Portfolio Selection Optimization with Adaptive Parameter Learning

This chapter investigates an adaptive online portfolio selection issue when high-frequency transaction costs are taken into account. To enhance the precision of return forecasting for risky assets, an adaptive online moving average approach is proposed, introducing a decaying factor which can be adjusted adaptively to reduce the prediction error. Subsequently, a return maximization model is established, with transaction costs incorporated into each sequential decision process. Furthermore, an online portfolio selection algorithm with adaptive parameter learning is developed to optimize the cumulative return. Numerical results demonstrate that the learning mechanism of adaptive parameters can effectively elevate the performance of strategies across multiple return and risk evaluation indicators.