Predicting daily stock prices in Mongolia using time series models

Unveiling the dynamics of Mongolia's stock market, this research delves into four critical companies in different sectors within the TOP 20 index in the Mongolian Stock Exchange, based on parameters such as market capitalization, average daily turnover, and concentration status. An in-depth analysis utilizes four diverse forecasting models - ARIMA (autoregressive integrated moving average), Prophet, NeuralProphet, and one of the deep learning model LSTM (Long short-term memory) - to predict stock prices across various terms (5, 22, and 66 business days) horizons. Employing the mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean squared error (RMSE) metrics, we compare the models' effectiveness, revealing their strengths and weaknesses for different prediction windows. The results indicate a significant challenge in achieving precise predictions, yet they also reveal a pattern: the ARIMA model performed better than others for short-term predictions, while the NeuralProphet model outperformed medium-term predictions in most of the companies examined. This study enhances our understanding of the Mongolian stock market and furnishes valuable insights for investors, traders, and analysts seeking optimal forecasting tools tailored to their specific needs and time frames.