An artificial intelligence approach to portfolio selection and management

This research addresses an important problem in finance, which is the portfolio selection and management problem. Modern portfolio theory and Markowitz efficient frontier start with a set of assets (securities) and generate an optimal weight combination for the optimal risky portfolio that lies on the efficient frontier. The current research work takes a step back, to identify which assets should be selected, in the first place, from a pool of available assets, and a step forward, to try to predict the expected returns for a better utilisation of the Markowitz efficient frontier. Artificial intelligence (AI) techniques are widely used in various fields of finance, which motivated the use of these techniques to find a quantitative and systematic method to construct an optimal portfolio. The genetic algorithms technique (GAs) is one of the AI techniques being successfully used to solve complex optimisation problems. GAs are deployed in this research to select the optimal portfolio based on maximising a composite objective function that maximises return, minimises risk and minimises cross-correlation between assets in the candidate portfolio. GAs are tested on two stock markets, the US stock market, represented by a pool of 40 US companies, and the Egyptian stock market, represented by a pool of 37 actively transacted companies. In both markets the generated optimal portfolio based on genetic algorithms was able to provide higher risk adjusted returns than the market index, in both the training period and the testing period. The neural networks technique is used to provide a better estimate for expected returns than the conventional historical average. It was found that, even in bearish market periods, the optimally selected portfolio, which was weekly managed using neural networks, was able to generate positive returns utilising the Markowitz efficient frontier. The research result has demonstrated the usefulness of applying the proposed AI approach represented by genetic algorithms and neural networks in active portfolio selection and management.