Corporate distress prediction in China: a machine learning approach

Abstract

Rapid growth and transformation of the Chinese economy and financial markets coupled with escalating default rates, rising corporate debt and poor regulatory oversight motivates the need for more accurate distress prediction modelling in China. Given China's historical, social and cultural intolerance towards corporate failure, this study examines the Special Treatment system introduced by Chinese regulators in 1998. Regulators can assign Special Treatment status to listed Chinese companies for poor financial performance, financial abnormality and other events. Using an advanced machine learning model known as TreeNet Ò we model more than 90 predictor variables, including financial ratios, market returns, macro-economic indicators, valuation multiples, audit quality factors, shareholder ownership/control, executive compensation variables, corporate social responsibility metrics and other variables. Based on out-of-sample tests, our TreeNet Ò model is 93.74 percent accurate in predicting distress (a Type I error rate of 6.26 percent) and 94.81 percent accurate in predicting active/healthy companies (a Type II error rate of 5.19 percent). Variables with the strongest predictive value in the TreeNet Ò model includes market capitalization and annual market returns, macro-economic variables such as gross domestic product growth, financial ratios such as retained earnings to total assets and return on assets; and certain non-traditional variables such as executive compensation.

Equity Financial Assets: A Tool for Earnings Management—A Case Study of a Chinese Corporation

“…We can, therefore, predict the amount of fluctuation in current profits that would occur under the new CAS 22 if all financial assets were classified as FVTPL. In China, any listed firms with two consecutive losses are designated as ST (Special Treatment) companies, which can create strong incentives for earnings management (Jiang and Jones, ). Classifying financial assets as FVTPL will create a great deal of pressure for companies to classify financial assets as long‐term equity investments, measured using the equity method to avoid potential volatility.…”

A machine learning analysis of citation impact among selected Pacific Basin journals

“…This study utilises an advanced machine learning method known as TreeNet Ò which is a commercial version of a widely used method known as the gradient boosting machine (Friedman, 2001;Jiang and Jones, 2018). Gradient boosting machines are used extensively in the literature and are among the most powerful methods available in terms of predictive performance (see Hastie et al, 2009).…”

“…Jones et al (2015Jones et al ( , 2017 provide evidence that the gradient boosting model and related machine learnings methods, including AdaBoost and random forests, strongly outperform conventional classifiers such as logit and probit (Hastie et al, 2009). They are a data analysis approach which uses predictive ability criteria for the selection and ranking of input variables (Jones, 2017a;Jiang and Jones, 2018). For instance, they are particularly useful in high dimensional settings (i.e., large number of predictor variables relative to sample size) and where there may be little theory driving the selection of alternative predictors.…”

“…Machine learning techniques do not rely on the rules of statistical inference (such as significance tests). They are a data analysis approach which uses predictive ability criteria for the selection and ranking of input variables (Jones, 2017a;Jiang and Jones, 2018). 3 The gradient boosting model (which is the conceptual foundation of TreeNet Ò ) is set out by Friedman (2001)…”

“…As noted by Jones (2017a) and Jiang and Jones (2018), the RVIs reported in Table 4 are scalars that measure the overall predictive power of a variable relative to all other predictors in the model. Partial dependence plots (or marginal effects) indicate the direction of the explanatory variables after controlling for all other model influences with respect to citation impact.…”