Abstract
A nonlinear portfolio model was formulated by applying a nonlinear prediction method and its prediction error to the Markowitz mean-variance portfolio model. Also, the Sharpe ratio, which is a typical evaluation function of portfolio optimization, was modified to adopt stock-trading commissions and the trading-unit system, which are inevitable for portfolio rebalancing in real investment. Then, we discussed the best rebalancing frequency from the viewpoint of the trade-off between prediction accuracy and rebalancing costs. By investment simulations based on real stock data, we confirmed that shorter-term rebalancing is more effective even if we are required to pay higher commissions because short-term nonlinear prediction works better to estimate future return rates and to reduce investment risks.
