AI-Driven NEPSE Forecasting

Abstract

Stock price prediction in emerging markets like Nepal remains a complex and challenging task due to high volatility, limited liquidity, and the influence of investor sentiment. This study investigates the performance of three deep learning models-Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Temporal Convolutional Networks (TCN)-for predicting stock prices using both historical Open, High, Low, Close, and Volume (OHLCV) data and financial news headlines. Sentiment scores are extracted from news headlines using FinBERT to evaluate their impact on prediction accuracy. The models are compared based on Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), R-squared (R²), and Mean Absolute Percentage Error (MAPE). Experimental results indicate that GRU outperforms LSTM and TCN in both scenarios, and integrating sentiment analysis further improves predictive performance. These findings highlight the potential of AI-driven models, combined with market sentiment, for informed investment decisions in the Nepalese capital market.