TFT -DQN -A Dual Framework Model for Stock Market Forecasting

The financial and economic outlook of any economy is effectively conveyed through the stock markets which in turn plays a very critical part in the global economy. Investors buy and sale stock to grow financially and earn investment returns and on the other hand, companies sell stock to gather funds. But then again, the share market is very volatile as the risk is high and there is uncertainty in it. Trends in the stock market can be accurately described as difficult tasks as the methods used in today's market are not successful enough in understanding the different aspects of how consistently changing markets work. This research seeks to build on the existing literature on stock market prediction and decision making in terms of trading by presenting a new model, Dynamic Temporal Fusion Transformers (D-TFT). The model integrates Temporal Fusion Transformers (TFT), Deep-Q- Networks (DQN) to enhance the speed and performance. Multi-step prediction is applicable due to the strong forecasting capabilities. Such strong forecasting capabilities are ideal because they enable the DQN, which is a reinforcement learning algorithm, with real time trading decision making capabilities that are influenced by the ongoing volatility of the financial market. Our findings suggest that while the TFT component exhibits strong proficiency in predicting various market trends, the DQN is able to utilize these predictions to optimize decision-making within real-time trading activities. The hybrid DTFT model achieves a Mean Absolute Percentage Error (MAPE) of 0.88%, significantly outperforming existing methods. This paper suggests a novel way of addressing additional challenges such as imprecision of predictive models, high uncertainties in stock market and improving the overall decision-making process.