Predicting Korean Recessions with Time-Varying Predictors

Abstract

This study examines the predictive ability of a wide set of variables to predict Korean recessions based on probit models. In doing so, we extend probit-based recession forecasting models in several important and novel ways. First, in addition to commonly used financial variables, we incorporate several macro leading variables as potential predictors of recessions. Second, our forecasts use the algorithm of dynamic model selection/averaging (DMS/DMA), which allows for specific predictors to switch over time in a data-based manner. Our main findings are as follows. First, in terms of both in-sample fits and out-of-sample forecasts, while financial indicators (such as interest rate spreads) are good predictors over short-horizons (i.e., one or three months ahead), some macro leading variables (such as commodity price index and job opening-to-application ratio) turn out to be useful over longer horizons. Second, forecasting using time-varying predictors (i.e., DMS/DMA models) performs well, beating individual best predictors for each forecast horizon. In addition, we show that forecasting using switching predictors outperforms the models that employ fixed predictors and the composite leading index, in most cases, especially in terms of the mean squared error (MSE). Third, we find strong evidence for predictor switching and illustrate how the performance of the key predictors has evolved over time.