Multi-Step-Ahead Stock Market Prediction Based on Least Squares Generative Adversarial Network

摘要

In this paper, the prediction of the stock market closing price using the least squares generative adversarial network (LSGAN) is addressed. In the data preprocessing phase, we perform feature generation by adding several technical indicators and oscillators. In addition, to decrease the noise in stock index and smooth the data, wavelet transformation is used. Removing outliers is also carried out by using z-score method. In discriminator, we use least-squares loss function for LSGAN in preference to the binary cross entropy in regular GANs. In order to evaluate GAN and LSGAN on the standard and poor's 500 (S&P 500) stock index, several simulation studies have been conducted. The results show that the proposed method outperforms the other ones in terms of the performance criteria.