Stock Market Prediction

Abstract

Abstract: The research explores the application of Transformer models with time embedding for stock price prediction, focusing on a selection of five diversified stocks from the New York Stock Exchange (NYSE): JP Morgan Chase Co. (JPM) from finance, Johnson Johnson (JNJ) from healthcare, Chevron Corporation (CVX) from energy, International Business Machines Corporation (IBM) from technology, and Procter Gamble Co. (PG) from consumer goods. The study rigorously evaluates the model's performance using metrics such as Mean Absolute Error (MAE), Mean Squared Error (MSE), and Mean Absolute Percentage Error (MAPE), indicating promising outcomes with low MSE and MAE values, albeit with a relative measure of prediction accuracy (MAPE) around 2.5\%. Research aims to improve the accuracy of stock price forecasts by refining predictive models for reducing the Mean Absolute Percentage Error (MAPE) for these NYSE stock price forecasts. This involves exploring advanced feature engineering techniques, alternative machine learning algorithms, ensembling methods, and adjustments to hyperparameters and model architecture to increase prediction accuracy and reduce errors in financial time series prediction. Study highlights the effectiveness of Transformer models with temporal embedding in stock price prediction for these specific stocks, while also acknowledging areas for further improvement to increase prediction accuracy and reduce errors, providing valuable insights into the potential of advanced deep learning techniques in financial market analysis and forecasting.