Research on Enterprise Financial risk Prediction Method Based on Regression Analysis

: The prediction and control of enterprise financial risk is an important research topic in the financial field. Over the years, the financial risks of enterprises have seriously affected the healthy development of enterprises, credit institutions, securities investors and even the whole country. However, from the current practice and theoretical research of enterprises, how to effectively evaluate and control the financial risk of Chinese enterprises is still very scarce. Therefore, this paper conducts detailed analysis on the prediction and control of financial risk based on regression analysis to better identify the financial risks of enterprises.


Introduction
At present, the operation and management mode of many large group companies is changing.The previous asset management are gradually transformed into the current capital management, and has entered the enterprise management stage which focus on finance [1].Finance plays a vital role in the development of enterprises.The life and death of an enterprise today and its subsequent growth are closely related to its financial status.With the further deepening of the theoretical system of socialist market economy in practice, Chinese enterprises are also facing more and more complex financial crises in the process of continuous innovation and development.The existence of these crises has greatly impacted the operation of enterprises, and has had a considerable impact on the social stability and economic development of the whole country [2].The high risks associated with economic activities urge enterprises to identify before the financial crisis they face, and to understand the nature of the crisis and predict the possible losses caused by risk accidents.On the basis of this identification and understanding, enterprises need to implement the most effective crisis prevention measures to minimize losses, avoid the deterioration of their financial situation and maintain their normal business activities.
Therefore, only by deeply analyzing and studying the financial crisis can the enterprise maintain a sound, stable and rapid growth.Foreign experts' research on financial risk is mainly based on Fitzpatrick's single variable early warning model and Altman's multivariable warning model.The accuracy of multivariate warning model is very high.In recent years, scholars in several countries have also carried out similar research on the basis of this model.However, the research on financial risk in China is still in its infancy, and most of the research on enterprise financial risk is still on financial indicators and managers' performance [3].On the basis of being familiar with the causes of financial risks, it is of great practical significance to create a relatively perfect risk control system [4].

Significance test of financial indicators
Non-parametric tests are used to determine whether data come from the same population when the population does not know whether it obeys the normal distribution and the distribution is unknown.There are numerous nonparametric testing methods.The Mann-Whitney U test with two independent samples is used in this paper.Mann-Whitney U test is made on the financial indicators of 32 ST companies and 32 normal companies in the t-3 years.Finally, the test results are as follows: x27 are greater than  , which means that their data are not significantly different between ST and non-ST companies and cannot be used to determine their difference, which means that they are to be excluded from the model is to be excluded from the model [5].In this way, there are 18 variables in the model.

Univariate Logistic Regression Analysis
To improve the goodness of fit of the model, we use Logistic regression analysis to check the binary relationship between each independent variable (the financial indicators selected for the first time above) and the dependent variable.We test the goodness of fit of the model for the significant financial indicators selected for the first time in t-3.The final test results are as follows:

Factor Analysis
The above nonparametric test and univariate regression analysis resulted in the selection of nine financial indicators with significant differences that are conducive to improving the model's prediction accuracy.Although it has been simplified, using these financial indicators to build models is still complicated and time-consuming, and there may be a strong correlation between them.[6] As a result, the factor analysis method is required to further simplify these financial indicators and use a few factors to describe the relationship between many indicators or factors.Based on data from the previous three years, this paper employs the factor analysis method to re-streamline the financial indicators chosen the second time.The following steps are roughly included in factor analysis: (1) Determine whether some of the original variables to be analyzed are suitable for factor analysis; (2) Rotate the constructed factor variables to make them more interpretable; and (3) Calculate the factor variable scores.Following confirmation, the following outcomes are obtained: The Table 3 shows the results of KMO test and Bartlett sphericity test.The value of KMO is 0.781.According to the standard given by statistician Kaiser, the value of KMO is greater than 0.6, which is suitable for factor analysis.In addition, the associated probability given by Bartlett sphericity test is 0.000, which is less than the significance level of 0.05.Therefore, the zero hypothesis of Bartlett sphericity test is rejected and considered suitable for factor analysis.The overall description of the original variables by the initial solution of factor analysis is shown in Table 4.The second column contains the variance contribution (eigenvalue) of factor variables, which serves as an index for determining the importance of factors.The first factor's eigenvalue is 5.02, indicating that it describes 5.02 of the total variance of the original variables 9, while the variance described by the later factors gradually decreases.The third column shows the variance contribution rate of each factor, and the fourth shows the cumulative variance contribution rate of factor variables, which shows the proportion of total variance described by the first M factors to total variance of the original variables [7].The cumulative contribution rate of the first three factors can reach 86.46 percent, indicating that the first three factors can be used for factor analysis of principal components.According to the score factor regression matrix calculated by the regression algorithm output by SPSS, the following factor score function can be obtained: { 1 = −0.047 1 + 0.27 2 + 0.011 7 − 0.153 13 − 0.052 14 − 0.028 16 + 0.268 18 + 0.274 19 + 0.268 21 2 = 0.391 1 − 0.078 2 − 0.242 7 + 0.498 13 + 0.168 14 + 0.35 16 − 0.084 18 − 0.101 19 − 0.072 21 3 = −0.09 1 + 0.004 2 + 0.852 7 − 0.272 13 + 0.358 14 − 0.027 16 − 0.002 18 + 0.021 19 − 0.001 21 (1) According to this score function, the information of nine variables can be converted into the information of three factors, that is, the variables simplified into three principal components.The following three factors are used for Logistic regression analysis.

Logistic Regression Analysis
Using the scoring functions of the above three factors, the new factor variables F1, F2 and F3 can be obtained by inputting the values of the original nine variables into the scoring function.The following results are obtained: In Table 5,B, S.E., Wald, df, Sig.,Exp(B) respectively represent variable coefficient, standard deviation, Wald score, degree of freedom, adjoint probability and coefficient logarithm.From Table 5, it can be seen that the probability of companionship of all three variables and constant quantities is very small, and if judged by the default significance level 0.05  = , all will be less than the significance level, i.e., they can meet the requirements.Therefore, it can be seen that the regression model obtained from the three factorial variables fits well.In addition, the logistic prediction probability model can be obtained from Table 5, namely ̂= (1.5462−5.54701−8.83342−6.37553)1+(1.5462−5.54701−8.83342−6.37553)(2) According to the probability model, if the calculated probability p<0.5, it can be predicted that the enterprise is risk-free; if the probability p>0.5, it can be predicted that the enterprise is risk-free, or in other words, it will be in danger of special treatment.In this way, this probability model can be used for prediction and inspection.

Testing and Prediction of Models
Using the above model formula, 64 modeling samples are brought into the calculation, and after statistics, the following results are obtained: Among them, sensitivity, also known as true positive rate, is the probability that the predicted result is also 1 among individuals actually classified as 1; Specificity, also known as true negative rate, is the probability that the predicted result is also 0 among individuals actually classified as 0; Missed diagnosis rate, also known as false negative rate, is the probability that the predicted result is zero among individuals actually classified as 1; Misdiagnosis rate, also known as false positive rate, is the probability that the predicted result is 1 among individuals actually classified as 0 [8]; The overall prediction accuracy rate is the rate of correct prediction results.Table 6 shows that the sensitivity =87.5%, specificity =96.9%, missed diagnosis rate =12.5%, misdiagnosis rate =3.1%, and the overall prediction accuracy rate is 92.2%, indicating that the overall prediction effect is very good.
In order to further prove the effectiveness of the model, we use the Z-Core model to re analyze the financial risks of 64 enterprises, and compare the results and the calculation, and after statistics, the following results are obtained: Through comparison in Table 7, the overall prediction effect of the two groups of models is similar.However, Zcore model is effective for financial early warning of Chinese enterprises, but its standard value is not suitable for China.We need to make some adjustments when applying this model.In contrast, our model is based on the Chinese market and is more suitable for the situation in China.In addition, the model we built is more brief and convenient to use.In conclusion, our model is a competitive method.
In order to better illustrate the representativeness and accuracy of the model, we selected another 48 enterprises for testing, and the results are as follows: Sensitivity =87.5%, specificity =91.67%, missed diagnosis rate =12.5%, misdiagnosis rate =8.33%, and overall prediction accuracy rate is 89.58%, which shows that this method has certain applicability and popularization.
Multiple regression analysis takes six selected indexes as independent variables and financial coefficient as observed variables, and uses stepwise regression method to study the factors that affect the financial risk of the company.The regression equation we attain is: Y=0.997x-1.

CONCLUSION
Nowadays, the financial risk of enterprises has caused many enterprises to go bankrupt or close to bankruptcy.Facing the problems brought by financial risks, we are required to avoid financial risks as much as possible with a positive attitude [9].In this paper, we establish a competitive financial risk prediction model based on regression analysis and verify its reliability and generalization.The management of financial crisis and establishment of financial early warning mechanism are essential for the sustainable and healthy development of enterprises.This requires enterprises to have a clear understanding of financial risks in the process of operation.Financial risks must be fully considered when making relevant decisions [10].Various financial information of the company should be frequently compared and analyzed.Finally, potential risk factors in the actual operation of the company should be clarified.

Table 2 .
Wald test of goodness of fit of financial indicators

Table 4 .
Contribution rate of factor variance

Table 5 .
Parameter Table of Regression Equation

Table 6 .
Comparison between predicted value based on our method and actual value of model

Table 7 .
Comparison between predicted value based on Z-core model and actual value