Abstract

Corporate bankruptcy and financial distress prediction is a topic of interest for a variety of stakeholders, including businesses, financial institutions, investors, regulatory bodies, auditors, and academics. Various statistical and artificial intelligence methodologies have been devised to produce more accurate predictions. As more researchers are now focusing on this growing field of interest, this paper provides an up-to-date comprehensive survey, classification, and critical analysis of the literature on corporate bankruptcy and financial distress predictions, including definitions of bankruptcy and financial distress, prediction methodologies and models, data pre-processing, feature selection, model implementation, performance criteria and their measures for assessing the performance of classifiers or prediction models, and methodologies for the performance evaluation of prediction models. Finally, a critical analysis of the surveyed literature is provided to inspire possible future research directions.

Introduction

Corporate bankruptcy prediction (BP) and financial distress prediction (FDP) have been a field of investigation by researchers around the world for nearly a half century. However, increasing attention has been paid to this evergreen subject since the Global Financial Crisis of 2008 (Alaminos et al., 2016), as predictions can have a significant impact on the decisions and returns of various stakeholders (Alam et al., 2021).

Corporate bankruptcy and financial distress events are not desirable (Gordon, 1971), as their legal and financial costs are prohibitive (Weiss, 1990). In addition, these events increase the expected costs for financial institutions, such as banks, to hedge against the risk of these events happening (Alnassar & Chin, 2015). To reduce exposure to risk and catch early warning signs, stakeholders including investors, bankers and governments are proactively looking for solutions to effectively analyze and predict corporate bankruptcy and financial distress events. The earliest study in terms of modern bankruptcy prediction can be traced back to 1932 when Fitzpatrick (1932) presented a successful way to distinguish between failed and healthy companies by analyzing 20 pairs of companies’ accounting ratios. Since the 1960s, several accounting-based statistical and probabilistic models have been proposed to predict corporate bankruptcy and financial distress (e.g., Beaver, 1966; Altman, 1968; Ohlson, 1980; Zmijewski, 1984; Zavgren, 1985). In addition, several studies focused on the identification of bankruptcy and financial distress drivers2 (e.g., Altman, 1968; Liang et al., 2016; Tobback et al., 2017; Tinoco et al., 2018; Mai et al., 2019), on one hand, and others focused on the design of methodologies for selecting such drivers (e.g., Tsai, 2009; Lin et al., 2014; Tian & Yu, 2017; Uthayakumar et al., 2020; Kou et al., 2021), on the other hand. Furthermore, a stream of literature investigated the causes of the under-performance of bankruptcy and financial distress prediction models such as data sample imbalance (e.g., Le et al., 2018; Veganzones & Séverin, 2018; Zoričák et al., 2020, Shen et al., 2020). In the era of the rise of machine learning and artificial intelligence technologies, the prediction of bankruptcy and financial distress has been lifted to a whole new level, as more and more new methodologies managed to improve prediction accuracy through the design features of new models (e.g., Ouenniche & Tone, 2017; Ouenniche et al., 2018a, 2018b, 2018c, 2019; Hosaka, 2019; Matin et al., 2019; Yuan et al., 2022), the type of implementation decisions such as over-sampling and under-sampling (e.g., Le et al., 2018; Sun et al., 2020; Shen et al., 2020; Du et al., 2020), and bagging and boosting (e.g., West et al., 2005; Zięba et al., 2016; Jones & Wang, 2019; Chen et al., 2020). On the other hand, another stream of research focused on reducing the computational requirements of prediction models or methods using parallel implementations that take advantage of today’s powerful computers, or to be more specific, GPUs (e.g., Huang & Yen, 2019; Le et al., 2019).

Conclusions and future research directions

The topic of corporate bankruptcy and financial distress prediction has attracted the attention of many researchers for several decades and continues to evolve with more and more advanced prediction methodologies and issue fixation solutions being proposed. This study contributes to this domain of research by providing an up-to-date state-of-the-art review, classification and critical analysis of the current BP and FDP literature, where six major research streams are identified and discussed, namely, the definition and coding of bankruptcy and financial distress; design of new prediction models/classifiers or new application of existing ones; design of new drivers or evaluation of existing ones; design or evaluation of feature selection methods; design of methodologies for the performance evaluation of prediction models; and issues affecting the performance of prediction models and related solutions. By painting the landscape of research and analyzing each research stream, this study would serve as a guide for researchers who intend to explore this field of study and/or contribute to its development.

Our analysis of the surveyed papers revealed a clear trend in terms of prediction methodologies of bankruptcy and financial distress, where more emphasis is put on advanced machine learning and artificial intelligence models such as ensemble learning models. Ensemble learning models and other methodological advances, such as DEA- and MCDA-based prediction methods, have contributed to further pushing the boundaries of research. Overall, it is fair to conclude that there is no single methodology that is better than the others, as each methodology has its strengths and weaknesses, some of which are design-related, and others are implementation decisions related, which have made the design and implementation of new ensemble models more popular.