1
The relevance of financial ratios in default prediction for
SMEs within transportation-related sectors
Yankova, Mariya Plamenova 10349642
June 2016 Bachelor Thesis
BSc Economics and Business, Economics and Finance track Thesis Supervisor: Mr R.C. Sperna Weiland
2
Statement of Originality
This document is written by Student Mariya Yankova who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3 Abstract
Small and medium enterprises (SMEs) represent a substantial part of the businesses in many developed and developing countries. Many factors, such as the simplified financial structure of SMEs make them different than the larger firms and therefore suggest that their default risk
should be evaluated and analyzed differently. Nevertheless, there is limited literature on testing which are the most suitable ratios for SME default prediction. This paper aims at analyzing the significance of several financial ratios as default probability determinants for SMEs in a time frame of 3 years, over the period 2012 to 2014. The analysis is based on small
and medium enterprises which are strongly related to the transportation sector for their businesses. A stepwise logistic regression model has been adopted in order to evaluate the available ratios. The results show that the ratio measuring liquidity is the best SME default
4 Introduction
Small and medium enterprises (SMEs) are considered one of the main driving forces of the economies in multiple countries (Modina & Pietrovito, 2014). According to the Organisation for Economic Co-operation and Development (OECD), SMEs represent over 97 percent of the total business population and they contribute to the economy through various channels, such as the creation of workspaces. Due to their simpler structure compared to larger firms, SMEs are more flexible in turbulent economic times, as well as more flexible in providing specialized customer service, which often makes them preferred to a larger firm (Altman & Sabato, 2007). However, their performance is proven to be more volatile as they could either grow into larger companies or go into default (Dietsch & Petey, 2004). These are one of the main reasons why SMEs are viewed as different from larger firms within the existing literature.
Besides the difference in setup, the default risk between the SMEs and their larger counterparts has also been proven to be different according to various researches. For example, Dietsch and Petey (2004) performed an analysis based on SMEs and larger firms from Germany and France and concluded that SMEs bear higher default risk; however they find that their correlation is lower than the one between larger firms, which could be a benefit in terms of diversification. Other studies, such as Kolari and Shin (2003) argue that default risk is lower for SMEs compared to larger firms due to their specialization and knowledge for specific market niches.
The different findings and conclusions within the existing literature prove the need for separate analysis on SMEs. The lack of official credit ratings available for them makes it difficult for banks and businesses to evaluate default risk and therefore they need to create their internal default prediction models and determine which potential default determinants to implement within that model, which is the issue this paper is aiming to answer. In the recent decade there has been increasing interest towards the findings of such studies, however for many industries such analysis has not been performed. Therefore this paper is performing an analysis on the financial ratios which could potentially be significant estimators of default probability for SMEs specifically within the sectors that are strongly dependent on
transportation. The conclusions of such analyses per industry are of interest to both banks and businesses, in order to be able to develop a better model for default evaluation of SMEs.
5
Nowadays, following the implementation of the Basel Accord, banks are required to have at least 20% of their SMEs portfolio classified as retail and furthermore lending to SMEs would benefit banks through lower capital requirements and portfolio diversification (Altman & Sabato, 2007). Furthermore, SMEs are often representing the main part of businesses’ customer portfolio (OECD, 2015) and therefore could potentially have high impact on their earnings. Nevertheless, there is limited financial information available for SMEs as they are not required to publish annual financial reports, which increases the value of such studies as banks and companies might not be able to perform evaluation of the ratios internally due to lack of financial data.
For the purpose of this research, data is taken from the internal database of TIP Trailer Services B.V., which is a company specialized in leasing and rental and is part of the HNA holding, one of the Fortune Global 500 companies. The company has the largest share of the transportation market in Europe and as such has a large customer portfolio of companies from various industries. Therefore this research is particularly focused on analyzing the financial ratios that could play a role in affecting the probability of default of firms which are strongly related to the transportation sector. The determinants of default are analyzed using a stepwise logistic regression for a sample of 200 SMEs over the period 2012 to 2014, which includes partially the aftermath of the recent financial crisis. The definition of default is following the Basel Accord and includes bankruptcies, loans past-due 180 days and ascertained losses. The focus of this paper is to detect which financial ratios are the best predictors of default event within one year, therefore the analysis is conducted on financial information from 2012 to 2014 for defaulting or non-defaulting firms within 2013 to 2015. The results provide valuable insights for the most significant predictors, specifically the ratio measuring the liquidity of a company. Compared to other studies, the findings of this article are specifically related to the industries which are dependent on the transportation sectors and therefore the conclusions could be used for implementation of default prediction model for SMEs from these industries.
The paper is structured in five sections. Section 2 provides a review of the existing literature on the topic. Section 3 describes the dataset, main statistics and the structure of the statistical model. Section 4 presents and discusses the findings of the research and finally, section 5 draws conclusions and presents suggestions for further research.
6 Literature review
Credit Ratings
There is substantial literature about default prediction where different methodologies have been developed and analyzed (Altman & Sabato, 2007). Before the recent financial crisis, most businesses were relying on customer’s credit rating for making business decisions- a statistical method of predicting delinquency or default, developed in the 1950s (Mester, 1997). However, the current credit crisis has shown that many investors relied too heavily on these ratings, which turned out to be inflated and non-realistic (Benmelech & Dlugosz, 2010). The literature research conclusions on the effectiveness of credit ratings in predicting
bankruptcy have become even more controversial after the recent financial crisis. One of the issues is the agency problem, caused by the fact that rating agencies like Moody’s are being paid by companies to calculate their credit rating and therefore such agencies are prone to artificially increasing the assigned rating (Becker & Milbourn, 2008). Furthermore, examples of the current financial crisis show that the initial credit ratings proved to be excessively optimistic, which is a misleading signal for investors (White, 2010). In response, banks and businesses have developed internal rating models for default prediction, on which the risk management heavily relies (Belás & Cipovová, 2011).
Prior to the recent financial crisis, related literature was mostly examining the methodology behind credit ratings and their capability of reducing the costs of asymmetric information (Grunert, Norden, & Weber, 2005). It was expected that such scoring is going to allow banks to expand their lending to SMEs due to the availability of a credit risk evaluation through the credit rating (Mester, 1997). Credit agencies such as Moody’s and Standard & Poor’s produce publicly available credit ratings, which have become standard variables for credit risk
management and control (Treacy & Carey, 2000).
Besides the establishment of external ratings, internal ratings were increasingly being
developed by banks and businesses in order to provide valuation based on both financial and non-financial factors available (Grunert, Norden, & Weber, 2005). According to Mester, most developed models have been using a maximum of 12 variables in order to yield the
combination with highest predictive power and in the same time avoid correlation between factors (1997). Alternative statistical techniques have been investigated among the related studies, such as the dichotomous classification test developed by Beaver (1967) and various univariate and multivariate models (Mester, 1997).
7 Default Prediction Studies
One of the first studies in the field was a research made by Beaver (1967), who used a set of financial ratios in order to develop an univariate model for default prediction (Altman & Sabato, 2007). Beaver defines default as the occurrence of either one of the following events: bankruptcy, bond default, overdrawing a bank account or failure to pay preferred stock dividend (1967). The analysis was based on data for industrial public companies, which according to Beaver represented over 90 percent of the invested capital at that time (1967). He examined the effectiveness of fourteen financial ratios and applied the model to a dataset of 158 firms from 38 different industries. Beaver applied a classification test which makes a dichotomous prediction of whether a firm is failed or not (Beaver, 1967). His finding is that the cash-flow to total debt ratio has best predictive power of the examined ratios and
furthermore, that ratio distribution for the defaulted firms deteriorates towards the point of default, whereas the non-defaulted firms maintained fairly stable ratios1.
Following Beaver’s model, Altman (1968) developed a multivariate model using a multiple discriminant analysis technique (MDA), which was an alternative, solving the inconsistency problem of the previously developed univariate model (Altman & Sabato, 2007). Furthermore the analysis was based on more financial information of the selected dataset than the one included in the univariate model, which is expected to increase the predictive power of the model. Altman classified the selected variables into five standard ratio categories, which are expected to be determinants of the firm’s performance and default probability – activity ratios, leverage, liquidity, profitability and solvency. He analyzed twenty-two potentially useful financial ratios, out of which he included 5 in the final model, which in combination were yielding the highest predictive power 2. Altman used data of firms from the
manufacturing industry and his statistical technique was replicated by many studies
afterwards using different datasets, as his model became prevailing among the existing default prediction models (Altman & Sabato, 2007).
1
Beaver (1967) suggests that due to the nature of his dataset, the results might not apply to non-public firms.
2
Altman (1968) used five ratios in his original model: working capital/total assets, retained earnings/total assets, EBIT/total assets, market value equity/BV of total debt and sales/total assets.
8
The extensive use of the model within the default prediction literature has pointed out that two of the main basic MDA assumptions are often violated when applied in practice3 (Altman & Sabato, 2007). Furthermore, due to the structure of the model, the standardized coefficients are not an indicator of the variables’ importance in relation to each other as they do not represent the slope of the regression (Altman & Sabato, 2007). Therefore, alternative models have been developed, attempting to create a statistical technique which would eliminate the existing issues.
The probit model has been used as an alternative to the MDA for default prediction for first time in a study by Zmijewski (1984). In his study, Zmijewski analyzes the two potential biases arising in the researches using MDA model and gives an alternative solution with the probit model (1984). Another solution was developed by Ohlson (1980), who applied the logit model to the analysis. Later studies have proven that the logit model has a better prediction power than the probit model (Altman & Sabato, 2007). One of the benefits of the logit model is that it does not require the application of the two main MDA basic assumptions (Altman & Sabato, 2007). Furthermore, the dataset does not need to be proportional among the selected time frame. Ohlson (1980) based his analysis on six years data of 2,163 firms, out of which 105 went bankrupt over the examined period. He used seven financial ratios and two binary variables as predictors.
More recent studies in the field, such as an analysis by Modina and Pietrovito (2014) rely on the logit model for modelling credit risk. Modina and Pietrovito find that economic variables are less relevant for default prediction than the SMEs capital structure and interest
expenditures (2014). They suggest that SMEs find it more difficult to respond to market globalization due to their low capitalization level and therefore lenders need to be evaluating their credit risk and exposure towards them with higher caution (Modina & Pietrovito, 2014). Despite the lower predictive power in comparison with the MDA model, the logit model has been preferred in most academic literature after Ohlson’s analysis (Altman & Sabato, 2007). One of the benefits of the logit model is that the yielded score is in the range from zero to one and therefore can be easily interpreted as default probability. However, there has been
criticism on the model due to the fact that the logit regression can lead to either positive or negative bias of the default probability due to its functional form (Altman & Sabato, 2007).
3Two basic assumptions MDA is based on are: (1) the independent variables are multivariate normally distributed; (2) the group dispersion matrices are equal for defaulted and non-defaulted firms.
9
Despite the criticism, the logit regression gives results which are easy to interpret with regards to the investigated issue of default prediction, due to the fact that the dependent variable can be binary and represent default and non-default and that the groups are discrete and
identifiable (Altman & Sabato, 2007). Another benefit of the logit model in terms of statistical structure is that each coefficient can be interpreted independently as the effect on the
dependent variable. SME studies
The introduction of credit scoring has allowed banks to increase lending to SMEs and diversify their portfolios (Mester, 1997). The implementation of the global regulatory framework Basel Accord has further encouraged focus on this segment, as new credit risk management techniques have been introduced4 (Altman & Sabato, 2007). Altman (2005) finds that the introduction of the Basel Accord would encourage banks to expand their SME portfolio, as they will benefit in terms of capital requirements. Being on average more than 97% of the firms worldwide, SMEs play a fundamental role in the economy and are the main source of working places (Altman, 2005). Furthermore, according to Altman they usually account for two-thirds of the GDP and are dynamic and responsive to different economic conditions (2005).
Due to the importance of SMEs for the economy, a lot of studies have focused on analysis of the effects of global regulatory framework, or on the different types of strategies and lending towards SMEs (Altman & Sabato, 2007). Berger and Udell (2006) analyze the nature of the issues which causes less lending to SMEs and provide insights into the various policies
available. Kolari and Shin (2003) use efficient frontier analyses in order to determine the risks and benefits of lenders towards SMEs. They conclude that the main benefit of SMEs is their expertise and specialization which reduces their risk of default (Kolari and Shin, 2003). Despite the availability of multiple analyses related to SMEs, there has not been much research analyzing the credit risk models for such firms (Altman & Sabato, 2007). Recently, Berger and Frame (2007) described the Small Business Credit Scoring (SBCS) technology for assigning a credit score to a small businesses applying for micro credits5 and the level of
4
For studies on the effect of the Basel Accord on SME lending, refer to e.g. Schwaiger (2002);
Saurina and Trucharte (2004); Udell (2004); Jacobson et al. (2005); Altman and Sabato (2005); Berger (2006).
5
10
availability of such. The SBCS is based on quantitative data and is combining consumer-specific and business information in order to create a score (Berger & Frame, 2007). For their analysis, Berger and Frame (2007) collected data from several consumer credit bureaus regarding the payment performance of the firm’s owner. They conclude that the lenders who implemented the SBCS have expanded their SME portfolio due to the availability of a better risk measurement tool.
Following the implementation of the Basel Accord, banks are required to have at least 20% of their SMEs portfolio classified as retail (Altman & Sabato, 2007). That is, if the total
exposure of the lender towards the firm is below one million, and it is managed on a pooled basis, the firm can be defined as a retail (Altman & Sabato, 2007). Due to the newly
implemented global policies, lending to SMEs brings benefits to the banks such as lower capital requirements and portfolio diversification and this further encourages them to develop methods for evaluating their credit risk.
In order to gain competitive advantage, several banks have already developed automated scoring systems for credit risk evaluation of allowing exposure up to €3 million for SMEs (Altman & Sabato, 2007). This gives them the flexibility of making lending decisions based on a standard tool which is using the available financial information for customer credit risk evaluation and creating a credit score. Several studies have proven the profitability and benefits of lending to small and medium-sized enterprises6 (Altman & Sabato, 2007). Kolari and Shin (2003) have found that failure risk is reduced among SMEs due to the fact that they are experts in specific niches of the market. However, other researches have contradictory conclusion- that SMEs bear higher default risk and therefore have to be treated separately when default risk is being analyzed (e.g. Dietsch and Petey, 2004; Saurina and Trucharte, 2004; Altman & Sabato, 2007).
Dietsch and Petey (2004) use a one-factor credit risk model in order to analyze the default probabilities of SMEs in Germany and France. They discuss the importance of asset
correlation7 in the modelling of credit risk and find that on average the asset correlations are positively related to probability of default on an industry level (Dietsch and Petey, 2004). Saurina and Trucharte (2004) perform an analysis on the policy effects of lending for Spanish SMEs and also confirm the theory that they have to be evaluated separately due to the higher
6
For a comprehensive analysis refer to e.g. Kolari and Shin (2003); Berger (2004). 7
11
default risk they carry (3.7% versus 0.65% for larger firms). Furthermore, the need of a separate tool for credit risk evaluation is supported by the finding of Saurina and Trucharte (2004) that more than 70% of the total credit exposure to firms is originating from exposure to SMEs.
According to Altman & Sabato (2007), many lenders have already implemented separate tools for credit risk analysis of SMEs in order to minimize their expected loss. However, in the existing literature on the matter there is not definite conclusion that implementing separate credit risk model for SMEs is beneficial for the lender. Bradbury (1992) finds that loan default of SMEs is reduced only when using univariate linear discriminant model and furthermore his research is based on 40 firms from New Zealand, which is not sufficient to prove this hypothesis (Altman & Sabato, 2007). Edmister (1972) performs an analysis on the value-added from various financial ratios but does not discuss the potential reasons for having a separate model for SMEs.
A recent study by Ciampi (2015) analyzes 934 SMEs from Spain using a logit regression including both financial ratios and characteristics of firm’s corporate governance. Ciampi finds that the results are different than for larger firms, which supports the statement that a separate analysis for SMEs is needed (2015). Furthermore, a comparison of the results with other research analyses which use only financial ratios shows that controlling for corporate governance characteristics increases the predictive power of the model (Ciampi, 2015). A research performed by Daily and Dalton (1994) using logit regression and a sample of 100 firms also finds that including corporate governance variables to the model increases is quality by 12% up to 3 years before failure.
Nowadays, as the global policies encourage lenders to expand their portfolio with more lending to SMEs by lowering the capital requirements, the need of a separate model for default prediction is increasing (Altman & Sabato, 2007). Therefore, the purpose of this paper is to define which financial ratios are the most suitable predictor of default among the SME companies in the transportation sector.
12 Methodology and Data
This section describes the model developed in this paper to evaluate which are the best determinants of the probability of default for SMEs out of a set of available financial ratios. Furthermore, it provides descriptive statistics of the dataset and definitions on the various terms used.
Data set
Most of the studies in the area use data from big firms for which financial data is publicly available and therefore easily accessible (Modina & Pietrovito, 2014). On the other hand, financial data for SMEs is more difficult to obtain, despite the fact that they are a big part of banks’ portfolios (Rikkers and Thibeault, 2009). For the purpose of this research, the data was collected from the internal database of TIP Trailer Services B.V., which is a company
specialized in leasing and rental and is part of the HNA holding, one of the Fortune Global 500 companies. The company has the largest share of the transportation market in Europe and as such has a diversified portfolio of SMEs and larger firms. In this paper, SME is determined following the Basel Accord definition8. Furthermore, a definition of default is adopted as defined by the Basel Committee9.
The analyzed portfolio contains customers from several industries, as defined by the SIC code, which is an official internationally accepted industry code. The distribution of the dataset per SIC code can be found in Appendix A. Due to the limitation of the available data and the industry-specific trends for the applied ratios, the findings of this research might not be applicable to SME companies from different industries.
The firm portfolio consists of eight European countries- Netherlands, Belgium, Spain, Germany, Austria, France, Poland and Romania. Out of the total dataset of 684 firms
available, 667 were defined as SME. Depending on the completeness of the data available, a total of 200 non-defaulted and 14 defaulted firms was randomly selected for the analysis. The fraction of defaulted firms against the non-defaulted is on average 7%, which is in line with
8
As defined in Basel II, a firm is considered SME if it has fewer than 250 employees and less than €50 million of sales. The same definition is used in the currently developed Basel III Accord. 9
The Basel Committee defines a firm as being default when it has been more than 90 days delinquent on its principal payments. Furthermore, the definition includes bankruptcy, loans due more than 180 days, bond default, or financial distress.
13
the findings of the existing literature (e.g. Altman & Sabato, 2007; Modina & Pietrovito, 2014). Financial data was collected over the period of three years (2012-2014). A firm has been considered as defaulted if the event of default has appeared within 1 year of the available financial data. Therefore, the occurrence of default event has been analyzed for the selected companies within the period from 2013 to 2015.
The composition of the selected dataset per year can be found in Table 1. The percentage of defaulted firms is higher in 2012 compared to 2014 due to the recent financial crisis and gradual economic recovery afterwards. Furthermore Graph 1 shows that the largest fraction of defaulted firms out of the analyzed countries is in Spain, as compared to lower defaults seen in Benelux region and Germany. Therefore, the data sample is reflecting the actual observed condition of the economy in the selected timeframe. Due to the fact that analyzing the financial crisis effect on the model is not within the scope of this research, only data from 2013 and 2014 was applied to the model described below.
Table 1: Sample distribution per year
Non-defaulted firms Defaulted firms
Number Percentage Number Percentage By year
2012 62 92.54% 5 7.46%
2013 70 93.33% 5 6.67%
2014 68 94.44% 4 5.56%
Total 200 93.46% 14 6.54%
Notes: The percentage of the distribution per year is the fraction of non-defaulted (defaulted) firms out of the total sample for each particular year.
Graph 1: Sample distribution per country of defaulted firms
Notes: The percentage of the distribution per country is the fraction of defaulted firms out of the total number of defaults in the sample.
14 Model construction
In line with the previous studies discussed above, the available financial data is divided into five categories – activity ratios, leverage, liquidity, profitability and solvency. Several
available financial ratios are related to each category and according to the available data, part of them have been selected for the current analysis (see Table 2). A statistical forward
stepwise selection method is applied in order to determine which combination of the available ratios leads to a model with the best predictive power.
Table 2: Set of indicators
Indicator Available ratios Ratios applied in the model
Leverage Short-term debt/Equity (book value) Short-term debt/Equity (book value) Equity (book value)/Total liabilities
Liabilities/Total assets
Liquidity Cash/Total assets Cash/Total assets
Working capital/Total assets Cash/Net sales
Profitability EBITDA/Total assets EBITDA/Total assets Net income/Sales
Retained earnings/Total assets Coverage Accounts payable/Sales
EBIT/Interest expenses
Retained earnings/Total assets
Activity
Accounts receivable/Liabilities EBITDA/Interest expenses Sales/Total assets
EBITDA/Interest expenses
Previous studies (e.g. Altman & Sabato, 2007; Modina & Pietrovito,2014) have applied a statistical forward stepwise selection method in order to determine the most efficient ratios (jointly significant) amongst the identified as potential predictors of default. In order to apply the stepwise selection method, Modina and Pietrovito (2014) perform a factor analysis10, which allows them to reduce the number of selected variables by extracting the statistically significant factors, affecting the correlation level between the variables. As the amount of available ratios within this research is limited, factor analysis will not be performed prior to applying the statistical forward stepwise selection model.
10Factor analysis is conducted on the original indicators and then the extracted factors together with each corresponding percentage of explained variance per factor is analyzed.
15
A study by Chen and Shimerda (1981) has shown that amongst more than 100 available ratios, about 50 percent have been selected for the model of at least one research. This fact supports the statement that there is no particular set of ratios which is proven to provide the highest default estimation accuracy. Furthermore, several recent studies have concluded that in order to create a model with a good prediction power, qualitative variables need to be used along with the financial ratios (e.g. Lehmann, 2003; Grunet et al., 2004). However, due to the limitations of the database available and the scope of this research, only quantitative variables will be applied to the model.
Prior to applying the selected ratios to the model described below, a test on their correlation was performed in order to ensure that the data quality is satisfactory. The results are presented in Table 3. No significant correlation between the variables has been found, which means that including them simultaneously in the regression model would not create any bias caused by variable correlation (or multicollinearity in case of perfect correlation).
Table 3: Explanatory variables correlation test
Leverage Liquidity Profitability Coverage Activity
Leverage 1 Liquidity -0.0880 1 Profitability -0.198 0.231 1 Coverage 0.00310 0.0849 0.248 1 Activity -0.0449 -0.113 0.127 0.0457 1 * p < 0.05, ** p < 0.01, *** p < 0.001 Logistic regression
For the purpose of this research, the selected five ratios from Table 2 will be used, which is in line with the majority of available literature on the subject as well as the available data. The dependent variable will be constructed as binary (0= defaulted;1=non-defaulted) and will therefore represent the Known Probability of Being Good (KPG)11.
In order to estimate the probability of default for the selected three year period, a logistic regression will be performed, using the described above financial ratios as explanatory variables an the binary variable (0 = defaulted; 1 = non-defaulted) as the dependent variable.
11
Using KPG as dependent variable is expected to ensure that the slopes and intercept will be positive since if the logit score is higher, probability of default is lower (binary variable closer to 1).
16
The model assumes that the observable variables are related to the independent (unobservable) default event. The logistic model would be structured as below: Equation 1: Logit(pi) = α + βxi1 + β2xi2 + … + βmxim
Where pi is the probability of default, α is the constant, xm is the set of selected explanatory
variables and βm are the explanatory variable coefficients. The equation will be estimated
through the stepwise method in order to determine which explanatory variables are most significant in combination within the model. The Maximum Likelihood Estimator (MLE) will be used for estimation of the explanatory variable coefficients. A higher coefficient would be interpreted as higher contribution of the variable in the default prediction.
The probability of default (pi) is derived from Equation 1 by converting it as follows:
Equation 2: p = 1
1+𝑒^𝑘
, where k = -(α+βx
i
1 + β2xi2 + … + βmxim)
The logistic equation will be estimated through the stepwise selection method in order to determine which ratios are jointly significant within the logistic model described above. Within this research the forward stepwise selection method is adopted. Therefore, the first model is consisting only of the intercept and depending on the significance of each variable when added to the model, a final model is developed where all explanatory variables are jointly significant. Significance for adding each variable to the model will be tested with the Likelihood-ratio test. The level of significance used for adding variables to the model is 0.05. Variables with lower significance are excluded from the final model.
After performing the testing as described above, the model with highest significance of the explanatory variables will be analyzed. The Type 1 and Type 2 errors will be presented and discussed. Type 1 error represents the amount of defaulting companies which would be classified as non-defaulting within the model by using a particular set of explanatory variables. Type 2 error represents the amount of non-defaulting companies which are classified as defaulting. The pseudo-R2 will not be analyzed in this paper as there is no conclusion within the literature regarding the quality of this test statistic12
12
17 Results
Logistic regression including all available ratios as explanatory variables
The results from the logistic regression including all available ratios for the selected years are summarized in Table 4. Only data from 2013 and 2014 is included in the final analysis as 2012 partially reflects the aftermath of the recent financial crisis, which is not within the scope of this research. Column (1) presents the marginal effect of the logit model with the five explanatory variables described in the Methodology and Data section for 2014, whereas column (2) presents the results of the regression with the data for 2013.
Table 4: Logistic regression output
(1) (2) VARIABLES 2014 2013 Leverage -22.05936** -1.87413 (10.304) (4.085) Liquidity 16.36551** 40.05354* (6.610) (21.134) Profitability 0.66346 0.01773 (3.437) (3.099) Coverage 0.14624 0.52411 (0.102) (0.513) Activity 0.03722*** 0.00458 (0.014) (0.011) Constant 4.08125*** 2.31493*** (0.739) (0.764) Observations 72 75
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
For both years all of the slopes have the expected sign and therefore the expected effect on the default probability. A positive sign is understood as a negative relationship between the predictor and the chance of being default (0 = defaulted; 1 = non-defaulted). It is expected that all explanatory variables except the leverage ratio will have a positive sign and therefore a negative effect on the default probability. The leverage ratio is expected to have a negative sign and that will be interpreted as having a positive effect on the default probability.
According to the results in Table 4, all explanatory variables have the expected sign.
18
error term is not proven to be constant across the dataset. That is in line with the methodology of other studies using financial ratios in a logistic regression (e.g. Modina & Pietrovito, 2014; Altman & Sabato, 2007).
The constant is significant in both 2013 and 2014 (p2014 = 0.001 and p2013 = 0.002). Liquidity
was also found to have a significant effect on the default probability in both years (p2014 =
0.013 and p2013 = 0.058). A positive sign means that the increase of liquidity reduces the
probability of being default, which is in line with the expectations. Other studies confirm the significance of the selected ratio representing liquidity within the described default prediction model (e.g. Altman & Sabato, 2007). In case the applied ratio representing liquidity is not available, Modina and Pietrovito (2014) suggest that Working capital/Total assets is another ratio which could be a good default predictor within the described logit model.
Forward Stepwise Logistic Procedure
In order to analyze whether a model with fewer or none of the available ratios would be significantly better than the complete model described above, the forward stepwise logistic method was applied. Firstly, the model including all available ratios as explanatory variables was tested against a null model (including only the intercept). For both 2013 and 2014 the complete model was proven to be significantly better than the null model at the 0.05 significance level (p2014=0.0482 and p2013=0.0003; see Test 6 in Table 5). Afterwards the
likelihood-ratio test was used to estimate whether a model with an additional explanatory variable is significantly better than the model without that variable and such test was
performed 5 times for each year in order to test each variable and then compare it against the null model. The results are presented in Table 5:
Table 5: Forward stepwise logistic regression test
Test № (2014) (2013) 1 Activity 0.1764 0.0301* 2 Liquidity 0.0324* 0.0148* 3 Leverage 0.0463* 0.0743 4 Coverage 0.3773 0.2870 5 Profitability 0.9050 0.0046** 6 Complete model 0.0482* 0.0003*** * p < 0.05, ** p < 0.01, *** p < 0.001
19
According to the results presented in Table 5, in 2014 liquidity and leverage are proven to have significant effect on the dependent variable at 0.05 significance level. For 2013, liquidity, activity and profitability are significant. The final model was constructed of all variables which were significant at the 0.05 significance level and tested against the null and the complete model. For 2014 a model including liquidity and leverage as explanatory variables was tested against the null model and it was found to be significant at the 5 percent level (p = 0.0374). For 2013, a model including activity, profitability and liquidity was tested against the null model and it was also found to be significant (p = 0.0001). Furthermore, for both years the models were significant against the complete model at 5 percent significance (p2014= 0.0402; p2014= 0.0213). Through the stepwise method it was proven that for both years
the models including several selected ratios out of the available set have in combination a better predictive power than the null model as well as the complete model.
Prediction accuracy
In order to estimate the prediction accuracy of the developed model, an analysis was performed on the correctly versus the incorrectly classified observations. Firstly, the prediction power of the model was calculated using Equation 2 as described in the
Methodology section. Furthermore, a test was performed on the final models for each year in order to find the size of the type 1 and type 2 errors. Type 1 error is referring to the number of firms which the model would classify as non-defaulting when they are actually defaulting. Type 2 error represents the amount of non-defaulting firms, which the model would classify as defaulting. The results can be found in the table below:
Table 6: Incorrectly classified observations
(2014) (2013)
Type 1 error 1.79% 1.49%
Type 2 error 3.51% 2.94%
Correctly classified 94.70% 95.57%
As all the estimated errors are below 5 percent, it can be assumed that the model has a good predictive power with low error probability. Furthermore, the final models have been proven to have a high prediction accuracy as in both years it is exceeding 90 percent. Therefore, it can be concluded that the selected financial ratios are successful default predictors.
20
data for constructing the financial ratios applied. The current analysis proves that from the analyzed ratios within the selected industries, the one representing liquidity is successful as a default predictor in both years and therefore a decrease in the liquidity could signalize higher probability of default. Nevertheless, these results might not be applicable to SME companies from other industries, due to the industry-specific trends for the applied ratios.
Conclusion
This research investigates the capability of financial ratios to predict default for SMEs through a sample of 200 firms, collected from the internal database of TIP Trailer Services B.V., which is a company specialized in leasing and rental and part of the HNA holding. As financial data for SMEs is difficult to obtain due to the lack of officially published annual reports, the research is aiming at using the available internal database in order to find the best financial ratios which are able to indicate a higher default probability of a firm. Therefore, the results are of help to banks and businesses for concentrating their effort in collecting and analyzing lower amount of financial variables which have the highest quality as default predictors.
The findings show that the ratio measuring liquidity is consistently significant default predictor for both 2013 and 2014. Therefore, it can be concluded that the decrease of cash holdings compared to the total assets of a firm indicates liquidity decrease and therefore higher probability of default. The results imply that banks and businesses should pay attention to the capital structure of SMEs and particularly to the amount of their cash holdings out of their total assets. A decrease in such liquidity measurement should indicate an increase in the default probability of a customer. The implications of this research are especially relevant for the period after the recent financial crisis as banks and companies are more cautious when dealing with SMEs despite the benefits in terms of diversification and are in search of the best predictors of default event.
As the financial data used within this research was of companies which are strongly related to the transportation sector, the conclusions might not apply to a different firm set. Another limitation of this research is the amount of available financial data applied to the analysis. The results could be improved by performing the analysis on a larger dataset and timeframe. The results are not proven to hold during turbulent times such as the recent financial crisis, as such an event increases the vulnerability of SMEs and they are assumed to apply different liquidity management.
21 Appendix A: Data distribution by SIC Code
SIC Code SIC Code Description # Companies
1623 Water, Sewer, Pipeline, and Communications and Power Line Construction 4
2082 Malt Beverages 27
2086 Bottled and Canned Soft Drinks and Carbonated Waters 6
2833 Medicinal Chemicals and Botanical Products 10
3663 Radio and Television Broadcasting and Communications Equipment 13
3679 Electronic Components, Not Elsewhere Classified 10
3711 MOTOR VEHICLES & PASSENGER CAR BODIES 1
4700 Transportation services 24
4731 Arrangement of Transportation of Freight and Cargo 9
4941 Water Supply 6
5013 Motor Vehicle Supplies and New Parts 3
5093 Scrap and Waste Materials 2
5122 Drugs, Drug Proprietaries, and Druggists' Sundries 3
5411 Grocery Stores 5
6513 Operators of Apartment Buildings 4
6722 Management Investment Offices, Open-End 44
7389 Business Services, Not Elsewhere Classified 20
7819 Services Allied to Motion Picture Production 9
22 References
Altman, E. I. (1968). Financial ratios, discriminant analysis and the prediction of corporate bankruptcy. The journal of finance, 23(4), 589-609.
Altman, E. I., & Sabato, G. (2007). Modelling credit risk for SMEs: Evidence from the US market. Abacus, 43(3), 332-357.
Altman, E., & Sabato, I. (2005). Effects of the New Basel Capital Accord on Bank Capital Requirements for SMEs. Journal of Financial Services Research,28(1), 15-42.
Beaver, W. (1966). Financial Ratios As Predictors of Failure. Journal of Accounting Research, 4, 71-111.
Becker, B., & Milbourn, T. (2008). Reputation and competition: evidence from the credit rating industry. Harvard Business School.
Belás, J., & Cipovová, E. (2011). Internal Model of Commercial Bank as an Instrument for Measuring Credit Risk of the Borrower in Relation to Financial Performance (Credit Scoring and Bankruptcy Models). Journal of Competitiveness, 3(4).
Benmelech, E., & Dlugosz, J. (2010). The credit rating crisis. In NBER Macroeconomics Annual 2009, Volume 24 (pp. 161-207). University of Chicago Press.
Berger, A. (2006). Potential Competitive Effects of Basel II on Banks in SME Credit Markets in the United States. Journal of Financial Services Research, 29(1), 5-36.
Berger, A., & Frame, W. (2007). Small Business Credit Scoring and Credit Availability*. Journal of Small Business Management, 45(1), 5-22.
Berger, A. N., & Udell, G. F. (2006). A more complete conceptual framework for SME finance. Journal of Banking & Finance, 30(11), 2945-2966.
Bradbury, M. E. (1988). The Effect of Analysts' Adjustments in the Context of Loan Assessment. Accounting & Finance, 28(1), 45-55.
Chen, K. H., & Shimerda, T. A. (1981). An empirical analysis of useful financial ratios.
Financial Management, 51-60.
Ciampi, F. (2015). Corporate governance characteristics and default prediction modeling for small enterprises. An empirical analysis of Italian firms. Journal of Business Research, 68(5), 1012-1025.
Daily, Catherine M., & Dalton, Dan R. (1994). Corporate governance and the bankrupt firm: An empirical assessment. (Research Notes and Communication).Strategic Management Journal, 15(8), 643.
23
Dietsch, & Petey. (2004). Should SME exposures be treated as retail or corporate exposures? A comparative analysis of default probabilities and asset correlations in French and German SMEs. Journal of Banking and Finance,28(4), 773-788.
Edmister, R. (1972). An Empirical Test of Financial Ratio Analysis for Small Business Failure Prediction. Journal of Financial and Quantitative Analysis, 7(2), 1477-1493.
Grunert, Norden, & Weber. (2005). The role of non-financial factors in internal credit ratings. Journal of Banking and Finance, 29(2), 509-531.
Hosmer, D. W., & Lemeshow, S. (2000). Introduction to the logistic regression model.
Applied Logistic Regression, Second Edition, 1-30.
Jacobson, T., Lindé, J., & Roszbach, K. (2005). Credit Risk Versus Capital Requirements under Basel II: Are SME Loans and Retail Credit Really Different? Journal of Financial
Services Research, 28(1), 43-75.
Kolari, J. W., & Shin, G. H. (2003). Assessing the profitability and riskiness of small business lenders in the banking industry. Available at SSRN 1017093.
Mathauer, I., & Imhoff, I. (2006). Health worker motivation in Africa: the role of non-financial incentives and human resource management tools. Human resources for health, 4(1), 24.
Mester, L. J. (1997). What’s the point of credit scoring?. Business review, 3, 3-16. Modina, M., & Pietrovito, F. (2014). A default prediction model for Italian SMEs: The relevance of the capital structure. Applied Financial Economics, 24(23), 1537-1554. Organisation for Economic Co-operation and Development. (2005). OECD SME and
entrepreneurship outlook. OECD.
Ohlson, J. (1980). Financial Ratios and the Probabilistic Prediction of Bankruptcy. Journal of Accounting Research, 18(1), 109-131.
Rikkers, F., & Thibeault, A. E. (2009). A structural form default prediction model for SMEs, evidence from the Dutch market. Multinational Finance Journal, 13(3/4), 229-264.
Saurina, J., & Trucharte, C. (2004). The Impact of Basel II on Lending to Small- and Medium-Sized Firms: A Regulatory Policy Assessment Based on Spanish Credit Register Data. Journal of Financial Services Research, 26(2), 121-144.
Schwaiger, W. S. (2002). Basel II: Quantitative impact study on Austrian small and medium-sized enterprises. Technical University of Vienna.
24
Treacy, W. F., & Carey, M. (2000). Credit risk rating systems at large US banks. Journal of Banking & Finance, 24(1), 167-201.
Udell, G. F. (2004). SME lending: Defining the issues in a global perspective. Kelley School
of Business.
White, L. J. (2010). Markets: The credit rating agencies. The Journal of Economic Perspectives, 24(2), 211-226.
Zmijewski, M. (1984). Methodological Issues Related to the Estimation of Financial Distress Prediction Models. Journal of Accounting Research, 22, 59-82. doi:1. Retrieved from