Stock trading using fuzzy neural networks

Abstract

The algorithm trading has be an increasing important trading method in today’s financial

market. With the development of technology, the performance of the algorithm trading

becomes very satisfactory to investors.

One important category of the algorithm trading family, the fuzzy neural network based trading system, has been studies in the project. Historical Stock Prices of HSI and IBM downloaded from Yahoo finance are utilized as the inputs to stock price forecasting system. Using the time-delayed prices difference approach, ANFIS in the Matlab was the network trained and to process historical stock prices to generate future prices.

A simply yet effective trading strategy, “LeadLag” is exploited to make the trading decision. The original historical prices of HSI and the predicted price based on the stock prices forecasting system in the upstream are loaded into the system. As the final result, the trading system with forecasting ability ended up with the capital value 3.54 times as the original value, and the system with optimized parameter has a dramatic final capital of the 294 times of the original one. Compared with the Buy and Hold strategy end value

of 1.36 times of the original capital and end value of 1.91 times for convention trading system without forecasting, the trading system demonstrate superior performance over these two traditional methods.

In order to validate the optimized parameter in the trading system, another set of

historical prices from the Dow Jones Industrial Average is used. The same trend in the return occurs. The trading system of optimized parameter has the final capital of 11.2 times of the initial value, compared with the buy and hold 2.35 times and conventional trading system 1.72 times.