Intelligent Forecast of Stock Markets to Handle COVID-19 Economic Crisis by Modified Generative Adversarial Networks

Abstract

Abstract Stock markets have voluminous data and are subjected to uncertainty. The coronavirus disease of 2019 (COVID-19) pandemic has hit the stock markets and the trends of stock markets have accelerated share prices of few companies and has also brought freefall to certain companies. This factor highlights the importance of technical analysis of the stock markets over fundamental analysis. So, the proposed robust model for financial forecasting is built based on the technical indicators and the fake price data generated over a period of time from the stock dataset by a novel architecture of modified generative adversarial network, which uses a dense recurrent neural network as the generator and a dense spectrally normalized convolutional neural network as the discriminator. The hyperparameters used in the network model follow the two-time-scale-update rule and they are tuned by using the Bayesian optimization technique. The feature importance of the technical indicators in predicting the performance by the stock market is enhanced by the XGBoost algorithm. The generative adversarial networks (GAN) used for forecasting in the previous works suffer from problems like mode collapse and non-convergence. So, the proposed work concentrates on building a GAN model, which is stable, robust and converges to Nash equilibrium. The generated GAN model is applied on stock data from the major 100 companies of the S&P 500 stock for a period of 20 years. The modified GAN model predicts prices precise ~99 percentage, which maximizes the stock returns. The proposed modified GAN model outperforms the baseline GAN model and other state of the art approaches of forecasting on comparison.