Incorporating textual and management factors into financial distress prediction: A comparative study of machine learning methods

• Author: Xiaobo Tang,Wenxuan Shi,Shixuan Li,Mingliang Tan
• DOI: http://doi.org/10.1002/for.2661
• Date: 01 August 2020
• Published date: 01 August 2020

RESEARCH ARTICLE

Incorporating textual and management factors into

financial distress prediction: A comparative study of

machine learning methods

Xiaobo Tang | Shixuan Li | Mingliang Tan | Wenxuan Shi

School of Information Management,

Wuhan University, Wuhan, China

Correspondence

Shixuan Li, School of Information

Management, Wuhan University,

No. 299 Bayi Road. Wuhan, Hubei, China.

Email: shixuan.li@hotmail.com

Funding information

National Natural Science Foundation of

China, Grant/Award Number: 71673209

Abstract

Financial distress prediction (FDP) has been widely considered as a promising

approach to reducing financial losses. While financial information comprises

the traditional factors involved in FDP, nonfinancial factors have also been

examined in recent studies. In light of this, the purpose of this study is to

explore the integrated factors and multiple models that can improve the pre-

dictive performance of FDP models. This study proposes an FDP framework to

reveal the financial distress features of listed Chinese companies, incorporating

financial, management, and textual factors, and evaluating the prediction per-

formance of multiple models in different time spans. To develop this frame-

work, this study employs the wrapper-based feature selection method to

extract valuable features, and then constructs multiple single classifiers,

ensemble classifiers, and deep learning models in order to predict financial dis-

tress. The experiment results indicate that management and textual factors can

supplement traditional financial factors in FDP, especially textual ones. This

study also discovers that integrated factors collected 4 years prior to the

predicted benchmark year enable a more accurate prediction, and the ensem-

ble classifiers and deep learning models developed can achieve satisfactory

FDP performance. This study makes a novel contribution as it expands the pre-

dictive factors of financial distress and provides new findings that can have

important implications for providing early warning signals of financial risk.

KEYWORDS

deep learning, financial distress prediction, machine learning, management factors, textual

factors

1|INTRODUCTION

Financial distress prediction (FDP) has attracted a high

degree of attention in both academic and industrial fields

over the last few decades as it is an effective tool of risk

management (Geng, Bose, & Chen, 2015; Wanke, Bar-

ros, & Faria, 2015). Such predictions can provide early

warning signals of financial risks for relevant agents in

the economy, which is important for stakeholders, listed

companies, and even the development of the economy

itself (Farooq & Qamar, 2019). Based on such signals, and

prior to a crisis, the relevant investors can realize invest-

ment strategy adjustments, and firms and governments

are enabled to develop remedial measures, thereby

avoiding financial losses to a certain extent (Wang,

Chen, & Chu, 2018). For these reasons, numerous studies

Received: 18 September 2019 Accepted: 12 January 2020

DOI: 10.1002/for.2661

Journal of Forecasting. 2020;39:769–787. wileyonlinelibrary.com/journal/for © 2020 John Wiley & Sons, Ltd. 769

have focused on the prediction of financial distress; spe-

cifically, exploring predictive factors and prediction

methods has been a key undertaking in this body of

research.

The factors that are involved in FDP include financial

and nonfinancial factors. Early studies in this field

mainly examined financial factors, including profitability,

solvency, operational capabilities, and finance structure

(Hua, Wang, Xu, Zhang, & Liang, 2007; Sun & Li, 2008).

Although financial factors do contribute significantly to

FDP, they only cover quantitative information, which

cannot comprehensively represent the status of compa-

nies. In light of this, some of the latest studies have

highlighted the limited predictive capacity of financial

variables, and have attempted to utilize certain non-

financial factors in predicting financial distress

(Cecchini, Aytug, Koehler, & Pathak, 2010; Hajek, Olej, &

Myskova, 2014). Specifically, using textual analysis

methods such as word frequency statistics and sentiment

analysis, qualitative nonfinancial predictive factors can

be extracted from unstructured texts, such as audit

reports and annual reports; these textual factors can then

supplement the traditional financial factors used in FDP

(du Jardin, 2016; Wang et al., 2018).

The prediction method of financial distress can, typi-

cally, be divided into two categories: statistical methods

and machine learning methods. Early research focused

on statistical methods, which mainly include linear dis-

criminant analysis (LDA), multiple discriminant analysis

(MDA), factor analysis (FA), and probabilistic modeling

(Chen, Ribeiro, & Chen, 2016; Kumar & Ravi, 2007;

Suntraruk, 2010). While these statistical methods play a

significant role in studies on FDP, when the restrictive

assumptions that these models depend on are not satis-

fied, their validity and applicability become limited

(Chen et al., 2016; Wang et al., 2018). Therefore, studies

in recent years have focused on using machine learning

methods for the purposes of FDP, such as the support

vector machine (SVM), decision tree (DT), and artificial

neural networks (ANN; Zhou et al., 2014; Geng et al.,

2015; Olson, Delen, & Meng, 2012). With the exception of

these single classifiers, studies have also constructed FDP

models based on ensemble classifiers (Liang, Tsai, Dai, &

Eberle, 2018; Wang, Hao, Ma, & Jiang, 2011). Moreover,

deep learning models have also received attention in the

field of FDP (Alexandropoulos, Aridas, Kotsiantis, &

Vrahatis, 2019).

While previous studies have clearly made contribu-

tions to the area of FDP, optimizing the predictive factors

and prediction models is still arguably in need of

improvement. In order to achieve this with regard to pre-

dictive factors, this study combines traditional financial

factors with management and textual factors; for

prediction models, this study constructs single classifiers,

ensemble classifiers, and deep learning models to predict

financial distress. To the best of our knowledge, this is

the first study to utilize three types of factors and both

traditional machine learning and deep learning models

to predict financial distress in listed Chinese companies.

In so doing, this study attempts to answer the following

research questions:

RQ1. Which are the key influencing factors in FDP of

listed Chinese companies?

RQ2. Which models perform better than others in FDP of

listed Chinese companies?

RQ3. How early can signs of financial distress be

predicted in listed Chinese companies?

In order to engage with these research objectives, this

study utilizes data from 3, 4, and 5 years ahead of the

predicted benchmark year, revealing changes in the key

predictive factors of different years, and constructing

multiple machine learning models to predict financial

distress. It is hoped that the results of this research can

provide a template for early warning mechanisms for rel-

evant economic agents, so that they can make the

corresponding efforts to avoid financial losses.

The main contributions of this study are as follows.

First, it provides a novel FDP method that integrates

multiple predictive factors, time spans, and classifier

models. Second, it highlights the discovery that man-

agement and textual features can play a significant role

in FDP for listed Chinese companies; specifically, the

study finds that these features can achieve optimum

performance 4 years prior to the predicted benchmark

year, and reveals the superiority of ensemble classifiers

and deep learning models. Finally, the methods described

in this paper may be seen as potential approaches that

can concretely be applied in companies or at the level of

economic policy towards avoiding the financial losses.

The remainder of this paper is organized as follows.

Section 2 outlines the relevant studies on the prediction

of financial distress. Section 3 describes the study frame-

work and explains the analysis approaches in detail,

followed by a presentation of the main research results in

Section 4. Following this, Section 5 provides a discussion

of the empirical results and answers to the research ques-

tions. The final section concludes this study and provides

directions for future research.

2|RELATED STUDIES

This section describes the different factors and models

used in studies of FDP, as summarized in Table 1,

770 TANG ET AL.