Statistical analysis of bankrupting and non-bankrupting stocks

Abstract

The recent financial crisis has caused extensive world-wide economic damage, affecting in particular those who invested in companies that eventually filed for bankruptcy. A better understanding of stocks that become bankrupt would be helpful in reducing risk in future investments. Economists have conducted extensive research on this topic, and here we ask whether statistical physics concepts and approaches may offer insights into pre-bankruptcy stock behavior. To this end, we study all 20092 stocks listed in US stock markets for the 20-year period 1989–2008, including 4223 (21 percent) that became bankrupt during that period. We find that, surprisingly, the distributions of the daily returns of those stocks that become bankrupt differ significantly from those that do not. Moreover, these differences are consistent for the entire period studied. We further study the relation between the distribution of returns and the length of time until bankruptcy, and observe that larger differences of the distribution of returns correlate with shorter time periods preceding bankruptcy. This behavior suggests that sharper fluctuations in the stock price occur when the stock is closer to bankruptcy. We also analyze the cross-correlations between the return and the trading volume, and find that stocks approaching bankruptcy tend to have larger return-volume cross-correlations than stocks that are not. Furthermore, the difference increases as bankruptcy approaches. We conclude that before a firm becomes bankrupt its stock exhibits unusual behavior that is statistically quantifiable. Copyright c © EPLA, 2012 Introduction. – How to predict bankruptcy before it occurs is an open challenge. The most recent financial crisis [1] was caused by sub-prime mortgages written in 2006, and it contributed to the Lehman demise in September 2008. The bankruptcies of many other corporations at that time also resulted in substantial losses to investors. The general consensus is that if we could accurately predict bankruptcy, i.e., identify a characteristic behavior exhibited by a stock before bankruptcy, it would help investors avoid such losses. Thus bankruptcy prediction is a topic of great interest, not only to investors, but also to researchers across a wide range of fields. Beginning as far back as 1966, the attempt to predict corporate failure has been an active topic of research [2–6]. Most of this research has attempted to predict bankruptcy by using such models as neural networks, logit, quadratic interval logit, support vector machine, and AdaBoost and Bankruptcy Risk [7–12], but these models depend upon (a)E-mail: liqian@bu.edu (b)E-mail: jphuang@fudan.edu.cn the availability of detailed financial information about the corporation being studied [3,4,7–20]. Because it is often difficult to obtain accurate internal financial information about a corporation in a timely fashion, these forecasting models are of limited utility [16]. Here we attempt to understand a corporation’s risk of bankruptcy by observing the market dynamics of the price of its stock. We hypothesize that because a stock price reflects the expectation of investors, an important factor in the pool of public information, analyzing stock price movement may provide important clues for predicting bankruptcy [21]. To test this hypothesis, we begin by examining data from the U.S. stock market, comparing the statistical properties of stocks approaching bankruptcy with those of stocks that are not [21–34]. The significant differences we find may prove useful in forecasting corporate bankruptcies. Database and variables. – Using the database from The Center for Research in Security Prices (CRSP), we collect the daily closing share prices and trading volumes

6 Figures and Tables

Cite this paper

@inproceedings{Li1a1982StatisticalAO, title={Statistical analysis of bankrupting and non-bankrupting stocks}, author={Qian Li1a and Fengzhong Wang and Jianrong Wei and Yuan Liang and Jiping Huang2b and H. Eugene Stanley}, year={1982} }