| Summary: | 碩士 === 國立政治大學 === 金融學系 === 106 === In the past, the models used in traditional financial econometric models need to have a deep experience in the relationship between variables, and understand the causal relationship between different variables. In traditional financial econometric models, most of them ignore long time distance or delays in time series, just focus on the variable change in short period.
In order to make the most information in multivariate time series and to further improve the accuracy of forecasting Taiwan's weighted stock price index, this paper uses the recurrent neural network in deep learning to perform multivariate forecasting on the stock price index in hopes of being able to extracts valid information from variable and resolves important events that were ignored at long time distance in past time series.
The Taiwanese weighted stock price index was selected as the subject of this study, and two models of recurrent neural network and vector autoregression were selected. These two models were used to predict the stock price index and compare the two in the forecast index. which performed. The final result shows that in our past historical data, the accuracy of the recurrent neural network is significantly better than the traditional Vector Autoregression model.
|
|---|
