Incorporating Correlated Markets' Prices into Stock Modelling with Neural Network

S.I. Ao (PRC)


Stock Index, neural network, VAR, econometrics, time series, stock modeling


It has been found that neural network is suitable for prediction in the stock markets, while it is also well known that neural network acts more or less like a black box. This is a criticism made by many experts. In this paper, the neural network is used along with the econometrics in the loose-coupling form. The purpose is to reveal more information from this hybrid system than that from working with neural network alone. Econometricians have employed many statistical methods to investigate the dynamics among the variables. VAR method has been developed for a while to investigate the relationship among the markets. Its advantage is its highly sophisticated testing of the hypothesis. With this VAR method, the result between the Asian countries with the US has been found to be positive in their recent research. While VAR method uses the whole sample period, in this paper, it has been suspected whether this kind of correlation among markets is time-independent. The result is that it is time-dependent, sometimes affected largely by the local/regional events while at other times highly correlated with the US markets. The variables studied by the VAR method are then serve as the input for the neural network to predict the stock price with highly correlated stock markets price. It is found that this hybrid system with a mixture of the financial engineering and financial economic can improve the predictability of the overall system by as much as 30% in my another paper [1] for some indices in the simplified basis.

Important Links:

Go Back