Forecasting Bankruptcy for organizational sustainability in Pakistan: Using artificial neural networks, logit regression, and discriminant analysis

Fraz Inam (Department of Business Administration, Air University Multan Campus, Punjab, Pakistan)
Aneeq Inam (Department of Business Administration, Air University Multan Campus, Punjab, Pakistan)
Muhammad Abbas Mian (Department of Business Administration, Air University Multan Campus, Punjab, Pakistan)
Adnan Ahmed Sheikh (Department of Business Administration, Air University Multan Campus, Punjab, Pakistan)
Hayat Muhammad Awan (Department of Business Administration, Air University Multan Campus, Punjab, Pakistan)

Journal of Economic and Administrative Sciences

ISSN: 1026-4116

Article publication date: 18 October 2018

Issue publication date: 11 September 2019

Abstract

Purpose

Considering the economic dimension of sustainability, the purpose of this paper is to analyze the risk of bankruptcy in the Pakistani firms of the non-financial sector from years 1995 to 2017.

Design/methodology/approach

Three techniques were used which include multivariate discriminant analysis (MDA), logit regression and multilayer perceptron artificial neural networks. The accounting data of firms were selected one year before the bankruptcy.

Findings

Findings were obtained by comparing and analyzing the methods which show that neural networks model outperforms in the prediction of bankruptcy. They further conclude that profitability and leverage indicators have the power of discrimination in bankruptcy prediction and the best variables to predict financial distress are also found and indicated.

Practical implications

Practically, this study may help the firms to better anticipate the risks of getting bankrupt by choosing the right method and to make effective decision making for organizational sustainability.

Originality/value

Three different techniques were used in this research to predict the bankruptcy of non-financial sector in Pakistan to make an effective prediction.

Keywords

Citation

Inam, F., Inam, A., Mian, M.A., Sheikh, A.A. and Awan, H.M. (2019), "Forecasting Bankruptcy for organizational sustainability in Pakistan: Using artificial neural networks, logit regression, and discriminant analysis", Journal of Economic and Administrative Sciences, Vol. 35 No. 3, pp. 183-201. https://doi.org/10.1108/JEAS-05-2018-0063

Publisher

:

Emerald Publishing Limited

Copyright © 2018, Emerald Publishing Limited


1. Introduction

Organizational sustainability can be achieved by giving importance to economic, social and environmental aspects (Kuhlman and Farrington, 2010; Najam et al., 2018; Ştefănescu-Mihăilă, 2015). Considering the economic aspects of an organization, different financial difficulties may lead toward bankruptcy (Franks et al., 2015; Hassanpour and Ardakani, 2017). Protecting bankruptcy of the firms plays a vital role in making the organization sustainable (Finley et al., 2006; Jan and Marimuthu, 2015). Bankruptcy or financial failure of a company is an incident that can result in considerable losses for suppliers, banks, shareholders and community (Gerstrøm and Isabella, 2015). Therefore, executives of the firm should be concerned in forecasting not only when they face failure, but also to know the reasons for the initiation of financial distress and this distress may lead toward bankruptcy (Altman et al., 2017; Dobbie et al., 2017).

Bankruptcy possesses behavioral character but still, it is possible to forecast it accurately by using different information on the company and market (Frade, 2012). Moreover, to predict bankruptcy, studies have used various methodologies and techniques but still there is a need for an efficient way of predicting bankruptcy (Barboza et al., 2017; Chou et al., 2017; Oliveira et al., 2017; Zhao et al., 2017). Despite previous studies and techniques used, this study highlights the results of different non-financial sectors of Pakistan from years 1995 to 2017 by using three different techniques, i.e. multivariate discriminant analysis (MDA), logit regression and multilayer perceptron artificial neural networks (ANNs). Comparatively, the findings revealed that ANNs were more appropriate than other predictive techniques.

The basic concept of this research is to indicate that executives make many of the decisions without analyzing the information related to the future perspective of the firm’s performance. In such uncertain conditions, risk increases and the firm faces crisis that leads to financial distress and even to bankruptcy. It became a matter of great concern for public institutions that are responsible for the maintenance of steadiness of financial markets (Bisogno et al., 2017). Besides, the regulators are seriously looking for minimizing the default/credit risk because bankruptcies not only affect the firm’s stakeholders and organizational sustainability but also generate a susceptible environment for the whole economy (Bernstein et al., 2017).

Forecasting and monitoring the credit risk of companies are the primary concerns in financial theory. Bankruptcy prediction is mostly essential for an extensive range of purposes which include loan security assessment, portfolio risk measurement, going concern estimation by auditors, and the valuing of the credit derivatives, bonds and further securities which are exposed to default risk (Hensher and Jones, 2007). However, the assessment and prediction of firm performance and the potential of financial distress and bankruptcy are not straightforward. There is a number of available alternative methods for the assessment and prediction of the financial position of firms. Some of the methods are qualitative in nature, while most of them are quantitative, relying on the information available in financial statements.

Valuable information related to the financial position of a firm and its operations’ quality is provided by financial statements. The basic objective of financial statements is to provide information useful for making economic decisions (Altman and Mcgough, 1974). The ability of a firm to maintain its operations as a going concern is the most essential concern of a firm in financial decision making. A firm may face multiple consequences when it is exposed to financial issues like bankruptcy or liquidation and fail to operate as a going concern. Financing problems and operating problems are the two key issues which are related to a going concern that stakeholders consider by observing the financial statements. Financing problem is defined as trouble in meeting financial obligations, while failure in continuous operating success is referred to as the operating problem.

Importantly, such indicators which result in bankruptcy or liquidation should be predictable so that they do not harm the stakes. Investors and creditors are interested in predicting the bankruptcy of companies because of the possibility of their own losses. Although the monitoring of the financial health of firms had always been a practice, in the last few decades, in the financial literature, the bankruptcy prediction of companies has changed into a principal research issue. Prediction of firm failure has been shifted from the conventional ratio analysis to modern day sophisticated tools. For this cause, different models have been constructed to date and each prediction model has its own strengths and weaknesses (Aziz and Dar, 2006).

Recent literature has suggested to improve the models of bankruptcy prediction (Barboza et al., 2017; Charitou and Trigeorgis, 2000; Chava and Jarrow, 2004a, b; Laitinen and Laitinen, 2000; Ohlson, 1980; Tsai, 2009). Additionally, the prediction models serve as functions to predict the continuation of business units’ activities using the financial ratios (Table AII) and among the different methods used to predict the bankruptcy, the ratio analysis is the most primitive one. Besides, the probability of bankruptcy in this method is estimated by a group of financial ratios, whereas the bankruptcy prediction model providing necessary warnings can make the companies aware of the occurrence of bankruptcy and help investors in identifying investment opportunities. Hence, the bankruptcy issue is discussed in this research using three bankruptcy prediction models. The ANN, used in this study, is a widely accepted method in bankruptcy prediction study which uses the benefits of technology and needs with no specific requirements for predictor variables. However, the present study uses the ANN as one of the bankruptcy prediction models and presents its comparison with logistic regression statistical modeling and multivariate discriminant analysis as well.

The factors which really cause bankruptcy in Pakistan are still under investigation. Therefore, there is a need to trace out the important indicators and factors by which bankruptcy can be predicted and appropriate measures can be adopted. In Pakistan, to check the companies for bankruptcy chances in future years with the most appropriate and accurate model is still in the process where companies can have a great distress in the economy as a whole or on a particular sector.

Concurrently, the main objectives of this research are to assess the bankruptcy issue and the effect of the determined financial ratios on Pakistan bankruptcies which happened in the last two decades of the non-financial sector in Pakistan and also compared the performance of each technique used for predicting bankruptcy. However, the current study is organized into following sections: Section 2 reviews various literature based on the relevant hypothesis related to bankruptcy in Pakistan and techniques to predict bankruptcy including ANN, BLR and MDA. Section 3 describes the data/sample taken and financial ratios applied to it, and it also outlines research methodology in detail. Section 4 discusses results, analyses, and interpretations. Finally, Section 5 offers conclusions and recommendations.

2. Hypotheses development

Corporate bankruptcy and distress analyses started a couple of centuries ago. In the start, subjective qualitative information was the base for the assessment of potential distress and bankruptcy. After that, the analyses in the twentieth century shifted from the financial conditions to financial statements analyses in which univariate ratio analysis was most popular. The initial studies on ratio analysis for distress bankruptcy forecast are recognized as the univariate studies. Therefore, these readings contained mostly examining individual ratios, and occasionally, by associating ratios of unsuccessful companies to those of prosperous firms. However, few studies were issued up to the mid-60s where age is known as comparatively rich in issued studies of business failures, in which researchers progressed further in the field.

In certain, Beaver (1966) found the forecasting and analytical ability of accounting information as forecasters of major happenings. However, Beaver recognized that several indicators differentiated between corresponding samples of insolvent and healthy firms until five years preceding liquidation. In an actual logic, his univariate study of several bankruptcy forecasters set the platform for the advancement of multivariate analyses techniques. Bankruptcy problems which an organization faces include liquidity, efficiency, equity-efficiency, debt default and funds storage, while operating problems include insufficient revenues, poor control on operations, deteriorated ability to operate and making operating losses. These problems can be expressed by the financial ratios which become the base for the following hypothesis:

H1.

Financial ratios are efficient indicators for predicting bankruptcy in the non-financial sector of Pakistan.

Ohlson (1980) presented logit models to forecast insolvency. The writer effectively established O-score via nine accounting variables representative of four factors (size, liquidity, performance and capital structure) with a data set of 2,163 companies (105 solvent and 2,058 healthy) over a 1970–1976 era. The Z-score and O-score established by Altman (1968) and Ohlson (1980), correspondingly, encouraged later investigators to discover the liquidation forecasting model with the finest foretelling capability. Shumway (2001) established a vigorous logit or hazard model for predicting insolvency. In accumulation, Shumway measured both classic accounting data and data of equity market to construct his model and he emphasized on the practicality of some formerly ignored market-driven variables like firm’s size, historical stock returns, and the distinctive standard deviation of returns to estimate bankruptcy. He claimed that his model is more reliable in forecasting bankruptcy. Zmijewski (1984) deplored that approximating models on non-random testers can result into subjective constraint and probability approximations if suitable valuation methods are not used. Zmijewski also used the additional attention-grabbing method of the logistic regression: the probit analysis or probit model to upkeep his results.

Altman (1968) challenged the univariate financial ratio examinations as a vulnerable approach to faulty analysis and recognized the first multivariate discriminant studies by joining a set of financial ratios in a linear multivariate basis and measured Z-score as a measure of insolvency. After two years, the primary multivariate research was issued by Altman (1968). With the renowned “Z-score,” which is a multivariate discriminant analysis (MDA) technique, Altman established the benefit of allowing for the entire profile of individualities which were mutual to the related companies, as well as the connections between these properties. Therefore, to this bigoted model, Altman was capable to categorize data into two differentiated groups: healthy firms and bankrupt firms. He also revealed a second benefit: if two clusters were considered, this analysis decreases the forecaster’s space dimensionality to a singular dimension.

Altman’s workings were then trailed by succeeding research works that applied analogous and corresponding models. A linear probability model was constructed by Meyer and Pifer (1970). This is a distinct case of OLS regression with dependent variables having dichotomous nature for bank liquidation forecasting. West et al. (1985) used the blend of factor analysis and logit approximation as a new method to estimate the state of singular institutions and to allocate each of them a possibility of being a problematic bank. He determined that the grouping of factor analysis and logit approximation was favorable in assessing the bank’s situation.

The above-stated arguments formulated the following hypotheses:

H2.

The multivariate discriminant analysis method is reliable in predicting bankruptcy in the non-financial sector of Pakistan using financial ratios.

H3.

Bankruptcy in the non-financial sector of Pakistan can be predicted in a reliable manner by binary logit regression using financial ratios.

The ANNs model is perhaps the furthermost extensively used model amongst the intellectual systems (Demyanyk and Hasan, 2010). This newest method compromises two exciting benefits contrary to classical statistical methods. First, neural networks as non-parametrical mockups do not depend on definite assumptions like the dispersal of forecasters or properties of statistics. The other benefit is the confidence on nonlinear methods, which proposes comprehensive opportunities for the analysis of complex data sets. Neural networks have now come to the forefront as the preferred method for bankruptcy prediction (Ravi Kumar and Ravi, 2007). The following hypotheses were made based on these arguments:

H4.

The method of ANNs forecasts bankruptcy in the non-financial sector of Pakistan efficiently using financial ratios.

Tam and Kiang (1992) in an early treatment of neural networks in bankruptcy research compared a variety of models. They studied MDA, Logit, K-nearest neighbor, a decision tree classification algorithm (ID3), a single-layer neural network, and a multi-layer neural network. The multi-layer network was the best for predicting bankruptcy using financial ratios one year ahead of bankruptcy. For two years ahead of bankruptcy, logit was the best in the same studies. Salchenberger et al. (1992), when considering the bankruptcies of thrifts, found that the BPNN significantly outperformed Logit. In a comparison of the BPNN with MDA, Coats and Fant (1993) found the BPNN to be generally better, although it had a wider variance in the classification result depending on the horizon used. Altman et al. (1994) considered 1,000 Italian firms in a bankruptcy study that compared a BPNN and MDA. For a one-year-ahead prediction, MDA appeared to perform slightly better than the BPNN. Efrim Boritz and Kennedy (1995) matched several methods, containing dissimilar BPNN training procedures, logit and MDA. Results of the comparisons were inconclusive. The BPNN has shown in many studies effectiveness in predicting bankruptcy that is mixed to good when compared to classical MDA and other approaches. Atiya (2001) developed novel predictors mined from the equity markets for a neural network system. They showed that the usage of these predictors, in accumulation to old financial ratios, gave a substantial enhancement in the bankruptcy forecast precision for the neural network. Forecasts were based on financial information three years in advance of bankruptcy. The previous research works suggest that the BPNN is a better choice when a target vector is available.

Following hypothesis is formulated based on the above arguments:

H5.

ANNs provide a better model for predicting Bankruptcy in the non-financial sector of Pakistan than what multivariate discriminant analysis and binary logit regression do.

3. Data and methodology

3.1 Participants and procedures

The population taken for this research is all those companies of the non-financial sector which were delisted by Pakistan Stock Exchange (PSE) of Pakistan because of winding up under court order/liquidation. It is a violation of PSE’s listing regulation No. 32 (1) (d). Also, the companies which were winded up by Securities and Exchange Commission of Pakistan in the period 1995–2016 were taken. Liquidation is the process by which the assets of a firm are distributed by a firm or a part of it is ended. The criteria for the sample selection were based on the listing period and the shares of the firm have been traded at PSE; the firm which was selected must be from the non-financial sector only as the bankruptcy environment of the financial sector is different, the availability of financial information of at least five years.

For the most accurate comparison between healthy and default companies’ groups, the sample in both the groups is selected based on relatively similar activities. The sample consists of two groups of 40 defaulted firms and 40 healthy firms from the period 1995–2016. These companies were selected from all the non-financial sectors of Pakistan including manufacturing and service firms.

In data pre-processing, the firms having incomplete data were excluded. Only those companies were included in our sample which had five years of published data. Thus, our sample of bankrupt firms was 40 and an equal number of non-bankrupt or healthy firms were taken as a sample of non-bankrupt companies of the similar time which remained healthy and operating. The testing sample taken was all the existing companies operating in all the non-financial sector of Pakistan. The total number of companies including in the testing sample was 371.

The sample data of bankrupt, healthy and current companies were extracted from different issues of “Balance Sheet Analysis (BSA) of Joint Stock Companies (JSC) listed on Pakistan Stock Exchange (PSE)” published by the State Bank of Pakistan (SBP). The period of bankrupt, healthy and current companies was 1995–2016 (Table I).

3.2 Financial indicators

Financial ratios for the year preceding bankruptcy are taken as bankruptcy indicators for the analysis of corporate bankruptcy. Ratios of similar time periods of healthy firms were taken, while the ratios of last operating year (2014–2016) of current firms were also considered. Ratios were calculated from the financial data of companies obtained from BSAs. Khani (2015) found that liquidity ratios are the best indicators which can discriminate whether the company will lead toward bankruptcy or not. Solvency ratios can also be used in the creation of a bankruptcy model (Miller, 2009).

The positive relationship between bankruptcy possibilities and activity ratios was also studied (Nam et al., 2008). One of the important financial indicators which significantly affect the corporate financial distress is the profitability ratio (Manalu et al., 2017).

The first thing assumed in this research is the integrity and accuracy of the historical data available from the reliable resources. As the data are used in the construction of the models, their reliability depends on it. So, the reliability of the future data is also assumed.

True reporting of firms in financial reports is also assumed as the performance of the models highly depends on it. Therefore, they are vulnerable to fraudulent reporting. It is also assumed that all the data in BSA published by SBP are valid and reliable. Some firms do fraudulent reporting of data and go into financial distress. Also, because of the confidentiality reasons, most of the firms do not provide the right information about their failure. It is believed that these are the firms which the model cannot detect.

The summary statistics for both groups are given in Table I. The mean of the indicators in the bankrupt firms shows completely different trends as that of the indicators in healthy firms. This shows that there are different behaviors in the indicators for companies which were bankrupted. Also, the standard deviation in the healthy firms is too low as compared to a large amount of variability in the bankrupt firms. Data are then finalized for the processing through ANN, MDA and BLR.

3.3 Methodology

To obtain the results and to predict the binary status of the existing companies from the previous data of both bankrupt and healthy firms, we selected three models to be studied: ANNs (multilayer perceptron), multivariate discriminant analysis and binary logistic regression (BLR). Dependent and independent variables taken for all the analysis were the same. The dependent variable was a binary variable named STATUS having the value 0 for healthy firms and 1 for bankrupt firms. Independent variables include QUICK, DEBEQ, PBTTA, ROE, CATA, CLTA, EQTA, CASHCL, CASHTD, QATA, FANW, SOLVEN, SALESFA, EBITCL, CFDEB, ROA, CURRENT, SIZE, WCTA, RETA, PAITTA, EQTD and SALTA. The number of firms and observations for all the analyses are the same. It is only differentiated during the process of each analysis according to significance and other validity estimations.

Three different models were analyzed in this study:

  1. Multivariate discriminant analysis (MDA).

  2. BLR.

  3. Multilayer perceptron artificial neural network (MLP NN).

4. Results

4.1 Multivariate discriminant analysis (MDA)

MDA model assigns every case to the group which obtains the highest value in the classification function. Here, the coefficients QUICK, PBTTA, SALES, QATA and CASHTD are smaller which means that these factors play a most important role when they are negative to lead a company to a bankruptcy status and they are most likely to default.

FANW is highly insignificant followed by ROE, CLTA, EQTA, QATA, CFDEB and WCTA. Wilks’ λ is another measure of a variable’s potential. Smaller values like SALTA, CURRENT, SALESFA, EQTD and EBITCL indicate the variables which are better at discriminating between groups. SALTA, CURRENT, EQTD, SOLVEN, SALESFA and EBITCL are the most significant indicators with the least value of Wilks’ λ and highest standardized coefficient values, showing that these six variables contribute the best in this model. The correlation is 0.915 which is quite good and indicates the good correlation between discriminant scores and group. The classification table shows the applied results of using the MDA model. Of the cases used in model creation, 24 of 24 companies which previously bankrupted are classified accurately. In total, 30 out of 30 healthy firms are classified accurately. Overall, 100 percent of the cases are categorized accurately. In all, 76.9 percent of these cases were acceptably categorized by the model. This means that overall the model is correct for about three out of four times. The 371 ungrouped companies are the potential companies. Thus, a model has been created to classify companies as future defaulters or non-defaulters using multivariate discriminant analysis (Table II).

4.2 Binary logistic regression (BLR)

The variable with the largest value statistic which has a significant value was added to the model at each step. At the last step, all the variables left were having the significance value more than 0.10. The method finally chooses two variables, i.e. SALTA and EBITCL. Generally, the model with the largest R2 statistic is considered “best.” Here, the R2 value is 0.808, which is very good. Practical results of BLR are shown in the classification table which shows that 26 out of 27 healthy companies were classified accurately. Overall, 94.1 percent of all the cases included were categorized correctly. From the unselected cases, 75.9 percent of them were accurately classified by BLR model constructed. This is three out of four times accurate as well. SALTA and EBITCL came out to be the best predictors of bankruptcy in this model. CLTA and CATA also contribute significantly if constant is not taken in the model. The model for predicting bankruptcy in Pakistani non-manufacturing companies using the logistic regression procedure is constructed.

4.3 Artificial neural networks (multilayer perceptron)

The case processing summary shows that 62 cases are assigned to the training sample and 18 to the holdout sample. The 371 cases excluded from the analysis are the prospective companies. The network information table displays information about the neural network and is useful for ensuring that the specifications are correct. Here, the number of units in the input layer is the number of factors and covariates which is 23. Likewise, a separate output unit is created for each category of STATUS, for a total of two units in the output layer. Automatic architecture selection has chosen three units in the hidden layer (Tables III, IV, V, VI, VII, VIII).

Cross-entropy error in the training was as minimum as 0.042 with 0 percent incorrect predictions. The training time taken was 15 sec. The incorrect prediction in holdout sample was 0 percent too. The estimation algorithm was stopped because the maximum number of epochs was reached. Ideally, training should stop because the error has converged. This summary shows that this model is 100 percent accurate with the cross entropy error of 0.042 (Table IX).

Of the cases used to create the model, 30 out of 30 companies which previously defaulted are classified correctly. In total, 32 out of 32 non-defaulters are classified correctly. Overall, 100 percent of the training cases are classified correctly, corresponding to the 0 percent shown incorrectly in the model of table summary. A better model should correctly identify a higher percentage of the cases. Classifications based upon the cases used to create the model tend to be too “optimistic” in the sense that their classification rate if inflated. The holdout sample helps to validate the model; here, 100 percent of these cases were correctly classified by the model. This suggests that, overall, the model is, in fact, correct for all the cases which are reasonably not possible. That is the reason why we corrected the model as it may have been over-trained. Here, we added the testing sample to correct overtraining. Then, the correction of overtraining was performed. The only change to the network information table is that the automatic architecture selection has chosen four units in the hidden layer.

From 62 cases originally assigned to the training sample, 12 have been reassigned to the testing sample. Rest 18 of the companies were kept as holdout sample. The training time taken was 62 seconds. Cross-entropy error of 4.269 is more than the previous model but is not much to resist the model from acceptance. The percentage of incorrect predictions in training sample was just 2 percent but 16.7 percent in a testing sample, which shows the overall accuracy of around 90 percent (Figure 1 and Tables X and XI).

The classification table shows that using 0.5 as the pseudo-probability cutoff for classification, the network does better at predicting non-defaulters than defaulters in the testing and holdout samples. However, in the training sample, it predicts defaulters and non-defaulters equally correct. Unfortunately, the single cutoff value gives a very limited view of the predictive ability of the network, so it is not necessarily very useful for comparing competing networks. Instead, we analyze the ROC curve (Table XII).

The importance of an independent variable is a measure of how much the network’s model predicts the value changes for different values of the independent variable. Normalized importance is simply the important values divided by the largest importance values and expressed as percentages. The importance chart is simply a bar chart of the values in the importance table, sorted in the descending value of importance. It appears that cumulative profitability (RETA) has the highest importance as a predictor of financial distress in this model of Neural Network. Also, CLTA has a greater effect on how the network classifies companies. PBTTA, EBITCL and FANW share nearly equal high importance too. CASHTD and SOLVEN have the lowest importance to this network. Using the Multilayer Perceptron, a network is constructed for predicting the probabilities that a company will default (Figure 2).

4.4 Comparison

According to the individual model’s results, MDA indicates 72 companies, BLR indicates 112 companies and ANN indicates nearly a half of the companies to be included in the situation of financial distress leading to bankruptcy. Some of the companies are shown in Table AI. It is also clear from the results that the textile sector is having the worst situation of all.

If we compare the accuracy of prediction of all three models, we can conclude that ANN is the best among all the three tested models with the lowest value of the error, most of the variables included and with a higher accuracy percentage of approximately 90 percent. BLR predicts nearly 85 percent accurate values and MDA stays at least with 75 percent prediction accuracy. This prediction does not include the exact prediction for the bankruptcy status but the higher probability to be bankrupted because of financial distress.

The statistics and results from MDA show that the best variables to forecast bankruptcy in Pakistan are RETA, SALTA, EBITCL, CURRENT and EQTD. Two of these variables are considered the best while predicting bankruptcy in the non-financial sector of Pakistan through BLR. These two variables are SALTA and EBITCL which contributed greatly to forecasting bankruptcy. Abbas and Ahmad (2011) also found these two variables the most appropriate in forecasting financial distress when they used multivariate discriminant analysis on the non-financial sector of Pakistan. The multilayer neural network model proved to be the best model for predicting bankruptcy in Pakistan’s corporate sector using financial ratios of one year prior to bankruptcy. The results are like the results of research conducted by Abbas and Ahmad (2011), who compared different models using one-year ratios for prediction. Our research contradicts with Altman et al. (1994) work in which MDA appeared to perform slightly better than MLP-NN. Also, the best predictors evaluated were different in that study. The best predictors chosen by MLP-NN in our research are RETA, CLTA and PBTTA. In line with the study of Pourreza et al. (2012), ANN found Cumulative profitability (RETA) as the best financial indicator for forecasting bankruptcy (Table XIII).

5. Conclusion

In this study, an analysis of corporate distress and bankruptcy was done on Pakistani companies included in the non-financial sector for the time span of 1995--2017, using three different methods, i.e. multivariate discriminant analysis, logistic regression and ANN (multilayer perceptron). In all the methodologies, the analysis was based on financial ratios and accounting predicting indicators which have been used and known as significant variables to explore bankruptcy status. In total, 23 different indicators (representative of each category of ratios) were introduced. However, according to model specifications, few of these indicators were selected and the analyses on indicators of bankruptcy were performed to test the forecasting ability of each variable used. The three different models used in this prediction vary with the prediction accuracy. ANN achieves the highest prediction accuracy of 88.9 percent while forecasting bankruptcies on the underlying sample. BLR achieves the lowest prediction accuracy score of about 75.9 percent which is even acceptable too. This study does not only estimate the best bankruptcy prediction model in Pakistan but also shows that many of the firms go bankrupt during the time of 1995–2017, depicting the signs of being faced by financial distress due to poor performance.

The key indicators to predict financial distress leading to bankruptcy in the corporate sector of Pakistan were found. SALTA, EBITCL and RETA proved to be the best predictors. Other indicators including CLTA, CURRENT, SOLVEN, SALESFA, EQTD and PBTTA were also found to be having an important role in forecasting bankruptcy.

It is suggested that Pakistan’s regulatory authorities must monitor and assess the significant financial indicator found in this research to keep the financial health of the company in a better position. Also, it is contended that the ANN model constructed in this study functions as an insight to assess difficult financial situations of a company and provides a wide scope of research for practitioners and students to construct a better distress forecasting model for Pakistan’s corporate sector.

6. Future recommendations

To predict bankruptcy, the technique of neural networks was adopted but as the technique can produce superior foresight, the analysts may find out the improvement areas in the neural networks. They may use better training methods and input. More efficient the input, the more effective the prediction is. To test bankruptcy, the future studies may use a different and related set of financial indicators. They may also cover the social and environmental aspects of sustainable organizations to analyze its impact on bankruptcy. They can also use more indicators in all the techniques (BLR, MDA and even ANN) which were used in this study. Moreover, the inclusion of variables like CEO’s tenure, CEO’s experience, family-owned business, characteristics of board of directors and top hierarchy can be useful in predicting bankruptcy more precisely. The data selection method can be slightly more adequate. As the data of some companies are not available and the indicators specified by the companies are less in numbers. That data can be traced out and collected to perform a better analysis by finding unique unexplored trends in the affected firms. Including more accounting data and variables in further studies could produce better predictor models than the estimated models in this study. The researchers can do sector-wise research and can make prediction models for each sector for better results. Also, probit analysis, a decision tree model, genetic algorithm NNs, AMOS and mixed complex models can be tested, applied and selected for the best prediction of the bankruptcy of firms in Pakistan. The exploration can also examine the financial sector of Pakistan.

Figures

Hidden layer activation function

Figure 1

Hidden layer activation function

Normalized importance

Figure 2

Normalized importance

Summary statistics of financial ratios

Bankrupt firms Healthy firms
Ratios n Mean Median SD Var Max. Min. n Mean Median SD Var Max. Min.
QUICK 40 0.37 0.11 0.75 0.57 3.60 −1.22 40 0.54 0.39 0.42 0.17 1.51 0.14
DEBEQ 40 −3.54 −1.48 17.02 289.84 10.44 −106.91 40 3.47 1.94 6.84 46.75 26.06 −2.76
PBTTA 40 0.80 −0.07 4.62 21.36 25.97 −2.43 40 0.03 0.01 0.08 0.01 0.23 −0.14
ROE 40 0.34 0.02 2.66 7.08 16.45 −1.75 40 −0.12 0.08 0.82 0.68 0.43 −2.98
CATA 40 0.34 0.21 0.33 0.11 1.00 0.00 40 0.45 0.45 0.16 0.03 0.74 0.14
CLTA 40 8.27 1.31 29.00 841.20 179.10 0.05 40 0.51 0.53 0.15 0.02 0.75 0.17
EQTA 40 −7.54 −0.59 28.91 835.96 1.00 −177.83 40 0.24 0.25 0.37 0.14 0.81 −0.61
CASHCL 40 0.07 0.00 0.32 0.10 2.00 0.00 40 0.11 0.04 0.22 0.05 0.82 −0.03
CASHTD 40 0.04 0.00 0.13 0.02 0.67 0.00 40 0.10 0.03 0.22 0.05 0.82 −0.03
QATA 40 0.23 0.19 0.36 0.13 1.00 −1.20 40 0.23 0.22 0.10 0.01 0.42 0.06
FANW 40 4.74 −0.11 44.68 1,996.45 267.00 −85.36 40 2.30 1.44 3.83 14.69 14.88 −1.39
SOLVEN 40 −0.03 −0.01 0.28 0.08 0.85 −1.02 40 0.15 0.05 0.25 0.06 0.94 −0.09
SALESFA 40 0.57 0.17 0.87 0.75 3.84 0.00 40 1.95 1.39 1.41 1.98 5.60 0.22
EBITCL 40 −0.19 −0.07 0.51 0.26 0.58 −2.08 40 0.14 0.08 0.21 0.05 0.69 −0.11
CFDEB 40 −0.03 −0.01 0.28 0.08 0.85 −1.02 40 0.64 1.11 0.36 0.13 −2.10 3.40
ROA 40 0.80 −0.07 4.63 21.41 26.01 −2.43 40 0.05 0.04 0.08 0.01 0.25 −0.07
CURRENT 40 0.46 0.19 0.78 0.61 3.60 0.00 40 1.01 0.96 0.59 0.35 2.78 0.22
WCTA 40 −7.93 −0.87 28.97 839.11 0.67 −178.74 40 −0.06 −0.01 0.21 0.04 0.38 −0.53
RETA 40 0.83 −0.07 4.62 21.31 25.97 −2.43 40 0.01 0.01 0.06 0.00 0.13 −0.14
PAITTA 40 0.80 −0.07 4.62 21.36 25.97 −2.43 40 0.02 0.01 0.08 0.01 0.20 −0.14
EQTD 40 0.31 −0.36 3.22 10.35 19.60 −0.99 40 0.70 0.33 1.06 1.12 3.52 −0.40
SALTA 40 0.26 0.04 0.38 0.15 1.49 0.00 40 0.94 0.76 0.59 0.35 2.58 0.17

Classification function coefficients

Status
Healthy Bankrupt
QUICK −3.498 −1.336
DEBEQ 1.057 −0.413
PBTTA −4.312 −1.858
ROE 7.029 −2.454
CATA 36.903 7.642
CLTA −0.031 0.003
CASHCL 641.051 57.34
CASHTD −676.668 −57.771
QATA −16.933 4.343
FANW 0.013 0.012
SOLVEN 51.421 1.387
SALESFA −1.208 −0.075
(Constant) −12.094 −2.596

Note: Fisher’s linear discriminant functions

Test of equality of group means

Wilks’ λ F df1 df2 Sig.
QUICK 0.869 7.84 1 52 0.007
DEBEQ 0.908 5.289 1 52 0.026
PBTTA 0.907 5.333 1 52 0.025
ROE 0.964 1.959 1 52 0.168
CATA 0.86 8.494 1 52 0.005
CLTA 0.964 1.952 1 52 0.168
EQTA 0.965 1.898 1 52 0.174
CASHCL 0.918 4.662 1 52 0.035
CASHTD 0.931 3.864 1 52 0.055
QATA 0.985 0.786 1 52 0.38
FANW 0.996 0.207 1 52 0.651
SOLVEN 0.833 10.414 1 52 0.002
SALESFA 0.736 18.691 1 52 0
EBITCL 0.798 13.167 1 52 0.001
CFDEB 0.95 2.73 1 52 0.105
ROA 0.892 6.293 1 52 0.015
CURRENT 0.681 24.326 1 52 0
SIZE 0.931 3.827 1 52 0.056
WCTA 0.962 2.031 1 52 0.16
RETA 0.935 3.596 1 52 0.063
PAITTA 0.911 5.082 1 52 0.028
EQTD 0.787 14.093 1 52 0
SALTA 0.672 25.344 1 52 0

Classification results: BLR

Predicted group membership
Status Healthy Bankrupt Total
Cases selected
Original
 Count 30
  Healthy 30 0
  Bankrupt 0 24 24
 %
  Healthy 100 0 100
  Bankrupt 0 100 100
Cross-validated
 Count
  Healthy 30 0 30
  Bankrupt 4 20 24
 %
  Healthy 100 0 100
  Bankrupt 16.7 83.3 100
Cases not selected
Original
 Count
  Healthy 7 3 10
  Bankrupt 3 13 16
  Ungrouped cases 273 98 371
 %
  Healthy 70 30 100
  Bankrupt 18.8 81.2 100
  Ungrouped cases 73.6 26.4 100

Notes: Cross validation is done only for cases in the analysis. In cross validation, each case is classified by the functions derived from all cases other than that case; 100.0 percent of selected original grouped cases correctly classified; 76.9 percent of unselected original grouped cases correctly classified; 92.6 percent of selected cross-validated grouped cases correctly classified

Model summary: BLR

Step −2 log likelihoods Cox and Snell R2 Nagelkerke R2
3 23.146a 0.605 0.808

Note: aEstimation terminated at iteration number 8 because parameter estimates changed by less than 0.001

Classification table

Predicted
Selected cases Unselected cases
Status Status
Observed 0 1 Percentage correct 0 1 Percentage correct
Step 1
Status 0 24 3 88.9 11 2 84.6
1 3 21 87.5 4 12 75
Overall percentage 88.2 79.3
Step 2
Status 0 24 3 88.9 11 2 84.6
1 3 21 87.5 4 12 75
Overall percentage 88.2 79.3
Step 3
Status 0 24 3 88.9 10 3 76.9
1 3 21 87.5 5 11 68.8
Overall percentage 88.2 72.4
Step 4
Status 0 26 1 96.3 11 2 84.6
1 3 21 87.5 5 11 68.8
Overall percentage 92.2 75.9

Notes: Selected cases validate EQ 1; Unselected cases validate NE 1; The cut value is 0.500

Case processing summary: ANN-MLP

n Percent
Sample Training 62 77.5
Holdout 18 22.5
Valid 80 100.0
Excluded 371
Total 451

Model summary: ANN-MLP

Training
Cross-entropy error 0.042
Percent incorrect predictions 0.0
Stopping rule used Training error ratio criterion (0.001) achieved
Training time 00:00:00.015
Holdout
Percent incorrect predictions 0.0

Note: Dependent variable: STATUS

Classification: ANN-MLP

Predicted
Sample Observed Healthy Bankrupt Percent correct
Training Healthy 32 0 100.0
Bankrupt 0 30 100.0
Overall percent 51.6 48.4 100.0
Holdout Healthy 8 0 100.0
Bankrupt 0 10 100.0
Overall percent 44.4 55.6 100.0

Note: Dependent variable: status

Case processing summary: ANN-MLP after correcting over-training

n Percent
Sample Training 50 62.50
Testing 12 15.00
Holdout 18 22.50
Valid 80 100.00
Excluded 371
Total 451

Model summary: ANN-MLP after correcting over-training

Training
Cross-entropy error 11.459
Percent incorrect predictions 11.3
Stopping rule used 1 consecutive step(s) with no decrease in errora
Training time 00:00:00.094
Testing
Cross-entropy error 0.173
Percent incorrect predictions 0.0
Holdout
Percent incorrect predictions 15.0

Notes: Dependent variable: status. aError computations are based on the testing sample

Classification: ANN-MLP after correcting over-training

Predicted
Sample Observed Healthy Bankrupt Percent correct
Training Healthy 24 0 100.0
Bankrupt 1 25 96.2
Overall percent 50.0 50.0 98.0
Testing Healthy 8 0 100.0
Bankrupt 2 2 50.0
Overall percent 83.3 16.7 83.3
Holdout Healthy 8 0 100.0
Bankrupt 2 8 80.0
Overall percent 55.6 44.4 88.9

Note: Dependent variable: status

Independent variable importance

Importance Normalized importance (%)
QUICK 0.039 40.30
DEBEQ 0.04 41.80
PBTTA 0.072 75.10
ROE 0.042 43.70
CATA 0.049 51.20
CLTA 0.078 81.40
EQTA 0.035 36.70
CASHCL 0.034 35.10
CASHTD 0.014 14.70
QATA 0.061 64.10
FANW 0.068 70.90
SOLVEN 0.012 12.20
SALESFA 0.018 18.50
EBITCL 0.071 74.40
CFDEB 0.019 19.80
ROA 0.047 48.80
CURRENT 0.035 36.50
SIZE 0.036 37.70
WCTA 0.028 29.20
RETA 0.096 100.00
PAITTA 0.028 29.50
EQTD 0.043 44.90
SALTA 0.037 39.20

List of companies identified which are to face financial distress in near future: (results)

(Colony) Sarhad Textile Mills Nazir Cotton Mills Ltd Sitara Peroxide Ltd
Adil Textile Mills Limited Olympia Textile Mills Ltd Aisha Steel Mills Ltd
Amtex Limited Redco Textiles Ltd Dost Steels Ltd
Annoor Textile Mills Limited Saleem Denim Industries Ltd Fateh Industries Ltd
Brothers Textile Mills Ltd Service Fabrics Ltd Pak Leather Crafts Ltd
Chenab Ltd Sind Fine Textile Mills Ltd Frontier Ceramics Ltd
D.M. Textile Mills Ltd Taha Spinning Mills Ltd Dadabhoy Cement Industries Ltd
Data Textiles Ltd Fateh Sports Wear Ltd Dandot Cement Co. Ltd
Fateh Textile Mills Ltd Moonlite (Pak) Ltd Dewan Cement Ltd
Gulistan Spinning Mills Ltd Al-Abid Silk Mills Ltd Zeal Pak Cement Factory Ltd
Gulistan Textile Mills Ltd Crescent Jute Products Ltd Dewan Automotive Engineering
Gulshan Spinning Mills Ltd Noor Silk Mills Ltd Dewan Farooque Motors Ltd
Hajra Textile Mills Ltd S.G. Fibers Ltd Ghani Automobiles Industries
Husein Industries Ltd Suhail Jute Mills Ltd Transmission Engineering
Karim Cotton Mills Ltd Abdullah Shah Ghazi Sugar Mills Telecard Ltd
Khurshid Spinning Mills Ltd Ansari Sugar Mills Ltd Wateen Telecom
Kohinoor Industries Ltd Morafco Industries Ltd World Call Telecom
Landmark Spinning Industries Shakarganj Foods Ltd Baluchistan Particle Board Ltd
Mehr Dastgir Textile Mills Agritech Ltd Central Forest Products Ltd
Mian Textile Industries Ltd Dewan Salman Fiber Ltd Johnson and Philips (Pakistan) Ltd
Mohammad Farooq Textile Pakistan PVC Ltd The Climax Engineering
Mubarak Textile Mills Ltd Shaffi Chemical Industries Ltd Haydari Construction Company
Globe Textile Mills Pace (Pakistan) Ltd

List of financial variables

Quick ratio (QUICK) Cash to total debt (CASHTD) Current ratio (CURRENT)
Debt equity ratio (DEBEQ) Return of assets (ROA) Sales/total assets (SALTA)
Profit before taxes/total assets (PBTTA) Earnings before interest and taxes/current liabilities (EBITCL) Working capital/total assets (WCTA)
Return on equity (ROE) Fixed assets turnover (SALESFA) Asset size (SIZE)
Current assets/total assets (CATA) Solvency ratio (SOLVEN) Retained earnings/total assets (RETA)
Current liabilities to total assets (CLTA) Fixed assets to net worth (FANW) Earnings before interest and taxes/total assets (EBITTA)
Owners’ equity to total assets (EQTA) Cash flow to total debt (CFDEB) Equity/total liabilities (EQTD)
Cash to current liabilities (CASHCL) Quick assets to total assets (QATA)

Appendix 1

Table AI

Appendix 2

Table AII

References

Abbas, Q. and Ahmad, A.R. (2011), “Modeling bankruptcy prediction for non-financial firms: the case of pakistan”, SSRN, doi: 10.2139/ssrn.1917458.

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, available at: http://doi.org/10.1111/j.1540-6261.1968.tb00843.x

Altman, E.I. and Mcgough, T.P. (1974), “Evaluation of a company as a going concern”, Journal of Accountancy, Vol. 138 No. 6, pp. 50-57.

Altman, E.I., Marco, G. and Varetto, F. (1994), “Corporate distress diagnosis: comparisons using linear discriminant analysis and neural networks (the Italian experience)”, Journal of Banking and Finance, Vol. 18 No. 3, pp. 505-529, available at: http://doi.org/10.1016/0378-4266(94)90007-8

Altman, E.I., Iwanicz-Drozdowska, M., Laitinen, E.K. and Suvas, A. (2017), “Financial distress prediction in an international context: a review and empirical analysis of Altman’s Z-score model”, Journal of International Financial Management & Accounting, Vol. 28 No. 2, pp. 131-171, available at: http://doi.org/10.1111/jifm.12053

Atiya, A.F. (2001), “Bankruptcy prediction for credit risk using neural networks: a survey and new results”, IEEE Transactions on Neural Networks, Vol. 12 No. 4, pp. 929-935, available at: http://doi.org/10.1109/72.935101

Aziz, M.A. and Dar, H.A. (2006), “Predicting corporate bankruptcy: where we stand?”, Corporate Governance, Vol. 6 No. 1, pp. 18-33, available at: http://doi.org/10.1108/14720700610649436

Barboza, F., Kimura, H. and Altman, E. (2017), “Machine learning models and bankruptcy prediction”, Expert Systems with Applications, Vol. 83, pp. 405-417, available at: http://doi.org/10.1016/j.eswa.2017.04.006

Beaver, W.H. (1966), “Financial ratios as predictors of failure”, Journal of Accounting Research Empirical Research in Accounting: Selected Studies, Vol. 4, pp. 71-111, available at: http://doi.org/10.2307/2490173

Bernstein, S., Colonnelli, E., Giroud, X. and Iverson, B. (2017), “Bankruptcy spillovers”, Working Paper Series No. 23162, National Bureau of Economic Research, available at: http://doi.org/10.3386/w23162

Bisogno, M., Cuadrado-Ballesteros, B. and García-Sánchez, I.M. (2017), “Financial sustainability in local governments: definition, measurement and determinants”, Financial Sustainability in Public Administrations, Exploring the Concept of Financial Health, Springer International Publishing, Cham, pp. 57-83, doi: 10.1007/978-3-319-57962-7_3.

Charitou, A. and Trigeorgis, L. (2000), “Option-based bankruptcy prediction”, available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=248709 (accessed 15 june 2018).

Chava, S. and Jarrow, R.A. (2004a), “Bankruptcy prediction with industry effects”, European Finance Review, Vol. 8 No. 4, pp. 537-569, available at: http://doi.org/10.1007/s10679-004-6279-6

Chava, S. and Jarrow, R.A. (2004b), “Bankruptcy prediction with industry effects”, Review of Finance, Vol. 8 No. 4, pp. 537-569, available at: http://doi.org/10.1093/rof/8.4.537

Chou, C.H., Hsieh, S.C. and Qiu, C.J. (2017), “Hybrid genetic algorithm and fuzzy clustering for bankruptcy prediction”, Applied Soft Computing Journal, Vol. 56, pp. 298-316, available at: http://doi.org/10.1016/j.asoc.2017.03.014

Coats, P.K. and Fant, L.F. (1993), “Recognizing financial distress patterns using a neural network tool”, Financial Management, Vol. 22 No. 3, pp. 142-155, available at: http://doi.org/10.2307/3665934

Demyanyk, Y. and Hasan, I. (2010), “Financial crises and bank failures: a review of prediction methods”, Omega, Vol. 38 No. 5, pp. 315-324, available at: http://doi.org/10.1016/j.omega.2009.09.007

Dobbie, W., Goldsmith-Pinkham, P. and Yang, C.S. (2017), “Consumer bankruptcy and financial health”, The Review of Economics and Statistics, Vol. 99 No. 5, pp. 853-869, available at: http://doi.org/10.1162/REST_a_00669

Efrim Boritz, J. and Kennedy, D.B. (1995), “Effectiveness of neural network types for prediction of business failure”, Expert Systems with Applications, Vol. 9 No. 4, pp. 503-512, available at: http://doi.org/10.1016/0957-4174(95)00020-8

Finley, D.S., Gralen, A. and Fichtner, L. (2006), “From bankruptcy to sustainability :stakeholder engagement and….”, International Journal of Arts Management, Vol. 9 No. 1, pp. 4-16.

Frade, C. (2012), “Bankruptcy, stigma and rehabilitation”, ERA Forum, Vol. 13 No. 1, pp. 45-57, available at: http://doi.org/10.1007/s12027-012-0249-9

Franks, J., Sussman, O. and Vig, V. (2015), “The privatization of bankruptcy: evidence from the shipping industry”, available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2880751 (accessed 9 May 2018).

Gerstrøm, A. and Isabella, L.A. (2015), “Understanding bankruptcy: how members of a bankrupted bank construe organizational death”, Illness Crisis and Loss, Vol. 23 No. 2, pp. 129-174, available at: http://doi.org/10.1177/1054137315575844

Hassanpour, S. and Ardakani, M.N. (2017), “The effect of pre-bankruptcy financial distress on earnings management tools”, International Review of Management and Marketing, Vol. 7 No. 3, pp. 213-219.

Hensher, D.A. and Jones, S. (2007), “Forecasting corporate bankruptcy: optimizing the performance of the mixed logit model”, Abacus, Vol. 43 No. 3, pp. 241-364, available at: http://doi.org/10.1111/j.1467-6281.2007.00228.x

Jan, A. and Marimuthu, M. (2015), “Bankruptcy and sustainability: a conceptual review on Islamic banking industry”, An International Journal Global Business and Management Research, Vol. 7 No. 1, pp. 109-138.

Khani, A.H. (2015), “Predicting bankruptcy by liquidity ratios analysis”, Jurnal UMP Social Sciences and Technology Management, Vol. 3 No. 2, pp. 372-380, available at: http://jsstm-ump.org/2015 Volume 3, Issue 2/372-380.pdf (accessed 24 May 2018).

Kuhlman, T. and Farrington, J. (2010), “What is sustainability?”, Sustainability, Vol. 2 No. 11, pp. 3436-3448.

Laitinen, E.K. and Laitinen, T. (2000), “Bankruptcy prediction: application of the Taylor’s expansion in logistic regression”, International Review of Financial Analysis, Vol. 9 No. 4, pp. 327-349, available at: http://doi.org/10.1016/S1057-5219(00)00039-9

Manalu, S., Octavianus, R.J.N. and Kalmadara, G.S.S. (2017), “Z-score approach and Zmijewski X-score on shipping service company”, Journal of Applied Management, Vol. 15 Nos 4/36, pp. 677-682, available at: http://doi.org/10.21776/ub.jam.2017.015.04.15

Meyer, P.A. and Pifer, H.W. (1970), “Prediction of bank failures”, The Journal of Finance, Vol. 25 No. 4, pp. 853-868, available at: http://doi.org/10.1111/j.1540-6261.1970.tb00558.x

Miller, W. (2009), “Introducing the morningstar solvency score, a bankruptcy prediction metric”, SSRN ELibrary, available at: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1516762

Najam, U., Inam, A., Awan, H.M. and Abbas, M. (2018), “The interactive role of temporal team leadership in the telecom sector of Pakistan: utilizing temporal diversity for sustainable knowledge sharing”, Sustainability, Vol. 10 No. 5, pp. 2071-1050.

Nam, C.W., Kim, T.S., Park, N.J. and Lee, H.K. (2008), “Bankruptcy prediction using a discrete-time duration model incorporating temporal and macroeconomic dependencies”, Journal of Forecasting, Vol. 27 No. 6, pp. 493-506, available at: http://doi.org/10.1002/for.985

Ohlson, J.A. (1980), “Financial ratios and the probabilistic prediction of bankruptcy”, Journal of Accounting Research, pp. 109-131, available at: http://doi.org/10.2307/2490395

Oliveira, M.D.N.T., Ferreira, F.A.F., Pérez-Bustamante Ilander, G.O. and Jalali, M.S. (2017), “Integrating cognitive mapping and MCDA for bankruptcy prediction in small-And medium-sized enterprises”, Journal of the Operational Research Society, Vol. 68 No. 9, pp. 985-997, available at: http://doi.org/10.1057/s41274-016-0166-3

Pourreza, A., Ahmadi, S., Soleimani, B., Vaghfi, S.H. and Salimi, M.B. (2012), “Corporate bankruptcy prediction using a logit model: evidence from listed companies of Iran”, World Applied Sciences Journal, Vol. 17 No. 9, pp. 1143-1148, available at: https://pdfs.semanticscholar.org/b7da/5ae2b9983d4668f64b9b2ad581eebacda5cd.pdf

Ravi Kumar, P. and Ravi, V. (2007), “Bankruptcy prediction in banks and firms via statistical and intelligent techniques – a review”, European Journal of Operational Research, Vol. 180 No. 1, pp. 1-28, available at: http://doi.org/10.1016/j.ejor.2006.08.043

Salchenberger, L.M., Cinar, E.M. and Lash, N.A. (1992), “Neural networks: a new tool for predicting thrift failures”, Decision Sciences, Vol. 23 No. 4, pp. 899-916, available at: http://doi.org/10.1111/j.1540-5915.1992.tb00425.x

Shumway, T. (2001), “Forecasting bankruptcy more accurately: a simple hazard model”, The Journal of Business, Vol. 74 No. 1, pp. 101-124, doi: 10.1086/209665.

Ştefănescu-Mihăilă, R. (2015), “Social investment, economic growth and labor market performance: case study – Romania”, Sustainability, Vol. 7 No. 3, pp. 2961-2979, available at: http://doi.org/10.3390/su7032961

Tam, K.Y. and Kiang, M.Y. (1992), “Managerial applications of neural networks: the case of bank failure predictions”, Management Science, Vol. 38 No. 7, pp. 926-947, available at: http://doi.org/10.1287/mnsc.38.7.926

Tsai, C.F. (2009), “Feature selection in bankruptcy prediction”, Knowledge-Based Systems, Vol. 22 No. 2, pp. 120-127, available at: http://doi.org/10.1016/j.knosys.2008.08.002

West, M., Harrison, J.P. and Migon, H.S. (1985), “Dynamic generalized linear models and Bayesian forecasting”, Journal of the American Statistical Association, Vol. 80 No. 389, pp. 73-83, available at: http://doi.org/10.2307/2288042

Zhao, D., Huang, C., Wei, Y., Yu, F., Wang, M. and Chen, H. (2017), “An effective computational model for bankruptcy prediction using kernel extreme learning machine approach”, Computational Economics, Vol. 49 No. 2, pp. 325-341, available at: http://doi.org/10.1007/s10614-016-9562-7

Zmijewski, M. (1984) “Methodological Issues Related to the Estimation of Financial Distress Prediction Models”, Journal of Accounting Research, Vol. 22, pp. 59-82, available at: https://about.jstor.org/terms

Further reading

Dietrich, J.R. (1984), “Discussion of methodological issues related to the estimation of financial distress prediction models”, Journal of Accounting Research, Vol. 22, pp. 83-86, available at: http://doi.org/10.2307/2490860

Mueller, U.G., Schultz, T.R., Currie, C.R. and Malloch, D. (2001), “The origin of the attine ant-fungus mutualism”, The Quarterly Review of Biology, Vol. 76 No. 2, pp. 169-197, available at: http://doi.org/10.1086/393867

Razmi, F., Mohamed, A., Chin, L. and Habibullah, M.S. (2015), “The effects of oil price and US economy on Thailand’s macroeconomy: the role of monetary transmission mechanism”, International Journal of Economics and Management, Vol. 9, Special Issue, pp. 121-141, available at: http://doi.org/10.5897/JAERD12.088

Acknowledgements

The authors of this article have not made their research dataset openly available. Any enquiries regarding the data set can be directed to the corresponding author.

Corresponding author

Aneeq Inam can be contacted at: aneeqinam93@gmail.com