Financial distress prediction models can be approximately classified into many categories. The investigation of corporate failure prediction models begins from univariate analysis (Beaver, 1966) and multivariate discriminant analysis (Altman, 1968). One of the classic works in the field of bankruptcy classification was provided by Beaver (1966). Beaver firstly employed dichotomous classification test to build financial distress prediction model.
This univariate analysis including bankruptcy indicators set the stage for the multivariate attempts, which replace several variables by one factor to detect failed firms.
The pioneering work in the area of bankruptcy prediction using multivariate techniques is generally contributed to Altman (1968). The multivariate discriminant analysis (hereafter called MDA) improves the drawback of univariate analysis which only uses one financial ratio as the variable in the model. The discriminant model included five explanatory variables that affect firm’s liquidity, profitability, leverage, solvency and activity and capture various financial dimensions of the firm. According to these predictable factors, Altman regression model calculates the discriminant score to distinguish whether the firm defaults or not. Very briefly, the variables in the regression model (called Z-score model) are: 1. Net working capital/total assets, 2. Retained earnings/total assets, 3. Earnings before interest and taxes/total assets, 4. Market value equity/book value of total debt, and 5. Sales/total assets. In the evidence from MDA model, Altman shows the discriminant score (Z-score) 2.675 as the cut-off point which could distinguish the sound firms from the default firms. If firm’s Z-score larger (smaller) than 2.675, the firm is classified as a non-failed firm (a failed firm).
Specifically, according to the sample of 33 bankrupt and non-bankrupt firms, Altman's linear MDA model was able to classify accurately 95 percent of the original sample using financial data one reporting period prior to bankruptcy. However, the accuracy of prediction in Z-score model declines as the length of time increasing. The classification accuracy
4
declined to less than 72 percent for data two years prior to bankruptcy and to 36 percent for data dating from five years before bankruptcy. Subsequent research (Deakin, 1972; Blum 1974; Sinkey, 1975) largely focused on improvements in the selection of explanatory variables which yielded the better result in terms of prediction accuracy over the 1968 Altman model.
The previous studies mostly use the 1968 Altman model as a benchmark because of its popularity in the literature. Later, Altman, Haldeman, and Naraynana (1977) constructed a second generation model with the enhancement to the original Z-score approach. Due to economical factors vary with time, the adjusted Z-score model called ZETA model incorporated seven significant variables with respect to business failures. The seven factors are Return on assets (ROA), Stability of earnings, Debt service, Cumulative profitability, Liquidity, Capitalization, and Size. The variables are respectively measured by (1) earnings before interest and taxes/total assets, (2) the standard error of estimate around a ten-year trend in ROA, (3) earnings before interest and taxes/total interest payments, (4) retained earnings/total assets, (5) current ratio, (6) common equity/total capital, and (7) total assets.
The ZETA model successfully enhanced the effectiveness in classifying bankrupt firms up to five years prior to failure on the 53 sample of manufacturers and retailers. The results show the prediction of accuracy is 96% in one year and 70% in five years prior to failure.
Generally, we use qualitative choice model when the dependent variables in the regression belong to discrete data, for example, dependent variable given 1 as failure and otherwise given 0. Ohlson (1980) firstly adopts Logit model to calculate the default probability. Logit model assumes that the probability of event happening follows Logistic distribution. The purpose of using Logit methodology is to avoid some well known problems related to MDA. The unprecedented assumption of distribution in financial distress prediction improved the drawback of MDA model which only can predict failure but cannot evaluate the default probability. The output of the application of MDA model is a score (Z-score) which is
5
indirectly related to decision policy of bankruptcy. Thus the misclassification may result from decision problem. Furthermore, there are certain statistical requirements in MDA model imposed on the distributional properties of the predictors. For instance, the variance-covariance matrices of the predictors should be the same for failed and non-failed firms groups. Also, the “matching” procedures in MDA model constrained the sample number.
Thus, the use of Logit analysis essentially avoids all of the problems discusses associated with MDA. That is why Ohlson can choose sample with 105 failed firms and 2058 non-failed firm in contrast with 53 firms in each groups. In Logit model, nine variables are: 1. Log( total assets/GNP price-level index), 2. Total liabilities/total assets, 3. Working capital/total assets, 4.
Current liabilities/current assets, 5. Bankruptcy dummy variable (one if total liabilities exceeds total assets, zero otherwise), 6. Net income/total assets, 7.Funds provided by operations/total liabilities, 8. Net income dummy variable (one if net income was negative for the last two years, zero otherwise), and 9. Change in net income. Under 0.5 as cut-off point, the predictions of accuracy are 96.12%, 95.55% and 92.84% related to the failure sample in period 1977, 1978, and 1977~1978 respectively.
Previous studies subsequently extend the application of Logit model to financial distress prediction. Lau (1987) classifies companies into five groups according to the soundness situation. Queen and Roll (1987) separate the eliminated firms into two groups according to the reason for emerge or default. Then analyze these firms via Logit model with five variables.
Hopewood, Mckeown and Mutchler (1994) state the prediction of Logit model consistent with the accountant’s opinion. Platt and Platt (1990) consider that the financial ratios would vary unsteadily over time because of economical factors such as business cycle, inflation, and interest rate. They assert that the accuracy of prediction would increase if focusing on the firms in the same industry. Hwang, Lee and Liaw (1997) predict the bankruptcy of bank in America during the period from 1985 to 1988 via Logit model with 48 financial ratios as variables. Kane, Patricia and Richardson (1998) investigate the influence of economic
6
recession on financial distress prediction. The evidence illustrates the importance of economics recession factor and shows the significance of cash flow/ total assets and net income/total assets in Logit model. Compared to the occurrence of event following Logistic distribution in Logit model, Probit model and Probabilistic model assume the occurrence of event following Normal distribution and Cauchy distribution respectively. The unprecedented application of Probit model to financial distress prediction originated with Zmijewski (1984).
However, in general, Logit model easily deals with the data, most papers usually construct financial distress prediction model based on Logit model.
The previous research on the failure of company mostly focuses on financial ratios to enhance the accuracy of financial distress prediction. However, only use firm’s internal information such as financial statement seems not enough to predict firm’s situation due to the significant effect of economical factors on these microeconomic variables (Platt and Platt, 1990; Kane, Patricia and Richardson, 1998). Suetorsak (2006) examines interactions between micro and macro variables in explaining the risk positions of East Asian banks. The analysis shows that macroeconomic policies significantly impacted the bank’s micro-economic decision. Suetorak (2006) states that macro conditions and government policies influences bank’s reactions to their microeconomic variables and the level of risk they take. Therefore, the macroeconomic factors are of importance in the investigation on the bankruptcy of firms.
In this paper, financial ratios combined with macroeconomic factors engage in the analysis of the default and failure companies to increase the accuracy of prediction.
7