Predicting Share Prices Using RNN and Its Variants

Abstract

Calculating share prices is extremely complex because financial markets are intrinsically nonlinear and volatile. The traditional statistical method is not very effective in capturing the complicated patterns that apparently appear in the time series data. Therefore, because they have proven to be highly promising in modeling sequential data by learning temporal dependencies, RNNs along with their advanced variants called LSTM and GRU, have gained popularity for modeling sequential data. This paper will explore the effectiveness of the many uses of RNNs and its derivatives toward price share prediction. The results are indicated in their performance on pre-existing historical stock market data, a comparison for their accurate predictive values, and highlights their attractiveness and limitations. Advanced architectures, such as LSTM, outperform traditional RNN variants by better handling long-term dependencies and overcoming problems with vanishing gradients. The results of this study are analyzed in the context of their implications and future research directions on financial forecasting in the conclusions.

DOI - https://doi.org/10.65525/SVUP.9788199778009.2026.163-171