Research On The Finance Warning Model Of China's Listed Companies Based On Deep Learning

With the development of economic globalization,the environment of companies' operation becomes unsure.To the existence and development of a company,financial condition is very important.The symptom of many companies' operation crises is financial crisis.However,Roma was not built in a day.The financial crisis also experiences a process.As a result,to be ready before financial crisis happens,it is important to establish an accurate financial early-warning model.The research about financial early warning model only focused on Z-score model,logistic model and neural network model in the past.However,these models all have their defects.With the advent of the big data era,people begin to use deep learning as the tool to predict financial crisis.In order to set up a more accurate model which can help firms analyze their finance condition scientifically,deep learning is used innovatively in the paper.As a technique which can realize artificial intelligence,deep learning gets great attention quickly because of its strong computing capability and accurate test results.Based on the review of the research of other scholars,the paper states the basic theory of financial risk and financial early warning.Then in the core chapters of the paper,it sets up an index system which includes 22 indexes and chooses 2066 listed firms' data as sample to train and 54 listed firms' data as test sample.Finally the above-mentioned steps show the warning accuracy rate of discrimination in two years before listed firms taking place the financial crisis is 90.74%,the rate in three years before listed firms taking place the financial crisis is 88.89%.So the model that is more close to the time that financial crisis happens have a better accuracy rate of discrimination.Then we set up and test a logistic model using the same sample and get the warning accuracy rate of discrimination in two years before listed firms taking place the financial crisis is 75.90% and the warning accuracy rate of discrimination in three years before listed firms taking place the financial crisis is 61.10%.So we can come to an important conclusion that deep learning model is better than the logistic model.