Predicting multistage financial distress: Reflections on sampling, feature and model selection criteria

• DOI: http://doi.org/10.1002/for.2588
• Published date: 01 November 2019
• Author: Muhammad Ali Jibran Qamar,Umar Farooq
• Date: 01 November 2019

RESEARCH ARTICLE

Predicting multistage financial distress: Reflections on

sampling, feature and model selection criteria

Umar Farooq

1,2

| Muhammad Ali Jibran Qamar

2

1

FAST School of Management, National

University of Computer and Emerging

Sciences (Faisalabad Campus), Faisalabad,

Pakistan

2

COMSATS, University Islamabad

(Lahore Campus), Lahore, Pakistan

Correspondence

Umar Farooq, FAST School of

Management, National University of

Computer and Emerging Sciences,

Faisalabad Campus, Pakistan

Email: farooq_umar57@yahoo.com

Abstract

This research proposes a prediction model of multistage financial distress

(MSFD) after considering contextual and methodological issues regarding sam-

pling, feature and model selection criteria. Financial distress is defined as a

three‐stage process showing different nature and intensity of financial prob-

lems. It is argued that applied definition of distress is independent of legal

framework and its predictability would provide more practical solutions. The

final sample is selected after industry adjustments and oversampling the data.

A wrapper subset data mining approach is applied to extract the most relevant

features from financial statement and stock market indicators. An ensemble

approach using a combination of DTNB (decision table and naïve base hybrid

model), LMT (logistic model tree) and A2DE (alternative N dependence esti-

mator) Bayesian models is used to develop the final prediction model. The per-

formance of all the models is evaluated using a 10‐fold cross‐validation

method. Results showed that the proposed model predicted MSFD with

84.06% accuracy. This accuracy increased to 89.57% when a 33.33% cut‐off

value was considered. Hence the proposed model is accurate and reliable to

identify the true nature and intensity of financial problems regardless of the

contextual legal framework.

KEYWORDS

bankruptcy, ensemble model, financial distress, prediction model

1|INTRODUCTION

In recent years, due to intense competition, dynamic

environment, and adverse economic conditions, more

firms are facing financial difficulties. A financially trou-

bled firm might generate inadequate returns or may not

fulfil its obligations to become bankrupt. These adverse

outcomes affect decisions and benefits of both internal

and external stakeholders (Branch, 2002). Creditors want

to ensure the repayment of their outstanding debt, share-

holders seek profits, employees want a secure position,

government collects taxes and society is concerned with

its social benefits. These stakeholders require accurate

imminent information of corporate health to make their

decisions accordingly and timeously.

Hence various studies have tried to develop early iden-

tification models of bankruptcy and financial distress.

However, these prediction models are criticized for their

contextual application and methodological issues. For

instance, Opoku, Amon, and Arthur (2015) found that

many prediction models forecast legal bankruptcy. Such

bankruptcy prediction models (BPMs) are contextual, as

bankruptcy laws, accounting standards, and economic

environments are country specific (Levratto, 2013). Simi-

larly, BPMs are criticized due to ex post definition of dis-

tress (Farooq, Qamar, & Haque, 2018), sampling methods

Received: 6 August 2018 Revised: 18 January 2019 Accepted: 8 March 2019

DOI: 10.1002/for.2588

632 © 2019 John Wiley & Sons, Ltd. Journal of Forecasting. 2019;38:632–648.wileyonlinelibrary.com/journal/for

(Balcaen & Ooghe, 2006), data mining techniques (Sun,

Li, Huang, & He, 2014) and modelling tools (Kumar &

Ravi, 2007). Such contextual and methodological issues

affect the accurate application of BPMs.

This paper intends to propose a forecasting model of

corporate health while considering such contextual and

methodological issues. This research follows a multistage

framework of financial distress as proposed by Farooq

et al. (2018). It is argued that financial distress is a preced-

ing stage of bankruptcy. Therefore, a prediction model of

financial distress can provide an ex ante solution regard-

less of the differences in the legal framework. However,

Farooq et al. divided financial distress into three adverse

stages: profit reduction (PR), mild liquidity (ML), and

severe liquidity (SL). They argued that at the first stage

firms face PR or ML issues. Subsequently, the continuity

of PR and ML create an SL problem that eventually

results in bankruptcy. Farooq et al. recommended that

future studies should develop a model to forecast such

multistage financial distress (MSFD) to predict accurate

and relevant future prospects.

This research extends the work of Farooq et al. (2018)

and proposes a model to predict PR, ML, and SL. How-

ever, special attention is given to methodological issues

regarding sampling, feature selection and modeling tech-

nique. For instance, the literature explored that cross‐

industry and imbalance data are two key sampling issues

that could affect the accuracy of a BPM (Sun et al., 2014).

To deal with these issues, industry‐adjusted data and an

oversampling method as recommended by Zhou (2013)

are applied. Similarly, selection of relevant variables after

applying an appropriate data mining technique is a vital

decision in developing a prediction model. This research

applied a “wrappers subset”data mining approach with

10‐fold cross‐validation, as suggested by Hall and Holmes

(2003), to extract relevant features from two sets of tradi-

tional financial statement and stock market indicators.

Similarly, an ensemble approach using three classifiers

of DTNB (decision table and naïve base), LMT (logistic

model tree), and Bayesian A2DE (averaged N dependence

estimator) via bagging optimization and one‐versus‐all

decomposition method is applied. These models are

selected after a systematic evaluation process. As case

research, data of Pakistani nonfinancial firms is selected.

In Pakistan, little attention is given to studying corporate

failure, especially from a prediction perspective. For the

last few years, several Pakistani nonfinancial firms have

shown a downward performance. In 2015, 47 nonfinan-

cial firms showed negative equity where total liabilities

(Rs. 531 billion) surpassed their total assets (Rs. 34337 bil-

lion), indicating an expected loss of Rs. 187.635 billion for

creditors if these firms went bankrupt (SBP, 2015). Such

high numbers of financially troubled firms and potential

losses make it imperative to study the early prediction

of financial distress for Pakistani nonfinancial firms. This

research will help stakeholders to realize the nature and

intensity of financial distress in a timely manner that will

assist in developing and applying remedial strategies

accordingly.

2|THEORETICAL BACKGROUND

The literature of bankruptcy prediction is rich and has

been studied under different research settings. For

instance, in their meta‐analyses, Kumar and Ravi (2007)

explored 128 models, Bahrammirzaee (2010) discussed

278 models, Abdou and Pointon (2011) analyzed 214

models and, more recently, Opoku et al. (2015) reviewed

137 prediction models of bankruptcy and financial dis-

tress. However, a prediction model is useful only if it fore-

casts financial health accurately as its incorrect

development and application could have multiple nega-

tive outcomes (Trabelsi, He, He, & Kusy, 2015). Prior

studies have attempted to develop accurate prediction

models but many of the techniques and tools are criti-

cized for their contextual and application perspectives

(Balcaen & Ooghe, 2006; Farooq et al., 2018; Opoku

et al., 2015). These issues and their relevant solutions in

the light of the literature are discussed in the following

sections.

2.1 |Definition of distress

Opoku et al. (2015) found that most prediction models are

developed to forecast a dichotomous variable containing

two categories of bankrupt and nonbankrupt. These stud-

ies define bankruptcy as a situation where a firm is

declared insolvent by the court under a specific legal

framework. However, BPMs are criticized due to their

contextual application, ignoring other indicators of dis-

tress, creating sample bias and notifying about bank-

ruptcy when it is too late or difficult to turn around

(Balcaen & Ooghe, 2006; Kumar & Ravi, 2007; Sun

et al., 2014). Therefore, contemporary literature empha-

sizes an ex ante approach and focuses on predicting

financial distress rather than bankruptcy. In financial dis-

tress, firms are unable to meet financial obligations but

not declared bankrupt by the court. Hence a financial dis-

tress prediction model (FDPM) is independent of the

legal framework and provides ex ante information of

expected default (Pindado, Rodrigues, & de la Torre,

2008).

Some studies also criticize FDPMs due to their reliance

on a single criterion (Farooq et al., 2018; B. Tsai, 2013;

Turetsky & McEwen, 2001). Turetsky and McEwen

FAROOQ AND QAMAR 633