Enhancing ALM

Behavioural models use available consumer data to estimate future behaviour in specific scenarios. A common assumption is to predict agents’ choices assuming their rationality. Unfortunately, the empirical evidence and recent development of behavioural economics, show how customers’ behaviour falls outside of what can be considered fact-based or rational. For example, when market rates increase, banks should expect a greater number of prepayments on floating rate mortgages, but according to empirical evidence this is not always the case. 

What are the main reasons behind this phenomenon? Prepayment models can help Banks to understand the main factors guiding customer behaviour in their loan portfolio.
Another example is about deposits, well known to be instruments without defined maturity and repricing rule. Behavioural model on Non-Maturing Deposits (NMDs) are generally calibrated on historical data in order to identify a behavioural maturity and a repricing rule based on the elasticity of the customer rate with respect to a market pivot rate.

The recent pandemic crisis has increased the interest in such models, as it has had a significant impact on customers’ behaviour. Taking into account the behavioural component it is possible to implement new and more flexible models, improving the reliability of metrics and the risk management process.
In this context the use of Machine Learning techniques can improve the predictive power of the models, compared to the behavioural model framework currently adopted by Banks.