1-minute version

  • In 2011 Emanuel Derman wrote his book Models Behaving Badly: In general models do not perform well in financial crises.
  • Risk and finance department will face a challenge in maintaining their behavioural models since these models are usually calibrated on ‘normal times’.
  • Behavioural models play a crucial role in, among others, ALM, pricing, capital calculations and IFRS 9.
  • Key question: How do behavioural models perform in the current economic environment with COVID-19 measures and an extremely low interest rate environment?
  • Examples where these models play a crucial role: payment and default behaviour, withdrawal behaviour non-maturing deposits, prepayment behaviour in mortgages.
  • Behavioural models need to cope with both risk and uncertainty.
  • It is ADC’s vision that expert judgement will play a more and more important role in the modelling of customer behaviour.

Behavioural modelling in unusual times

In 2011 Emanuel Derman wrote his book Models Behaving Badly. Very relevant at the time since models used in the financial sector were not behaving as they should do in the financial crisis that started in 2008. In the current economic environment with extremely low interest rates and the impact of the COVID-19 measures on businesses and the financial services industry, this topic is as relevant as ever.

A fair number of models in the domain of finance, risk and ALM departments in the Financial Services industry, model the behaviour of their customers. Three primary examples in banking are:

  1. Payment and default behaviour in credit risk models used for the calculation of regulatory capital, pricing spreads and the loss provision under IFRS 9,
  2. Withdrawal behaviour in variable rate savings accounts (non-maturing deposits), and
  3. Prepayment behaviour in mortgage and loan portfolios both used to estimate the duration and other interest rate sensitivity measures.

Insurance companies also have behavioural models in place. They are also large investors in mortgages and more traditionally, insurance companies model the lapse behaviour of their policyholders for ALM, hedging and capital purposes.

Pension funds face different challenges in behavioural modelling regarding the interaction between low solvency ratios, strategic asset allocation, indexation, and potential cutting policy (behaviour of policy makers in the pension fund).

Are the behavioural models used in the Financial Services industry still behaving? Or are they – quoting Derman – behaving badly?

Risk and finance departments are pondering whether the models in place are still robust and provide risk parameter estimates that make sense given the current economic environment. As most of these models are estimated based on historical information, the key question is, of course: is the past is a good prediction for the future?

Going back to the examples described above, the following difficulties may arise:

1. Credit risk models

When banks apply measures to help customers survive the current crisis by giving them temporary solutions like delaying loan repayments or interest payment holidays, this will immediately have an impact on default and forbearance data and hence on risk parameter estimates Please refer to ADC insight – How will COVID-19 affect banks? The stress test as a diagnostic tool? – for more information. How will banks – going forward – deal with such a structural break in data and modelling?

2. Non-maturing deposits

Some replication portfolios (for example, the constant margin model) that model the interest rate sensitivity of non-maturing deposits, tend to estimate a relatively high duration in the current economic environment. Reasons can be that 1. (some) customers save more (less opportunities to spend money in lock-down period) and 2. in the current low-interest rate environment client rates are stickier since clients rate have been close to 0% for quite some time. Several questions arise for modellers, among others:

  1. If duration estimate from the current models is high, will this be a correct reflection of the interest sensitivity of non-maturing deposits?
  2. Are the customers behaviour and price elasticity components in the models a good reflection of what will happen in an increasing interest rate/inflation environment?
  3. The difference between current accounts and non-maturing deposits is becoming less clear: both products do not have a maturity and client rates are 0% or close to 0%. If there is a large difference between duration estimates of these products, and if yes, how can this be explained?

Related to question 2): this is important since if interest rates suddenly increase and the duration of non-maturing deposits is high, a bank cannot increase client rates without loss of interest rate margin.

3. Mortgages

Clients usually prepay more on their mortgage in a decreasing interest rate environment. The current interest rate is already record low. Again, several questions for risk modellers:

  1. How can we estimate prepayment behaviour for mortgages with already extremely low client rates? Will it be likely that customers with older mortgage loans that have not prepaid yet will prepay in the future?
  2. Suppose house prices drop in the current economic environment, how will this affect prepayment behaviour?
  3. What are the implications of new approaches in behavioural modelling for the company’s hedge strategies?

Derman explains in his book that modelling risk is different than modelling uncertainty. It is important that models in general, and behavioural models in particular, do not only model the risks observed in the past but also should contain an element of expert judgement. If you are interested in what ADC’s vision is on these matters, please contact us.

Frans Boshuizen

Frans is Sector Lead Financial Services at Amsterdam Data Collective. Having worked for some of the Netherland’s leading financial institutions in senior management roles, he helps clients navigate through challenging times. He is a pragmatic and empathetic leader who applies his solid background in academia to generate real value from data-science initiatives.

Get in touch

Latest Insights