Can board diversity predict the risk of

financial distress?
Umair Bin Yousaf, Khalil Jebran and Man Wang

Abstract Umair Bin Yousaf is based

Purpose – The purpose of this study is to explore whether different board diversity attributes (corporate at the School of
governance aspect) can be used to predict financial distress. This study also aims to identify what type of Accounting, China Internal
prediction models are more applicable to capture board diversity along with conventional predictors. Control Research Center,
Design/methodology/approach – This study used Chinese A-listed companies during 2007–2016. Dongbei University of
Board diversity dimensions of gender, age, education, expertise and independence are categorized into Finance and Economics,
three broad categories; relation-oriented diversity (age and gender), task-oriented diversity (expertise Dalian, China. Khalil Jebran
and education) and structural diversity (independence). The data is divided into test and validation sets. is based at the School of
Six statistical and machine learning models that included logistic regression, dynamic hazard, K-nearest
Business Administration,
neighbor, random forest (RF), bagging and boosting were compared on Type I errors, Type II errors,
Dongbei University of
accuracy and area under the curve.
Finance and Economics,
Findings – The results indicate that board diversity attributes can significantly predict the financial
Dalian, China. Man Wang is
distress of firms. Overall, the machine learning models perform better and the best model in terms of
based at the School of
Type I error and accuracy is RF.
Accounting, China Internal
Practical implications – This study not only highlights symptoms but also causes of financial distress,
which are deeply rooted in weak corporate governance. The result of the study can be used in future Control Research Center,
credit risk assessment by incorporating board diversity attributes. The study has implications for Dongbei University of
academicians, practitioners and nomination committees. Finance and Economics,
Originality/value – To the best of the authors’ knowledge, this study is the first to comprehensively Dalian, China.
investigate how different attributes of diversity can predict financial distress in Chinese firms. Further, this
study also explores, which financial distress prediction models can show better predictive power.
Keywords China, Machine learning, Financial distress, Board diversity, Relation-oriented diversity,
Task-oriented diversity, Structural diversity
Paper type Research paper

1. Introduction
A firm’s financial distress causes a substantial cost to investors and creditors including, but not
limited to, loss of sales and profits, reduced dividends, legal costs, high cost of further capital and
credit, tax avoidance, inability to issue new securities and the opportunity cost of positive NPV
projects (Bhattacharjee and Han, 2014; Habib et al., 2018; Zhou, 2019). Accordingly, much
attention has been paid to financial distress on both academic and practical ends. A vast majority
of financial distress prediction (FDP) literature can be classified into statistical and machine
Received 21 June 2020
learning models that use accounting and market ratios as predictors (Altman, 1968; Barboza et al., Revised 26 August 2020
5 October 2020
2017; Campbell et al., 2008; Hillegeist et al., 2004; Lohmann and Ohliger, 2019; Ohlson, 1980; 9 October 2020
Shumway, 2001; Taffler, 1983; Wang, 2017; Zmijweski, 1984). Several studies have explained the 17 November 2020
Accepted 8 December 2020
relationship between corporate governance and financial distress (Abdullah, 2006; Chaganti et al.,
1985; Daily and Dalton, 1994; Darrat et al., 2014; Elloumi and Gueyié, 2001; Fich and Slezak, The authors are thankful to
Gagan Deep Sharma
2008; Lajili and Zéghal, 2010; Lee and Yeh, 2004; Li et al., 2008; Muranda, 2006; Parker et al., (Associate Editor) and two
2002; Shahwan, 2015; Udin et al., 2017). However, incorporating corporate governance measures anonymous reviewers for many
insightful comments and
into FDP models have received less attention in the literature. suggestions.

DOI 10.1108/CG-06-2020-0252 © Emerald Publishing Limited, ISSN 1472-0701 j CORPORATE GOVERNANCE j

 Moreover, it has been realized that the accuracy of standard accounting-based FDP
models has dropped significantly in recent times (Beaver et al., 2012). Therefore,
researchers argue for additional predictors to be incorporated into FDP models (Altman
et al., 2010; Beaver et al., 2012).
Over the years, the diversity of the board of directors has captured a considerable debate
in corporate governance literature. Based on upper echelons (Hambrick and Mason, 1984),
agency (Jensen and Meckling, 1976) and resource dependence theories (Pfeffer and
Salancik, 1978), we argue that diversity is an essential element for an organization’s
success and if we incorporate board diversity attributes in our prediction models, the FDP
ability improves significantly. The purpose of this study is not to prove a causal relationship,
but to assess the predictive power of board diversity in predicting financial distress.
Considering the limitations of some methods and to add robustness to our results, we
compare statistical models (static and dynamic) with machine learning models. Thus, this
study also explores, which FDP models better use corporate governance information along
with macro-economic, accounting, market and growth information to predict financial
distress (FD).
Prior studies have mostly focused on one or two features of board diversity when examining
the relationship with FD (Kristanti et al., 2016; Manzaneque et al., 2015; Mittal and Lavina,
2018; Santen and Donker, 2009; Zhou, 2019). This study uses different facets of board
diversity, namely; gender, age, education, expertize and independence and categorizes
them into three broad dimensions of relation-oriented diversity, task-oriented diversity and
structural diversity to predict FD.
We used a sample of Chinese firms from 2007 to 2016. China provides a unique context to
explore the association between board diversity and FD. China has undergone an
enormous transition from a centrally planned to a market economy over the past three
decades. However, unnecessary protection of state-owned enterprises (SOEs), limited
access to financial resources that mainly concentrated on SOEs, poor protection of
creditors’ rights and weak legal infrastructure of bankruptcy make China a unique context
that cannot be generalized to Anglo-American studies (Bhat et al., 2019; Bhattacharjee and
Han, 2014; Sabbaghi, 2016; Wang and Deng, 2006).
Our study contributes to the literature in several dimensions. First, we integrate board
diversity with financial distress risk assessment and explore the predictive power of a wide
range of board diversity attributes in financial distress risk assessment. In the financial
distress risk assessment, the main focus is on predictive ability rather than causation.
However, our study allows us to predict more accurately as we explore causes (and not the
symptoms) of financial distress, which are deeply rooted in weak board diversification.
Second, most of the previous studies considered only one or two dimensions of
diversity, such as age or gender (Adams and Ferreira, 2009; Francoeur et al., 2008;
Mittal and Lavina, 2018; Talavera et al., 2018; Ullah et al., 2019; Zhou, 2019). However,
our study takes into account five key diversity facets and categorizes them into relation-
oriented (gender and age), task-oriented (education and expertize) and structural
(independence) diversity attributes. This is the first study to collectively incorporate
relation-oriented (gender and age), task-oriented (education and expertize) and
structural diversity (independence) attributes into FDP models.
Third, our study expands the FDP literature by considering the best predictors from
accounting, market, growth, macroeconomic and corporate governance variables by using
stepwise regression on feature selection. Finally, we further add to financial distress
literature by comparing popular static, dynamic and machine learning models. This study
uses a unique definition of financially distressed firms in the context of China where normally
special treatment (ST) stocks are used as a proxy of financially distressed firms.

j CORPORATE GOVERNANCE j
This study proceeds as follows. In Section 2, we review the literature on board diversity,
financial distress and FDP models. In Section 3, we discuss data and methods. In Section
4, we present the results including receiver operating characteristic (ROC) curves and
predictive accuracy of six FDP techniques. Section 5 concludes the study with a discussion
on findings, limitations and implications.

2. Literature review
Upper echelons theory (Hambrick and Mason, 1984) connotes that an organization is a
reflection of its top managers. The choices of individual top managers have a significant
impact on organizational outcomes (Kaur and Singh, 2019). Board is the top decision-
making body in a contemporary organization (Simpson and Gleason, 1999). Boards are
responsible for authorizing key strategic and financial decisions, such as recruitment and
evaluation of top executives e.g. CEOs, mergers and acquisitions, dividend policy and
changes in the capital structure (Adams and Ferreira, 2007; Clarke and Branson, 2012).

2.1 Board diversity

Board diversity refers to the heterogeneity of the board members, which may include
different attributes, such as age, gender, education, expertize, independence, tenure,
ethnicity, religion and nationality. The literature has classified board diversity into distinctive
categories using perspectives such as skills, experiences, duties, observable and non-
observable factors, personality, demographic, cognitive abilities and cohort membership
(Aggarwal et al., 2019; Bernile et al., 2018; Bhat et al., 2019; Ferna ndez-temprano and
Tejerina-gaite, 2020; Jackson et al., 1995; Jebran et al., 2020; Kim and Starks, 2016;
Milliken and Martins, 1996; Mustafa et al., 2017). Joshi and Jackson (2003) provide a
scheme for categorizing personal attributes of individuals at the top management team. The
scheme divides personal attributes into relation-oriented and task-oriented diversity.
Relation-oriented diversity refers to attributes that are less related to the job and imperative
in shaping interpersonal relationships. Task-oriented diversity attributes are highly related to
the job. Likewise, Cummings (2004) defines structural diversity attributes as the one, which
is related to positions or roles of members such as outsiders vs insiders and independent vs
other directors. Following the literature (Aggarwal et al., 2019; Bhat et al., 2019; Cummings,
2004; Jebran et al., 2020; Milliken and Martins, 1996), we classify gender and age into
relation-oriented diversity, education and expertize as task-oriented diversity and
independence as structural diversity.

2.2 Board diversity and financial distress

Two authoritative views explain the effect of board diversity on firm performance; namely,
agency view and resource-based view (Aggarwal et al., 2019). The agency view reflects
upon the monitoring role of the board and postulates that a diverse board enhances the
monitoring role of management, as it consists of directors from different backgrounds with
discrete opinions (Carter et al., 2003; Erhardt et al., 2003). The resource dependence view
(Pfeffer and Salancik, 1978), on the other hand, suggests that diverse boards are in a better
position to perform an advisory role because heterogeneous members bring high-quality
resources in the form of skills, knowledge, information and outside connections at their
disposal (Ali et al., 2014; Ferreira, 2010; Gul et al., 2011). Thus, based on agency and
resource-based views, it is argued that a more diverse board has better monitoring and
advisory capabilities, which can ultimately improve performance and mitigate financial
distress.
The literature shows mixed results concerning the effect of board diversity on financial
distress. Darrat et al. (2014) show that the presence of female directors mitigates financial
distress. Kristanti et al. (2016) studied Indonesian family firms and found a negative

j CORPORATE GOVERNANCE j
 association between gender diversity and financial distress. However, Salloum and Azoury
(2012) found an insignificant relationship between the presence of female directors and
financial distress. Santen and Donker (2009) also reflected that gender diversity has an
insignificant relationship with financial distress. Khaw et al. (2016) argue that as compared
to women, men are likely to take excessive risks. Platt and Platt (2012) showed an increase
in the average age of both board and CEO reduces the chances of bankruptcy. Ferna ndez-
temprano and Tejerina-gaite (2020) conclude that age diversity has a positive impact on
financial performance in Spain.
Wilson et al. (2014) found that there are fewer chances of insolvency when boards have
more experienced directors, whereas Salloum et al. (2013) found insignificant evidence to
support the argument that an insufficient experience affects financial distress. Adnan et al.
(2016) investigated the association of education diversity with firm performance and
concluded that education diversity has a negative relation with firm performance. Similarly,
Fernandez-temprano and Tejerina-gaite (2020) found a negative effect of education
diversity on the performance of directors. However, no previous study measured the effect
of education on financial distress.
Most of the prior studies (Appiah and Chizema, 2016; Darrat et al., 2014; Fich and Slezak,
2008; Manzaneque et al., 2015; Salloum et al., 2013; Wang and Deng, 2006) show the
presence of independent directors significantly reduces financial distress. Similarly, Lajili
and Zéghal (2010) indicate that financially distressed firms have shorter stays of
independent directors as compared to financially healthy firms. However, Chaganti et al.
(1985), Daily and Dalton (1994), Robinson et al. (2012) and Freitas Cardoso et al. (2019)
revealed that the difference in terms of outside (independent) directors between failed and
non-failed firms is insignificant. Based on the discussion so far, it is assumed that board
diversity can predict financial distress risk. Hence, we propose the following hypotheses:
H1. Board relation-oriented diversity (age and gender) can predict financial distress.
H2. Board task-oriented diversity (education and expertize) can predict financial
distress.
H3. Board structural diversity (board independence) can predict financial distress.

2.3 Models
Since the pioneering work of Altman (1968), many researchers have developed different
prediction models (Alaka et al., 2018; Barboza et al., 2017; Li and Wang, 2018; Sun et al.,
2014; Tsai et al., 2014; Wu et al., 2010). After examining these models, we divide them into
three broad categories of static, dynamic and machine learning techniques. The first two
categories are statistical models. Although there is an extensive list of available models
ranging from simple regression to neural networks, support vector machines and other data
mining techniques (Barboza et al., 2017; Farooq and Qamar, 2019; L. Bellovary et al., 2007;
Shah et al., 2020; Sun et al., 2014; Tsai et al., 2014; Wang et al., 2018), a total of six models
are selected based on popularity and uniqueness. The selected models for comparison are
logistic regression (static); dynamic hazard (dynamic); random forest (RF), bagging,
boosting and K-nearest neighbors (KNN) (machine learning) techniques.
2.3.1 Logistic regression. Firstly used by Ohlson (1980) in FDP, logistic regression is the
most extensively used statistical technique in FDP literature for reasons such as efficiency,
high interpretability, fewer requirements of computational resources and less tuning (Süsi
and Lukason, 2019). The technique does not require input features to be scaled. Logistic
regression is easy to regularize and provides calibrated predicted probabilities. It does not
require holding key assumptions of normality, homoscedasticity and measurement level,
which are commonly required in linear regression models (Shah et al., 2020).

j CORPORATE GOVERNANCE j
2.3.2 Dynamic hazard. Shumway (2001) applied a dynamic hazard model for the first time
in FDP. It measures financial distress risk at each point of time rather than a year before
financial distress. The model uses a discrete-time setting in survival analysis when applied
in different software i.e. R, Stata, etc. It is used to assess the expected time duration until
one event happens. The dynamic hazard model is superior to Cox proportional hazard
model, as its time-varying covariates are measured at times when financial reports are
presented while Cox proportional hazard assumes time covariates continuously (Li et al.,
2020). Moreover, the dynamic hazard model does not have problems such as selection
bias, non-adjustment of the period at risk and no account of duration dependence or age of
the firm. Such weaknesses are common in static or single-period models (Shumway, 2001;
Sun and Li, 2011; Zhou, 2015).
2.3.3 Random forest. The RF technique, also known as generalized classification and
regression trees, is a combination of tree predictors such that each tree depends on the
values of a random vector sampled independently and with the same distribution for all
trees in the forest (Breiman, 2001). Although RF’s precision is similar to AdaBoost, yet it is
more robust because it allows noise and the presence of outliers in training. Another
advantage of the RF technique is, it provides the importance of each variable in explaining
the dependent variable (Barboza et al., 2017; Maione et al., 2016).
2.3.4 Bagging. The bagging technique, also called bootstrap aggregating, is a method for
generating multiple versions of a predictor and using them to get an aggregated predictor.
The model uses independent classifiers that process slices of the data and combine these
slices through model averaging (Breiman, 1996). Bagging creates subsets of data that
reduce noise and get relatively high accuracy as compared to decision trees (Wang et al.,
2012).
2.3.5 Boosting. Boosting technique is based on the idea of Kearns and Vazirani (1994), that
posed the question of “whether a ‘weak’ learning algorithm, which performs just slightly
better than random guessing in the PAC model can be ‘boosted’ into an arbitrarily accurate
‘strong’ learning algorithm.” Boosting uses the base prediction rule repeatedly on the initial
set. The gbm package in R software calculates Boosting based on Friedman’s Gradient
Boosting Machine (Friedman et al., 2000).
2.3.6 K-nearest neighbors. It is a non-parametric supervised machine learning model. It has
a tuning parameter, k. The k parameter determines how the model is trained, instead of
learning through training. Commonly based on the Euclidean distance between a test
sample and the specified training samples, the KNN model assumes similar things exist in
close proximity (Li and Wang, 2018).

3. Data and methods

3.1 Sample
We used the data from Chinese A-listed companies listed at the Shanghai stock exchange
and the Shenzhen stock exchange, which is available on China Stock Market and
Accounting Research Database from 2007 to 2016. This period is chosen mainly because
major corporate governance reforms took place in April 2005 in China (Khaw et al., 2016).
Other reasons include the membership of China in the world trade organization in 2001 and
arrangements to protect the interest of minority shareholders just before 2007. Additionally,
the approval of new standards of accounting in China in 2005 and the introduction of non-
tradable shares reforms in 2005 also justify restricting the final study to the 2007–2016
period. Considering the different regulatory requirements and capital structure of financial
institutions, the sample data comprising only non-financial listed companies. To train
models and to add robustness, the sample is divided into training and validation sets. As
financial distress is a “protracted process of decline” and “downward spiral” (Daily and
Dalton, 1994; Hambrick and D’Aveni, 1988), we used five years, three years, one year and

j CORPORATE GOVERNANCE j
 contemporaneous year observations for each financially distressed and healthy firm in our
models.
Prior literature used different definitions of financial distress. A detail of these is provided in
Table 1. Studies conducted in the Chinese context mostly use ST stocks as a proxy for
financial distress. However, Sun et al. (2014) suggested that future studies should not rely
on single criteria when defining financial distress. Hence, we used a novel definition of
financial distress in the sense that we do not simply rely on ST stocks. Financially distressed
firms in the current study are those firms that meet at least two of the below-mentioned
conditions:

䊏 ST company.
䊏 Negative profit current year or decreasing profit for two years.
䊏 Interest coverage ratio of less than one.
䊏 Decreasing or negative net worth.

By doing this, we were not only able to extract actual FDFs from ST stocks but also select
firms that did not go through ST procedure but are in a financially distressed situation. We
also confirm our logic by calculating Altman’s (1968) z-score for these firms. Financially
distressed companies were either in “gray” or “distressed” zones.
Based on the definition set above, we identified 169 financially distressed firms. Out of
them, 34 companies had missing data, thus excluded. Healthy firms were randomly
selected from listed companies that do not fall under the definition of financial distress. Our
final sample comprised 295 firms of which 135 are financially distressed and 160 are
healthy. Using the 90/10 rule (Veganzones and Séverin, 2018), we further divided the
sample into training (n = 268) and validation sets (n = 27). This bifurcation is necessary;
firstly because machine learning models may suffer from overfitting problems in the training
phase (Barboza et al., 2017). Second, the predictive ability of different models depends on
performance on unseen data supplied in the validation set (Sun et al., 2014). Figure 1
depicts the sample selection process.
Using the aforementioned procedure, 135 FDFs are segregated into two parts; 122 firms for
training and 13 firms for validating the models, respectively. Besides, out of 160 healthy

Table 1 Financial distress definitions

Author and year Definition of financial distress

Doumpos and Zopounidis (1999) A condition where a firm is unable to pay its creditors, preferred stock shareholders, suppliers,
etc. or the firm file for bankruptcy
Bose (2006) The situation where the stock price of a company is less than 10 cents
Farooq et al. (2018) There are three stages of FD, namely, profit reduction(PR), mild liquidity(ML) and severe
liquidity (SL)
Salloum and Azoury (2012) The situation where the interest coverage ratio is less than 1 for two successive years or where it
is under 0.8 in a year
Bhattacharjee and Han (2014) Financial distress is comprising three conditions, namely, interest coverage is below 0.7 in the
given or prior year, reduction in fixed assets in the given or coming year, a decrease in share
capital in the given year
Manzaneque et al. (2015) The FDF meets the two conditions, namely, EBITDA is less than financial expenses for two
successive years; reduction in the market price for two continuous periods
Daily and Dalton (1994) Firms that filed Chapter 11 bankruptcy
Fich and Slezak (2008) Altman’s Z score < 1.81 and interest coverage ratio (ICR) < 1
Li et al. (2020);Wang and Deng (2006), Special treatment (ST) stocks are considered financially distressed
Zhoua et al. (2012)
Pessarossi and Weill (2013) FD is measured by two ratios, namely, total debt to total assets ratio (measures leverage),
current assets to current liabilities ratio (measures liquidity)

j CORPORATE GOVERNANCE j
 Figure 1 Sample selection process

Seng the criteria for selecng financially distressed firms

ST, net loss, interest coverage, networth

Selecng the FD firms that meet the criteria

169 Chinese A-listed firms are selected.

Deleng firms with missing values

34 firms with missig data are excluded. 135 FDFs remained in final sample

Selecng healthy firms

Random sampling is used to select healthy firms, keeping in mind to select at
least same number as FDFs from each industry. 160 healthy firms are selected.

Spling the sample into train and test sets

Using 90/10 criteria, sample firms (n = 295) are divided into train (n = 268) and validaon sets
(n = 27)

Train the models and validate them

through test set

firms, the train set has 146 firms and the validation set has 14 firms. Thus, to train our
models, we used an imbalanced set of 122 financially distressed firms and 146 financially
healthy firms. The use of an imbalanced data set where the minority class represents less
than 20% can jeopardize the results of FDP models (Veganzones and Séverin, 2018),
notwithstanding our sample has more than 45% firms in the minority class. Moreover, we
applied the synthetic minority oversampling technique (SMOTE) to verify if sample
imbalance affects results. The results of the balanced set using SMOTE were roughly the
same (the results from SMOTE are not reported for brevity, but are available on request).
The industry-wise division of the sample is provided in Table 2.

3.2 Methods
The study used six statistical and machine learning techniques; namely, logistic regression,
dynamic hazard, RF, bagging, boosting and KNN. The comparison of these techniques is
made in two steps. In the first step, accounting, market, growth and macro-economic
predictors (which were selected in stepwise regression) are used. In the second step, we
added board diversity attributes to these variables. Thus, we compare statistical and
machine learning techniques with and without board diversity facets. The results are
analyzed in terms of Type I error, Type II error, accuracy and area under the receiver
operating characteristic curve (AUC-ROC).
Type I error means the fraction of FDF that were predicted to be healthy. Type II error refers
to the fraction of healthy/non-financially distressed firms that were predicted to be financially
distressed. Accuracy of the models is calculated by dividing the accurate classifications

j CORPORATE GOVERNANCE j
 Table 2 Industry-wise sample distribution
Industry name HF FDF Training set Validation set

Commerce
Food and beverage 1 1
Hotels 1 1
Other wholesale 1 0 1
Retail trade 4 2 6
Wholesale and retail trade 6 5 9 2
Wholesale of material, energy and electric equipment 1 1
11 10 18 3
Conglomerates
Agriculture 1 1 2
Conglomerates 1 2 3
Decoration 3 2 5
Forestry 2 2
Graziery 1 0 1
Support services for farming, forestry, animal husbandry and fishery 1 0 1
7 7 11 3
Industrials
Automobile manufacturing 1 1 2
Beverages 4 4 6 2
Chemical fibre manufacturing 1 0 1
Chinese medicines manufacturing 1 0 1
Coal mining and quarrying 1 1
Communication apparatus manufacturing 1 0 1
Computer and related equipment manufacturing 1 0 1
Steam, electric power and hot water supply and generation 2 5 5 2
Electrical equipment/machinery manufacturing 11 9 20
Components and appliance (electronic) 2 0 2
Ferrous metal smelting and extruding 1 4 5
Manufacturing (food) 2 2
Processing (food) 6 3 9
Manufacturing of garment and related fabric products 1 1
Production and supply of gas 2 2
Machinery manufacturing (general) 3 6 8 1
Information technology (IT) 10 11 20 1
Manufacture of petroleum, chemical, rubber and plastic products 2 2 3 1
Manufacturing (medicine) 15 1 14 2
Metal products 4 0 3 1
Nonferrous metal mining 1 1
Metal smelting, drawing, rolling and extruding 3 4 7
Mineral products (non-metallic) 5 3 7 1
Manufacturing (other) 2 0 1 1
Paper products 1 0 1
Petroleum processing and coking 1 2 2 1
Plastics manufacturing 1 0 1
Printing 2 0 2
Raw chemical materials and chemical products 9 13 20 2
Rubber manufacturing 2 0 1 1
Special equipment manufacturing 3 3 6
Support services for mining 2 2
Textile 1 2 3
Processing of timber, palm, rattan, bamboo and grass items 2 1 3
Manufacturing (transportation equipment) 13 7 19 1
Civil engineering construction 6 1 7
Real estate 3 7 10
119 98 200 17
Public utility
Air transportation 1 1
(continued)

j CORPORATE GOVERNANCE j
 Table 2
Industry name HF FDF Training set Validation set

Arts 1 1 2
Communication service 1 0 1
Computer application service 9 8 16 1
Health care, nursing care services 2 1 3
Other public services 5 6 8 3
Postal services 1 0 1
Professional and scientific research services 2 1 3
Public facilities services 1 1 2
Support service for transportation 1 1
Water transportation 1 0 1
23 20 39 4
Grand total 160 135 268 27

(sum of True positives and true negatives) by a total number of observations selected as
under:

TP þ TN
Accuracy ¼ ; (1)
TP þ FP þ TN þ FN

where true positive (TP) indicates financially distressed firms predicted correctly, false
positive (FP) means incorrectly predicted financially distressed firms, true negative (TN)
refers to healthy firms predicted correctly and false negative (FN) are the healthy firms
predicted incorrectly. Sensitivity is the probability that a model will, given a cut-off point,
classify a firm as FDF when it is indeed financially distressed. Specificity is the probability
that a model classifies a firm as healthy when it is indeed healthy. For financial distress,
there is a preference for low Type I error or higher sensitivity because of substantial losses
to investors. As we used an imbalanced set, the cut-off point was set as a percentage of the
FD firm-year observations divided by total firm-year observations (Cenciarelli et al., 2018).
AUC-ROC shows how much a model is capable of distinguishing between financially
distressed and healthy firms. The use of AUC-ROC is found frequently to compare different
models in FDP studies (Barboza et al., 2017; Le et al., 2018). A higher AUC value is
preferable as it represents a greater discriminating ability of the model. A model with 0.5
AUC is equal to random guessing and a model with 1 AUC value is the best one.
All the predictive variables are included at their original values. The procedure of using
original data without any transformation is also followed by different researchers (Barboza
et al., 2017; Tsai et al., 2014). This procedure may hamper the predictability of our models,
but we intend to check the capability of FDP techniques without undergoing any data
treatment.
We used different packages in R software to run the models. Logistic regression is
executed using glm function, dynamic hazard using dynamichazard package, RF and
bagging using rfsrc, boosting using gbm and KNN using class package. For bagging and
RF, same package is used as it is a special case of a RF where mtry is equal to the number
pffiffiffiffi
of predictors, p. For RF, we use mtry equal to p.
3.2.1 Variable selection. To construct our baseline model, we used stepwise regression
using accounting, market, growth, macro-economic and corporate governance variables
from prior literature. Corporate governance variables were left out, as these variables did
not add predictive power to our models. The variables used in stepwise regression are
mentioned in Table 3. To avoid multi-collinearity issues, variables with VIF values greater
than 3 are dropped.

j CORPORATE GOVERNANCE j
 Table 3 Variables used in step-wise regression
Variable name Calculations Authors

Accounting ratios
Return on assets Net income divided by total assets Bellovary et al. (2007), Ohlson (1980)
(ROA)
Current ratio Current assets divided by current liabilities Bellovary et al. (2007), Zmijweski (1984)
Net working capital current assets minus current liabilities and then Altman (1968), Bellovary et al. (2007); Ohlson
ratio divided by total assets (1980), Wu et al. (2010)
EBIT to Total assets Earnings before interest and taxes (EBIT) divided by Altman (1968), Bellovary et al. (2007); Wu et al.
ratio total assets (2010)
Operating return on Operating profit divided by total assets Bellovary et al. (2007)
assets (OROA)
Debt ratio Total liabilities divided by total assets Bellovary et al. (2007), Ohlson (1980); Shumway
(2001), Zmijweski (1984)
Interest coverage ratio Earnings before interest and taxes (EBIT) divided by Bellovary et al. (2007), Tinoco and Wilson
interest expense (2013); Zhou (2019)
Retained earnings to Retained earnings divided by total assets Altman (1968), Altman et al. (2016)
total assets ratio
Asset turnover ratio Sales divided by total assets Altman (1968), Wu et al. (2010)
Net income to total Net income divided by total liabilities Zmijweski (1984)
liabilities ratio
Market ratios
Relative size Log of the number of outstanding shares multiplied Campbell et al. (2008), Chan et al. (2016); Darrat
by year-end share price then divided by total market et al. (2014), Oz and Simga-Mugan (2018);
value Shumway (2001), Wu et al. (2010)
Sigma The standard deviation of each firm’s daily stock Campbell et al. (2008), Chan et al. (2016); Darrat
return over the past 3 months et al. (2014), Oz and Simga-Mugan (2018);
Shumway (2001), Wu et al. (2010)
Excess return Cumulative annual return in year t-1 minus the value- Campbell et al. (2008), Chan et al. (2016); Darrat
weighted index return in year t-1 et al. (2014), Oz and Simga-Mugan (2018);
Shumway (2001), Wu et al. (2010)
Growth ratios
Growth in sales (GS) Sales in current year minus sales in prior year, then Altman et al. (2016), Barboza et al. (2017);
divided by sales in prior year Carton and Hofer (2006), Darrat et al. (2014);
Lohmann and Ohliger (2019)
Growth in assets (GA) Total Assets(TA) in Year t minus TA in year t-1 then Altman et al. (2016), Barboza et al. (2017);
divided by TA in year t-1 Carton and Hofer (2006)
Growth in return on Growth in ROE= ROE in Year t minus ROE in year t-1 Barboza et al. (2017), Carton and Hofer (2006)
equity (GROE) then divided by ROE in year t-1
Macro-economic
variables
Growth in GDP The annual percentage growth rate of GDP at Bhattacharjee and Han (2014), Cole (2009); Li
market prices based on constant local currency et al. (2020)
Unemployment rate The share of the labor force that is without work but Bhattacharjee and Han (2014), Cybinski (2001);
available for and seeking employment Li et al. (2020)
Inflation rate The percentage price change in the economy as a Li et al. (2020), Liu (2004); Mensah (1984),
whole in a year Tinoco and Wilson (2013)
Real Interest rate The lending interest rate adjusted for inflation as Bhattacharjee and Han (2014), Cole (2009);
measured by the GDP deflator in a year Cybinski (2001), Liu (2004); Mensah (1984),
Tinoco and Wilson (2013)
Corporate
governance variables
Board Size Calculated as the total number of directors on the Adams and Ferreira (2009), Chaganti et al.
board (1985); Daily and Dalton (1994), Darrat et al.
(2014); Fich and Slezak (2008), Salloum et al.
(2013); Wang and Deng (2006)
CEO Duality It is a dummy variable that equals 1 if the CEO is Chaganti et al. (1985), Daily and Dalton (1994);
also the chairman of the board and 0 otherwise Fich and Slezak (2008), Li et al. (2020)
State ownership Calculated as the proportion of shares held by the Bhattacharjee and Han (2014), Li et al. (2008); Li
state et al. (2020), Wang and Deng (2006)

j CORPORATE GOVERNANCE j
The literature on credit risk prediction models shows that the FDP techniques are not
required to control for other effects. Consequently, no control variables are added to the
models. The variables included in the models are considered as potential predictors.

3.2.2 Measurement of diversity. Table 4 shows the measurement of the variables used in the
study. We use the percentage of independent directors to represent structural diversity.
Further, we use Blau’s (2000) index to calculate relation-oriented and task-oriented board
diversity values as under:
X
D ¼1 pi 2 ; (2)

where D represents the index of diversity, p shows the fraction of each category and i refers
to index categories. Table 4 provides the definitions of variables. Relation-oriented diversity
(RELATION_D) is measured by adding age diversity (D_AGE) and gender diversity
(D_GENDER) and task-oriented diversity (TASK_D) is calculated by adding education
diversity (D_EDUCATION) and expert diversity (D_EXPERT).

Table 4 Measurement of variables

Variable Symbol Measurement

Dependent variable
Financial Distress FD ST stocks with a net loss, decreasing net worth or below 1 interest coverage
ratio (any two conditions). A dummy variable denoted by 1 for financial
distress firm and 0 otherwise is used
Predictive variables
Age diversity D_AGE Age is divided into 5 categories to measure age diversity. These are: less
than or equal to 40, 41–49, 50–59, 60–69 and 70 years or more
Gender diversity D_GENDER Male and female categories are used to measure gender diversity.
Education diversity D_EDUCATION It is measured by five categories, namely, 1 = Technical secondary school
and below, 2 = Associate degree, 3 = Bachelor, 4 = Master and 5 = PhD
Expert diversity D_EXPERT It is measured using five categories, namely, 1. Financial = ACCA, Financial,
Analyst, Arbitrator, CPA, auditor, Tax
2. Management = Accountant, Manager, Management, Engineer,
Supervisor, Economist 3. Legal= Lawyer, Judge, 4. Consulting = Consultant,
Counsel
5. Others (i.e. research, technology, medical, etc.)
Relation-oriented diversity RELATIONS_D It is the sum of age diversity (D_AGE) and gender diversity (D_GENDER)
Task-oriented diversity TASK_D It is the sum of tenure diversity (D_TENURE) and education diversity
(D_EDUCATION)
Structural diversity STRL_D It is calculated by dividing the independent directors by the total number of
directors on the board
Accounting variables
Net profit to total assets NPTL Net profit divided by total assets
Operating profit to total assets OPTA Operating profit divided by total assets
Working capital to total assets WCTA Working capital divided by total assets
Growth variables
Growth in sales GRS Sales in the current year minus sales in the previous year, then divided by
sales in the previous year
Market variables
Lag ex return EXRETURN Cumulative annual return in year t-1 minus the value-weighted SHSE/SZSE
index return in year t-1
Relative size R_SIZE Log of the fraction of number of outstanding shares multiplied by the year-
end share price to total market value
Macro-economic variables
GDP growth rate GDP_G The rate at which GDP grows annually at market prices, based on constant
local currency
Real interest rate REAL_IR The lending interest rate adjusted for inflation as measured by the GDP
deflator

j CORPORATE GOVERNANCE j
 3.3 Data description and correlations
Table 5 shows descriptive statistics. Panel A represents descriptive statistics for financially
distressed firms; Panel B shows descriptive statistics of healthy firms and Panel C represents
summary statistics for the complete sample. The mean values of NPTL, OPTA and WCTA are
negative for financially distressed firms indicating severe profitability and liquidity problems. The
mean value of GRS for healthy firms is much better than financially distressed firms indicating that
growth is a good predictor. Similarly, diversity attributes on average have higher mean values for
healthy firms indicating that higher board diversity has a negative relation with financial distress.
Table 6 shows the spearman’s rank correlation among variables. All the predictive variables
except EXRETURN are significantly correlated with the dependent variable. We keep this
variable in our models as it adds predictive power to our models. We re-confirm its
predictive power by using the importance function in RF model. The importance function
indicates the explanatory power of each independent variable in the model. The correlation
results indicate that board diversity, profitability, liquidity, market, growth and
macroeconomic variables all have a negative relationship with financial distress.

4. Results
Table 7 shows the outcomes for six models from statistical (static and dynamic) and machine
learning techniques divided into training and validation sets. The comparison of the models is
done on the basis of Types I and II errors, accuracy and area under the ROC curve (AUC-ROC).
Table 7 makes a comparison of six models based on predictors included. Panel A
compares models on accounting, market, growth and macroeconomic predictors. Panel B
includes board diversity predictors; relation-oriented, task-oriented and structural diversity,
in addition to accounting, market, growth and macroeconomic predictors.
The overall accuracy of static, dynamic and machine learning models improves by
including board diversity attributes in training and validation. Hence, the results support our
hypothesis. A substantial rise in the AUC percentage of the dynamic hazard model by
adding board diversity attributes indicates that board diversity attributes are time
covariates. In other words, board diversity variables are explanatory variables that change
with time. They reveal the changing health of the firm, which deteriorates before default.
Machine learning models; RF, bagging and boosting outperform static (logistic regression)
and dynamic (dynamic hazard) statistical models, on the validation set except for KNN. As
mentioned earlier, we did not tune the models to maximize the results. Moreover, no data
normalization was done. KNN is sensitive to outliers. KNN also faces the problem of the
curse of dimensionality. Besides, the technique depends greatly on distances between
points. As the number of dimensions increases, the distances become less representative,
which results in poor performance of KNN. The value of k if not tuned properly may also
cause underfitting in the model implementation phase.
In the training phase, the RF and bagging models show 100% AUC. This result was anticipated,
as these models use decision trees that often cause overfitting of models in the training phase.
To understand whether these are the best techniques, we need to consider the validation set.
Although the boosting technique has the highest accuracy in train and validation sets, its
Type I error is sizable. In risk management, loss to investors and lenders in terms of Type I
error is much higher as compared to Type II error. So the better technique in terms of
accuracy and minimum Type I error is the RF. In terms of Type I error, RF and KNN both
have the same lowest Type I error of 7.69. However, KNN has the highest Type II error rate
of 93.67 in test sets. The overall accuracy of KNN is 18.48, which is the lowest in all models.
The accuracy of the logistic regression (static) model drops significantly in the validation
set, which indicates its limitations on unseen data. Although it does not require a linear

j CORPORATE GOVERNANCE j
 Table 5 Descriptive statistics
Panel A
FDF NPTL EXRETURN R_SIZE GRS OPTA WCTA GDP_G REAL_IR RELATIONS_D TASK_D STRL_D
Min. 7.390 1.102 0.000 0.995 18.918 25.366 6.737 2.302 0.287 0.138 0.250
First qu. 0.159 0.219 0.000 0.229 0.111 0.248 7.300 1.002 0.752 0.714 0.333
Median 0.034 0.037 0.000 0.048 0.033 0.025 7.860 3.581 0.859 0.980 0.364
Mean 0.132 0.081 0.000 0.052 0.155 0.314 8.958 2.074 0.863 0.911 0.369
Third qu. 0.020 0.210 0.000 0.143 0.007 0.209 9.654 4.285 0.982 1.156 0.400
Max. 0.683 2.925 0.002 10.08 0.222 0.765 14.231 5.532 1.218 1.351 0.600
sd. 0.482 0.564 0.000 0.814 0.987 1.957 2.081 2.768 0.153 0.294 0.051
Panel B
HF NPTL EXRETURN R_SIZE GRS OPTA WCTA GDP_G REAL_IR RELATIONS_D TASK_D STRL_D
Min. 0.118 1.221 0.000 0.659 0.089 0.483 6.737 2.302 0.356 0.125 0.091
First qu. 0.072 0.167 0.000 0.028 0.035 0.094 6.905 3.242 0.748 0.899 0.333
Median 0.134 0.035 0.000 0.135 0.062 0.201 7.769 3.758 0.840 1.087 0.333
Mean 0.236 0.154 0.000 0.216 0.079 0.220 8.000 2.972 0.848 1.003 0.371
Third qu. 0.275 0.334 0.000 0.289 0.105 0.366 7.860 4.285 0.944 1.175 0.400
Max. 2.065 2.997 0.007 6.784 0.391 0.705 14.231 5.532 1.190 1.350 0.667
sd. 0.294 0.539 0.001 0.492 0.061 0.196 1.380 2.153 0.139 0.246 0.061
Panel C
Comp. obs. NPTL EXRETURN R_SIZE GRS OPTA WCTA GDP_G REAL_IR RELATIONS_D TASK_D STRL_D
Min. 7.390 1.221 0.000 0.995 18.918 25.366 6.737 2.302 0.287 0.125 0.091
First qu. 0.003 0.190 0.000 0.065 0.010 0.040 6.905 0.259 0.748 0.823 0.333
Median 0.066 0.000 0.000 0.086 0.032 0.158 7.769 3.581 0.847 1.053 0.333
Mean 0.083 0.123 0.000 0.148 0.018 0.002 8.399 2.598 0.854 0.965 0.370
Third qu. 0.177 0.286 0.000 0.240 0.080 0.316 9.551 4.285 0.960 1.170 0.400
Max. 2.065 2.997 0.007 10.081 0.391 0.765 14.231 5.532 1.218 1.351 0.667
sd. 0.424 0.551 0.001 0.650 0.648 1.297 1.770 2.467 0.145 0.270 0.057

j CORPORATE GOVERNANCE j
 Table 6 Correlation analysis
Variables 1 2 3 4 5 6 7 8 9 10 11 12

FD 1
NPTL 0.495 1
EXRETURN 0.007 0.114 1
R_SIZE 0.240 0.437 0.120 1
GRS 0.292 0.405 0.202 0.306 1
WCTA 0.279 0.526 0.022 0.232 0.162 1
OPTA 0.491 0.942 0.124 0.515 0.445 0.495 1
RELATIONS_D 0.042 0.036 0.044 0.169 0.033 0.099 0.002 1
TASK_D 0.125 0.180 0.007 0.003 0.068 0.154 0.155 0.083 1
STRL_D 0.050 0.042 0.009 0.098 0.062 0.017 0.019 0.014 0.062 1
GDP_G 0.102 0.193 0.018 0.230 0.088 0.260 0.185 0.059 0.013 0.049 1
REAL_IR 0.003 0.060 0.122 0.117 0.151 0.123 0.064 0.015 0.023 0.031 0.523 1
Notes: *p < 0.05; **p < 0.01; ***p < 0.001

association between independent and dependent variables, the method needs to have a
linear relationship between independent variables and log odds. Thus, the decision surface
of logistic regression is linear. This property of logistic regression reduces its predictability
as compared to machine learning models.
Figures 2 and 3 show ROC curves for six models with and without board diversity attributes. An
increase in an area under the ROC curve from Figures 2 to 3 supports our hypotheses that the
inclusion of relation-oriented, task-oriented and structural diversity attributes improves the
prediction ability of FDP models. The machine learning models of RF, bagging and boosting show
significant superiority over statistical models of logistic regression and dynamic hazard, except
KNN, which is because of the reasons discussed. Although it is hard to select a single preferred
model from the curves below, RF, bagging and boosting seems to be promising candidates.

5. Discussions and conclusions

This study shows how the inclusion of different features of board diversity improves the FDP
of different statistical and machine learning models. Five board diversity attributes are
categorized into task-oriented diversity (expertize and education), relation-oriented diversity
(age and gender) and structural diversity (independence) categories. Although, the prime
focus of the study is to answer the question “can board diversity predicts financial
distress?,” and not to develop a causal relationship, yet the results show that there is an
association between board diversity and financial distress. Descriptive statistics,
correlations analysis and coefficients of logistic regression and dynamic hazard models (not
reported for brevity) indicate that the relationship between board diversity and financial
distress is negative. Our results are robust for two reasons. We use different models
simultaneously to check the FD prediction ability of board diversity attributes. Second, we
use a validation set, which also ratifies robustness.
The results of the study are consistent with prior studies (Li et al., 2020; Zhou, 2019; Law
Chapple et al., 2012; Mittal and Lavina, 2018; Fich and Slezak, 2008), which document that
the director characteristics can influence financial distress status of firms. The study results
also support the view that boards play their dual role: monitoring and advice in handling a
financial distress situation (Adams and Ferreira, 2009; Aggarwal et al., 2019; Daily and
Dalton, 1994). How much each role contributes could be a future research direction in the
said relationship. Finally, the findings are consistent with the view that board diversity
improves board performance because of the qualities brought by directors of unique
backgrounds (Ararat et al., 2015; Kagzi and Guha, 2018; Mahadeo et al., 2012).

j CORPORATE GOVERNANCE j
 Table 7 Results of the prediction models

Panel A
Train set Type I error Type II error Sensitivity Specificity Accuracy AUC (%)
Logit 15.57 17.84 84.43 82 82.46 91.96
Dynamic Hazard 9.84 19.35 90.16 80.65 81.92 92.20
K-Nearest Neighbor 0 89.95 51.64 10.05 15.58 83.85
Random Forest 0 8.92 100 91.08 92.27 100
Bagging 0 8.67 100 91.33 92.48 100
Boosting 22.95 1.89 77 98 95 98
Validation set Type I error Type II error Sensitivity Specificity Accuracy AUC(%)
Logit 30.77 31.65 69.23 68.35 68.48 76.83
Dynamic Hazard 38.46 25.32 61.54 74.68 72.83 54.23
K-nearest neighbor 7.69 93.67 92.31 6.33 18.48 50.6
Random forest 7.69 24.05 92.31 75.95 78.26 89.53
Bagging 15.38 20.25 84.62 79.75 80.43 88.7
Boosting 46.15 7.59 53.85 92.41 86.96 87.5
Panel B
Train set Type I error Type II error Sensitivity Specificity Accuracy AUC(%)
Logit 9.84 17.34 90.16 82.66 83.66 92.39
Dynamic Hazard 7.38 19.35 92.62 80.65 82.24 93.75
K-nearest neighbor 0 89.82 51.64 10.18 15.69 83.61
Random forest 0 8.05 100 91.95 93.02 100
Bagging 0 8.54 100 91.46 92.59 100
Boosting 22.95 1.13 77.05 98.87 95.97 98.7
Validation set Type I error Type II error Sensitivity Specificity Accuracy AUC(%)
Logit 30.77 26.58 69.23 73.42 72.83 80
Dynamic Hazard 23.08 24.05 76.92 75.95 76.09 73.6
K-nearest neighbor 7.69 93.67 92.31 6.33 18.48 50.6
Random forest 7.69 20.25 92.31 79.75 81.52 89.99
Bagging 15.38 20.25 84.62 79.75 80.43 89.9
Boosting 30.77 6.33 69.23 93.67 90.22 89.1

Board diversity has drawn wide academic interest in many fields over the past two
decades, most of which have documented that one or two features of board diversity are
related to corporate success. In terms of FDP, the current study has tested a good range of
board diversity facets as predictors of financial distress risk, using a post-corporate
governance reforms period, considering Chinese A-listed firms.
Our study provides several advantages. First, we put a special focus on variable selection.
After reviewing business failure literature over the 35 last years’ period, Balcaen and Ooghe
(2006) concluded that there is little agreement on which variables perform the best in
distinguishing between non-failed and failed firms. Thus, variable selection is very important
in FDP studies. We use stepwise regression to find out the best predictors among
accounting, market, macroeconomic, growth and corporate governance variables. Second,
we used a novel definition of financial distress by not only relying on ST stocks but also
considering deteriorating net-worth, below one interest coverage ratio and negative/
decreasing profitability when selecting financially distressed firms. Third, as we have not
used any data transformation technique and variables are taken as they are; it can be
inferred that machine learning methods can generate significant classification accuracy,
relative to conventional methods such as logistic regression, and dynamic hazard models.
Fourth, we use a wide range of board diversity facets to measure the financial distress of
firms. Such a wide range of variables is not normally incorporated into FDP models.
Along with the benefits, it is imperative to highlight some limitations. Although we examined
five important attributes of board diversity, there are other facets of board diversity such as
tenure, nationality, religion and ethnicity that were not considered. Future studies should
highlight how these diversity aspects affect FDP ability. Second, as we did not tune the

j CORPORATE GOVERNANCE j
 Figure 2 Comparison of models without board diversity variables

Figure 3 Comparison of models including board diversity variables

algorithms applied in different R packages, the benefits of the full capacity of the models
were not realized. Nevertheless, simple settings of the algorithms show the superiority of
machine learning models. Upcoming studies may draw attention on the difference in
prediction ability once these models are tuned.

j CORPORATE GOVERNANCE j
Our study provides important implications. First, our results provide implications for the
board’s nomination committee. The nomination committee largely remains in charge of the
board’s effectiveness such as director succession, recruitment, education and evaluations.
In light of the findings of the study, the nomination committee should focus on enhancing
board diversity in terms of gender, age, education, expertise and independence.
Regulators may formulate different corporate governance codes for financially distressed
firms by incorporating age, gender, education, expertise and independence diversity of the
board. This is because the board diversity facets have different influences on healthy firms.
For academics, in terms of corporate governance, the results highlight important insights about
board diversity, which is an essential aspect of corporate governance mechanism. While prior
studies mostly consider one attribute of board diversity, we argue that boards have different
facets and future studies should consider multiple facets of diversity when studying their effects
on firm performance. In terms of financial distress, our study adds to the discussion that non-
financial aspects are powerful predictors of financial distress. As board diversity variables (non-
financial predictors) are weakly correlated with accounting, market, growth and macro-
economic predictors, they bring incremental information when combined with other categories
of predictors in FDP studies. Accordingly, future research can also point out how the attributes of
the CEO (another aspect of upper echelons) help in predicting financial distress.
Our study has important implications for practitioners. Practitioners can assess credit risk
by considering the information of firms from accounting, market, growth and board diversity
aspects to arrive at a decision. Our study also adds to the literature that explores the
performance of machine learning models in comparison with conventional statistical
techniques in the credit industry. Although machine learning models do not explain a causal
relationship, they predict financial distress more accurately and can be used extensively in
the credit industry.
Finally, our study also offers implications for society in general. Society is comprised of
individuals with an assortment of identifications, i.e. ethnicity, nationality, age groups,
gender, education level to name a few. Having voices of such subgroups to the decision-
making body of the organization is ethically and strategically correct. Many studies
including ours suggest that the practice of enhancing diversity in organizations can
significantly safeguard the interest of a wide range of stakeholders.

References
Abdullah, S.N. (2006), “Board structure and ownership in Malaysia: the case of distressed listed
companies”, Corporate Governance, Vol. 6 No. 5, pp. 582-594.
Adams, R.B. and Ferreira, D. (2007), “A theory of friendly boards”, The Journal of Finance, Vol. 62 No. 1,
pp. 217-250.
Adams, R.B. and Ferreira, D. (2009), “Women in the boardroom and their impact on governance and
performance”, Journal of Financial Economics, Vol. 94 No. 2, pp. 291-309.
Adnan, M.F., Sabli, N., Rashid, M.Z.A., Hashim, A., Paino, H. and Abdullah, A. (2016), “The impact of
educational level of board of directors on firms’ performance”, in Abdullah, M., Yahya, W.K., Ramli, N.
and Mohamed, S.R. (Eds.), Regional Conference on Science, Technology and Social Sciences (RCSTSS
2014), Springer, Singapore, pp. 37-48.
Aggarwal, R., Jindal, V. and Seth, R. (2019), “Board diversity and firm performance: the role of business
group affiliation”, International Business Review, Vol. 28 No. 6, available at: https://doi.org/10.1016/j.
ibusrev.2019.101600
Alaka, H.A., Oyedele, L.O., Owolabi, H.A., Kumar, V., Ajayi, S.O., Akinade, O.O. and Bilal, M. (2018),
“Systematic review of bankruptcy prediction models: towards a framework for tool selection”, Expert
Systems with Applications, Vol. 94, pp. 164-184.
Ali, M., Ng, Y.L. and Kulik, C.T. (2014), “Board age and gender diversity: a test of competing linear and
curvilinear predictions”, Journal of Business Ethics, Vol. 125 No. 3, pp. 497-512.

j CORPORATE GOVERNANCE j
 Altman, E.I. (1968), “Financial ratios, discriminant analysis and the prediction of corporate bankruptcy”,
The Journal of Finance, Vol. 23 No. 4, pp. 589-609.
Altman, E.I., Sabato, G. and Wilson, N. (2010), “The value of non-financial information in small and
medium-sized enterprise risk management”, The Journal of Credit Risk, Vol. 6 No. 2, pp. 1-33.
Altman, E.I., Iwanicz-Drozdowska, M., Laitinen, E.K. and Suvas, A. (2016), “Financial and non-financial
variables as long-horizon predictors of bankruptcy”, The Journal of Credit Risk, Vol. 12 No. 4, pp. 49-78.
Appiah, K.O. and Chizema, A. (2016), “The impact of board quality and nomination committee on
corporate bankruptcy”, Advances in Accounting, Vol. 35, pp. 75-81.
Ararat, M., Aksu, M. and Cetin, A.T. (2015), “How board diversity affects firm performance in emerging
markets: evidence on channels in controlled firms”, Corporate Governance: An International Review,
Vol. 23 No. 2, pp. 83-103.
Balcaen, S. and Ooghe, H. (2006), “35 Years of studies on business failure: an overview of the
classic statistical methodologies and their related problems”, The British Accounting Review,
Vol. 38 No. 1, pp. 63-93.
Barboza, F., Kimura, H. and Altman, E. (2017), “Machine learning models and bankruptcy prediction”,
Expert Systems with Applications, Vol. 83, pp. 405-417.
Beaver, W.H., Correia, M. and McNichols, M.F. (2012), “Do differences in financial reporting attributes
impair the predictive ability of financial ratios for bankruptcy?”, Review of Accounting Studies, Vol. 17
No. 4, pp. 969-1010.

Bellovary, L.J., Giacomino, D. and Akers, M. (2007), “A review of bankruptcy prediction studies: 1930-
Present”, Journal of Financial Education, Vol. 33 No. Winter, pp. 1-42.
Bernile, G., Bhagwat, V. and Yonker, S. (2018), “Board diversity, firm risk, and corporate policies”,
Journal of Financial Economics, Vol. 127 No. 3, pp. 588-612.
Bhat, K.U., Chen, Y., Jebran, K. and Memon, Z.A. (2019), “Board diversity and corporate risk: evidence
from China”, Corporate Governance: The International Journal of Business in Society, Vol. 20 No. 2,
pp. 280-293.

Bhattacharjee, A. and Han, J. (2014), “Financial distress of Chinese firms: microeconomic, macroeconomic
and institutional influences”, China Economic Review, Vol. 30 No. November, pp. 244-262.

Blau, J.R. (2000), “Group enmity and accord”, Social Science History, Vol. 24 No. 2, pp. 395-413.
Bose, I. (2006), “Deciding the financial health of dot-coms using rough sets”, Information & Management,
Vol. 43 No. 7, pp. 835-846.
Breiman, L. (1996), “Bagging predictors”, Machine Learning, Vol. 24 No. 2, pp. 123-140.
Breiman, L. (2001), “Random forests”, Machine Learning, Vol. 45 No. 1, pp. 5-32.
Campbell, J.Y., Hilscher, J. and Szilagyi, J. (2008), “In search of distress risk”, The Journal of Finance,
Vol. 63 No. 6, pp. 2899-2939.
Carter, D.A., Simkins, B.J. and Simpson, W.G. (2003), “Corporate governance, board diversity, and firm
value”, The Financial Review, Vol. 38 No. 1, pp. 33-53.

Carton, R.B. and Hofer, C.W. (2006), Measuring Organizational Performance: Metrics for
Entrepreneurship and Strategic Management Research, Edward Elgar Publishing limited, Cheltenham.

Cenciarelli, V.G., Greco, G. and Allegrini, M. (2018), “External audit and bankruptcy prediction”, Journal
of Management and Governance, Vol. 22 No. 4, pp. 863-890.

Chaganti, R.S., Mahajan, V. and Sharma, S. (1985), “Corporate board size, composition and corporate
failures in retailing industry”, Journal of Management Studies, Vol. 22 No. 4, pp. 400-417.

Chan, C.Y., Chou, D.W., Lin, J.R. and Liu, F.Y. (2016), “The role of corporate governance in forecasting
bankruptcy: pre- and post-SOX enactment”, The North American Journal of Economics and Finance,
Vol. 35, pp. 166-188.

Clarke, T. and Branson, D. (Eds.). (2012), The SAGE Handbook of Corporate Governance, SAGE
Publications Ltd, 10.4135/9781446200995.

Cole, R.A. (2009), “Predicting bank failures using a simple dynamic hazard model”, 22nd Australasian
Finance and Banking Conference (AFBC 2009), pp. 1-29.

j CORPORATE GOVERNANCE j
Cummings, J.N. (2004), “Work groups, structural diversity, and knowledge sharing in a global
organization”, Management Science, Vol. 50 No. 3, pp. 352-364.
Cybinski, P. (2001), “Description, explanation, prediction - the evolution of bankruptcy studies?”,
Managerial Finance, Vol. 27 No. 4, pp. 29-44.
Daily, C.M. and Dalton, D.R. (1994), “Corporate governance and the bankrupt firm: an empirical
assessment”, Strategic Management Journal, Vol. 15 No. 8, pp. 643-654.
Darrat, A.F., Gray, S., Park, J.C. and Wu, Y. (2014), “Corporate governance and bankruptcy risk”, Journal
of Accounting, Auditing and Finance, Vol. 31 No. 2, pp. 1-40.
Doumpos, M. and Zopounidis, C. (1999), “A multicriteria discrimination method for the prediction of
financial distress: the case of Greece”, Multinational Finance Journal, Vol. 3 No. 2, pp. 71-101.
Elloumi, F. and Gueyié, J.P. (2001), “Financial distress and corporate governance: an empirical analysis”,
Corporate Governance: The International Journal of Business in Society, Vol. 1 No. 1, pp. 15-23.
Erhardt, N.L., Werbel, J.D. and Shrader, C.B. (2003), “Board of director and committee diversity and firm
financial performance”, Corporate Governance, Vol. 11 No. 2, pp. 102-111.
Farooq, U. and Qamar, M.A.J. (2019), “Predicting multistage financial distress: reflections on sampling, feature
and model selection criteria”, Journal of Forecasting, Vol. 38 No. 7, pp. 632-648.
Farooq, U., Jibran Qamar, M.A. and Haque, A. (2018), “A three-stage dynamic model of financial
distress”, Managerial Finance, Vol. 44 No. 9, pp. 1101-1116.
Fernandez-Temprano, M.A. and Tejerina-Gaite, F. (2020), “Types of director, board diversity and firm
performance”, Corporate Governance: The International Journal of Business in Society, Vol. 20 No. 2,
pp. 324-342.
Ferreira, D. (2010), “Board diversity”, in Baker, H.K. and Anderson, R. (Eds), Corporate Governance: A
Synthesis of Theory, Research, and Practice, John Wiley & Sons, Inc., pp. 225-242.
Fich, E.M. and Slezak, S.L. (2008), “Can corporate governance save distressed firms from bankruptcy? An
empirical analysis”, Review of Quantitative Finance and Accounting, Vol. 30 No. 2, pp. 225-251.
Francoeur, C., Labelle, R. and Sinclair-Desgagné, B. (2008), “Gender diversity in corporate governance
and top management”, Journal of Business Ethics, Vol. 81 No. 1, pp. 83-95.
Freitas Cardoso, G., Peixoto, F.M. and Barboza, F. (2019), “Board structure and financial distress in
Brazilian firms”, International Journal of Managerial Finance, Vol. 15 No. 5, pp. 813-828.
Friedman, J., Hastie, T. and Tibshirani, R. (2000), “Additive logistic regression: a statistical view of
boosting”, The Annals of Statistics, Vol. 28 No. 2, pp. 337-407.
Gul, F.A., Srinidhi, B. and Ng, A.C. (2011), “Does board gender diversity improve the informativeness of
stock prices?”, Journal of Accounting and Economics, Vol. 51 No. 3, pp. 314-338.
Habib, A., Costa, M.D., Huang, H.J., Bhuiyan, M.B.U. and Sun, L. (2018), “Determinants and consequences of
financial distress: review of the empirical literature”, Accounting and Finance, doi: 10.1111/acfi.12400.
Hambrick, D.C. and D’Aveni, R.A. (1988), “Large corporate failures as downward spirals”, Administrative
Science Quarterly, Vol. 33 No. 1, pp. 1-23.

Hambrick, D.C. and Mason, P. A. (1984), “Echelons: the organization as a reflection of its top managers”,
Academy of Management Review, Vol. 9 No. 2, pp. 193-206.

Hillegeist, S.A., Keating, E.K., Cram, D.P. and Lundstedt, K.G. (2004), “Assessing the probability of
bankruptcy”, Review of Accounting Studies, Vol. 9 No. 1, pp. 5-34.

Jackson, S.E., May, K.E. and Whitney, K. (1995), “Understanding the dynamics of diversity in decision
making teams”, in Guzzo, R.A. and Salas, E. (Eds), Team Effectiveness and Decision Making in
Organizations, 1st ed., Pfeiffer, pp. 204-261.
Jebran, K., Chen, S. and Zhang, R. (2020), “Board diversity and stock price crash risk”, Research in
International Business and Finance, Vol. 51 No. October 2019, p. 101122.

Jensen, M.C. and Meckling, W.H. (1976), “Theory of the firm: managerial behavior, agency costs and
ownership structure”, Journal of Financial Economics, Vol. 3 No. 4, pp. 305-360.

Joshi, A. and Jackson, S.E. (2003), “Managing workforce diversity to enhance cooperation in
organizations”, in A.West, M., Tjosvold, D. and G.Smith, K. (Eds), International Handbook of

j CORPORATE GOVERNANCE j
 Organizational Teamwork and Cooperative Working, John Wiley & Sons Ltd, The Atrium, Southern Gate,
Chichester, West Sussex PO19 8SQ, England, pp. 277-294.
Kagzi, M. and Guha, M. (2018), “Does board demographic diversity influence firm performance?
Evidence from Indian-knowledge intensive firms”, Benchmarking: An International Journal, Vol. 25 No. 3,
pp. 1028-1058.
Kaur, R. and Singh, B. (2019), “Do CEO characteristics explain firm performance in India?”, Journal of
Strategy and Management, Vol. 12 No. 3, pp. 409-426.
Kearns, M.J. and Vazirani, U.V. (1994), An Introduction to Computational Learning Theory, MIT Press
Cambridge, MA, London, England.
Khaw, K.L.H., Liao, J., Tripe, D. and Wongchoti, U. (2016), “Gender diversity, state control, and corporate
risk-taking: evidence from China”, Pacific-Basin Finance Journal, Vol. 39, pp. 141-158.
Kim, D. and Starks, L.T. (2016), “Gender diversity on corporate boards: do women contribute unique
skills?”, American Economic Review, Vol. 106 No. 5, pp. 267-271.
Kristanti, F.T., Rahayu, S. and Huda, A.N. (2016), “The determinant of financial distress on Indonesian
family firm”, Procedia - Social and Behavioral Sciences, Vol. 219, pp. 440-447.
Lajili, K. and Zéghal, D. (2010), “Corporate governance and bankruptcy filing decisions”, Journal of
General Management, Vol. 35 No. 4, pp. 3-26.
Law Chapple, L., Kent, P. and Routledge, J. (2012), “Board gender diversity and going concern audit
opinions”, SSRN Electronic Journal, No. February, doi: 10.2139/ssrn.1979040.
Le, T., Lee, M.Y., Park, J.R. and Baik, S.W. (2018), “Oversampling techniques for bankruptcy prediction:
novel features from a transaction dataset”, Symmetry, Vol. 10 No. 4, doi: 10.3390/sym10040079.
Lee, T.S. and Yeh, Y.H. (2004), “Corporate governance and financial distress: evidence from Indian
companies”, Corporate Governance, Vol. 12 No. 3, pp. 378-388.
Li, Z., Crook, J., Andreeva, G. and Tang, Y. (2020), “Predicting the risk of financial distress using
corporate governance measures”, Pacific-Basin Finance Journal, No. September 2019, p. 101334.

Li, Y. and Wang, Y. (2018), “Machine learning methods of bankruptcy prediction using accounting
ratios”, Open Journal of Business and Management, Vol. 06 No. 01, pp. 1-20.

Li, H.X., Wang, Z.J. and Deng, X.L. (2008), “Ownership, independent directors, agency costs and
financial distress: evidence from Chinese listed companies”, Corporate Governance: The International
Journal of Business in Society, Vol. 8 No. 5, pp. 622-636.
Liu, J. (2004), “Macroeconomic determinants of corporate failures: evidence from the UK”, Applied
Economics, Vol. 36 No. 9, pp. 939-945.

Lohmann, C. and Ohliger, T. (2019), “Using accounting-based information on young firms to predict
bankruptcy”, Journal of Forecasting, Vol. 38 No. 8, pp. 803-819.

Mahadeo, J.D., Soobaroyen, T. and Hanuman, V.O. (2012), “Board composition and financial
performance: uncovering the effects of diversity in an emerging economy”, Journal of Business Ethics,
Vol. 105 No. 3, pp. 375-388.

Maione, C., De Paula, E.S., Gallimberti, M., Batista, B.L., Campiglia, A.D., Barbosa, F. and Barbosa, R.M.
(2016), “Comparative study of data mining techniques for the authentication of organic grape juice based
on ICP-MS analysis”, Expert Systems with Applications, Vol. 49, pp. 60-73.
Manzaneque, M., Priego, A.M. and Merino, E. (2015), “Corporate governance effect on financial distress
likelihood: evidence from Spain”, Revista de Contabilidad, Vol. 19 No. 1, pp. 111-121.
Mensah, Y.M. (1984), “An examination of the stationarity of multivariate bankruptcy prediction models: a
methodological study”, Journal of Accounting Research, Vol. 22 No. 1, p. 380.

Milliken, F.J. and Martins, L.L. (1996), “Searching for common threads: understanding the multiple effects
of diversity in organizational groups”, Academy of Management Review, Vol. 21 No. 2, pp. 402-433.

Mittal, S. and Lavina, (2018), “Females’ representation in the boardroom and their impact on financial
distress: an evidence from family businesses in India”, Indian Journal of Corporate Governance, Vol. 11
No. 1, pp. 35-44.

Muranda, Z. (2006), “Financial distress and corporate governance in Zimbabwean banks”, Corporate
Governance: The International Journal of Business in Society, Vol. 6 No. 5, pp. 643-654.

j CORPORATE GOVERNANCE j
Mustafa, A.S., Ahmad, A.C. and Chandren, S.A. (2017), “Board diversity and audit quality: evidence from
Turkey Akademia Baru board diversity and audit quality: evidence from Turkey”, Journal of Advanced
Research in Business and Management Studies, Vol. 6 No. 1, pp. 50-60.
Ohlson, J.A. (1980), “Financial ratios and the probabilistic prediction of bankruptcy”, Journal of
Accounting Research, Vol. 18 No. 1, p. 109.
Oz, I.O. and Simga-Mugan, C. (2018), “Bankruptcy prediction models’ generalizability: evidence from
emerging market economies”, Advances in Accounting, Vol. 41, pp. 114-125.
Parker, S., Peters, G.F. and Turetsky, H.F. (2002), “Corporate governance and corporate failure: a survival
analysis”, Corporate Governance: The International Journal of Business in Society, Vol. 2 No. 2, pp. 4-12.
Pessarossi, P. and Weill, L. (2013), “Choice of corporate debt in China: the role of state ownership”, China
Economic Review, Vol. 26, doi: 10.1016/j.chieco.2013.03.005.
Pfeffer, J. and Salancik, G.R. (1978), “The external control of organizations: a resource dependence
perspective”, Stanford Business Classics, 1st ed., Stanford University press, CA.
Platt, H. and Platt, M. (2012), “Corporate board attributes and bankruptcy”, Journal of Business
Research, Vol. 65 No. 8, pp. 1139-1143.
Robinson, D., Robinson, M. and Sisneros, C. (2012), “Bankruptcy outcomes: does the board matter?”,
Advances in Accounting, Vol. 28 No. 2, pp. 270-278.
Sabbaghi, O. (2016), “Corporate governance in China: a review”, Corporate Governance: The
International Journal of Business in Society, Vol. 16 No. 5, pp. 866-882.
Salloum, C. and Azoury, N. (2012), “Corporate governance and firms in financial distress: evidence from a
Middle Eastern country”, International Journal of Business Governance and Ethics, Vol. 7 No. 1, pp. 1-17.
Salloum, C.C., Azoury, N.M. and Azzi, T.M. (2013), “Board of directors’ effects on financial distress
evidence of family owned businesses in Lebanon”, International Entrepreneurship and Management
Journal, Vol. 9 No. 1, pp. 59-75.
Santen, B. and Donker, H. (2009), “Board diversity in the perspective of financial distress: empirical
evidence from The Netherlands”, Corporate Board Role Duties and Composition, Vol. 5 No. 2, pp. 23-35.
Shah, K., Patel, H., Sanghvi, D. and Shah, M. (2020), “A comparative analysis of logistic regression”,
Augmented Human Research, Vol. 5 No. 1, pp. 1-16.
Shahwan, T.M. (2015), “The effects of corporate governance on financial performance and financial
distress: Evidence from Egypt”, Corporate Governance: The International Journal of Business in Society,
Vol. 15 No. 5, pp. 641-662.
Shumway, T. (2001), “Forecasting bankruptcy more accurately: a simple hazard model”, The Journal of
Business, Vol. 74 No. 1, pp. 101-124.
Simpson, W.G. and Gleason, A.E. (1999), “Board structure, ownership, and financial distress in banking
firms”, International Review of Economics & Finance, Vol. 8 No. 3, pp. 281-292.
Sun, J. and Li, H. (2011), “Dynamic financial distress prediction using instance selection for the disposal
of concept drift”, Expert Systems with Applications, Vol. 38 No. 3, pp. 2566-2576.

Sun, J., Li, H., Huang, Q.H. and He, K.Y. (2014), “Predicting financial distress and corporate failure: a
review from the state-of-the-art definitions, modeling, sampling, and featuring approaches”, Knowledge-
Based Systems, Vol. 57, pp. 41-56.
Süsi, V. and Lukason, O. (2019), “Corporate governance and failure risk: evidence from Estonian SME
population”, Management Research Review, Vol. 42 No. 6, pp. 703-720.
Taffler, R.J. (1983), “The assessment of company solvency and performance using a statistical model”,
Accounting and Business Research, Vol. 13 No. 52, pp. 295-308.

Talavera, O., Yin, S. and Zhang, M. (2018), “Age diversity, directors’ personal values, and bank
performance”, International Review of Financial Analysis, Vol. 55, pp. 60-79.

Tinoco, H.M. and Wilson, N. (2013), “Financial distress and bankruptcy prediction among listed
companies using accounting, market and macroeconomic variables”, International Review of Financial
Analysis, Vol. 30, pp. 394-419.

Tsai, C.F., Hsu, Y.F. and Yen, D.C. (2014), “A comparative study of classifier ensembles for bankruptcy
prediction”, Applied Soft Computing Journal Computing, Vol. 24, pp. 977-984.

j CORPORATE GOVERNANCE j
 Udin, S., Khan, M.A. and Javid, A.Y. (2017), “The effects of ownership structure on likelihood of financial
distress: an empirical evidence”, Corporate Governance: The International Journal of Business in
Society, Vol. 17 No. 4, pp. 589-612.
Ullah, I., Fang, H. and Jebran, K. (2019), “Do gender diversity and CEO gender enhance firm’s value?
Evidence from an emerging economy”, Corporate Governance, Vol. 20 No. 1, pp. 44-66.

Veganzones, D. and Séverin, E. (2018), “An investigation of bankruptcy prediction in imbalanced

datasets”, Decision Support Systems, Vol. 112 No. August, pp. 111-124.
Wang, N. (2017), “Bankruptcy prediction using machine learning”, Journal of Mathematical Finance,
Vol. 7 No. 4, doi: 10.4236/jmf.2017.74049.
Wang, Z.J. and Deng, X.L. (2006), “Corporate governance and financial distress: evidence from Chinese
listed companies”, The Chinese Economy, Vol. 39 No. 5, pp. 5-27.
Wang, G., Chen, G. and Chu, Y. (2018), “A new random subspace method incorporating sentiment and
textual information for financial distress prediction”, Electronic Commerce Research and Applications,
Vol. 29, pp. 30-49.
Wang, G., Ma, J., Huang, L. and Xu, K. (2012), “Two cedit scoring models based on dual strategy
ensemble trees”, Knowledge-Based Systems, Vol. 26, pp. 61-68.
Wilson, N., Wright, M. and Altanlar, A. (2014), “The survival of newly-incorporated companies and
founding director characteristics”, International Small Business Journal: Researching Entrepreneurship,
Vol. 32 No. 7, pp. 733-758.
Wu, Y., Gaunt, C. and Gray, S. (2010), “A comparison of alternative bankruptcy prediction models”,
Journal of Contemporary Accounting & Economics, Vol. 6 No. 1, pp. 34-45.

Zhou, L. (2015), “A comparison of dynamic hazard models and static models for predicting the special
treatment of stocks in China with comprehensive variables”, Journal of the Operational Research Society,
Vol. 66 No. 7, pp. 1077-1090.

Zhou, G. (2019), “Financial distress prevention in China: does gender of board of directors matter?”,
Journal of Applied Finance & Banking, Vol. 9 No. 6, pp. 127-153.
Zhoua, Y.A., Kim, M.H.Y. and Ma, S. (2012), “Survive or die? An empirical study on Chinese ST firms”,
International Conference of the American Committee for Asian Economic Studies (ACAES), Melbourne,
Australia: Deakin University, pp. 1-32.
Zmijweski, M.E. (1984), “Methodological issues related to the estimation of financial distress prediction
models”, Journal of Accounting Research, Vol. 22, pp. 59-82.

About the authors

Umair Bin Yousaf is a Doctoral Student at the School of Accounting, Dongbei University of
Finance and Economics, China Internal Control Research Center, Dalian, China.
Khalil Jebran is an Assistant Professor at the School of Business Administration, Dongbei
University of Finance and Economics, Dalian, China. He has published research papers in
international journals including Corporate Governance: An International Review, Emerging
Markets Finance and Trade, Research in International Business and Finance, Asia-Pacific
Journal of Financial Studies and International Journal of Bank Marketing. Khalil Jebran is the
corresponding author and can be contacted at: [email protected]

Man Wang is a Professor at the School of Accounting, Dongbei University of Finance and
Economics, China Internal Control Research Center, Dalian, China. She has published
research papers in many reputed journals including China Finance and economic review,
Journal of Cleaner Production and International Journal of Financial Studies.

For instructions on how to order reprints of this article, please visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details: [email protected]

j CORPORATE GOVERNANCE j