비상장 중소기업의 부실예측과 현금보유수준

This paper addresses whether the level of cash holdings(CASH, hereafter) is useful in predicting loan defaults for unlisted SMEs in Korea. While the necessity of incorporating liquidity measures in predicting a business failure is emphasized in almost all the models developed for that purpose, the prime measure of liquidity, CASH, is rarely considered both in academe and in practice. Research findings of two papers are instrumental in launching this study. One is Altman and Sabato(2007, A&S hereafter), which replace working capital ratio with CASH as a liquidity proxy in their model to assess credit risk for U.S. SMEs. The other is Na and Kim(2008, written in Korean), which suggest that Korean SMEs with higher level of cash holdings tend to possess characteristics of operationally and financially more sound firms. The main claim of this paper is that CASH would have a robust negative relation with the probability of loan default, which is the incidence of business failure in this study. To see if the claim comes across as an empirical regularity, this paper initially examines the sign and the significance level of CASH in the model (model 1 in Tables 5～7) that utilizes five financial ratios including CASH as proposed in A&S. Since there is a possibility that CASH would not be selected in a model which starts from a different set of financial ratios, this paper collects variables from diverse sources of established provenance. This is the approach taken in Xu and Zhang(2009). While they predict business failures of Japanese listed firms with variables suggested in Altman(1968), Ohlson(1980), and distance-to-default literature, this paper predicts business failures of Korean unlisted SMEs with variables from five sources：Altman's model as modified for SMEs by Begley et al.(1996), Ohlson(1980), Beaver et al.(2005) which employed Shumway(2001)'s estimation method, distance-to-default practice, and finally A&S. A total of 19 variables are collected from these sources. The estimation is proceeded in two phases. Firstly, the final version of the estimated model whose classificatory performance is at least comparable to A&S's should include CASH as one of its predictors. Secondly, CASH in that model (model 6 in Tables 5～7) should display significantly negative coefficient. With limited ability to capitalize on the direct provision of funds from capital market, SMEs continuously solicit loans from financial institutions to stay in business and to support their growth activities. They can only secure new loans by maintaining good credit standings. When a firm defaults any part of its current loan portfolio, opportunity for tapping funds will vanish away and viability as a sustainable entity will shortly be challenged. Loan defaults can occur at any point in time during the year. Time lapsed from a specified starting point -survival time- is continuous by nature. The probability for a firm to survive up till it defaults a loan may best be estimated by the proportional hazards regression or the Cox regression. In the case of listed firms, this probability measured on a timely basis will be estimated by variables capturing up-to-date information from capital market where their securities are traded, as well as by financial ratios. Risk analyses of unlisted SMEs, however, can not benefit from continuous supply of information from capital market. They generally rely on annual financial statements. A much confronted setting in practice is the one in which a loan officer predicts a business failure within an upcoming year with currently available set of financial ratios for the firm. Survival time is reduced to an observation of failure event for the year in concern. When continuous durations are grouped into occurrences of failure events during regularly spaced intervals of time, their probabilities can properly be assessed by the complementary log-log transformations of the event probabilities according to Kalbfleisch and Prentice(2002). In our complementary log-log regression, the response variable is expressed, quite literally, as log[-log(1-p)], where p stands for the probability that a failure occurs in a specific year. When the set of explanatory variables includes year dummies, regressing the response variable on this set is tantamount to estimating parameters of the proportional hazards regression(Allison(1995, 1999), Beck et al.(1998)). This regression is referred to as Clog-log regression in the literature. Data is extracted from the database maintained for and used in rendering guarantees by Korea Credit Guarantee Fund (KODIT). Quasi-population of this study is 98,816 observations for the years 2001 through 2006 from 31,592 unlisted SMEs with total assets of at least ￦1 billion and annual sales of at most ￦60 billion(approximately US$50 million equivalents). Their fiscal years end at the end of December and their industries encompass manufacturing, construction, wholesale, retail, or services other than of financial nature. Roughly 30% of the quasi-population (29,644 firm/years) is selected as an estimation sample and 30% again of the remainder in the quasi-population (20,751 firm/years) is selected as a holdout sample, while stratified random sampling technique is applied to keep default ratio in each sample at 5.5%, the most current five-year average figure for newly issued guarantees by KODIT to the unlisted SMEs. When the estimation sample is partitioned into two groups, i.e. default firm/years and non-default or solvent firm/years, CASH in the former group is significantly lower than that in the latter group with a t-statistic of -12.18. Further, the classification result of univariate logistic regression is about 56% concordant with the actual default status. These results reported in <Table 4> hint at potential importance of CASH in distinguishing delinquent accounts from active ones. The five variables including CASH identified in A&S are all significant and all their coefficients have the expected signs in a Clog-log regression applied to the estimation sample. Specifically, a higher level of CASH is found to be associated with a lower probability of default. For comparison purpose, concordant % is reported in <Table 5>. A&S (model 1) obtains 72.4%, which is an improvement of about 4% points over the result reported in a contemporary study which employs the same database. Model 6 is the final version of stepwise Clog-log regressions. The result reported in <Table 5> shows that a total of seven financial ratios are selected to be effective in estimating loan defaults. All of them are correctly specified in terms of the signs of the coefficients and their levels of significance. While model 1 and model 6 are different in their fine details, they share certain common features. On such feature to be noticed is the consistently significant role of CASH in estimating defaults for unlisted SMEs. A preferred measure of discriminatory power under the Basel II regime is AUC(area under the ROC curve) which combines the model's records of Hit Rates and False Alarm Rates. The figures for AUC seem to suggest that model 6(75.1%) performs slightly better than model 1(73.1%). When the models 1 and 6 are estimated with industry dummies as exhorted in Chava and Jarrow(2004), the dummy variable for construction firm/years registers significance. Noticing this result, separate estimations of each model are performed for construction sub-sample and for non-construction sub-sample. Results of industry-specific regressions shown in Table 6 also support the importance of CASH in estimating default probabilities. The estimated coefficients from each model are applied to the holdout sample with 20,751 firm/years and to its industry-specific sub-samples. Validation results are satisfactorily in concert with the estimation results. <Table 7> supports the claim that CASH exerts systematic influence in predicting defaults. All these pieces of empirical evidence seem to collaborate in evincing the usefulness of CASH in predicting loan defaults of unlisted SMEs in Korea. This study does not claim that CASH should be regarded as one of the predictors for business failures for all classes of firms. It is instead an attempt to recognize, at least in assessing the credit risk for unlisted SMEs, the importance of CASH, which had not been attended to until A&S ascertained its proper role. One of the caveats of this study is the lack of qualitative characteristics represented in the models examined. An interesting remedial extension would be the study which endeavors to see if the role of CASH in predicting business failure of SMEs is replicated in a setting where financial variables are competing with reasonable proxies for qualitative characteristics linked to business sustainability.