Since the global financial crisis of 2008, there has been a growing concern of government and private sectors in Korea over the crisis and the bankruptcy of shipping companies that are sensitive to changes in the economic cycle. As a result, the government has come up with various measures to support them. Shipping companies play important roles in import and export logistics of Korea and they are also resources which could be mobilized in case of emergency as a defense force from the military side. In addition, shipping industry posts biggest balance surplus in international services trade, and it''s failure could have a large economic impact on directly related shipping sub-industries and indirectly related industries, or, even further, shipbuilding and financial institutions.
In spite of the importance, there has been no discussion on the applicable bankruptcy prediction model for shipping companies so far, and in particular, it has not been developed bankruptcy prediction model as an early warning system which reflects the characteristics of the shipping companies with a strong correlation with macro-variables. As a result, from the point of view of credit-risk evaluation, financial institutions have not reflected such characteristics except when they assess the extent of insolvency of the shipping companies, resulting in no more sophisticated credit-risk evaluation system. Moreover, as many people points out in terms of shipping policy making before and after the global financial crisis, there has not been enough policy responses to the crisis, in many cases, leading to a situation in which many shipping companies have been insolvent. Such insolvency of shipping companies have surely had a negative impact on the national economy.
For these reasons, this study estimates the bankruptcy prediction models by reflecting the characteristics of the shipping industry. While previous studies estimated the models utilizing mostly financial ratios variable, this study added the business cycle variables and macro-variables to build the model a dynamic one.
As for the statistical methods, in addition to the discriminant analysis(DA) model(Altman, 1986) and logit analysis(LA) model(Ohlson, 1980) that are most commonly used in bankruptcy prediction models, the soft-computing technology, or neural network analysis(NNA) was used to estimates the models. Based on these three analysis methods, this study evaluated HR(Hit Ratio) for each model and compared among each results of analysis. In addition, for each model, the dynamic bankruptcy prediction models including the economic variables and the business cycle variables are estimated.
The samples utilized in the estimation of the models are 166 shipping companies which include 128 non-defaulters and 38 defaulters between 2000 and 2011. On the other hand, they are classified the financial statements of shipping companies during the same period, and take advantage of 1,075 firm-year observations with 815 non-defaulters and 260 defaulters. In particular, the samples consist of the estimation samples from 2000 to 2010 and the holdout samples in 2011 to verify the discriminant power.
First, 15 financial ratios variables for DA are selected in the previous studies and 11 variables in this study. t-test is performed for totally selected 26 financial ratio variables. Except for the five non-significant variables by t-test is performed to estimate the models. In the estimation results of the DA model, ten variables are shown significant in the financial model. On the other hand, it shows that 9 variables in DA considering macro-variables are significant and in particular, shipping variables, or bulk carrier fleet variable(BULSHIP) is found as a significant. Finally, in the estimation of bankruptcy prediction models that take into account the business cycle variables, it could be seen that BDI growth rate is a significant variable representative of shipping business cycles and 11 variables are finally selected.
Secondly, it is estimated the financial models with 21 financial ratios variables except the five non-significant variables in t-test results by LA. The financial model is estimated using the stepwise selection method. However, it is impossible to estimate the model using the stepwise selection method that includes the variables business cycle variables and macro-variables because many variables have been removed, and so the models are estimated by the input method. Business cycle model and macro-model by LA are estimated including the eight variables, respectively.
Third, the NNA included initially selected 26 variables of financial model without reflecting the result of t-test assuming logical processes that differ from statistical methods. Further, the NNA is analyzed able to sort between the estimation samples and holdout samples by the statistical program and so the model is estimated without distinguishing the estimation sample and holdout sample. However, some statistically significant variables from the existing models are selected for estimating bankruptcy prediction model considering marco-variables and business cycle variables.
In other words, the bankruptcy prediction models by introducing 6 variables evaluated to have a high degree of importance of NNA(except the same variable in a significant variable in DA), 9 variables of the DA and 2 variables of LA model are estimated.
Their features and comparison of bankruptcy predictive power of each model are as follows.
First, the MLP(Mutilayer Perceptron) as a sort of NNA is high HR(Hit Ratio) to be defined bankruptcy company as the bankruptcy in financial models, HR shows a 79% in the process of MLP. Then it is 76.6% in DA and 73.7% in LA.
Secondly, the main features of the macro-model is selected the bulk carrier fleet(BULSHIP) and ship price(TSD) as a significant variables. And BDI growth rate(BDIR) related with the dry bulk shipping in the business cycle model is selected as a significant variables. As a result, the variables related with dry bulk shipping plays important roles in bankruptcy prediction.
Third, MLP''s predictive power is the best in the competition models. HR shows 82.3% in the estimation of the model by MLP with the training samples. HR in the models considering business cycles variables show 78% in the DA model which improved bankruptcy predictive power than the macro-model.
Fourth, total predictive power of the MLP financial model is 92.6%, and it shows the most outstanding performance by showing a 92.9% in the macro and business cycle models.
Fifth, MLP method in bankruptcy prediction shows excellent performance as a whole, but it does not mean that there are no useful statistical methods such as DA and LA. Recent studies compare the performance of statistical methods and the artificial neural networks analysis and improve the predictive power by applying the various methodologies in a hybrid manner. In the construction of the MPL model, this study take advantage of the financial ratio variables that are estimated as significant by LA and DA and showed such results.
The estimated models presented above are limited to include the business cycles and macro-variables on the basis of financial models. As a result, there are able to contain some of the considerations of credit risk analysis C. TH. Grammenos(2010) pointed out. Therefore, in order to propose an integrated model(IM) for the prediction of bankruptcy of shipping company, it is considered as non-financial variables which C. TH. Grammenos(2010) and South Korea Credit & Evaluation(2005) suggested. The additional non-financial variables that should be taken into account to estimate the integrated model are included, based on the C. TH. Grammenos(2010)''s credit risk assessment of shipping companies.
Thus, contribution point in the study is as follows.
First, it is bankruptcy prediction models based on relatively long-term time-series data.
The second is that it is bankruptcy prediction models that reflects the characteristics of a shipping company, taking into account the variable of business cycles and macro-variables.
Third, the methods of analysis is diversified with the help of statistical analysis, or LA and DA and soft computing technology.
Fourth, overcoming the limitations of the financial model by including the business cycle variables and macro-variables, the model is expanded to the non-financial models beyond financial models and the more sophisticated methods so that the bankruptcy prediction model of shipping companies can be estimated.
The policy implications on this study are as follows.
First, the bankruptcy of the shipping companies can affect the national security in addition to the economic impact of such related industries. Therefore, the government must build an early warning system for the shipping company. By estimating the bankruptcy prediction model, a framework that can respond to impact variables in the future in this study is presented, but the early warning system which applicable to practice should be build. For this reason, it is required to ensure the organization''s human resources and the operation of the early warning system .
Second, it is necessary to improve credit evaluation system of shipping companies of financial institutions. It is believed that financial institutions value that credit risk of shipping companies is high and they have a self-sustaining credit evaluation system in each financial institution. Thus, if the fluctuations of the ship price occurs, financial institutions may request additional collateral to shipping companies. That is, if financial institutions makes a request for additional collateral in recession, the financial situation of shipping company can be worse. Therefore, there are needs to perform reasonable rating of the ship price to avoid a situation of sudden financial difficulties shipping company. In this case, the expected probability of default of shipping companies will be lower.
Third, regarding the risk management of shipping companies, a scheme that allows a group of experts to participate actively in policy making is required. When considering only the financial statements, there is a risk of distortion in risk assessment of shipping companies. Hence a non-financial model in this context should be considered and it is important to participate experts in the field of risk evaluations for constructing of non-financial model.

Keywords : shipping company, bankruptcy prediction, credit risk, early warning system, financial ratio, discriminant analysis, logit analysis, neural network analysis.