Estimating the Worst Scenario of Net Cash Outflow of Non-Maturity Deposits Based on Quantile Regression

Abstract

The article considers models that are the most commonly used to estimate the cash outflows of nonmaturity deposits: core level of deposit balances, Geometric Brownian Motion (GBM), and Cash Flow at Risk (CFaR). These models use a one-dimensional normal distribution of the probabilities of balances or cash flows. We generalize the above-mentioned cash outflow models in linear quantile regression. The properties of quantile regression are particularly attractive for assessing the liquidity risk of non-maturity deposits. Quantile regression determines quantile directly. It does not depend on the type of distributions of deposit balances or cash flows; it is resistant to outliers that are typical of outstanding deposits; captures fat tails of the underlying distribution of variables; does not assume that variance is constant. To construct a quantile regression, we use a conditional twodimensional empirical distribution of cash outflows and deposit balances for corporate non-maturity deposits denominated in hryvnias under both normal and crisis conditions. Statistical tests confirmed the satisfactory goodness-of-fit for the model and the hypothesis of linear dependence of cash flows on the current level of deposit balances. As a result, the constructed quantile regression has no areas of the values of deposit balances in which liquidity risk would be underestimated. The developed methodology for quantile measure of deposit outflow will be useful both for banking supervision and banks.