Intraday stock price patterns and predictability using artificial intelligence techniques: The case of Microsoft in growing, stable and declining markets

Artificial intelligence (AI) techniques are applied to intraday stock prices for shares of Microsoft. An artificial neural network (ANN) is used to predict daily returns for shares during 1999 based on intraday opening session (IOS) data. Results are compared with linear regression forecasts. Complex neural trading systems using neural networks and genetic algorithms (GA) are built and tested using IOS data alone, then in combination with daily market indicators during three periods of relatively differing market conditions---1995 (stable market), 1999 (growing market), and 2000 (declining market). Linear regression methods are found to outperform ANNs in making daily forecasts during a period where substantial changes in market conditions occur. In more stable periods, ANNs clearly outperform linear regression in terms of exhibiting significantly lower average prediction error. Additionally, neural trading systems are found to generally provide significantly higher returns than a buy-and-hold position when tested out-of-sample during growing and declining markets, but not in a stable market.