Revisiting Early Warning Signals of Corporate Credit Default Using Linguistic Analysis

Abstract

We apply computational linguistic text mining (TM) analysis to extract and quantify relevant Chinese financial news in an attempt to further develop the classical early warning models of financial distress. Extending the work of Demers and Vega (2011), we propose a measure of the degree of credit default, referred to in this study as the ‘distress intensity of default-corpus’ (DIDC), and investigate the predictive power of this measure on default probability by incorporating it into the signaling model, along with the classical financial performance variables (the liquidity, debt, activity and profitability ratios). We also apply the ‘naïve probability of the Merton distance to default’ model (Bharath and Shumway, 2008) for our robustness analysis. A logistic regression (LR) model is constructed to better integrate the DIDC and financial performance variables into a more effective early warning signal model, with the incorporation of DIDC into the LR model revealing a significant reduction in Type I errors and an apparent increase in classification accuracy. This provides proof of the effectiveness of the additional information from TM on the financial corpus, whilst also confirming the predictive power of TM on credit default. The major contribution of this study stems from our potential refinement of early warning models of financial distress through the incorporation of information provided by related media reports.

Keywords: Credit default, Financial distress, Early warning, Linguistic analysis, Media, Logistic regression

JEL Classification: G33, C10, G14