Elsevier

Information Sciences

Volume 254, 1 January 2014, Pages 98-110
Information Sciences

Going-concern prediction using hybrid random forests and rough set approach

https://doi.org/10.1016/j.ins.2013.07.011Get rights and content

Abstract

Corporate going-concern opinions are not only useful in predicting bankruptcy but also provide some explanatory power in predicting bankruptcy resolution. The prediction of a firm’s ability to remain a going concern is an important and challenging issue that has served as the impetus for many academic studies over the last few decades. Although intellectual capital (IC) is generally acknowledged as the key factor contributing to a corporation’s ability to remain a going concern, it has not been considered in early prediction models. The objective of this study is to increase the accuracy of going-concern prediction by using a hybrid random forest (RF) and rough set theory (RST) approach, while adopting IC as a predictive variable. The results show that this proposed hybrid approach has the best classification rate and the lowest occurrence of Types I and II errors, and that IC is indeed valuable for going-concern prediction.

Introduction

The going-concern principle is one of the most important accounting assumptions in the preparation of financial statements. According to this principle, an entity or organization will continue its operations into the foreseeable future, at least, or in perpetuity. A going-concern opinion implies that the entity is not at risk of liquidation or even of reducing the scale of its operations substantially, whether voluntarily or involuntarily. Although auditors are not responsible for predicting bankruptcy or future events, according to Chen and Church [15], “going-concern opinions are useful in predicting bankruptcy and provide some explanatory power in predicting bankruptcy resolution.” Thus, going-concern prediction has been the focus of rigorous research efforts for decades. In particular, several researchers have suggested prediction models to aid auditors in conducting going-concern assessments of firms.

Prior studies on going-concern prediction are based primarily on conventional statistical techniques [22], [34] such as univariate statistical methods, multiple discriminant analysis (MDA), and logit and probit analyses. These conventional statistical methods, however, have some restrictive assumptions such as the linearity, normality, and independence of predictor or input variables. Considering that the violation of these assumptions occurs frequently within financial data [18], the methods have intrinsic limitations in terms of effectiveness and validity. Recently, Bellovary et al. [6] had undertaken an extensive review of going-concern prediction.

Artificial intelligence (AI) approaches such as inductive learning are less vulnerable to these violations. Moreover, AI aims to identify valid, novel, potentially useful, and understandable correlations and patterns in data [6]. AI can be an alternative solution to classification problems, given that data mining has shown to have better predictive capability than conventional statistical methods of going-concern prediction [2], [23], [26], [35], [37], [40]. Although the above-mentioned AI techniques have generally been shown to be effective in going-concern prediction, they are not without limitations.

First, since McKee [42] used financial ratios for going-concern prediction, most studies have primarily used financial ratios as independent variables. Financial ratios, originating in the financial statements of firms, can reflect some characteristics of a corporation. However, in the current knowledge era, the core competences of firms are derived from the knowledge and skills of their employees, and the value of intellectual capital (IC) now exceeds that of some tangible assets. Moreover, numerous researchers have recognized that the nature of a firm’s IC plays an important role in its risk of bankruptcy [57], [59]. Despite the growing importance of IC for firms in the knowledge era, it is usually excluded from early prediction models. Therefore, in this study, we believe that IC, which reflects the status of the corporation in going-concern predictions, will be a decisive factor that influences the predictive capability. Second, the studies mentioned above show that different researchers have used different independent variables as inputs for going-concern prediction. However, a few of the researchers used independent variables as a module of their going-concern prediction. These researchers did not pay much attention to finding and selecting important independent variables. Moreover, few studies employed these variables to generate the appropriate rules for going-concern decisions.

Recently, rough set theory (RST) [48], [49], [50], [51] is a relatively new approach in AI that has been extensively used for knowledge reasoning and knowledge acquisition. Using the concepts of lower and upper approximations in rough sets, the knowledge hidden in the information systems can be unraveled and expressed in the form of ‘‘if …, then …” decision rules [1], [47], [50], [51], [60], [62]. The extracted rules are easily interpretable, permitting complex relationships to be represented in an intuitive and comprehensible manner. The rules establish a relationship between descriptions of objects based on attributes and their assignment to specific classes. Moreover, the rules can be used for the classification of new objects [36]. Recently, it has found its application in a wide variety of fields including credit rating [17], [64], business failure [7], [61], knowledge acquisition [60], market decision-making [48], and early warning [7], [54], etc. Therefore, we attempt to investigate the effectiveness of RST approach in conducting the going-concern prediction tasks and to predict the characteristics of going-concern so decision-makers can understand the rules of going-concern.

Moreover, to determine and select important independent variables in the development of a going-concern prediction model, the random forest (RF) method is used in this study. RF is a relatively newer ensemble method that combines trees grown on bootstrap samples of data and a random subset bagging of predictor variables [10]. During the randomization of features, RF can provide an importance index of independent variables by calculating accuracy and the Gini index. Furthermore, the importance index captures the interactions among predictors through the randomizations of predictors [40]. In terms of robustness to outliers and noise, and calculation time, RF is superior to other machine learning methods such as bagging or boosting.

In order to make great use of the advantages of RF in preprocessing the business data, and further improve the classification accuracy of the RST predictor model, RF + RST is proposed to predict the going-concern in this work. Moreover, we examine whether an assessment of a corporation’s IC conveys any useful additional predictors in going-concern prediction. First, we use IC as a predictive (independent) variable. Second, RF is used to conduct variable selection because of its reliability in obtaining the significant independent variables. Third, the obtained significant independent variables from RF are used as inputs for the RST model. Fourth, we generate meaningful rules using RST for going-concern prediction. Fifth, to validate the effectiveness of our model, comparative experiments are conducted. Finally, to examine the effect of IC, we also compare the obtained results to see whether the model including IC gives better classification accuracy.

The remainder of this paper is organized as follows: In Section 2, we conduct a literature review about going-concern prediction and IC. In Section 3, we present the methodologies used in previous research, which are relevant to our paper for RF and RST. In Section 4, we describe the experimental design of this study. In Section 5, we summarize and discuss the empirical results. Finally, in Section 6, we present the conclusions of this study and discuss the future research directions.

Section snippets

Going-concern

Going-concern prediction as a concept remains one of the most controversial areas of the auditing profession and has received much criticism since its origin in the eighteenth century. Lenard et al. [37] state that the auditor must provide the annual audit report on the financial condition of a company, which is consolidated with the company’s financial statements. One of the important things that an audit report should address is the likelihood of the survival of the company (remain a

Random forest

RF is another advanced method of machine learning. The classification is achieved by constructing an ensemble of randomized classification and regression trees (CART) [10]. The RF algorithm uses a combination of independent decision trees to model data and measure variable importance [10]. Each decision tree in a forest is constructed using a bootstrap sample from the data. Approximately one-third of the data instances are not used to grow the tree; these instances are termed the out-of bag

Experimental process

In the study, we propose a hybrid model (RF + RST), which combines RF and RST to improve the accuracy rate of going-concern prediction. Moreover, we investigate whether IC is useful as an additional predictor in going-concern opinions. Initially, we incorporate IC as a potential predictive variable. Then, we use RF to perform variable selection because of its reliability in obtaining the relative importance of the predictive variables. Subsequently, we use the important predictive variables

Importance ranking of variables

To determine the relative importance of the potential predictive variables, we employed the RF available in the R package random forest [39] to calculate the value of variable importance. This implementation is based on the original Fortran code authored by Leo Breiman, the inventor of RF.

Following Liaw and Wiener [39], our preliminary tests indicated that the performance of RF barely depends on the actual value of its hyper-parameters within a large interval, which is consistent with the

Conclusions

Corporate going-concern prediction plays a significant role in accounting and audit decisions. In the current knowledge era, the core competences of firms are derived from the knowledge and skills of their employees, and the value of IC now exceeds that of some tangible assets. In current prediction models, many researchers have focused on the financial ratios rather than IC. In this study, we use IC as the predictive variable and propose a hybrid model that combines RF and RST (the LEM2

Acknowledgments

The authors would like to thank the Editor-in-Chief and reviewers for their useful comments and suggestions, which were very helpful in improving this manuscript.

References (67)

  • P. Verwijmeren et al.

    Employee well-being, firm leverage, and bankruptcy risk

    Journal of Banking & Finance

    (2010)
  • Y. Wu et al.

    A comparison of alternative bankruptcy prediction models

    Journal of Contemporary Accounting & Economics

    (2010)
  • Z. Xiao et al.

    The prediction for listed companies’ financial distress by using multiple prediction methods with rough set and Dempster– Shafer evidence theory

    Knowledge-Based Systems

    (2012)
  • C.-C. Yeh et al.

    A hybrid approach of DEA, rough set and support vector machines for business failure prediction

    Expert Systems with Applications

    (2010)
  • C.-C. Yeh et al.

    A hybrid KMV model, random forests and rough set theory approach for credit rating

    Knowledge-Based Systems

    (2012)
  • E.H. Aboul et al.

    Rough sets data analysis in knowledge discovery: a case of Kuwaiti diabetic children patients

    Advance in Fuzzy Systems

    (2008)
  • P. Barnes et al.

    The auditor’s going concern decision: some UK evidence concerning independence and competence

    Journal of Business Finance & Accounting

    (1993)
  • B.K. Behn et al.

    Further evidence on the auditor’s going-concern report: the influence of management plans

    Auditing: A Journal of Practice & Theory

    (2001)
  • T. Bell et al.

    Empirical analysis of audit uncertainty qualifications

    Journal of Accounting Research

    (1991)
  • J.L. Bellovary et al.

    A review of going concern prediction studies: 1976 to present

    Journal of Business & Economic Research

    (1976)
  • G. Biau et al.

    Consistency of random forests and other averaging classifiers

    Journal of Machine Learning Research

    (2008)
  • N. Bontis

    Intellectual capital: an exploratory study that develops measures and models

    Management Decision

    (1998)
  • L. Breiman

    Random forests

    Machine Learning

    (2001)
  • Y. Cao et al.

    Predicting financial distress of Chinese listed companies using rough set theory and support vector machine

    Asia-Pacific Journal of Operational Research

    (2011)
  • J.V. Carcello et al.

    Audit committee composition and auditor reporting

    Accounting Review

    (2000)
  • J.R. Casterella et al.

    Modeling the audit opinions issued to bankrupt companies: a two-stage empirical

    Journal of Business Finance & Accounting

    (2000)
  • C.C. Chang, C.J. Lin, LIBSVM: a library for support vector machines, 2001. <http://www.csie.ntu.edu.tw/cjlin/libsvm>...
  • K.C.W. Chen et al.

    Default on debt obligations and the issuance of going-concern opinions

    Auditing: A Journal of Practice & Theory

    (1992)
  • D. Cornier et al.

    The auditor’s consideration of the going concern assumption: a diagnostic model

    Journal of Accounting, Auditing & Finance

    (1995)
  • M. Daubie et al.

    A comparison of rough sets and recursive partitioning induction approaches: an application to commercial loans

    International Transactions in Operational Research

    (2002)
  • E.B. Deakin

    A discriminant analysis of predictors of business failure

    Journal of Accounting Research

    (1972)
  • P. De Saa-Perez et al.

    A resource-based view of human resource management and organizational capability development

    International Journal of Human Resource Management

    (2002)
  • R. Diaz-Uriarte et al.

    Gene selection and classification of microarray data using random forest

    BMC Bioinformatics

    (2006)
  • Cited by (101)

    • Impacts of crisis on SME bankruptcy prediction models’ performance

      2023, Expert Systems with Applications
      Citation Excerpt :

      Performance metrics of the developed models were better for a combination of qualitative indicators and financial ratios than only financial ratios. This study is in this meaning in line with existing studies (Bragoli et al., 2021; Tang et al., 2020; Xie et al., 2011; Yeh et al., 2014). However, only a minimum of these differences was statistically significant.

    View all citing articles on Scopus
    View full text