Forecasting Bankruptcy More Frequently: Information Update via High Frequency Data

Abstract

This paper considers the econometric problems arising from using outdated data in a regression model in which the independent variable is observed less frequently than the dependent variable. Specifically, OLS estimates may suffer from a form of omitted variable bias if outdated data is correlated with information during the time no observation takes place. We claim that using data correlated with the independent variable but with a shorter observation period to update the independent variable can eliminate the bias, as well as reducing uncertainty in estimating the dependent variable. We test the theory with an empirical model of bankruptcy forecast for medium sized firms. We present a more accurate default forecast model that updates the average change in firms financial standing with monthly business cycle information. Financial institutions may use the monthly estimates to monitor losses on their loan portfolios more accurately and more frequently.