Mean reversion is an important property for constructing efficient on-line portfolio selection strategy. The existing strategies mostly suppose that the mean reversion is multi-period symmetric or single-period asymmetric. However, the mean reversion is multi-period and asymmetric in the real market. Taking this into account, on-line strategies based on multi-period asymmetric mean reversion is proposed. With designing multi-piecewise loss function and imitating passive aggressive algorithm, we

more »

... propose a new on-line strategy WMAAMR. This strategy runs in linear time, and thus is suitable for large-scale trading applications. Empirical results on four real markets show that WMAAMR can achieve better results and bear higher transaction cost rate.