governance for ERM InsuranceERM Seminar, Thursday, 16 May 2013 Andrew Hitchcox, CRO, Kiln Group
Different sorts of data in an insurance company Data for transactions with clients: Sales / conduct issues Product information Money to & from client Data for financial statements: Actuals / cash Outstandings / forecasts Data for risk management: Current exposures / amounts at risk Forecasting future risk amounts / probabilities SII balance sheet Internal Model / ORSA Subject matter for my talk 2
Forecasting future risk amounts / probabilities Future assets / investment returns: Internal cashflows and asset holdings External market movements Future liability flows: Premiums / claims Amounts of risk / exposure Central forecasts / trend values: Variability around expected trend Probabilities: 1/10, 1/100, 1/200, 1/500 outcomes These are data sets for IM / ORSA 3
Data to support risk management Extract from Metro 14/05/13: Partnership targets 1.4bn London float. Parnership s business model relies on 18 years of data collection that allows it to expertly judge life expectancy and offer higher guaranteed retirement income to people with health impairments. Transactions: Data items used as themselves Risk management needs more than data: Expert judgement Assumptions Making modelling choices Need data to support these activities Often more than transaction data 4
Data quality support for ERM [more than just transactions] Asset valuations Basic quality: transactions / internal data Good quality: risk / exposures / benchmarks / external data Data quality criteria * Use only internal data to value assets, such as book values or cash flow analysis, that rely on significant grouping of data * Apply data in the "mark-tomarket" or "mark-to-model" valuations, taken from a market maker or provider of market information * Use justifiable assumptions to complete holes in data due to the absence of market observations 5
Data quality issues for ERM [more than just transactions] Market risk modelling Basic quality: transactions / internal data Good quality: risk / exposures / benchmarks / external data Data quality criteria * Asset data generically grouped into buckets based on similar characteristics, or proxy data which may not have been vetted or adjusted * Apply data in the "mark-tomarket" or "mark-to-model" valuations, taken from a market maker or provider of market information * Use justifiable assumptions to complete holes in data due to the absence of market observations 6
Data quality issues for ERM [more than just transactions] Credit risk modelling Basic quality: transactions / internal data Good quality: risk / exposures / benchmarks / external data Data quality criteria * Buckets" applied to data, or use proxy data without quality control * Collect historical data used to generate key risk driver data without characteristic information * Differentiate credit exposures by default risk and loss severity * The time period to which the data pertain may include at least one severe economic or several credit downturns 7
Data quality issues for ERM [more than just transactions] Mortality risk modelling Basic quality: transactions / internal data Good quality: risk / exposures / benchmarks / external data Data quality criteria * Buckets" exposure data (e.g., product, exposure levels, age, and sex) * For unmodeled business, scale up results based on volumes * More granular approach to data grouping granularity of data used to derive the assumptions, e.g. suitably detailed with statistics on cause of death, geography, and population type 8
Data quality issues for ERM [more than just transactions] Underwriting risk modelling Basic quality: transactions / internal data Good quality: risk / exposures / benchmarks / external data Data quality criteria * Collect and maintain historical loss and policy data, including data from the acquired and divested businesses * Include development scenarios for large losses, for instance, minimum, expected, and maximum ultimate-loss scenarios * External data sources used to supplement internal data, including external benchmarks or proxies to help compensate for insufficient data, particularly catastrophe-loss data 9
Data quality issues for ERM [more than just transactions] Operational risk modelling Basic quality: transactions / internal data Good quality: risk / exposures / benchmarks / external data Data quality criteria * Use only industry data without quality control * Conduct internal workshops to identify operational risk loss scenarios of frequency and severity * Maintain an internal database of loss history * Take loss data from crossreferenced sources to ensure consistency., e.g. external loss databases containing industry losses 10
Data quality issues for ERM [more than just transactions] Dependencies / correlations Basic quality: transactions / internal data Good quality: risk benchmarks / external data Data quality criteria * Limited time period of data that do not represent adequately tail dependencies * External and internal data from a range of sources to capture dependencies at a variety of confidence levels, including tail dependencies * Procedures for the cleaning and interpolation or extrapolation of rare data 11
Setting priorities for data quality initiatives So much you could improve: Need to prioritise efforts Very tempting to rush in and start checking!: Need to prioritise efforts Pull as well as push: Set priorities by looking at outputs from data processes Don t just look at inputs 12
Setting priorities for data quality initiatives Look at schematic data flows of end-to-end process: Where is data transformed and why Where are your controls and why Data dictionaries good for detailed disciplines: But don t overdo / polish too much Concentrate on hand-overs between departments / processes How we prioritised our data quality efforts Use Test / burning issues Just do it : create the process and criteria afterwards Prioritise the data sets, not the data items: Kiln example DQS (Data Quality Scorecard) [see next slide +1] Kiln example qualitative commentary on results [next slide +2] 13
Data governance for ERM: - concluding remarks Not just transaction data sets: But also risk / exposure data sets Not just internal data sets: But also external data sets So many data items: Prioritise data quality efforts by data set impact on outputs from balance sheet and Internal Model 14