Introduction

To begin this study, a traditional Malay quote: “Kecil-kecil cili api,” in English—Small is the fiery chilli—do not underestimate something just because it is small.

More than 99% of registered businesses in many countries around the world are SMEs, including developed countries such as Germany (Britzelmaier et al. 2013), developing economies like Indonesia (Kuncoro 2008) and newly industrialized countries like Malaysia (Mohd Aris 2007). SMEs are also engines of economic growth, employing a significant portion of the workforce and contributing to their nation’s GDP (Ardic et al. 2012). In Malaysia, as of 2014, SMEs contributed 35.9% of the Malaysian GDP (SME Corporation Malaysia 2015). Since the 2000s, countries around the world have begun to devote more attention to the development of their local SMEs, acknowledging the important role that they play in the national economy (Britzelmaier et al. 2013). SMEs are also big employers in their respective countries. In Malaysia, 97.3% of enterprises are categorized as SMEs and employ up to 65% of the Malaysian working population (SME Corporation Malaysia 2013, 2015). In recognition of the importance of SMEs, governments around the world have instituted measures to assist these businesses through grants, government-backed venture capital, debt relief and credit guarantees (Britzelmaier et al. 2013; Samujh et al. 2012).

However, SMEs are considered high-risk borrowers by lending institutions due to their size, volatility and opacity (Berger & Frame 2007; Meisenzahl 2014). Research has shown that due to the high degree of information asymmetry associated with SMEs, loans given to SMEs are charged a high level of interest and are often guaranteed by the personal assets of the business owner (Kirschenmann & Norden 2012). Click or tap here to enter text. We find banks subject SMEs to high interest rates and restrictive loan covenants based on their size because they do not have a reliable means of estimating the required returns on SMEs’ risk. SME owners have a low perception of financial risk, which is concerning as Avery & Bostic, (1998) show that SME loans are guaranteed by their business owner’s personal assets, leaving the business owner in a situation of unlimited liability in the event the SME is not able to repay the loan. SME entrepreneurs themselves do not seem to consider financial risk to be a key issue for them (Brenner et al. 2010) SMEs can reduce the amount of opacity by pricing and disclosing the risk on the required returns associated with their financing, allowing them to obtain financing from financial institutions at more favorable terms (Kirschenmann and Norden 2012).

A question arose here—how to measure or calculate SME’s financial risk? Calculation of financial risk, which can be translated into the investor’s required return, is still under-utilized among finance providers, meaning that operators of SMEs are depriving themselves of a useful tool for monitoring and predicting their business performance in the future (Palliam 2005). Firm characteristics such as capital structure (Baxamusa and Jalal 2014), ethnicity (Bates and Robb 2013), maturity of relationship with banks (Kirschenmann and Norden 2012) and firm size (Glennon and Nigro 2005) have been shown to affect SMEs’ loan maturity, access to capital and interest rates payable.

The majority of the literature regarding SME risk involves the issue of survivability and how to predict SME survivability (Abdullah et al. 2015). Efforts made to quantify SME risk have been limited to case studies (Britzelmaier et al. 2013; Palliam 2005) and theoretical papers (Moro & Nolte 2012). SME development is important to the economy of both developed and developing nations and this study provides the information needed to better understand how financial risk can be valued in SMEs and how business characteristics affect that risk.

To fill the gap, this study contributes to the literature by answering the following two research questions.

1). What is a suitable measure of SME risk?

2). What is the relationship between SME risk and its characteristics?

This study answers question 1 by demonstrating the use of the pure-play beta (PP beta), accounting beta and the probability of survival models (Model 1 and 2) as methods of measuring financial risk for SMEs.

In answering question 2, the authors examine the relationship between industry, firm size, location, business ethnicity, and business gender and the risk faced by small and medium-sized enterprises (SMEs). The results indicate that the type of industry the SME belongs to does not affect its overall risk, as measured by the PP beta, Accounting beta and probability of survival models 1 and 2. Similarly, firm size and location do not appear to have a significant impact on the risk faced by the business. However, the results do show that there is a significant relationship between the gender categories and financial risk. The results also indicate that there is no significant relationship between business ethnicity and SME risk. Overall, the findings of this research suggest that the traditional factors that are often used to assess the risk of SMEs, such as industry, firm size, location, and business ethnicity, may not be as important as previously thought.

We use Malaysia as the region of study due to its status as a developing nation and for access to financial datasets on SMEs in the region. Malaysian SMEs represent a generalisable subset of SMEs in developing and developed nations that have a preponderance of family-owned businesses and have policies grounded in affirmative action. As of 2014, SMEs contributed 35.9% of Malaysia’s GDP and employed 65% of the total employed workforce (SME Corporation Malaysia 2015). SMEs in the service sector contributed 58.6% of the SME GDP contribution in 2014, followed by SMEs in the manufacturing sector at 21.7% (SME Corporation Malaysia 2015).

Studies on risk in Malaysian firms look at it from an attitudinal perspective (Salleh and Ibrahim 2011). In their research, Salleh & Ibrahim (2011) looked at the risk attitudes prevalent among SME owners of differing demographic characteristics. It is found that characteristics such as gender, age and education did not affect any difference towards their risk-taking propensity as measured by the Risk Attitudes Inventory (Calvert 1993).

There remains a gap in research looking at financial risk as it relates to Malaysian SMEs. Risk being the volatility returns from the firm to its owners. This type of risk is closely related to business failures and is of interest to policy makers. In this area, we find that Malaysian businesses have factors which play a part in their financial risk.

First, about 60–70% of SMEs in Malaysia are family owned. There is a high level of concentration of ownership in the hands of one party among Malaysian SMEs. The largest shareholders are often members of the same family, with the head of the family often holding the highest level of shares, which can create principal–principal agency conflict between majority and minority shareholders (Rachagan and Satkunasingam 2009).

Second, prior to the independence of Malaysia in 1957, the British administration practiced a policy that hampered the Bumiputra (sons of the soil), the indigenous people of Malaysia, from owning and developing land, limiting them only to paddy fields and aquaculture. While the Bumiputra (the majority of whom are Malay) could own land and pass it down from generation to generation, this land mostly remained undeveloped in the early stages of independence. This created an environment post-independence where the Bumiputra are left behind economically compared to the other ethnicities, most notably the Chinese (Whah 2010). On May 13, 1969, a string of racial riots across Kuala Lumpur caused the Malaysian government to rethink their economic policies which led to the formation of the New Economic Policy (NEP) in the 1970s. Through the NEP, the People’s Trust Council, known by its Malaysian acronym, MARA, is established to provide financial and social support to the Bumiputra. Hence, it can be argued that race/ethnicity of the business owner would have an effect on the financial risk of the SME in Malaysia.

The remainder of this paper is organized as follows. Sect. "2). What is the relationship between SME risk and its characteristics?" presents the literature review. Section 3 describes the data, sample selection and regression framework. Section 4 reports the empirical results and analysis. Section 5 concludes.

Literature review

Information on SME survivability is usually of great interest to financial institutions that provide loans to SMEs. This is especially prevalent with the development of credit rating scores, which are widely used by lenders to determine the creditworthiness of an SME (Berger and Frame 2007; Dias Duarte et al. 2017). To banks, the risk associated with investing in SMEs is the potential that the SME may not be able to meet their financial obligations in terms of interest payments. Conversely, the risk of an SME being unable to service its financial obligations can result in bankruptcy proceedings instituted against the SME, cutting short its operational lifespan. To this end, several researchers have also considered the risk of bankruptcy for SMEs in making financial decisions for SMEs (Dichev 1998). Sometimes the strict credit scores used by the financial industry push SME owners into unconventional streams of finance, such as borrowing from friends and family, bootstrapping and even loan sharks, creating additional risks to the SME owner (Prina 2015).

Researchers have been working on different methods to bridge the gap between these different risk perspectives as SMEs continue to play prominent roles in their local economies. SME risk research is often concerned with how these different perspectives can be used together in developing a more holistic overview of SME risk. Research into financial risk in the literature is still dominated by qualitative measurements of risk, based on how risk is perceived by entrepreneurs, judging it by the intensity of financial risk that is experienced by them (Belás et al. 2018). Our study digs further on the relationship between financial risk and SMEs by demonstrating a significant result.

Owners and investors alike are concerned with the risk that the SME will incur a loss as a result of poor financial or operational decision making. This risk of loss creates the motivation to develop methods to mitigate the risk via the use of hedging instruments, insurance policies and investment in internal controls (Hammoudeh & McAleer 2013). There are costs associated with such activities and this can be termed the ‘price’ the SME has to pay to manage the risk. Risk is an abstract concept, which deals with the uncertainty that surrounds an entity. From a financial point of view, risk is the volatility of return. More importantly, it is usually associated with the probability of making a loss on investments made in a business or a project (Jensen 2005; Lintner 1965). This is known as the ‘downside risk’ as financial risk also includes the probability of making a profit on the investment (the upside risk) (Hammoudeh and McAleer 2013). There is a risk inherent in every investment and investors will seek a return commensurate with the level of risk they are facing. The general rule is the higher the risk associated with the investment, the higher the required return (Kirschenmann and Norden 2012).

In our study, we are able to investigate further on measuring SMEs risk using Malaysia’s context setting. We find that, in Malaysia, the level of debt an SME has does not necessarily indicate a higher level of risk compared to its peers. SME risk is significantly impacted by profitability. In addition, SME risk is not significantly associated with factors such as size, firm age, and business ethnicity. Our results provide new and consistent evidence to support the theoretical links.

Risk measures used in the research

In adapting market-based models for SMEs, the first step is to estimate the beta of the SME. For listed companies, beta is treated as the financial elasticity, or the correlated relative volatility of the firm’s asset returns to market returns and is usually represented as

$$\beta = \frac{Covariace({r}_{a},{r}_{b})}{Variance\left({r}_{b}\right)}$$
(1)

where \({r}_{a}\) is the return on the stock price of the firm and \({r}_{b}\) is the market return.

However, in calculating the beta for unlisted firms, deriving the stock price of the firm is impossible, as in the absence of publicly traded stocks, there is no way to estimate the return derived from the trading of unlisted stocks. Because of this, two methods stand out in the literature as reliable means of estimating the beta for SMEs. They are the pure-play beta method (Fuller & Kerr 1981) and the accounting beta method (Hill and Stone 1980).

The accounting beta. The accounting method first emerged as a way to benchmark the performance of the market beta against accounting information and is calculated as the accounting return of the firm being evaluated (Ball & Brown, 1969). Hill and Stone (1980) suggest using purely accounting data to derive a non-market beta with the following formulas for accounting-based betas: the operating beta and the income beta (Ball & Brown, 1969; Hill & Stone 1980):

$$Operating\, \beta =\frac{COV ({ROA}_{i, }{ROA}_{m})}{VAR ({ROA}_{m})}$$
(2)
$$Income\, \beta =\frac{COV \left({IR}_{i, }{IR}_{m}\right)}{VAR \left({IR}_{m}\right)}$$
(3)

where

$$ROA\, (Return\, on\, Assets)= \frac{Earnings\, before\, interest\, and\, tax}{Total\, Assets}$$
$$IR\, (Income\, Received)= \frac{Net\, Income+Non-recurring\, adjustments\, to\, net\, income}{Total\, Assets}$$

i = company/security m = market

(Vos 1992) is one of the first researchers to look at the use of the accounting beta to estimate the risk faced by unlisted businesses. Using the accounting beta calculation proposed by Hill and Stone and Ball and Brown, Vos implements the approach using the following formula (Ball & Brown, 1969; Hill & Stone 1980; Vos 1992):

$$Accounting\, \beta =\frac{{\Delta ROE}_{i}}{{\Delta ROE}_{m}}$$
(4)

where

$$ROE (Return\, on\, Equity)= \frac{Earnings\, before\, interest\, and\, tax}{Total\, Equity}$$

i = company/security m = market

This allows the accounting beta to be calculated on a year-on-year basis using only two consecutive years as a starting point.

There are limitations, as using accounting betas to determine the cost of capital for small and unlisted businesses may be inaccurate(St-Pierre and Bahri 2006) The accounting betas, however, seem to be a more reliable estimate for listed companies and companies going for IPOs (Almisher and Kish 2000). The literature suggests that the accounting beta can be used to support the derivation of the market beta as opposed to being a standalone measure by itself.

The pure-play beta. The term ‘pure-play’ is used in portfolio management to describe a situation where only specific, targeted industries are used to build a portfolio, hence the portfolio risk only reflects the idiosyncratic risk faced by that target industry and is not influenced by risk from other industries. Historically, the PP beta is developed as a means of estimating the cost of equity of individual divisions within multidivisional companies (Fuller and Kerr 1981). As individual divisions are not listed on the stock market, the pure-play beta has been used as a means of calculating risk for these respective divisions. Fuller and Kerr (1981) propose that the beta of a multidivisional firm is the sum of the weighted average beta of each of its component divisions (Fuller and Kerr 1981):

$${\beta }_{j}= {\sum }_{i}{W}_{ij}{\beta }_{ij}$$
(5)

where \({\beta }_{j}\) is the beta of the multidivisional firm, \({W}_{ij}\) is the weight assigned to each division ‘i’ calculated as the divisional sales divided by the total sales of the multidivisional firm, and \({\beta }_{ij}\) is the beta of each division.

In order to derive \({\beta }_{ij}\), listed companies of roughly the same size as the division and within the same industry are used as market proxies. An unlevered beta is first calculated from the proxy’s capital structure in order to remove the ‘capital structure’ portion of the company’s beta (Fuller and Kerr 1981; Hamada 1969).

The unlevered beta is then ‘relevered’ with the capital structure of the multidivisional firm to calculate the adjusted beta for the respective division. Pure-play beta has its limitations in that investors using the beta are expected to have well diversified portfolio of investments. This may not always be the case for small business owners.

Probability of survival. Using the probability of survival to price risk for SMEs. Where SME survivability is said to be the main concern of SME owners, predicting business survivability can be a useful measure of SME risk, although arguably, the risk being measured is more of an operational risk. Probability models for estimating return to the investor have been discussed in the literature, where the key input is the probability of the business’s survival instead of beta (Moro and Nolte 2012). The model proposed by Cheung is as follows:\({k}_{e}=\frac{{R}_{f}}{p}\) + \(\frac{1-p}{p}\) (6)where \({k}_{e}\) is the cost of equity; \({R}_{f}\) is the risk-free rate of return; p is the probability of business survival.

In the literature, some authors have discussed means of predicting business survivability using logit regressions and financial ratios (Altman and Sabato 2007). One of the most popular models to emerge from this school of thought is the Z-score model. In their paper, Altman and Sabato (2007) develop a model for predicting the survivability of SMEs using a sample of US companies. This model is later adapted by Abdullah et al. (2015) in their research on predicting the survivability of Malaysian SMEs. The output of Abdullah et al.’s model is a percentage of the survivability of the business, as given by the following formula:

\(Log\left(\frac{PD}{1-PD}\right)= a+b\left(total\, liabilites\, to\, total\, assets\right)-c (EBITDA\, to\, total\, assets)\) PS1.and a second model, which includes firm age as a predictor:

\(Log\left(\frac{PD}{1-PD}\right)= a+b\left(total\, liabilites\, to\, total assets\right)-c\left(EBITDA\, to\, total\, assets\right)-d(Ln Firm Age)\) PS2.where PD is the probability of default.

It should be noted that the model does not specifically mention the best way to predict a business’s survivability, instead offering an example of an industry-wide average survival rate.

Business characteristics which affect business risk

Financial performance

A business’s financial performance is important for serving the interests of its stakeholders, such as investors and financiers. Various ratios, such as EBIT, sales/asset, efficiency, liquidity, and gearing ratios, are used to measure a company’s performance (Almisher and Kish 2000). The financial performance of a business is also closely tied to its ability to remain a going concern and is a key factor in determining the risk associated with investing in the business (St-Pierre and Bahri 2006; Vos 1992). Research has shown that businesses with lower liquidity and EBIT ratios have a higher likelihood of failure, and that financial performance can affect the pricing of the risk associated with returns on investments in small and medium-sized enterprises (SMEs). Research in the literature have also discussed how debt and access to debt affects the recovery and survivability of SMEs, which suggests a connection between financial performance and SME risk (Didier et al. 2021; Lawless et al. 2015; Neville & Lucey 2022).

Experience of the business owners

Education and experience can have a positive effect on the success and survivability of a business. Research has shown that higher levels of education and more years of working experience can lead to better business outcomes (Bates 1990; Moro and Nolte 2012; Robinson and Sexton 1994). Experience is also an important factor in reducing the risk of business failure. The literature suggests that experience can be seen as a form of informal education and is crucial for business survival and success (Corbett 2005; Knotts 2011; Kolb and Kolb 2005). The education and experience level of the small and medium-sized enterprise (SME) owner may be linked to the pricing of risk associated with required returns on investments in SMEs as it can affect the business’s survival and success.

Concentration of ownership

Concentration of ownership, measured by largest percentage of shareholding held by the largest shareholder, can affect the presence of agency cost in a business, which can negatively impact SME risk (Banchit & Locke 2011; Hewa Wellalage & Locke 2014; Jensen & Meckling, 1976). However, using the largest share percentage alone is an inadequate measure of the concentration of ownership. Typically, information regarding the staffing of managerial positions in the business, whether they are family or non-family, can also help determine the level of managerial influence concentrated in the hands of the owner (Bai., et.al., 2021). In the absence of such information, concentration of ownership can be seen as a good yardstick for risk faced by a small business.

Gender

In developing countries, a large proportion of SME owners are female, and they run their own cottage industries to supplement their families’ incomes (Jothilakshmi et al. 2009; Kyaw and Routray 2006; Ondoro and Omena 2012). However, studies have shown that gender affects the kind of industries that entrepreneurs choose to get involved in. Male entrepreneurs preferred capital-intensive retailing and manufacturing work while female entrepreneurs are more service-oriented (Dunn and Shore 2009; Verheul and Thurik 2001). This difference in choice of industry may be defined by cultural and gender bias and it may affect women’s access to and associated costs of capital. Gender and gender-based discrimination relates to the access to and pricing of finance (Wellalage and Locke 2017). Female entrepreneurs are at a disadvantage in terms of networking, choice of industry and level of commitment, which in turn affects their business’s risk, and therefore, the type of capital they can gain access to will affect the pricing of the risk associated with the required returns on investments made in female-owned SMEs (Verheul and Thurik 2001).

Ethnicity

Cavaluzzo and Cavalluzzo, (1998) found that SME owners from ethnic minorities are unfairly discriminated against by banks in terms of disbursement of loans, particularly in areas with low bank competition. Bates and Robb, (2013) also found that SME owners from ethnic minorities pay higher interest rates. In Malaysia, government-linked financial institutions such as MARA only provide loans to members of the majority ethnic Bumiputra group. Discrimination based on ethnicity can affect access to capital, which in turn affects the ability of a business to operate and survive, further increasing its risk profile and the pricing of risk associated with the required returns of investing in that SME.

Firm size and age

Firm size is an important indicator of firm risk, and it can affect the likelihood of an SME defaulting on its loan guarantees and the volume and maturity of loans given to businesses by banks and financial institutions (Glennon and Nigro 2005; Kirschenmann and Norden 2012). Firm size also affects the duration and cost of finance a business has access to and forms an important factor used in the multifactor models used to estimate the pricing of portfolio risk (el Kalak and Hudson 2016; Fama and French 1996; Hymer and Pashigian 1962; Yasuda 2005).

Firm age is defined as the number of years a business has been in existence. Many researchers have found that growth prospects tend to slow down as a firm age and that younger businesses are more susceptible to business failure than older businesses (Bernardt and Muller 2000; Evans 1987; Lundvall and Battese 2000). Younger firms tend to have greater difficulty in accessing capital due to their unproven track record and lack of transparency in their financial statements (Neely and van Auken 2012; Thornhill and Amit 2003). There is a link between firm age and firm failure, which in turn affects the level of risk associated with the business, and the literature has shown that there is a relationship between firm age and the pricing of the risk associated with the required return on the SME.

Firm industry

The industry in which a business operates can affect its access to financing and its level of risk, which in turn can affect the pricing of the risk associated with the required returns on investments made in that business (Degryse & van Cayseele 2000; Fong 1990; Jahan-Parvar et al. 2013). Literature has shown that banks tend to view service-oriented businesses less favorably, indicating some barriers to capital for entrepreneurs in the service industry (Verheul and Thurik 2001). Research has also found that different industries face different levels of risk and have different betas, which can affect the cost of capital or hurdle rates relevant to their investment decisions (Cox and Griepentrog 1988; Fields & Kwansa 1993).

Geographic location

Geographic location can affect the profitability of a business due to factors such as proximity to finance providers, suppliers, and customers (Dunn and Shore 2009; Sikligar 2008). Research has found that businesses closer to financial centers have lower costs of capital (el Ghoul et al. 2013; Harvey 2004). Special Economic Zones (SEZs) have been set up in several countries to stimulate economic growth in those areas, but their success varies (Aggrawal 2006; Farole, 2010; Saleh and Ndubisi 2006). Depending on their location, businesses experience different levels of economic growth and access to finance, affecting the risk associated with investing in SMEs (Singh and Nejadmalayeri 2004). The literature supports a link between geographic location and SME risk but has not analyzed the effect on an SME's beta. Our results show evidence to support the link as well, they are (1) the level of debt an SME has does not necessarily indicate a higher level of risk compared to its peers; (2) profitability significantly affects SME risk; and (3) there is no significant association between SMEs’ risk and factors such as their size, age, and business ethnicity in Malaysia. Our research provides strong implications to investors, customers, suppliers, employees, and the SME society.

Data and regression framework

Data and methodology

To test the above-mentioned measures and models, we download the data from the Malaysian Companies Commission, known by their Bahasa Malaysia acronym as ‘SSM’ (Suruhanjaya Syarikat Malaysia). The database consists of the annual company reports, which all registered businesses in Malaysia have to file, regardless of their ownership structure and size. As a result, relevant financial and non-financial information, even from smaller businesses, is available through this database.

As of 2021, there are a total of 7,901,167 businesses registered with the Companies Commission as per the Companies Commission website (https://www.ssm.com.my/Pages/Home.aspx). Of this total, about 97.2% are classified as SMEs.

For the purposes of this research, a request is made for company data, specifically asking for observations with the following characteristics:

Sales of less than RM50,000,000 a year (to ensure that the business falls under the SME definition as set by the Malaysian Government) has the following information available:

Capital structure

Profitability

Shareholder’s equity

Average age of owners

Ethnicity of business owners

Gender of business owners

Location of the business premises

Age of the business/firm

Sub-industry to which the business belongs

From the criteria mentioned above, an unbalanced raw panel dataset is obtained, containing 400 companies and observations covering the years 2005 to 2014. Based on our estimates, this represents roughly 0.01% of the SMEs registered in Malaysia. In arriving at the final sample size, several data cleaning and editing procedures had to be undertaken. These steps included:

Removal of ‘Dormant’ companies—some companies in the dataset are marked as ‘dormant’: this implies that the business is no longer in operation and thus observations containing the word ‘dormant’ are removed.

Classification of industries—companies are sorted into their respective industrial categories, which initially consisted of manufacturing, services, primary agriculture, construction and mining.

Removal of technically insolvent companies—while the companies in the dataset are formally still in operation, several companies are found to have had negative equity for consecutive years of operation (brought about by consecutive losses). Years where companies had negative equity are removed from the sample, despite such businesses formally still being going concerns. This research treated them as technically insolvent, because keeping them in the sample would skew the results of the regression model and create unnecessary bias in the results.

Where there are no values for a given variable in an observation, that observation is removed. This applied to observations that had missing values in fields such as ‘Total Sales’, and ‘Total Assets’, because having a null figure in any of these fields does not make logical sense. In addition, observations which appeared to be obviously incorrect, such as where the profit after tax figure is higher than the total sales figure, or there are negative total sales figures, are also removed. It is likely that such errors are caused by human errors during data input.

In the dataset, some observations had similar company IDs and years of reporting, because the company in question had published two financial reports in that time period. This can be attributed to a change in financial reporting periods or a need to fulfill certain legal reporting requirements. To prepare the dataset for panel data analysis, these repeated time variables had to be removed. In most instances, the period ending 31 Dec of the observed year is selected as the observed variable and the other observation is deleted.

After cleaning, editing and removing observations that are not relevant to the research, a final sample consisting of 303 companies over the years 2005 to 2014 resulting in 1541 observations is derived (Tables 1, 2, 3, and 4).

Table 1 Summary statistics of the main variables
Full size table
Table 2 Summary statistics of other variables
Full size table
Table 3 Summary statistics of owner and firm age variables
Full size table
Table 4 Summary statistics of risk measures
Full size table

It is interesting to note the existence of negative betas calculated under PP beta and accounting beta. Our interpretation is that SMEs with negative betas have a return volatility profile which runs counter to that of the market. The mean beta for PP beta and accounting beta are 5.22 and 0.82, respectively, denoting that on average, SMEs in Malaysia have a return profile that matches the market return. However, the discerning SME investor will find opportunity to hedge and diversify their investment via these negative beta entities.

Besides, the independent variables selected for this study are the SME characteristics that have been identified in the literature as having the potential to affect SME risk. Using the GMM regression, this research aimed to determine to what extent a relationship between SME risk and these characteristics exists. By running GMM regressions on each of the three dependent variables with the same set of independent variables, this research compared, analyzed and contrasted the different measurements of risk and their relationship with SME characteristics. This enables a better understanding of what affects the different measurements of risk and creates avenues for debate regarding the supposed impact that these identified characteristics have on SME risk. These independent variables are described in Table 5:

Table 5 Predictor variables description
Full size table

Control variables

In this research, a selection of control variables is regressed against the risk measures to determine whether there is any influence from factors not identified in the independent variables. The control variables used in this research are:

Total assets

Total equity

Total debt

These control variables are chosen because, taken together, they reflect the overall value of a company, and it is important to consider whether there is any additional impact from these figures on the risk measures.

Selection of the regression model

In the process of selecting an appropriate regression model, this research first considered the use of the Ordinal Least Squares (OLS) regression, adapted for use with panel data as it is a commonly used and generalisable regression model. However, given that the dependent variables measured in this research are all non-parametric in distribution, the OLS regression cannot be used.

Quantile regression is then evaluated as a potential better selection to match the dataset in this research. Unfortunately, the dataset faces a high threat of endogeneity, especially in the form of loop causality, as the dataset contains financial figures which affect one another. Because of this endogeneity, quantile regression is not a suitable method for this research.

The next alternative considered is quantile instrumental variable regression. This regression model combines the semi-parametric nature of the quantile regression together with instrumental variables to mitigate the threat of endogeneity. However, the difficulty with this regression method is that the identification of instrumental variables is extremely challenging. Given that the dataset in this research in limited in terms of variables, the quantile instrumental variable regression is unsuitable.

The generalized method of moments dynamic data panel regression is eventually chosen as the regression model for this research as it is a semi-parametric regression which utilizes the lag of the variables tested in the model as instrumental variables. This satisfies the non-parametric distribution of the data as well as mitigates the threat of endogeneity. This regression model is further explained in the following sections.

Results and analysis

Summary statistics

In order to answer the research questions, we now compare three of the methods used to measure SME risk in the literature, which are the pure-play beta, the accounting beta and the probability of survival method. In the case of the probability of survival method, the model used is the modified Z-score model employed by Abdullah et.al (2014): in their research, two models are used, and this research used both models in its analysis.

A correlation analysis is carried out on these three variables to determine whether they are highly correlated with each other or if they are sufficiently different to constitute independent measures of risk. A comparison of the averages of each variable across the five categories of Sub-Industry, Business Ethnicity, Size Category, Location and Business Gender is carried out to better understand the distribution of risk across these different categories. To investigate the relationship between SME risk and its characteristics, the dynamic panel system GMM estimator is chosen as the regression model. This model allows for panel data to be regressed against a single dependent variable and uses lag observations of independent variables as instrumental variables in mitigating the threat of endogeneity in the regression model.

Table 6 shows the correlation between the risk measures (PP beta, accounting beta, PS Model 1, PS Model 2) and the characteristics analyzed in this research. The table shows that PP beta is, at the 0.05 level, significantly correlated with all the variables tested, including the other risk measures as well. This implies that PP beta has a relatively reliable degree of predictability in determining the values of the other variables. It can also be argued that from these results, PP beta seems to be the most reliable form of risk measure for SMEs in this dataset. However, since variables such as the debt-to-equity ratio, total equity, total debt and total assets are key input variables in the calculation of PP beta, it is expected that these variables would be strongly correlated with it.

Table 6 Correlation matrix
Full size table

The accounting beta is only significantly correlated only with the debt-to-equity ratio and the total equity ratio, which is expected given that the accounting beta is calculated based on the year-on-year change in return on equity. However, somewhat interestingly, the accounting beta is not significantly correlated with return on equity in this case. The accounting beta is not significantly correlated with any of the other variables (except PP beta) and does not seem to be a good supplementary indicator of risk that can be used alongside other measures of risk at this point.

The probability of survival (PS) model 1 predicts the survivability of an SME based on the input figures of the debt to asset ratio and the return on asset ratio. In terms of correlation, it is highly correlated with all the other variables, except for accounting beta, owner age and the concentration of ownership. This low correlation with accounting beta is likely due to the PS model being more focused on the interaction between assets, debt and return, whereas the accounting beta is built upon the change in the return on equity. The return on asset figure is highly correlated with the PS model due to it being an input figure in its calculation. The correlation matrix implies that the probability of business survival, treated as a potential risk to the SME, has a certain degree of correlation with the SME characteristics which affect SME risk.

The PS model 2 has essentially the same results as the PS model 1, with the main difference being the addition of firm age as an input figure in its calculation. Abdullah (2014) finds that the PS Model 2 has a greater degree of accuracy in predicting the survival rate of a SME compared to Model 1. In the correlation matrix given, the PS Model 2 is highly correlated with all the variables, save for accounting beta and ROE. The reasons for its lack of correlation with accounting beta are similar to that for the PS Model 1. Interestingly, PS Model 1 has a significant correlation with ROE, but PS Model 2 does not. This implies that the addition of the firm age variable skews the prediction of the PS Model 2 to favor the age of the business and its owner (Firm and Owner age) over the returns generated by the business. Nonetheless, the results of the correlation matrix suggest that PS Model 2 is a better measure of SME risk than PS Model 1.

The initial results given by the correlation matrix suggest that PP beta is the most accurate measure of SME risk for this dataset. Given its significant correlation with all the variables across the board, it seems that for this selection of financial and numerical SME characteristics, PP beta is the best measure of risk for the purposes of calculating the required return on an SME or determining its cost of capital. However, given the shortcomings of the correlation matrix, categorical variables cannot be measured in this analysis, hence this research uses the dynamic data panel GMM regression to determine the relationship between the risk measures and SME characteristics.

GMM regression framework

GMM regression is a type of regression analysis that helps to mitigate issues of endogeneity in the dataset. In conjunction with the use of a panel dataset, the GMM regression can use lag instrument variables to deal with endogeneity, instead of relying on the identification of external instrument variables, as is the case with instrumental variable regressions. Following is our model specification:

$${y}_{it}={\alpha }_{1}+{\kappa }_{1}{y}_{it-1}+\beta {X}_{it-1}+ \gamma {Z}_{it-1}+{\eta }_{i}+{\varepsilon }_{it} GMM\, Model$$

where ‘y’ represents the dependent variables being tested, the lag of which is also tested in the model with a lag function of ‘t-1’ (Lag 1).

‘X’ represents the independent variables in the equation, and each risk measure is regressed against a slightly differing mix of independent variables to avoid regressing them against components of their calculation. The independent variables are also tested in the model with a lag function of ‘t-1’. This is done to prevent an overidentification of instruments (Wintoki et al. 2012). ‘Z’ represents the control variables put in the regressions and these include the total debt, equity and asset figures, respectively, although given that some of them are components of the various risk measures used, they are not used in all of the four models.

Robustness testing for collinearity, exogeneity and overidentification

The Arellano-Bond AR(1) and AR(2) are tests for first-order and second-order serial correlation in the first-differenced residuals, with the null hypothesis of no serial correlation: a rejection of this at the 0.05 level means that there is serial correlation in the model. The Sargan and Hansen test of overidentification is conducted with the null hypothesis that all instruments are valid: a rejection of this at the 0.05 level means that the instruments are over-identified and may be invalid. The Hansen test is only available when using the two-step and is preferred to the Sargan test when interpreting the results (Blundell and Bond 1998). The Diff-in-Hansen test of exogeneity is conducted under the null hypothesis that instruments used for the equations in levels are exogenous: a rejection of this at the 0.05 level means that the instruments are not exogenous. Hansen values of 1.000 are highly suspect, even if they indicate an acceptance of the null hypothesis and can indicate a lack of completeness in data or other issues with the model (Blundell and Bond 1998; Wintoki et al. 2012).

The instruments used in the GMM estimations are D_it and E_it, which represent Total Debt and Total Equity, respectively. These variables are taken from the dataset as well but not tested for in the regression. These variables are selected as instrumental variables because they are used in calculating the PP beta variable.

The two-step method of calculation is used together with a Windsor robustness correction measure which is more accurate and less biased than the first-step method (Blundell and Bond 1998; Wintoki et al. 2012). The lag instruments used in the equation are ‘collapsed’ instead of generated for each time period in order to prevent over-confidence (Roodman, 2006). Table 7 shows the results of the above-mentioned collinearity, exogeneity and overidentification tests.

Table 7 Tests of collinearity/exogeneity/overidentification
Full size table

The model uses the System GMM method (Arellano and Bover 1995; Blundell and Bond 1998) which is more asymptotically efficient than the Difference GMM as it makes use of forward orthogonal deviations instead of first differences in estimating the identity matrix used in the equation. Table 8 shows the results of the GMM panel regression framework.

Table 8 Regression table
Full size table

Industry—As shown in Table 8, the PP beta is based on using a proxy beta from a business from a similar industry listed on the market; therefore, any sectoral differences should be captured through the PP beta measurement. However, the regression results for PP beta show that there are no significant differences between the industries, implying that the type of industry the SME belongs to does not affect its overall risk. Looking at the accounting beta and both PS Models, which represent a performance and survivability point of view, respectively, this seems to be the case as well. The magnitude of the coefficient between PP beta, PS Models 1, 2 and industry is also small. Interestingly there is a very large coefficient value of 187.7 between the accounting beta and industry, indicating an impact on the risk measured by accounting beta. However, the p-value for the accounting beta regression indicates no significant relationship between the two.

In their work regarding the survivability of SMEs, Abdullah et al. (2014) specifically mention that their work in the development of the PS Models is limited to manufacturing companies only. However, the present results seem to indicate that it does not matter what type of industry the SME is in as this has no significant relationship with the level of risk it is facing.

Firm size—The results indicate that neither firm size nor its L1 significantly impacts the risk faced by the business across all the risk measurement models that we have tested. Unfortunately, due to the opacity of smaller businesses, they are still ranked as being more risky than larger businesses; however, these results shed some light on the operations of small businesses, showing that they are not significantly more risky than larger businesses, and supporting the findings of Berger and Frame (2007) in developing a credit scoring method which is fairer towards small businesses.

Location—We find there is no significant relationship between geographic location and all four of the risk measures, indicating that it is possible that the reported business address may not be the actual physical location of the business. Furthermore, the majority of the observations analyzed in this dataset are service businesses (1,377), which can include home-based offices that can theoretically operate from anywhere using internet connectivity.

Business ethnicity—The only risk measure that is significant is the accounting beta measure, at the 10% level. For the other risk measures, the relationship is not significant. However, PP beta and both the PS Models do not suggest any significant relationship between being a purely Chinese-owned business and SME risk. This has several implications. First, it challenges the notion that Malaysian Chinese have an easier time setting up/running a SME as compared to different Malaysian ethnicities. Second, despite all the assistance given by the Malaysian government, there is no evidence from this research that suggests this assistance has made it easier for Malays to set up and run a business, otherwise there would have been a significant relationship between business owner ethnicity and SME risk. While it is true that in migrant communities especially, members of the same ethnicity pool their finances together in order to start businesses and make investments (Brenner et al. 2010) this does not give them a significant advantage/disadvantage as compared to other races. Rasheed, (2004) finds that in the USA, ethnicity affects access to capital, with white entrepreneurs getting more favorable loans than black entrepreneurs. However, while this suggests that access to finance might be easier for certain ethnicities, it does not mean that a business owned by any ethnicity is any more or less risky than a business owned by a different ethnicity.

Business gender—It is surprising to see that there is a significant relationship between the gender categories for PP beta and PS Model 1. The relationship as analyzed against accounting beta and Ps model 2 is not significant at the 5% level. In terms of financial risk, the gender of the business owners, regardless of if it is predominantly male or female, should not affect the risk faced by the SME. While previous studies have pointed out that the gender of the business owner can affect things like access to loans and obtaining favorable interest rates (Rasheed 2004), a study by Verheul and Thurik (2001), conducted in the Netherlands, points out there is no significant evidence that women face discrimination when it comes to getting a loan. This point of view is also shared by Cavalluzzo et al., (2002) who find that their sample of female business owners in the USA did not face discrimination when it came to access to capital. These findings run counter to the results from the regression table, which indicate that gender does significantly affect SME risk as calculated by PP beta and the PS Model 1.

Capital structure—Interestingly, for the accounting beta, the coefficient value is negative for both the current and L1 observation, indicating that the higher the debt-to-equity ratio, the lower the risk as measured by accounting beta. Given that the coefficient value is negative, this creates an interesting perspective on how debt affects risk. The general opinion in the literature is that higher debt leads to higher risk and most banks are reluctant to lend to highly geared companies (Anderson et al. 2003; Berger and Udell 1998, 2002). However, the results presented in this dataset imply that higher geared companies have, in fact, a lower risk (according to accounting beta). However, the magnitude of the coefficient is quite close to ‘0’ so the overall impact that gearing has on accounting beta is not large. Debt finance is the most popular form of finance for SMEs, as debt, unlike equity, generates interest expenses that can be used to offset tax costs, which is a common practice in many regions (López-Gracia and Sogorb-Mira 2008). Furthermore, many SMEs are family-owned businesses; generating finance through equity would often mean having to dilute the control of the business, which many SME owners would like to avoid (Gomez 2007; Martin et al. 2017). Having a high amount of debt does not necessarily make a SME riskier than its counterparts. Having a high amount of debt can indicate that the SME has a reasonably large amount of finance that it can use for business and expansion purposes.

Profitability—The profit ratio, representing the performance aspect of profitability, is measured as a percentage of profit after tax over total sales. The current observation for the profit ratio is significantly related to the PP beta and the PS model. This indicates that the profit ratio only captures the risk for the current year of observation for an SME, meaning that a poor performance for a business will manifest in its risk value immediately, and the same applies for a good performance. The negative coefficient value for profit ratio with PP beta indicates that as profits go down, risk goes up. The coefficient value between profit margin and PS Model 1, however, is positive, which makes sense as the PS Model captures the survivability of a business, and a more profitable business would have a higher survivability rate in the future. The findings largely mirror that of previous research, indicating performance is an important aspect of risk and return for listed businesses (Moro and Nolte 2012). Regarding SMEs, performance is used to predict survivability and credit eligibility (Altman and Sabato 2007; Berger and Frame 2007; Hewa Wellalage and Locke 2012), which can also impact SME risk. In addition, from an efficiency perspective, the return on assets ratio is significantly related to the PP beta for its current observation. A positive coefficient value indicates that a higher return is significantly related to a higher risk. This is in line with the risk-return literature, which argues that the higher the risk, the higher the return (Sharpe 1964; Treynor 1965). It is important to note that with the measurements of accounting beta and PS Model 2, the relationship is not significant.

Firm age—As we can see, firm age is not significantly related to the accounting beta, PP beta or PS model 1 (PS Model 2 is not run in this regression because of the presence of firm age as a component of the calculation of PS Model 2). This means that firm age does not affect financial risk and return. In addition, the development of credit scoring practices that are biased towards older firms may be creating unnecessary challenges for new business start-ups.

Owner age—Looking at the relationship between the PS Model 1 and owner age, it is significantly related for both the current observation and its L1. The coefficient for the current observation is positive, indicating that the older the average owner, the more likely the business is to survive. This is in line with existing research, which finds that most SME owners are between 45 and 55 years of age (N. H. Ahmad et al. 2010; Cope 2005). However, this relationship is not significant when regressed against the PP beta, accounting beta and PS model 2.

Concentration of ownership—The largest share percentage held by the largest shareholder in the business is used as a means of measuring the level of ownership held in the hands of one individual. The results show there is no significant relationship between any of the risk measures and the largest share percentage.

Comparison between the risk measures

It is important to note that the different risk measures are significant for different variables. No single variable is significantly with all risk measures. In summarizing the results for the different risk measures used, we have developed the following correlation summary table (Table 9):

Table 9 Correlation summary
Full size table

In comparing the different risk measures, it is important to first review the methods and justifications behind the use of each risk measure. The PP beta reflects the relative level of risk faced by the SME, benchmarked against its closest market-listed proxy. Therefore, to a certain degree, it reflects the risk faced by that sector and the risk faced by an SME operating within that particular size category for that sector in any given year. The accounting beta measures the risk as a function of the historical performance of a single SME benchmarked against the performance of the investment portfolio (in this case, the portfolio is taken to be the entire dataset). Therefore, the accounting beta gives an overview of the risk associated with an SME’s performance, compared to other SMEs within the same portfolio. Both the PS Models are functionally similar, with the main difference being the use of firm age in PS Model 2. Otherwise, both PS Models use the input figures of return on assets and the debt to asset ratio. Despite their similarities, however, the models do not share a single independent variable that significantly relates to both, meaning that the addition of firm age changes the parameters of the probability of survival of a firm.

Overall, the PP beta has the highest number of correlations and significant relationships with SME characteristics. This is not to say, however, that it is the best or the only method of risk measurement that should be used. From an investor’s perspective, using the PP beta to estimate an SME’s risk can give a good indication of its future performance (Profit margin, Return on Assets, Return on Equity). Table 16 indicates that the gender balance of SME owners can also affect the PP beta, which in turn will affect its performance variables. Therefore, if used from an investor’s return perspective, the PP beta is a very effective measure of SME risk.

The accounting beta has the fewest number of correlations and significant relationships with SME characteristics. It is only significantly related to the debt-to-equity ratio, a variable which is not regressed against the other risk measures because it shared input figures with them. This means that in the event that other risk measures are unsuitable for use, the accounting beta can be a reliable indicator of risk associated with SME gearing. For SMEs looking to increase their debt capacity or banks looking to lend to small businesses, the accounting beta is a useful indicator of the risk faced by the SME. However, due to its low correlation with other SME characteristics and risk measures, it is not very reliable in providing an overall risk measure for any SME.

Looking at the probability of survival, it is worth noting that while risk does not necessarily equate to business survival, the reason why these models are chosen is that they can be used as input figures for cost of equity calculations, which are comparable to the cost of equity calculations using CAPM (of which beta is an input figure) (Cheung 1999). As such, they can be compared side by side with the beta figures (PP beta and accounting beta), but it should be remembered that they predict the likelihood of survival of the SME only within the next year. (Abdullah et al. 2015), who developed the PS Models, indicated that the PS Model 2 has better predictability than the Model 1. However, this research tested both models and from Table 16, each model relates differently to different SME characteristics. Model 1, which only uses financial indicators, closely mirrors the relationships of PP beta, with a significant relationship for business gender and profit margin. More importantly, it captures the effect that owner age has on the survivability of the business, with older owners being more likely to succeed. Model 2, on the other hand, is only significantly related to the return on equity and the control variable, total equity. It is interesting that with the addition of firm age, business gender, profit margin and owner age do not impact the risk of survival. However, given that extant research strongly indicates that these three factors do affect SME risk, this research would argue that the PS Model 1 is more in line with findings from the literature (Moro and Nolte 2012; Singh and Zammit 1999; Wahyudin et al. 2016).

To this end, this research found that PP beta is the most suitable measure of SME risk for most investor-related appraisal processes. However, it must be noted that using it in conjunction with a survival prediction model allows for a more holistic view of business operations, as the likelihood of business failure is also an important aspect of SME investment/operations.

Conclusion, implications and limitations

SMEs inherently face more risk than larger companies. This risk is attributed to resource constraints as well as managerial and operational issues that are endemic to SMEs. However, this risk has not been sufficiently quantified; nor is there a significant amount of research on its relationship with SME characteristics. This study has demonstrated the most feasible methods of calculating SME risk and their application to a panel dataset of SMEs. More importantly, the relationship between the risk calculated by these models and SMEs’ characteristics is analyzed, drawing attention to the areas that are vital to managing the level of risk exposure faced by SMEs.

This research empirically estimates the relationship between SME risk and characteristics. This particular aspect of the research has yielded some very interesting findings: (1) having a high amount of debt does not necessarily make a SME riskier than its counterparts; (2) profitability significantly affects SME risk; and (3) size, firm age and business ethnicity are not significantly associated with SME’s risk in Malaysia.

The significant relationships between risk and profitability and owner age are interesting; however, they are very much in line with what the literature suggests. What is of greater importance is the significant relationship between risk and gender which implies that female business owners operate riskier businesses as compared to men and as such would require a higher level of return to compensate for this added level of risk. As a developing nation, gender equality is an important issue for Malaysia. This finding is of particular importance to policy makers in creating policies to specifically benefit female business owners. This is an especially interesting development for further research in the field of female business ownership.

Some of the non-significant relationships are important as well. Most noticeably, the lack of any significant relationship between ethnicity and risk suggests that the ethnicity of the business owner has no bearing on the risk faced by the business. Given that Malaysia practices affirmative action policies that favor those of Bumiputra descent, these findings indicate that other ethnicities are not significantly worse off for it. Conversely, the perceived dominance of the ethnic Chinese in the Malaysian economy is a subject of frequent debate. These results also suggest that being Chinese does not significantly affect the risk of doing business in Malaysia, which suggests that Chinese business owners do not have it easier than business owners of other ethnicities. This finding is of interest to future research in ethnic studies based in Malaysia and in other countries around the world.

The limitation of this study is the dataset, as the research could only be carried out on variables available in the dataset provided. While the dataset is comprehensive in recording various financial and non-financial details of SMEs in Malaysia, a more expansive database could allow for the testing of other characteristics such as owner’s experience, the use of Islamic finance and the impact of the cost of debt on SME risk. With such information, this research would have been able to develop further insights into other factors and their relationships with SME risk and return.