A Bayesian estimation approach of random switching exponential smoothing with application to credit forecast

Abstract

We introduce an efficient Markov Chain Monte Carlo sampler in precision-based algorithms for the estimation of the Random Switching Exponential Smoothing model, a versatile forecasting mechanism for time series data characterized with changing trends. Through a series of simulation experiments, RC-MCMC exhibits superior parameter estimation accuracy, particularly for datasets featuring low persistence trends. Furthermore, an empirical evaluation using the Bank for International Settlements’ quarterly time series data on the non-financial sector’s total credit relative to GDP validates the findings. The out-of-sample results indicate that the proposed approach outperforms its counterparts in estimating and forecasting accuracy for trending time series data.