Behavioral technology credit scoring model with time-dependent covariates for stress test

European Journal of Operational Research 242 (2015) 910–919

Contents lists available at ScienceDirect

European Journal of Operational Research journal homepage: www.elsevier.com/locate/ejor

Decision Support

Behavioral technology credit scoring model with time-dependent covariates for stress test Yonghan Ju, Song Yi Jeon, So Young Sohn∗ Department of Information and Industrial Engineering, Yonsei University, 134 Shinchon-dong, Seoul 120-749, Republic of Korea

a r t i c l e

i n f o

Article history: Received 6 May 2014 Accepted 22 October 2014 Available online 13 November 2014 Keywords: Technology credit scoring Stress test Time-varying covariate Survival analysis

a b s t r a c t Technology based loan default is related not only to technology-oriented attributes (management, technology, profitability and marketability), and firm-specific characteristics but also to the economic situation after the loan. However, the default phenomenon for technology based loan has not reflected the change of economic situation. We propose a framework of utilizing a time varying Cox hazard proportional model in the context of technology based credit scoring. The proposed model is used for stress test with various scenarios of lending portfolio and economic situations. The results indicate that the firms with higher management score than average have the lower loan default rates than the firms with higher profitability or marketability score than average due to the effect of manager’s knowledge and experience and fund supply ability when they are exposed under the same economic condition. In scenario test, we found the highest default rate under stable exchange rate with high consumer price index. Moreover, firms with a high level of marketability factors turn out to be significantly affected by economic conditions in terms of technology credit risk. We expect the result of this study can provide valuable feedback for the management of technology credit fund for SMEs. © 2014 Elsevier B.V. All rights reserved.

1. Introduction Small and medium enterprises (SMEs) that have a new technology play vital roles in various industries, and their performance is an essential part of the national economy (Moon & Sohn, 2010). However, many SMEs experience financial difficulties. To survive in a fast-paced business environment, SMEs with a new technology need to raise funds for the commercialization of their technology (Jeon & Sohn, 2008; Ju & Sohn, 2014a). In order to support SMEs with high growth potential, various technology credit guarantee programs have been made available to those firms that attain high technology scores. However, high loan default rates have been reported with critical losses of technology credit fund. Inadequate and inaccurate evaluations were claimed to be associated with these serious problems (Moon & Sohn, 2010). To improve technology-based credit evaluations, many efforts have been made to develop more advanced technology credit scoring models. Previous investigations that focused on the development of advanced credit scoring models are outlined below. Using logistic regression, Sohn, Moon, and Kim (2005) suggested an improved technology credit scoring model based on the factors of individual evaluation attributes. Moon and Sohn (2010) extended ∗

Corresponding author. Tel.: +82 2 2123 4014; fax: +82 2 364 7807. E-mail addresses: [email protected] (Y. Ju), [email protected] (S. Y. Jeon), [email protected] (S.Y. Sohn). http://dx.doi.org/10.1016/j.ejor.2014.10.054 0377-2217/© 2014 Elsevier B.V. All rights reserved.

it by not only considering technology-related attributes but also environmental conditions such as firm-specific characteristics and the economic environment at the time of the application. Also, Moon and Sohn (2011) proposed a survival model for loan default that considers the total perception scoring phenomenon. Ju and Sohn (2014b) proposed a survival model that can predict the probability of loan default by a start-up SME based on technology-oriented attributes, firm characteristics, and the economic environment at the time of lending. The expected loss for a given time was obtained based on the loan default probability, exposure at default, and the loss in the event of default. These previous studies are associated with a technology credit scoring model that predicts loan default by SMEs at a given time, considering the evaluation attributes at the time of the application for funding. However, after the loan is made, attributes can change over time (Ammann & Suss, 2009; Bellotti & Crook, 2009). Therefore, timevarying covariates need to be accounted for when developing a behavioral technology credit scoring model. Moreover, potential interaction effects between a firm and the economy need to be reflected in the model. Some related papers that considered time-varying covariates are as follows. However, none of the time-varying covariate studies were done in the context of a technology credit scoring model. An examination of the interaction effects among time-varying economic indicators, firm characteristics, and technology attributes evaluated at the time of the application would have significant implications for improving technology credit funds.

Y. Ju et al. / European Journal of Operational Research 242 (2015) 910–919

In this paper, we suggest a behavioral technology credit scoring model that reflects the time-varying characteristics that are associated with the loan recipient firms for general manufacturing industry. In order to adopt time-varying variables, we employ the Cox proportional hazard model. The Cox proportional hazard (PH) model (Cox, 1972) is commonly utilized to predict loan default over time (Bellotti & Crook, 2009; Brown & Larson, 2007; Van den Poel & Lariviere, 2004). Additionally, we apply the proposed model to stress tests in an effort to analyze the impact of economic changes on a technology credit fund. A stress test is a popular tool in the area of risk management to assess the potential impact of dynamic situations that might impact the financial sector (Bellini 2013; Coffinet, Pop, & Tiesset, 2012; Huang, Zhou, & Zhu 2009). We expect that the proposed model will provide guidelines for the effective management of a technology credit fund under a changing economic environment. This paper is structured as follows. In Section 2, we review previous studies related to the Cox proportional hazard (PH) model with timevarying covariates. In Section 3, we propose a behavioral technology credit scoring model. In Section 4, we conduct a stress test based on historical scenarios. In Section 5, we discuss our study results and suggest further research. 2. Cox proportional hazard model with time-varying covariates The Cox hazard regression model proposed by Cox (1972) has been widely used in survival analyses involving time-to-event data with censoring, and it has been extended to accommodate time-varying covariates (Mata & Portugal, 1994). In this paper, we apply the Cox proportional hazard model to represent changing economic conditions after a loan to technology-based firms. We define the event as a default by a SME on a technology credit loan. In this section, we briefly introduce the Cox proportional hazard model with time-varying covariates applied to our data. The time-varying proportional hazard model estimates the relationship between the hazard rate (the likelihood of SME default), λ(t) and a number of explanatory variables (three groups of input variables: the technology-oriented attributes, the firm-specific characteristics, and the economic indicators), z(t) which are permitted to vary over time. The proportional hazard function is specified so that the explanatory variables shift an underlying baseline hazard function, λ0 (t), up or down. The time-varying proportional hazard function is expressed as follows:

λ(t; z(t)) = λ0

(t)eβ z(t).

(1)

In this equation, β is the set of coefficients to be estimated. Cox (1972) describes how β can be estimated by maximizing the partial likelihood function of the probability of default observed in the sample. β is estimated from inferences on the conditional probability of defaulting in a given time period (Boyson, 2003). It is assumed that there is a sample of n SMEs, k of which default on their loans during the observed period with default times of t1 < t2 < · · · < tk . The assumption of this model is that each default occurs in a different time period, with the defaults ordered from 1 to k chronologically. The remaining n–k SMEs are censored and have no loan default times during the sample period. However, these SMEs could default on their loan at some time after the sample period ends. Let zi (t) be z(t) for the SME with default time ti and let zj (t) be z(t) for each SME at risk at time ti . Assign ci equal to 1 if SME i defaults in observed period, otherwise zero. Ri is the set of SMEs at risk of loan default in period i. The partial likelihood function to be maximized is

L(β) =

n  i=1

 

eβ zi (ti ) j∈Ri

eβ zj (ti )

 ci .

(2)

This allows the use of the maximum likelihood method to estimate β without needing to know the baseline hazard. However, in

911

order to estimate survival probabilities, the baseline hazard is needed. This can be estimated based on the parameter estimates β given by the maximum likelihood estimation and using an estimate for the integrated baseline hazard given by Andersen (1992), as follows:

λ0 (t) =

 ti ≤t



Ci j∈Ri

eβ zj (tj )

.

(3)

With this information, the survival probability at time t can be given in terms of the hazard function:

  S(t) = P (T ≥ t) = exp −

t 0



λ(u) du .

(4)

There are several studies related to time-varying proportional models. Using the Cox regression model with time-dependent covariates, Bellotti and Crook (2009) tested whether the probability of default of an individual is affected by macroeconomic conditions such as bank interest rates, the unemployment index, and earnings after the loan. Malik and Thomas (2010) proposed a hazard rate model to predict the probability of default by a consumer. They used Cox proportional hazard rate models, and their study results showed that the rates of default by consumers are significantly affected by macroeconomic variables such as interest rates, the GDP (Gross Domestic Product), and the CPI (Consumer Price Index). Their model can be used to predict consumer credit risk in loan portfolios. Thomas (2000) has also described the important impact that dynamic economic conditions can have on credit risk. In that study, four general macroeconomic variables were analyzed. Three macroeconomic variables that were shown to be significant on the default behavior of consumers were the CPI (specifically the rate of inflation), interest rates, and GDP growth. In this paper, we use the Cox PH model with time-varying economic covariates to predict defaults by SMEs on technology credit loans. Malik and Thomas (2010) suggested that consumer-specific ratings (behavioral scores) incorporate macroeconomic variables to construct a consumer default probability model. Their model showed that consumer default is significantly associated with macroeconomic factors. Carling, Jacobson, Linde, and Roszbach (2007) used the Cox proportional hazard model to explain the survival time to default for borrowers of a major Swedish bank. The authors estimated the expected survival time of firms using macroeconomic explanatory variables. 3. Empirical study 3.1. Data and variables description In order to develop a behavioral technology credit scoring model with time-varying covariates, we apply the Cox proportional hazard (PH) model to our empirical data set based on three groups of input variables: technology-oriented attributes, firm-specific characteristics, and economic indicators. The technology-oriented attributes employed in the evaluations of these firms were divided into four sub-groups: management, technology, marketability, and profitability, which contain a total of 16 individual attributes, as displayed in Table 1 (Moon & Sohn 2010; Sohn et al., 2005) and Appendix A. These attributes are observed at the time of the application for the loan. The data set consists of 4566 cases that obtained a credit guarantee by a technology scoring system implemented in Korea between 1999 and 2004. Obtained cases were evaluated by an expert committee sent by Technology Credit Guarantee Fund that screens SMEs. A total of 1327 firms experienced loan default during this period, and the remaining firms were censored cases. Censored cases represented not only SMEs successfully repaid their loan but also they are under technology credit guarantee. According to the scorecard in Table 1, an applicant firm was assessed in terms of the 16 attributes, all of which had a pre-assigned relative weight of 5 or 10. Those evaluated on a tenpoint Likert scale are regarded to be worth two times more than those

912

Y. Ju et al. / European Journal of Operational Research 242 (2015) 910–919 Table 1 Technology-oriented evaluation attributes (Sohn et al., 2005). Factors

Variable name

Attributes

Management

P1 P2 P3 P4 P5

Knowledge management Technology experience Management ability Fund supply Human resources

Technology

P6 P7 P8 P9 P10

Environment for technology development Output of technology development (e.g. patents, certifications) New technology development Technology superiority Technology commercialization potential

5 5 5 10 10

Marketability

P11 P12 P13

Market potential Market characteristics Product competitiveness

5 5 10

Profitability

P14 P15 P16

Sales schedule Business progress (newa ) or amount of sales (oldb ) Return on investment (newa ) or profitability (oldb )

10 5 5

a b

Scale 5 5 5 5 5

New: companies less than three years old. Old: those more than three years old.

Table 2 The result of the EFA of the technology-oriented attributes. F1

F2

F3

F4

F5

F6

Manager’s knowledge and experience

Human resources and environment for technology development

Product competitiveness and technology superiority

Sales schedule and return on investment

Business progress

Management ability

F7

F8

F9

F10

F11

F12

Output of technology development

Market potential

Market characteristics

New technology development

Technology commercialization potential

Fund supply

Table 3 SME-specific characteristics (Moon & Sohn, 2010). Attributes

Explanation

1. Stock market listed 2. External audit 3. Investment by foreigners 4. Professional manager 5. Venture capital company 6. INNO-Biz 7. Production stage 8. Joint company

Listed on the KOSPI or KOSDAQ market = 1, otherwise = 0 External audit = 1, otherwise = 0 Investment by foreigners = 1, otherwise = 0 Separation between owner and manager = 1, otherwise = 0 Certified by SMBAa = 1, otherwise = 0 Certified by SMBAa = 1, otherwise = 0 After pilot production stage = 1, otherwise = 0 Consortium = 1, otherwise = 0

a

SMBA: small and medium business administration.

evaluated on a five-point scale. The final score is the simple sum of the score of each attribute, and a score of 100 points denotes the highest score (Moon, Kim, & Sohn, 2011). Many attributes may cause problems due to the potential for multi-collinearity. Accordingly, we used an explanatory factor analysis (EFA) to eliminate the multi-collinearity among them (Coldrick, Longhurst, & Hannis, 2005; Farrukh, Phaal, Probert, Gregory, & Wright, 2000; Hernriksen & Traynor, 1999; Moon & Sohn, 2010). Before the EFA, we rescale technology superiority, technology commercialization potential, product competitiveness, and the sales schedule on a five-point scale in order to eliminate weighted values. All attributes are measured on a five-point scale, with larger being better. As a result of the EFA of the technology-oriented attributes, we obtain 12 factors as shown in Table 2 that can explain 87.65 percent of the variation of the original 16 technology-oriented variables and detailed information is shown in Appendix B. In addition, we used characteristic variables of the SMEs; these variables are binary variables with 0 (non-conformance) or 1 (conformance) for each characteristic, as represented in Table 3. Next, as shown in Table 4, we considered seven economic environment variables that are potentially associated with the credit risk

of firms. They are used as time-varying covariates that are observed monthly. 3.2. Survival analysis To propose a behavioral technology scoring model, we use the Cox proportional hazard model and analyze the hazard rate in accordance with the time-varying economic environment over time. We expect that the inclusion of time-varying covariates will improve the prediction performance. The significance level used was 5 percent, and the selected results are displayed in Table 5. The estimated coefficients of the explanatory variables represent the associated effect on the hazard rate. When the variable is assigned with the parameter of a positive sign, it is associated with an increased hazard rate in terms of loan default of a firm. F8 (market potential), ECO3 (consumer price index), and stock market listing turn out to have positive associations with the loan default. This result indicates that the default rate of SMEs is increased when market potential is high because they would be faced with

Y. Ju et al. / European Journal of Operational Research 242 (2015) 910–919

913

Table 4 Economic indicators. Variable name

Economic indicators

ECO1 ECO2 ECO3 ECO4 ECO5 ECO6 ECO7

Monthly change rate of KOSPI (Korean Composite Stock Price Index) The operation index of SMEs The CPI (Consumer Price Index) The three-year earnings rate of national bonds The won-to-dollar exchange rate GDP (Gross Domestic Product) growth rate Unemployment rates

Table 5 Cox proportional hazards regression with stepwise selection results. Parameter∗

Estimate

Standard error

Chi-square

P value

F1∗ F2 F3 F4∗ F5∗ F6 F7 F8∗ F9 F10 F11 F12∗ ECO1 ECO2 ECO3∗ ECO4 ECO5∗ ECO6 ECO7 Stock market listed∗ External audit∗ Investment by foreigners Professional manager Venture capital company∗ INNO-Biz∗ Production stage Joint company

− 0.37593 − 0.00731 − 0.02534 − 0.07501 − 0.08571 − 0.02531 − 0.02646 0.13809 − 0.02764 0.02289 − 0.01676 − 0.17684 0.000562 0.00239 0.05514 0.07206 − 0.0017 − 0.01778 − 0.0406 0.73146 − 0.26479 − 0.19454 − 0.46614 − 1.42239 − 0.32815 − 0.01549 − 0.02093

0.0304 0.02927 0.02731 0.02789 0.0322 0.02837 0.02925 0.02791 0.02793 0.03208 0.02827 0.02791 0.000633 0.00666 0.0264 0.07883 0.000619 0.01976 0.0687 0.20945 0.11358 0.35782 0.35677 0.16268 0.09351 0.06245 0.07079

152.8975 0.0624 0.8611 7.2326 7.0854 0.796 0.818 24.4868 0.9797 0.5092 0.3513 40.1441 0.7888 0.1292 4.3606 0.8358 7.5238 0.8095 0.3493 12.196 5.4352 0.2956 1.7071 76.4521 12.3154 0.0615 0.0874

<0.0001 0.8027 0.3534 0.0072 0.0078 0.3723 0.3658 <0.0001 0.3223 0.4755 0.5534 <0.0001 0.3745 0.7193 0.0368 0.3606 0.0061 0.3683 0.5545 0.0005 0.0197 0.5867 0.1914 <0.0001 0.0004 0.8041 0.7674

−2log L: 20382.248 (without covariates), 19895.835 (within covariates). Root mean square error: 1.61. ∗ P-value < 0.05.

disadvantageous condition due to intensive participation of large enterprises in the area of the high market potential (Moon & Sohn, 2010). The effect of ECO3 can be interpreted that SMEs can suffer from reduced consumption activities due to an increased consumer price index. In addition, the SMEs listed in a stock market were positively associated with loan default. In 1999, the Korean government attempted to boost venture capital businesses in their efforts to mitigate the effects of the Asian financial crisis that occurred in 1997. Many SMEs were listed on stock market such as KOSDAQ (Korean Securities Dealer Automated Quotation). However, the economic boom did not last for long. At the end of 2000, this attempt caused major losses. The large positive association between a stock market listing and the hazard rate of loan default reflects such a phenomenon (Ju & Sohn, 2014b; Moon & Sohn, 2010). On the other hand, the F1 (manager’s knowledge and experience), F4 (sales schedule and return on investment), F5 (business progress), ECO5 (the won-to-dollar exchange rate), F12 (fund supply), External audit, Venture registration, and INNO-Biz registration have negative association with loan default rate of SMEs. 4. Stress test Credit risk analysis is an important issue related to loan portfolio management, because it plays a crucial role in financial institutions

(Jakubik, 2007). Stress tests based on the estimation of a technology credit risk in various scenarios have been used as an important tool in the area of financial risk management. These tests seek to measure the sensitivity of a group of institutions, or even an entire financial system, to economic shocks (Hilbers & Jones, 2004; Virolainen, 2004). In this paper, we apply the behavioral technology credit scoring model developed in Section 3 to predict the loan default rate for use in a stress test. To measure the stress test results over time, the time-varying values of variables in Table 5 need to be explored under various scenarios. For unknown economic indicators (ECO3 and ECO5), we apply historical scenarios by adopting previous economic situations to which technology credit loans were exposed. Historical scenarios are used here because it is difficult to set up long-term monthly economic indicators reflecting economic conditions that can interact with each other in the future. First, we set up a scenario which represents the unstable economic situation of the global financial crisis which started in the middle of 2008. In order to include this crisis in this economic scenario, we consider the period from October 2007 to September 2009. During this period, all economic indicators including the consumer price index (ECO3) and exchange rates (ECO5) reflect a climate of global financial crisis. Second, we consider the period from December 2009 to November 2011. As in the previous scenario, we consider two years before and after December 2010. This situation represents a stable economic

914

Y. Ju et al. / European Journal of Operational Research 242 (2015) 910–919

Fig. 1. The trend of ECO3 (consumer price index).

Fig. 2. The trend of ECO5 (the won-to-dollar exchange rate).

situation. During this period, foreigners purchased blue-chip stocks on a large scale because various Korean economic indicators showed an increase in economic growth. In accordance with this situation, the KOSPI (Korean Composite Stock Price Index) increased significantly and the overall economic environment improved. According to Kim, Seo, and Sohn (2011), the default rate of SMEs is closely related to macroeconomic conditions. Figs. 1 and 2 show the changes of economic variables ECO3 and ECO5 that turn out to be significant, as displayed in Table 4. In the case of ECO3 (consumer price index) as displayed in Fig. 1, the period from October 2007 to September 2009 reflected a low

consumer price index. On the other hand, the period from December 2009 to September 2011 reflected a relatively high consumer price index. A high consumer price index can depress consumption activities (Cunado & Pérez de Gracia, 2005). Fig. 2 shows two phases of activity related to the won-to-dollar exchange rate. During the period from October 2007 to September 2009, the won-to-dollar exchange rate fluctuates significantly. Initially during this period, the won-to-dollar exchange rate increased steadily for 12 months. However, after this 12 month period, a dramatic increase occurred that lasted for the next six months, but then the rate decreased after that. In contrast, the period from December 2009 to September 2011, which represents the

Y. Ju et al. / European Journal of Operational Research 242 (2015) 910–919

915

Table 6 The twelve scenarios for the stress tests. Scenarios #

Technology characteristics

Economic condition

Sample size

Scenario 1-1

The management factor (F1 and F12) is higher than average

1) Unstable economy representing the situation in the period from October 2007 to September 2009 (covering the global financial crisis) with a low consumer price index 2) Stable economic situation representing the situation in the period from December 2008 to November 2011 with a high consumer price index 3) Unstable economy representing the situation in the period from October 2007 to September 2009 (covering the global financial crisis) with a high consumer price index 4) Stable economic situation representing the situation in the period from December 2008 to November 2011 with a low consumer price index

1330

The profitability factor (F4 and F5) is higher than average

1) Unstable economy representing the situation in the period from October 2007 to September 2009 (covering the global financial crisis) with a low consumer price index 2) Stable economic situation representing the situation in the period from December 2008 to November 2011 with a high consumer price index 3) Unstable economy representing the situation in the period from October 2007 to September 2009 (covering the global financial crisis) with a high consumer price index 4) Stable economic situation representing the situation in the period from December 2008 to November 2011 with a low consumer price index

988

The marketability factor (F8) is higher than average

1) Unstable economy representing the situation in the period from October 2007 to September 2009 (covering the global financial crisis) with a low consumer price index 2) Stable economic situation representing the situation in the period from December 2008 to November 2011 with a high consumer price index 3) Unstable economy representing the situation in the period from October 2007 to September 2009 (covering the global financial crisis) with a high consumer price index 4) Stable economic situation representing the situation in the period from December 2008 to November 2011 with a low consumer price index

2042

Scenario 1-2 Scenario 1-3 Scenario 1-4

Scenario 2-1 Scenario 2-2 Scenario 2-3 Scenario 2-4

Scenario 3-1 Scenario 3-2 Scenario 3-3 Scenario 3-4

second phase does not have significant fluctuations in the exchange rate in comparison to the period from October 2007 to September 2009. The average exchange rate of the period from October 2007 to September 2009 is 1158.637, and the won-to-dollar exchange rate for the period from October 2007 to September 2009 is in the range of 914.81–1453.35. The average won-to-dollar exchange rate for the period from December 2009 to September 2011 is 1132.02 and the won-to-dollar exchange rate for the period from October 2007 to September 2009 ranges from 1058.49 to 1214.02. That is, the average won-to-dollar exchange rate of the two situations during 24 months is similar; however, the fluctuation ranges of the exchange rates differ. Based on the this information, we set up four economic situations using macroeconomic indicators: 1) an unstable economic situation (exchange rate) with a low consumer price index, 2) a stable economic situation (exchange rate) with a high consumer price index, 3) an unstable economic situation with a high consumer price index, and 4) a stable economic situation with a low consumer price index. According to the results of the behavioral technology credit scoring model in Section 3, F1 (manager’s knowledge and experience), F4 (sales schedule and return on investment), F5 (business progress), F8 (market potential), F12 (fund supply), ECO3 (consumer price index), ECO5 (the won-to-dollar exchange rate), stock market listing, External audit, Venture registration, and INNO-Biz registration have significant associated with the rate of loan default. F1 and F12 represent the management of the technology and Factor 8 represents the marketability of the technology; F4 and F5 show the profitability of the technology. Thus, we cluster the entire sample into three groups according to the values of F1, 4, 5, 8, and 12, and then we consider the twelve situations that reflect changes in the four economic situations, as displayed in Table 6. Scenarios from 1-1 to 1-4 consist of SMEs that were evaluated as having higher management scores than average under the four economic conditions. Scenarios from 2-1 to 2-4 consist of SMEs that were evaluated as having higher marketability scores than average under the four economic conditions. Scenarios from 3-1 to 3-4 consist of SMEs that were evaluated as having higher profitability scores than average under the four economic conditions. In all scenarios, the scores of the remaining technology-oriented at-

tributes and levels of firm-specific characteristics retained the original values for each firm. Next, we estimate the technology credit risk based on the timevarying behavioral technology credit scoring model. In general, credit risk is obtained as a product of the risk probability of default (PD), the loss given default (LGD), and the exposure at default (EAD) (Altman & Sabato, 2005; Altman & Saunders, 2001; Claseens, Krahnen, & Lang, 2005; Jeon & Sohn, 2008; Kolbe & Zagst, 2010; Pirotte & Vaessen, 2008). PD is obtained from a stress test based on the result of the behavioral technology credit scoring model. EAD represents the loss that depends on the bank’s exposure to the borrower at the time of default; we used the loan amount as the EAD value. LGD was not considered because SMEs are assigned the same LGD value in general. To compare the technology credit risk in terms of the portfolios described by our twelve scenarios, we adjust the amount provided in each scenario by the total credit guarantee fund, as follows. In the first step, we calculate the proportion of loans assigned to SMEs by the total credit guarantee funds assigned to all recipient SMEs (4566 firms) in each scenario. In the second step, the actual guaranteed funds assigned to each SME are divided by this ratio to represent the EAD for each SME. EAD multiplied by the loan default probability at time t of each scenario is defined as the technology credit risk at time t. The total technology credit risk during 24 months under four economic scenarios in terms of technology characteristics such as management, profitability or marketability factor is summarized in Table 7. The stress test results based on the twelve proposed scenarios are displayed in Figs. 3–5 in terms of the loan default probability for two years. The results show the changes in the loan default probability according to the scenario. The x-axis represents the elapsed month as t while the y-axis shows the probability of loan default at each t. In the case of SMEs with technology associated with a high level of management factors, the loan default probability of both scenario 1-2 and scenario 1-4 tended to increase during the elapsed 24 month period. The average loan default probabilities of the four scenarios in Fig. 3 are 8.7 percent, 13.4 percent, 12.1 percent, and 9.7 percent. In the case of SMEs with technology associated with a high level of

916

Y. Ju et al. / European Journal of Operational Research 242 (2015) 910–919 Table 7 The average loan default probability and total technology credit risk (unit: US dollar, $). The average loan default probability (percent)

The total technology credit risk (US dollar)

Scenario

Technology characteristics

Scenario 1-1 Scenario 1-2 Scenario 1-3 Scenario 1-4

The management factor (Factors 1 and 12) is higher than average

8.7 13.4 12.1 9.7

3,769,823,587 5,827,529,229 5,258,422,781 4,187,737,593

Scenario 2-1 Scenario 2-2 Scenario 2-3 Scenario 2-4

The profitability factor (Factors 4 and 5) is higher than average

12.4 18.5 16.8 13.7

3,371,327,790 5,079,730,147 4,615,800,887 3,726,278,834

Scenario 3-1 Scenario 3-2 Scenario 3-3 Scenario 3-4

The marketability factor (Factor 8) is higher than average

15.9 23.0 21.2 17.5

9,258,035,838 13,582,777,238 12,437,389,097 10,175,441,219

0.400

Default probability

0.350 0.300 0.250 scenario 1-1 scenario 1-2 scenario 1-3 scenario 1-4

0.200 0.150 0.100 0.050 0.000

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Elapsed month Fig. 3. Stress test results of funded firms with high management factors (Factors 1 and 12).

0.500 0.450

Default probability

0.400 0.350 0.300 scenario 2-1

0.250 0.200

scenario 2-2 scenario 2-3

0.150

scenario 2-4

0.100 0.050 0.000

1

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Elapsed month

Fig. 4. Stress test results of funded firms with high profitability factors (Factors 4 and 5).

profitability factors, the average loan default probabilities of scenarios 5-8 in Fig. 4 are 12.4 percent, 18.5 percent, 16.8 percent, and 13.7 percent. In the case of SMEs with technology associated with a high level of marketability factors, the average loan default probabilities of the four scenarios in Fig. 3 are 15.9 percent, 23.0 percent, 21.2 percent,

and 17.5 percent. The results of the stress test are represented in Table 7. In terms of technology-oriented attributes, firms that are evaluated as having a high score on marketability factors showed higher loan default rates than the other firms that were evaluated as having a

Default probability

Y. Ju et al. / European Journal of Operational Research 242 (2015) 910–919

0.600 0.550 0.500 0.450 0.400 0.350 0.300 0.250 0.200 0.150 0.100 0.050 0.000

917

scenario 3-1 scenario 3-2 scenario 3-3 scenario 3-4

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Elapsed month Fig. 5. Stress test results of funded firms with high marketability factors (Factor 8).

high score on management or profitability in all economic scenarios. The loan default rate of scenario 3-2, in particular, is the highest. This result reflected the ECO3 (consumer price index), ECO5 (the won-todollar exchange rate), and F8 (market potential) that have a positive sign on the loan default. It can be interpreted that the firms having high market potential would have high loan default rates when economy is depressed due to high consumer price index and stable exchange rate. According to the result of estimated total technology credit risk in Table 7 the SMEs with technology associated with a high level of marketability factors have the highest technology credit risk under scenario 3-2, because it provided the highest average loan default probabilities. On the other hand, scenario 2-1 resulted in the lowest technology credit risk compared to other economic situations. These results imply that firms with good profitability capabilities respond well to fluctuating economic environments, such as an exchange rate fluctuation. However, although firms have good profitability capabilities, these firms have difficulty dealing with a high technology credit risk when economic consumption activity slows down. The SMEs with technology associated with a high level of marketability have higher loan default probabilities under a stable situation with a high consumer price index (Scenario 3-2) than more dynamic economic situations; they also showed the highest total technology credit risk. In addition, compared with other SMEs, firms with technology associated with a high level of marketability factors turn out to be significantly affected by economic conditions in terms of technology credit risk. Overall, our stress test indicated that profitability of technology plays an important role in decreasing default under various economic circumstances.

5. Conclusion Technology credit guarantees have been issued to SMEs based on a high degree of growth potential, as determined by a technology credit scoring model. Many previous studies attempted to improve the technology credit scoring model using technology-oriented factors, firm-specific characteristics, and the economic conditions at the time of funding. However, they did not consider varying economic conditions over time after the loan application. This type of approach is unrealistic because the effects of economic situations can change after the loan application has been submitted. In this paper, we suggested a technology credit scoring model with time-varying covariates using the Cox proportional hazard model. Additionally, stress

tests reflecting changes in economic situations, such as stable and unstable fluctuations in terms of the consumer price index, were performed. The stress test results showed that the loan default rates of firms that were evaluated as having high scores in marketability factors were higher under various economic situations than other SMEs that were evaluated as having high scores in management or profitability. Firms with a high management, profitability, or marketability score have a low loan default probability under an unstable economic situation with a low consumer price index, as compared to other economic situations. Firms with a high marketability score have higher loan default probability than the other firms under a stable economic situation with a high consumer price index (scenario 3-2). In the results of the survival analysis, ECO3 is positive with regard to loan default, and F8 (market potential) is related to marketability evaluation attributes (scenario 3-2). This result can be explained by the fact that when funding are supported to firms which received the loan depending on good evaluation of market potential under depressed economic situation, those firms may be suffering from loan default. This result provides a measure of risk management pertaining to appropriate funding for a loan portfolio of SMEs. In addition, in terms of average loan default probability, SMEs with high marketability were resistant to change in the economic conditions while SMEs with high management score were sensitive. The results and process of our model can be applied to many other areas. However, limitations also exist. In this study, we analyzed empirical data from 1999 to 2004. Future study can extend recent data to establish a more accurate technology credit scoring model. For further research, financial variables of the firms and different industries can be added to the model with the help of a richer data. In addition, one can analyze the model in terms of resurrection after the loan default in order to build up more adequate support policies for technology based SME, when such data are available. Further studies that account for changes in the levels of technology-oriented competitiveness are also necessary.

Acknowledgment This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIP) (2013R1A2A1A09004699).

918

Y. Ju et al. / European Journal of Operational Research 242 (2015) 910–919

Appendix A. Examples of technology oriented attributes (excerpt). Factors

Attributes

Sub attributes

Description

Management

Knowledge management

Management of technical expert

 Evaluating incentive system for technical expert  Evaluating management by objective  Existing risk management team  Experience of overcoming a crisis

Ability to cope with crisis Technology experience

 Employment history in same field

Level of technology experience of CEO Level of technological knowledge

 Technological certification level

Appendix B. Results of the factor analysis of technology-oriented attributes. Indicator

Factor1

Factor2

Factor3

Factor4

Factor5

Factor6

Factor7

Factor8

0.0512 − 0.0221

− 0.0095 0.00766

0.13331 − 0.0837

0.05073 0.0439

− 0.0311 0.02089

Factor9

Factor10

Factor11

Factor12

0.00849 0.00727

0.03846 − 0.0341

0.04949 − 0.0349

0.00637 0.03189

P2 P1

0.90723 0.90554

− 0.0074 0.13544

− 0.0086 0.02852

P5 P6

− 0.0398 0.23323

0.8182 0.75408

0.06017 0.11305

0.09392 0.03333

0.05815 0.02144

0.35625 − 0.1274

− 0.0391 0.26909

0.07612 0.07906

0.02803 0.07051

0.04125 − 0.0464

0.0549 0.10995

0.04915 0.14106

P13 P9 P14

− 0.02 0.04931 0.00495

− 0.0111 0.18937 0.13007

0.85434 0.75061 0.04011

0.12597 − 0.0096 0.91978

0.02291 0.03783 − 0.0081

0.18575 − 0.227 0.02346

− 0.0321 0.18395 − 0.0396

− 0.0432 0.13443 − 0.0118

0.11442 0.04276 0.00833

0.12573 − 0.0487 0.10476

0.03375 0.04567 0.06445

− 0.0747 0.14848 0.05505

P16 P15

0.05175 − 0.0112

− 0.0713 0.0694

0.15023 0.02311

0.56581 0.07226

0.40533 0.95737

0.17301 0.01984

0.16378 − 0.0297

0.18709 0.003

0.14656 0.01508

− 0.0026 0.04235

0.07896 0.0167

0.04064 0.04343

0.0368

0.00181

0.09858

0.20483

0.0253 − 0.005

0.07748 0.02658

0.01663 0.03198

0.00539 0.01444

0.01535 0.97835 0.071 0.01655

0.01004 0.06856 0.97488 0.04222

0.03675 0.01513 0.04194 0.94733

P3

− 0.0029

0.05331

0.15226

P7 P11

0.08233 − 0.0119

0.13345 0.11031

0.09549 0.05354

0.09053 0.02455 0.05795

P12 P8 P10 P4

0.01361 0.00431 0.01303 0.03477

0.06943 − 0.0016 0.11722 0.13459

0.12995 0.07124 0.06278 0.03995

0.06515 0.0981 0.09666 0.07286

0.04498 − 0.0096 0.0258 0.03547 0.04245 0.02946 0.05275

0.86738

0.0861

0.07994 0.03614

0.94738 − 0.0127

0.03464 0.00467 0.0946 0.19643

0.02728 0.07337 0.02027 0.00859

References Altman, E. I., & Sabato, G. (2005). Effect of the new Basel capital accord on bank capital requirements for SMEs. Journal of Financial Services Research, 28(1-3), 15–42. Altman, E. I. & Saunders, A., (2001). Credit ratings and the BIS reform agenda The NYU Salmon Center/University of Maryland Research Paper, pp. 1–38. Ammann, M., & Suss, S. (2009). Asymmetric dependence patterns in financial time series. The European Journal of Finance, 15(7-8), 703–719. Andersen, P. K. (1992). Repeated assessment of risk factors in survival analysis. Statistical Methods in Medical Research, 1, 297–315. Bellini, T. (2013). Integrated bank risk modeling: A bottom-up statistical framework. European Journal of Operational Research, 230(2), 385–398. Bellotti, T., & Crook, J. (2009). Credit scoring with macroeconomic variables using survival analysis. Journal of the Operational Research Society, 60(12), 1699–1707. Boyson, N. M. (2003). Why do experienced hedge fund managers have lower returns? (Working Paper). Purdue University. Brown, J., & Larson, C.E. (2007). The issue of retail credit risk seasoning and its impact upon Basel II PD estimation. Retrieved from http://www.promontory.com/assets/ 0/78/110/168/c19f3640-58b5-4ebe-afe3-abddff624f0e.pdf Coffinet, J., Pop, A., & Tiesset, M. (2012). Monitoring financial distress in a high-stress financial world: The role of option prices as bank risk metrics. Journal of Financial Services Research, 44(3), 229–257. Coldrick, S., Longhurst, P., & Hannis, J. (2005). An R&D options selection model for investment decisions. Technovation, 25(3), 185–193. Cox, D. (1972). Regression models and life tables (with discussion.). Journal of the Royal Statistical Society, 34, 187–220. Carling, K., Jacobson, T., Linde, J., & Roszbach, K. (2007). Corporate credit risk modeling and the macroeconomy. Journal of Banking and Finance, 31, 845–868. Claseens, S., Krahnen, J., & Lang, W. W. (2005). The Basel II reform and retail credit markets. Journal of Financial Services Research, 28(1-3), 5–13. Cunado, J., & Pérez de Gracia, F. (2005). Oil prices, economic activity and inflation: Evidence for some Asian countries. The Quarterly Review of Economics and Finance, 45(1), 65–83. Farrukh, C., Phaal, R., Probert, D., Gregory, M., & Wright, J. (2000). Developing a process for the relative valuation of R&D programmes. R&D Management, 30(1), 43–54.

0.03852 − 0.0153 0.975 − 0.004 0.02672 0.03421 0.01519

0.9802 0.01509 0.0106 0.03821

Henriksen, A. D. P., & Traynor, A. J. (1999). Practical R&D project-selection scoring tool. IEEE Transactions on Engineering Management, 46(2), 158– 170. Hilbers, P., & Jones, M. T. (2004). What If . . . ? Stress testing uses different scenarios to determine how vulnerable countries’ financial systems are to shock. Finance and Development-English Edition, 41(4), 24–27. Huang, X., Zhou, H., & Zhu, H. (2009). A framework for assessing the systemic risk of major financial institutions. Journal of Banking and Finance, 33, 2036– 2049. Jakubik, P. (2007). Macroeconomic environment and credit risk. Czech Journal of Economics and Finance, 57(1-2), 60–78. Jeon, H. J., & Sohn, S. Y. (2008). The risk management for technology credit guarantee fund. Journal of the Operational Research Society, 59(12), 1624–1632. Ju, Y. H., & Sohn, S. Y. (2014a). Updating a credit-scoring model based on new attributes without realization of actual data. European Journal of Operational Research, 234(1), 119–126. Ju, Y., & Sohn, S. Y. (2014b). Stress test for a technology credit guarantee fund based on survival analysis. Journal of the Operational Research Society Published online, 12, March. Kim, S. S., Seo, J. Y., & Sohn, P. D. (2011). SMEs’ capital structure behavior on cash holdings to Korean financial crisis: The evidence from emerging market. African Journal of Business Management, 5(34), 13095–13115. Kolbe, A., & Zagst, R. (2010). Valuation of reverse mortgages under (limited) default risk. The European Journal of Finance, 16(4), 305–327. Malik, M., & Thomas, L. C. (2010). Modelling credit risk of portfolio of consumer loans. Journal of the Operational Research Society, 61(3), 411–420. Mata, J., & Portugal, P. (1994). Life duration of new firms. Journal of Industrial Economics, 42(3), 227–245. Moon, T. H., & Sohn, S. Y. (2010). Technology credit scoring model considering both SME characteristics and economic conditions: The Korean case. Journal of the Operational Research Society, 61(4), 661–675. Moon, T. H., & Sohn, S. Y. (2011). Survival analysis for technology credit scoring adjusting total perception. Journal of the Operational Research Society, 62(6), 1159–1168. Moon, T. H., Kim, Y., & Sohn, S. Y. (2011). Technology credit rating system for funding SMEs. Journal of the Operational Research Society, 62(4), 608–615. Pirotte, H., & Vaessen, C. (2008). Residual value risk in the leasing industry: A European case. The European Journal of Finance, 14(2), 157–177.

Y. Ju et al. / European Journal of Operational Research 242 (2015) 910–919 Sohn, S. Y., Moon, T. H., & Kim, Y. H. (2005). Improved Technology scoring model for credit guarantee fund. Expert Systems with Applications, 28(2), 327–331. Thomas, L. C. (2000). A survey of credit and behavioural scoring: forecasting financial risk of lending to consumers. International Journal of Forecasting, 16(2), 149–172.

919

Van den Poel, D., & Larivière, B. (2004). Customer attrition analysis for financial services using proportional hazard models. European Journal of Operational Research, 157(1), 196–217. Virolainen, K. (2004). Macro stress testing with a macroeconomic credit risk model for Finland (Bank of Finland Discussion Paper, 18).