The modelling of non-maturing deposits: can Machine Learning algorithms help improve the prevailing methods?

Details

Supervisors

Henrard, Luc

Faculty

Faculté des sciences économiques, sociales, politiques et de communication

Degree label

Master [120] en sciences économiques, orientation générale

Abstract

Some types of accounts, such as current accounts or savings accounts, do not have a fixed maturity date. They are called ‘non-maturing deposits’. They constitute a big part of the financing of depository institutions. NMD are characterized by two embedded options: one for the client who can decide to withdraw partly or all money from his/her account, and one for the bank who can decide to adjust the level of the interest rate. They are usually respectively referred to as the volume option and the rate option. These two options render the modelling of NMD particularly hard, due to the uncertainty they entail. The interest rate risk and liquidity risk of NMD constitute a large part of the Asset-Liability Management risk of banks. The current low and decreasing interest rate environment contributes to the challenge NMD modelling represents for the risk management and hedging strategy of banks. A widely used modelling method is the replicating portfolio approach. Three sub-models, the market rates model, the deposit rate model and the volume model typically characterize such approach. The three models are interrelated to each other. In the current literature and practice, the deposit rate model and the volume model are formulated respectively as linear multiple regressions on various market rates. One of the big issues with replicating portfolios is that they are usually calibrated on historical observations and thus ‘benefit’ from the falling interest rates, which is currently contributing to increase the estimated deposit duration. Besides, it is reasonable to say that a sudden rise in the market rates would not affect the deposit rate and volumes the way it does when the market rates are decreasing. Firstly, this report hence tries to identify other explanatory variables than market rates such as macroeconomic indicators, alternative indices on financial instruments and returns of other investment strategies that could explain the deposit rates and volumes. Modelling the savings rate and volume on other variables as well could mitigate the adverse effects a rise in market rates can have on the deposit volumes, the profitability and the hedging strategy of banks. Secondly, in this report we imagine that machine learning techniques can help to better model the future evolution of deposit rates and volumes. We want to improve the quality of the model as compared to a model derived from standard statistical techniques banks currently use. An algorithm is able to identify among many variables the ones that can explain the evolution of deposit rates and volumes and their impact on these variables. The goal is to focus quantitatively on decision trees and random forests. And then identify whether the performance of the model is enhanced by the introduction of ML techniques.