| Other Abstract | Since the 1960s, Financial Early Warning Analysis of Small and Medium enterprises has been rapidly researched and promoted. There are some methods
about financial warning, for example, Univariate Analysis, Multiple Discriminant
Analysis Model, Linear Probability Model, Logistic Model, Neural Network Model,
Survival Analysis and the model based on market value an so on. Each approach
has advantages, however, we should take the numerical characteristics of sample
data into account. In this thesis, Cox is the research method. The Cox model have
the following advantages: First, The Cox model composed by the parameters and
non-parameters, It combines the advantages of parametrics and non-parametrics
model. It is a typical semi-parametric model, It does not require the specific distribution
of non-parameters. The Cox model conforms the features of realistic data.
Second, Martin(1977) firstly used Logistic models in bankruptcy warning. Because
Logistic model can solve the problem of non-normal distribution of independent
variables and the calculation is simple, many scholars applied the Logistic model
to predicting the financial crisis, However, The logistic model need to satisfy
the following condition, that is, the independent variables without the presence of
collinearity number of samples is greater than the number of regression parameters,
the Cox model does not have such requirement. Lastly, In the observation,
some samples are unable to provide complete information, but we would’t like to
give up the data, Cox models can take full advantage of these incomplete data,
II
then, the model will be more accurate.
In this thesis, we choose the appropriate variable selection method, Lasso, to
select the desired variables. The 70% sample is used for modeling, and the 30%
samples are used for prediction, combining with the analysis method of Cox, the
financial early warning model of quoted SMEs was established. The accuracy of the model is 84%, there is a certain reference value. |
