A Hybrid LSTM–Attention Framework for Multivariate Stock Forecasting with Macroeconomic Indicators

Abstract

There are numerous factors that can influence the prices of the stock, such as the company-specific indicators of performance and macroeconomic factors. Furthermore, such a combination of interdependent and intricate financial and economic variables makes it intrinsically hard to predict such price movements. In order to manage this complexity, the current research incorporates the data of the stock market, as well as the principal macroeconomic indicators, into a single prediction model.

It is proposed to use a hybrid model based on LSTM and Attention, whereby the LSTM element models the time-related factors in a periodic data set, and the Attention element distinguishes and highlights the most significant factors in the share market prices. This combination approach will allow the model to identify the timing of changes in the market as well as the foregoing effects that are less significant. Empirical evidence shows good explanatory strength, where the R 2 value is about 0.97 and a low level of error of prediction. Quadruplication. The backtesting also shows that the model has a good payoff on investment, and it continues to outperform the traditional forecasting methods.