Model risk in unusual corners - PowerPoint Presentation

Model risk in unusual corners the case of home price indices (HPI) Edinburgh chapter 14 FEBRUARY 2017

IntroductionVijay Krishnaswamy is a Partner at True North Partners, a specialist consulting boutique18 years of front-office, finance and risk management experience Previously, worked at the FSA, KPMG, Standard Chartered, ANZ Co-editor of “Managing Illiquid Assets”, published by RiskBooks incl. a chapter on model riskSet up model governance framework for the FSA (now PRA)

“Index”Back to basicsThe role of modelsImplicationsOverview of talk

Source: Oxford Dictionary Index /ˈ ɪndɛks/

(Re-)Valuation of assets Credit scoring Pricing Capital reserving, P&L provisioning/ IFRS 9Stress testingBenchmark for pricingOverall barometer of the marketAsset in itselfThe use of HPIs in mortgage portfolios is very commonVariety of sources: Banks (Halifax, Nationwide,…)Property firms (Knight Frank etc…)Government (Land Registry)Why have indices become important? Easy, quick, inexpensive Exposure to a diverse portfolio Highly prevalent…. … and abused EU Benchmarks Regulation Uses Outcomes

1 ordinary share of UnileverTrade 1: 3310p  latest valuationUpdate to FTSE indices Trade 2 (within a few milliseconds/ end of day): 3350Change is 1.2%Update to FTSE indices intra-day/ end-of-dayEnabling feature:Trades 1 and 2 involve an identical shareWhere do HPIs come from?Source: LSE website

House at <address>Trade 1: Sold at £ 200k  latest valuationUpdate to HPI Trade 2: Sold next month at £ 220kChange is 10% in 1 month Update to HPI of the month Where do HPIs come from?Assumptions needed:Same house has to be sold again the following monthHouse has not changed intrinsically during the month What happens if we relax these assumptions?

Houses:Trade 1: Sold at £ 200k  latest valuationUpdate to HPI Trade 2: Sold next month at £ 250kChange is 25% in 1 month Update to HPI of the monthDoes not mean a 25% increase in asset price Why?Where do HPIs come from?Result: The analogy to equity prices breaks downForming an index on home prices in the same way is not possible

Create a “standardised” house Serves as a reference point for indexingTake each house sale transaction and “break-down” into constituents House price = m * Value of a bedroom + n * Value of a garden + …. Re-assemble the standardised house in the base period with the latest prices All this is achieved through a regression model…Enter: models to the rescue

Enter: models to the rescue Source: IHS Markit

Enter: models to the rescue Source: IHS Markit

Enter: models to the rescueSource: IHS Markit

First built in the late 1970s/ early 1980sFirst published 1983Initial explanatory power: high 80sSteady decline in model explanatory power; latest: ~70% Why? If the bank’s portfolio is not similar to the portfolio used to build the index, then the bias can be significantE.g. granular geographical variations Reference portfolio: disclosures are limitedSo how good is this model?

And not all indices are equal… Aspect Halifax/ Nationwide HPI Land Registry HPICoverageMix-adjusted Yes No However, how much does the mix change? Halifax Registry Nationwide

Blind use of indices can add to model riskMore a model than a “traditional” indexModel risk framework Identify the index as a source of model risk?Familiarise with (whatever) available documentation? Be aware of pros and cons of different HPIs Some implications of using HPIs

Raise awareness of model risk in HPIsUnderstand how HPIs compare with equity price indices and the differencesThe role of models in making HPIs feasibleUnderstanding what drives the performance of such models Similar issues apply in the case of indices built on assets that are non-identical Recognise the use of a HPI as a source of model risk Recap

The information contained herein is proprietary. © 2016 True North Partners LLP www.tnp.eu Questions?