Application Of Functional Data Analysis To Stock Price Forecasting

Forecasting stock prices has been regarded as one of the most challenging tasks, because the stock price is influenced by many factors, such as public policy, changes in economic environment or traders’ anticipation, or trading techniques and investment psychology. There has been lots of research work on stock price prediction, Models, like ARIMA, need to impose stationarity or linearity assumption, which may not always hold in practice. Therefore, it is necessary to propose a general model that requires fewer assumptions. For example, Wang and Leu(1996) proposes to use neural network model to relax the linearity assumption, but this is applied to model low-frequency observations. As the development of the financial market advances, it becomes more important to study techniques for high-frequency data. In this paper, combining the idea of autoregressive regression and nonparametric regression, we propose a mixed model for opening price of the stock. The nonparametric part is the integration of the daily price multiplies an unknown function. In order to estimate the nonparametric part, we apply functional principle component analysis on the daily stock price. In our empirical study, we use CSI300data and the results seem to indicate that our mixed model could perform better than the traditional autoregressive model.