Support vector machine-based and crisis-pertaining forecasts of a subset of foreign currency- denominated bank deposits in Türkiye

This paper presents support vector machine-based forecasts of a subset of the banking system’s foreign currency-denominated deposit-growth for a crisis-inclusive period in Türkiye. Forecasts concerning such periods pose challenges that may not always be efficiently handled within the confines of conventional statistical methods. This brings out a need to make recourse to alternative methods, one of which is employed in this paper. The method employed in the paper belongs to a particular group of machine learning/artificial intelligence algorithms known as support vector machines, which could yield successful results in a wide range of cases. We demonstrate that proper employment of support vector machines leads to a reasonably high degree of accuracy in forecasting and produces, with a small margin of error, real-value-replicating trajectories of the target variable in question. Accurate forecasts of foreign currency-denominated deposit growth rates at crisis-inclusive junctures could be of practical significance to the policy designers attempting to limit, in an optimal manner, the magnitudes(s) or growth(s) of the foreign currency-denominated deposits within the banking system. This article shows how the objective of practical significance in question could be achieved with an alternative method.