| Summary: | 碩士 === 國立東華大學 === 應用數學系 === 93 === In order to predict financial distress, many economists and financier have studied this topic for a long time
by using the multivariate discriminant analysis(Altman, 1968), the logit model(Ohlson, 1980), the probit model(Zmijewski, 1984), and the survival analysis(Shumway, 2001) et al. Among so many methods, Shumway refered to the logit and probit model as static models, which only use the single-period data of the company with multiple-period historical data.
Cox is the first one to deduce the proportional hazard model in the survival analysis. Many researchers used the Cox model on the topic of predicting financial distress,
but dismissed the discrete type of the variables in the way of gathering the empirical materials, and taked them as continous variables into the predicting model.
In this thesis, the discrete-time proportional hazard model(Allison, 1982) is applied to predict financial distress. I take the time-dependent variables as of the discrete type, in order to agree with the way of gathering the empirical data. By the type of empirical data, partition the duration time into [0,infinity)= U [k-1,k),
and define [k-1,k) to be the k-th period, which the unit of
k can be year, quarter days, month, or day(defined by the interval of gathering data).In addition, using the sampling method of survival analysis(Cox and Oaks, 1984) to gather all historical data of the companies. Then analyze and predict the out-of-sample probability of the companies.
So, this is as refered to be a dynamic model.
In this thesis, I estimate the values of the parameters of the logit model, the probit model, and the discrete-time proportional hazard model with three sets of indenpendent variables which incorporate Altman's(1968) 5 financial ratios variables, Zmijewski's(1984) 3 financial ratio variables, and Shumway's(2001) 2 financial ratio variables and market-driven variables integrating with Shumway's(2001) variable of the natural log of firm age.
To compare the ability of the model's predicting financial distress, and find out whenever having the best effects with one of the three models. Empirical studies demonstrate that the discrete-time proportioanl hazard model is more accurate of out-of-sample forecasts than the logit model and the probit model.
|
|---|
