Advanced-forecasting-of-stock-prices-with-LSTM-networks-with-attention-based-mechanisms

The stock market is renowned for its intricacy and volatility, and investors have long relied on technology to assist them in making trading decisions. This paper investigates the application of LSTM neural networks with attention mechanisms and wavelet transformations for stock price forecasting, based on the impressive findings of a prior study. The objective of the first portion of this research duplicates the results of the authors by constructing the same WLSTM+Attention model and analysing its performance using the same stock market data. In addition, important information on the profitability of a trading strategy based on this model is missing from the author's work. Therefore, the second phase of this project develops an algorithmic investment strategy based on the model's predictions in order to generate profitability metrics to support the model’s predictions. The results show that replication of the authors work was not possible, due to numerous proven flaws discovered in their work that effect the validity of their results, however, the evaluation of three separate datasets proved successful with accurate model predictions resulting in a coefficient of determination of 0.997 and trading strategy performance reaching an average profitability of 16.608%. Therefore, this study supports the application of LSTMs, attention mechanisms and wavelet transformations in financial forecasting and investment strategies and is highly competitive with existing research.