Bankruptcy prediction with financial systemic risk

ABSTRACT

Financial systemic risk – defined as the risk of collapse of an entire financial system vis-à-vis any one individual financial institution – is making inroads into academic research in the aftermath of the late 2000s Global Financial Crisis. We shed light on this new concept by investigating the value of various systemic financial risk measures in the corporate failure predictions of listed nonfinancial firms. Our sample includes 225,813 firm-quarter observations covering 8,604 US firms from 2000 Q1 to 2016 Q4. We find that financial systemic risk is incrementally useful in forecasting corporate failure over and above the predictions of the traditional accounting-based and market-based factors. Our results are stronger when the firm in consideration has higher equity volatility relative to financial sector volatility, smaller size relative to the market, and more debts in current liabilities. The combined evidence suggests that systemic risk is a useful supplementary source of information in capital markets.

KEYWORDS:

• Bankruptcy prediction
• Systemic risk
• Hazard model

JEL CLASSIFICATIONS:

• G33
• G32
• C35
• G10

Acknowledgement

We are grateful to the editor Professor Chris Adcock, especially two anonymous referees, and seminar participants at International Finance and Banking Society (IFABS) Asia Conference 2017 (Ningbo, China), the 8th Financial Engineering and Banking Society (FEBS) International Conference 2018 (Rome, Italy), the 30th Anniversary Conference of China Economic Association (CEA) 2018 (Edinburgh, UK), University of Sheffield, University of Leicester, for helpful comments and discussions. We would also like to thank Stefano Giglio, Bryan Kelly and Seth Pruitt for generously sharing their code and data with the public. All errors are our own.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 See e.g. Ivashina and Scharfstein (2010), Lemmon and Roberts (2010), Chava and Purnanandam (2011).

2 Available at: https://fred.stlouisfed.org/.

3 Considering the existence of the time gap between systemic risk in the financial sector and its transporting to individual firms, we take one month/quarter lag for all systemic risk measures and our proposed systemic risk variables in predictions depending on our observation interval.

4 We would like to thank Giglio, Kelly, and Pruitt for sharing these measures data and code, which can be downloaded from https://sethpruitt.net/research/downloads/.

5 Note size concentration is calculated based on the top 100 financial institutions (Giglio, Kelly, and Pruitt 2016), and WSF is calculated based on the top 20 banks (López-Espinosa et al. 2012).

6 Appendix Table A.1b lists the details of indicators used in this study.

7 Net Income to Market-valued Total Assets (NIMTA) is an alternative profitability measure (e.g. Campbell, Hilscher, and Szilagyi 2008; Gupta and Chaudhry 2019). As far as we are concerned, NIMTA is a ‘half market variable, as it is constructed by dividing net income by market value of total assets. However, we construct pure accounting variables which enable us to make a comparison of the performance of different types of measures (e.g., accounting/market/systemic risk measures) in the prediction.

8 If less than 20 institutions are available, we construct volatility measures from all available institutions.

9 See Appendix Table A.1a for the systemic risk measures in detail.

10 The sample exceptions are January 2000 to January 2010 for the measurement of GZ and January 2000 to September 2010 for intl. spillover (See Giglio, Kelly, and Pruitt 2016 for more details).

11 The Internet Appendix reports the correlation matrix of variables used by Shumway (2001) and Campbell, Hilscher, and Szilagyi (2008) and our novel systemic risk measures.

12 See Appendix 2.

13 Note we do not report the results of one-quarter prediction here since they can see the same trends as two-quarter predition shows. However, they are presented in the Internet Appendix.

14 We do not report results of one-quarter prediction here.

15 Available at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3423328.

16 AUC is between 0 and 1, which captures the relationship between the type I and type II errors. See the Internet Appendix for detail.

17 See Cox and Snell (1989), Nagelkerke (1991) and Efron (1978).

18 Note that we use the version suggested by Osius and Rojek (1992).

19 In the Internet Appendix, we illustrate the coefficients of predictors with 83,795 monthly observations, which cover the period from January 2000 to January 2012.

20 See also in the Internet Appendix. There are 44,536 firm-year observations from 2000 to 2016 in yearly predictions.