Does size matter in predicting SMEs failure?

• Author: Mariachiara Barzotto,Amir Khorasgani,Jairaj Gupta
• DOI: http://doi.org/10.1002/ijfe.1638
• Published date: 01 October 2018
• Date: 01 October 2018

RESEARCH ARTICLE

Does size matter in predicting SMEs failure?

Jairaj Gupta

1

| Mariachiara Barzotto

2

| Amir Khorasgani

3

1

Department of Finance, University of

Birmingham, Birmingham, UK

2

Newcastle University Business School,

Newcastle University, Newcastle, UK

3

School of Economics, Finance and

Accounting, Coventry University,

Coventry, UK

Correspondence

Jairaj Gupta, Department of Finance,

University of Birmingham, Birmingham

B15 2TY, UK.

Email: J.Gupta@outlook.com

JEL Classification: G32; G33

Abstract

This study acknowledges the diversity between micro, small, and medium‐

sized firms while predicting bankruptcy and financial distress of the United

States small and medium‐sized enterprises. Empirical findings suggest that

survival (failure) probability increases (decreases) with increasing firm size

and firms in different size categories have varying determinants of bankruptcy,

whereas factors affecting their financial distress are mostly invariant.

Magnitude of significant covariates changes across the size categories of both

bankrupt and financially distressed firms. Further, operating cash flow

information does not add any marginal increment in prediction performance

of multivariate hazard models above baseline models developed using

information from income statements and balance sheets. This result holds for

failure likelihood of small and medium‐sized enterprises and their respective

size categories.

KEYWORDS

bankruptcy, financial distress, operating cash flow, SMEs, survival analysis

1|INTRODUCTION

In developed economies, small and medium‐sized enter-

prises (SMEs) are, relative to turnover, disproportion-

ately linked to employment rates. In the United States,

for instance, almost half of all employees work for

enterprises with fewer than 250 employees. SMEs play

a crucial role in the global economy and are pivotal to

the economic growth and development of a country

(Bosma & Levie, 2010), as well as to poverty reduction

(Koshy & Prasad, 2007). Therefore, a detailed under-

standing of the factors affecting SMEs survival is

important for policymakers, firms, and capital suppliers.

The attention devoted to SMEs survival has constantly

increased over the years, particularly after the

financial crisis in 2008–2009. Indeed, the revised Basel

capital accords and national governments have placed

greater emphasis on understanding the credit risk attri-

butes of SMEs.

Notwithstanding, the extensive literature on the

performance and financial distress of SMEs, the factors,

and the extent to which they affect SMEs failure likeli-

hood across size categories are still overlooked. As

argued in a recent study by Altman, Iwanicz‐

Drozdowska, Laitinen, and Suvas (2017), bankrupt

and nonbankrupt firms have different boundaries due

to their size (small and large), which affect the accu-

racy of prediction when data from one size category

is used for another size category. Building upon the

previous evidence showing that size affects access to

finance (e.g., Beck & Demirgüç‐Kunt, 2006), we pro-

pose a development in modelling financial distress

and bankruptcy in the United States. More specifically,

we address this issue by exploring whether insolvency

and financial distress likelihood varies across size

categories of the United States SMEs, by looking at

the factors affecting SMEs failure likelihood in three

subcategories of SMEs (namely, micro, small, and

Received: 14 August 2017 Revised: 30 March 2018 Accepted: 20 June 2018

DOI: 10.1002/ijfe.1638

Int J Fin Econ. 2018;23:571–605. © 2018 John Wiley & Sons, Ltd.wileyonlinelibrary.com/journal/ijfe 571

medium

1

). Few studies explore the differences amongst

the subcategories of SMEs. Analysing financial and

nonfinancial factors affecting UK SMEs bankruptcy

across company size, Gupta, Gregoriou, and Healy

(2015) show that the credit risk characteristics of micro

firms significantly differ from SMEs as a whole.

Accordingly, they suggest that they should be treated

separately for better pricing of credit risk and devising

effective credit policies. In the light of this discussion,

we expect the default characteristics to vary across

SMEs size categories. We draw upon and advance this

study by (a) using firms'annual sales turnover, which

is a preferred/more appropriate proxy of firms'size

than number of employees; (b) providing distinct evi-

dence for two default definitions: bankruptcy (based

on Chapter 7/11 filings) and financial distress (based

on a firm's ability to honour its financial commitments,

and the value of its net worth); (c) exploring operating

cash flow marginal discriminatory power across size

categories; and (d) examining the presence of statisti-

cally significant differences in the magnitude of mutu-

ally significant covariates in the model for all SMEs,

and models for respective size categories, via the statis-

tical (Wald) test of equivalence of coefficients.

Based on our empirical analysis in the context of the

United States SMEs, using annual firm level financial

information obtained from the Compustat database (from

1990 to 2014), we conclude that all SMEs are not the

same. More specifically, the determinants of bankruptcy

vary across different size categories of SMEs. Earnings

are only found to be important for the largest size cate-

gory, as is also the case for the ratio of assets to liabilities.

Financial expenses are almost always found to be signifi-

cant, but the size of its effect varies, especially in

reference to micro firms. We also present compelling evi-

dence that estimated coefficients differ between models

for financial distress estimated across SMEs as a whole,

and the varying size categories. Forecasters would there-

fore be advised that distinct models for bankruptcy or

financial distress should be specified not in reference to

SMEs as a whole, but rather in consideration of the

different size classifications. In contrast to the work by

Gentry, Newbold, and Whitford (1987) and Gilbert,

Menon, and Schwartz (1990), we do not find that cash

flow contributes to an understanding of bankruptcy.

However, the results do compliment the findings of

Charitou, Neophytou, and Charalambous (2004) for the

UK in explaining financial distress using cash flow from

operations (CFO).

There are a number of differences in the estimated

determinants of financial distress as opposed to bank-

ruptcy. Firms with greater holdings of cash and short‐

term investments are less likely to face financial distress.

Taxes are consistently found to have an effect on financial

distress, but this is not the case for bankruptcy, where

only the model across SMEs as a whole provides evidence

of a significant effect. There is also evidence that the

effect of the ratio of current assets to current liabilities

is different across different classes of firm when

predicting financial distress. The value of trade debt pre-

dicts financial distress, concordant with the findings of

Hudson (1986) and Beck, Demirgüç‐Kunt, Laeven, and

Maksimovic (2006). It is possible that the value of trade

credit is reduced as a firm appears more likely to file for

bankruptcy, explaining the different result. Bankruptcy

and financial distress are distinct events and separate

modelling of them shall lead to improved risk pricing. A

similar conclusion might be reached in reference to the

consideration of different size categories of firms.

The remainder of the paper is structured as follows:

Section 2 presents a literature review on bankruptcy pre-

diction and survival analysis, which is the foundation for

SMEs failure prediction analysis, and includes the poten-

tial effect of firm size and operating cash flow on SMEs

likelihood of entering financial distress and bankruptcy.

Section 3 outlines the empirical methods, including an

explanation of the dataset and covariates. Results and

discussion are reported in Section 4. Finally, Section 5

draws conclusions and policy implications.

2|LITERATURE REVIEW AND

HYPOTHESES DEVELOPMENT

This section reviews the past studies on bankruptcy pre-

diction and survival analysis and is the foundation for

our SMEs failure prediction analysis. The discussion also

includes the potential effect of firm size and operating

cash flow on SMEs likelihood of entering financial

distress and bankruptcy.

2.1 |Approaches to SMEs failure

prediction

The principal source of external funding for SMEs is debt

and, more specifically, bank lending. However, lenders

face problems in forecasting loan performance. This issue

has been exacerbated over the years due to the presence

of less favourable economic environments, particularly

after the financial crisis in 2008–2009. Such conditions

also lead to restricted and overpriced credit. Credit risk

incorrectly or inadequately measured can generate

detrimental effects for SMEs, banks, and the wider econ-

omy. Notwithstanding, the importance of understanding

and forecasting insolvency for SMEs, until the last

decade, research in this area has been scant compared

572 GUPTA ET AL.

with the study on larger firms. This debate, to which the

present paper aims to contribute, has mainly focused on

improving banks'estimation and treatment of credit risk

for SMEs.

There is an extensive literature, spanning more than

three decades, on business failure prediction (Balcaen &

Ooghe, 2006). This literature includes various credit risk

models, which are mainly derived from two approaches:

the Altman (1968) model, which uses accounting‐based

indicators, and the Merton (1974) model, based on

market information. Although the Merton (1974) model

has significant advantages, the unavailability of market

information in the case of unlisted companies deems it

inapplicable for the majority of SMEs (e.g., Pompe &

Bilderbeek, 2005). Prediction of bankruptcy using

accounting information began with the seminal work by

Beaver (1966), who employed financial (accounting) ratios

in a univariate model to predict failure. Shortly thereafter,

the seminal multivariate (Z‐score) model was developed

by Altman (1968). Altman's (1968) study concludes that

traditional ratio analysis is not a reliable approach and

should be replaced by multivariate discriminant analysis

(MDA), as a more sophisticated tool for predicting default

events. Following Altman (1968), a vast number of studies

has applied the MDA statistical method to predict firms'

default. More recently, Altman et al. (2017) analysed the

performance of the Z‐score model in 31 European and

three non‐European countries. The authors argue that

the Z‐score model performs well in most countries,

reaching a prediction accuracy of approximately 90%

(when associated with additional country specific vari-

ables, or 75% otherwise). However, Ohlson (1980) chal-

lenged Altman's (1968) Z‐score model and raised some

critical issues with the predictive efficiency of the MDA

technique. To mitigate/overcome technical issues of previ-

ous models, Ohlson (1980) proposed logistic regression

technique instead of MDA, and thereafter, it remains a

popular choice (e.g., Altman & Sabato, 2007; Gupta,

Wilson, Gregoriou, & Healy, 2014).

Most bankruptcy prediction models are based on

single period classification, with multiple period bank-

ruptcy data. Given the fact that firms change through

time, the bankruptcy probabilities produced by MDA or

logistic models might be biased and inaccurate. Zavgren

(1985) finds that in traditional default prediction models,

the coefficients'signs of the bankruptcy indicators may

change in the years prior to failure. Luoma and Laitinen

(1991) extend this claim by showing that not only the

coefficients'signs change before failure but also the

values of the coefficients. Evidence provided in these

studies seems to suggest that traditional cross‐sectional

default prediction models are not valid, as the underlying

failure process does not remain stable over time.

Conversely, survival analysis models have the ability to

address these changes and hence are more suitable to

modelling the dynamic process such as bankruptcy

prediction. However, Luoma and Laitinen (1991) con-

clude that the survival analysis approach slightly

underperforms compared with discriminant analysis and

logistic analysis in bankruptcy prediction. Laitinen and

Kankaanpaa (1999) implemented a comparative study to

test the performance of various bankruptcy prediction

models. Their analysis indicates that hazard models have

better predictive power for 2‐and 3‐year predictions,

whereas logistic analysis shows superior performance

for 1‐year prediction. However, they conclude that the

differences in models'predictive powers are not statisti-

cally significant. Nevertheless, more recent studies shed

light on the superior performance of the hazard models.

Shumway's (2001) study was one of the first to employ a

large sample of about 2,000 firms, spanning over 31 years.

He found that half of the accounting variables used in

previous models by Altman (1968) and Zmijewski (1984)

are not significant indicators of bankruptcy likelihood.

Moreover, the accuracy of the hazard model substantially

increased when using both market‐based and accounting‐

based indicators to predict business failures. Laitinen

(2005) also found that the classification accuracy of the

proportional hazard model in the years prior to the firms'

default is superior to other statistical models used by

credit institutions. Employing the complete database of

UK listed firms between 1979 and 2009, Bauer and

Agarwal (2014) tested the performance of two hazard

models (Campbell, Hilscher, & Szilagyi, 2008; Shumway,

2001) against the traditional accounting‐based Z‐score

model (Taffler, 1983) and Merton's contingent claims‐

based model (Bharath & Shumway, 2008). They report

clear evidence regarding the miscalibration of the Z‐score

model and contingent‐claim based model, whereas the

average default probabilities of hazard models are closer

to observed default rates. They also find that the Z‐score

model and contingent claim‐based approach clearly

underperform, whereas the receiving operator character-

istics (ROCs) analysis highlights no significant differences

between the two hazard models.

The use of qualitative information presented a further

development in modelling firms'credit risk (e.g.,

Lehmann, 2003). Analogously, Grunert, Norden, and

Weber (2005) and Tsai, Lee, and Sun (2009) report that

nonfinancial factors present a useful supplement to finan-

cial factors in credit rating. In the context of SMEs,

Altman, Sabato, and Wilson (2010) report improvement

in models'classification performance after accounting

for qualitative information of UK SMEs.

Until recently, less academic attention has been

devoted to SMEs in the failure prediction literature. This

GUPTA ET AL.573