Threshold mixed data sampling logit model with an application to forecasting US bank failures

Abstract

This paper introduces a threshold mixed data sampling logit (TM-logit) model, which allows for a threshold effect of independent variables sampled at different frequencies on the log-odds of dependent variable. We propose model estimation procedure and develop test statistics for relevance of high-frequency predictors, threshold effect, and equal weighting scheme. We also suggest a test statistic for the difference in forecasting accuracy between two competing models. We then extend the model to the framework with a covariate-dependent threshold (CDTM-logit) and propose estimation procedure and test statistic for threshold constancy. Monte Carlo simulations are conducted to assess the finite sample performance of the proposed estimation procedure and test statistics. The simulation results show that the estimation procedure performs well and test statistics have good size and power properties in finite samples. We apply the proposed model to predict US bank failures, and the empirical results indicate that the TM-logit and CDTM-logit models have good forecasting performance.