Global Stock Market Prediction Based on Stock Chart Images Using Deep Q-Network
Prof. Priyanka Kumbhar, Aniruddha V. Kasar, Shreyas M. Borkar, Aniket S. Surve
1 Prof. Priyankya Kumbhar, Information Technology , P.G.Moze College Of Engineering
2Aniruddha Kasar, Information Technology , P.G.Moze College Of Engineering
3Shreyas Borkar, Information Technology , P.G.Moze College Of Engineering
4Aniket Surve, Information Technology , P.G.Moze College Of Engineering
---------------------------------------------------------------------***---------------------------------------------------------------------
ABSTRACT- In this study, we present a novel approach to global stock market prediction using a Deep Q-Network (DQN) with a Convolutional Neural Network (CNN) function approximator. Our model is designed to analyze stock chart images as input to make predictions on the future movements of stock prices. Remarkably, our model not only demonstrates profitability when applied to the US stock market, where it was trained, but also consistently yields positive returns in 31 different countries over a span of 12 years.
We exclusively trained our model on historical data from the US stock market and then evaluated its performance on diverse international stock markets. Our findings reveal that the portfolios constructed based on our model's predictions typically generate returns ranging from 0.1% to 1.0% per transaction, prior to considering transaction costs, across the 31 countries tested. These results suggest the presence of patterns in stock chart images that exhibit consistent correlations with stock price movements on a global scale.
Moreover, our study demonstrates the remarkable transferability of our model's predictive capabilities. Even when trained and tested on data from different countries with varying market characteristics, our model consistently demonstrates the ability to forecast future stock prices effectively. This suggests that artificial intelligence-based stock price forecasting models can be employed in relatively small and emerging markets, even in cases where limited historical data is available for training. In summary, our research underscores the potential of combining deep reinforcement learning techniques with CNNs for stock market prediction. Our model not only exhibits profitability in multiple global markets but also highlights the existence of universal patterns in stock chart images that transcend geographical boundaries. This research opens up exciting possibilities for the application of AI-based forecasting models in diverse and data-constrained financial markets, ultimately enhancing investment decision- making processes.
KEY WORDS : Stock Prediction, Live Chart Scraping, Web Scraping, CNN , LSTM, Data analysis
