Modeling Non-Maturing Deposits: A Procedure for the Determination of the Minimal Time Series Length Required for Model Calibration and Back-Testing

Abstract

This paper proposes a procedure for the determination of the minimal length of the historical time series of daily deposit variations in accordance with an institution’s specific risk tolerance. In a previously released paper we developed a methodology to ascertain an institutional specific confidence level to be used for the same purposes. As the formula for the determination of the confidence level depends on the length of the deposit variations time series, we complete the procedural construct by proposing a means to estimating the minimal length of the daily deposit volume variations vector in accordance with Principle 9 of the EBA recommendations for liquidity risk management best practices. As the formula determining the minimal length of the volume time series will depend on the institutional specific confidence level to be used (as described in the previous paper), we will augment our procedure with a recursive method which converges to optimal values for both parameters. The procedure centers on the identification of structural breaks in the volume dynamics distributions. The resulting vectors of constant maturities are identified via distributional power statistical tests. We further illustrate the application of the proposed procedure using a dataset of daily demand deposit volumes from a fictional European savings bank.