In this thesis, a model of bankruptcy prediction conditional on financial statements is presented. In addition to financial ratios which captures the aspects of liquidity, solidity and profitability a list of supplementary variables is introduced. The number of years since the establishment of the firm, the size of the firm and several industry characteristics are found useful predictors.
Apart from giving a discussion on the suggested variables, the choice of functional form is discussed: First, with reference to Laitinen & Laitinen (2000), the rate of compensation between financial ratios are discussed: The specification most commonly applied for the bankruptcy prediction model imply that the rate at which two variables can substitute another holding predicted risk unchanged will be constant. If the aspect captured by single financial ratios is considered less a substitute for any other aspect as this ratio grows, this restriction may not be appropriate. Even if the different aspects are believed to be constant substitutes; if variables are suspected to be less relevant as measures of the different aspects as the variables are measured at ''extreme'' values, the model should still allow for flexible rates of compensation. Specifically, the structure of constant compensation will make predictions sensitive to non-credible outliers. A business analyst put on the task of giving a subjective evaluation of the firm is not likely to let judgements be severely affected by a single ratio taking on odd values. In general we would like the statistical model to behave in a similar manner. A specification of the logit model that allows for flexible rates of compensation is motivated. The specification suggests that at any given level of probability estimates, the marginal effect of a variable may decline as the variable deviates from some critical value.
Second, by questioning the direct connections between financial ratios and the particular outcome of bankruptcy, a model structure which determine an upper bound on probability estimates is explored. By reference to a simple model of misclassification, the specification distinguishes between the probability of bankruptcy and the probability of insolvency. Where as the predicted probabilities of bankruptcy can be evaluated empirically, the event of insolvency is not observable. Nevertheless; conditional on the model structure, probabilities can be derived for this event as well. The model is estimated and estimation results are discussed.
An evaluation is given on the ability of the model to measure the over-all development in credit risk for the Norwegian limited liability sector. Individual probabilities of bankruptcy are multiplied with the firm’s debt to generate a prediction of expected loss in absence of recovered values. This measure is then aggregated. A regression of total loan losses for the Norwegian banking sector over the years 1989-2000 is then conducted. The fitted model is found to explain 80% of the variation in total loan losses in this period. In particular, the model does fit well with the massive loan losses that were recorded in 1991.
Preferably, a model of bankruptcy prediction should seek to capture the effect of macro variables. However, as the time dimension of the relevant panel data commonly will be limited, identifying time specific effects will be difficult. With reference to an aggregation property of the probit model, a suggestion is given how to estimate time-specific effects on aggregate data as a means to identify macro coefficients that can be included in the micro-level model.