Building A Trading Strategy Via Multilayer Perceptron Neural Network

The author uses a multilayer perceptron neural network to build a trading strategy with the5minutes data of the CSI300stock index futures. In detail, I construct a trading strategy based on the predicted results which is provided by the multilayer perceptron network structure. After the backtesting of this trading strategy, we get a backtesting result. Then this result compared with the other result which is obtained by using wavelet filtering techniques to preprocess the input data.The results base on the three months unseen data show that the trading strategy based on multilayer perceptron neural network can generate a positive income. Further, the accumulated assets generated on the forecast period8is the most stable, while the final cumulative asset value on the forecast period32may be higher, but more volatile. In contrast, the performance on the forecast period16is the worst one. Horizontal comparison of the results of two data processing way, the results of the wavelet filter pretreatment better than the raw data results, only on the forecast period32.