Bankruptcy risk prediction and pricing : unravelling the negative distress risk premium

• Main Author: Bauer, Julian
• Other Authors: Agarwal, Vineet
• Published: Cranfield University 2012
• Subjects: 332.75
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.558681

In sharp contrast to the basic risk-return assumption of theoretical finance, the empirical evidence shows that distressed firms underperform non-distressed firms (e.g. Dichev, 1998; Agarwal and Taffler, 2008b). Existing literature argues that a shareholder advantage effect (Garlappi and Yan, 2011), limits of arbitrage (Shleifer and Vishny, 1997) or gambling retail investor (Kumar, 2009) could drive the underperformance. Herein, I test these potential explanations and explore the drivers of distress risk. In order to do so, I require a clean measure of distress risk. Measures of distress risk have usually been accounting-based, market-based or hybrids using both information sources. I provide the first comprehensive study that employs a variety of performance tests on different prediction models. My tests are based on all UK non-financial firms listed in the Main market segment of the London Stock Exchange (LSE) between September 1985 and October 2010. It includes 22,217 observations with 2,428 unique firms of which 202 went bankrupt. I find that hybrid models clearly outperform the accounting-based z-score (Taffler, 1983) and the market-based model of Bharath and Shumway (2008) (BS). Hybrid models forecast bankruptcies more accurately and they subsume the bankruptcy related information of z-score and BS. While there is little to distinguish between the hybrids in forecasting accuracy, tests of differential misclassification costs show that the highest economic value is delivered by the most parsimonious hybrid model of Shumway (2001) (Shum). The forecasting accuracy between z-score and BS depends on the sample period while both carry complementary bankruptcy related information. My study provides confirmatory evidence on the puzzling negative distress risk premium. Moreover, my tests show that the distress risk premium is independent of the distress risk proxy (Shum, z-score or BS). Remarkably, z-score –the weakest bankruptcy prediction model - subsumes the return related information of Shum and BS in cross-sectional tests suggesting that it might not be distress risk per se that is priced. My results provide no evidence that the potential explanations in the existing literature are able to account for the distress puzzle. As such, I find no confirmation for the shareholder advantage effect. Although the characteristics of firms with high limits of arbitrage and gambling features are shared by distressed firms, tests provide no evidence for their pricing relevance or their impact on the distress risk premium. This is the first study that deconstructs the distress measures into their component parts to unravel the distress risk premium and shows that the profitability components of Shum and z-score drive the premium. The composite measure without the information carried by profitability is insignificant in the pricing tests. In time-series regressions, I show that the pricing information carried by a profitability factor is able to reduce the distress risk premium. Portfolio analysis identifies low distress risk-high profitability firms as the key driver of the mispricing. The distress anomaly is not driven by distress risk but by profitability. Another major contribution is the use of the three approaches to assess distress risk. Together with the full range of major performance measures, I provide the first comprehensive test of the competing approaches. This study has important implications for the future research agenda on both, how we measure distress risk and the pricing of distress risk.