LSTM FORECASTING ON THE MACEDONIAN STOCK EXCHANGE (MSE)

Abstract

Stock market forecasting defines a guiding role for investors decision making, therefore it is an important topic of research in academia and finances. With the advancement of artificial intelligence and deep learning, its inclusion into the financial world became imperative. Machine learning algorithms focus on training models over large amounts of available stocks data and making an accurate stock forecast is challenging as the market is volatile. However, it is necessary for reducing risk in market decisionmaking. This research attempts to explore the potential of a forecasting model using Long Short-Term Memory (LSTM) in order to predict stock prices for several companies traded on the MSE. This multivariate time series forecasting, focuses on the maximum, minimum, starting and last stock price. We train the model using the monthly reports provided by the MSE for the past 15 years and in order to understand the long term market, we forecast the next several months. The conducted experiments focus on observing the error in order to assess the consistency between the predicted and the actual stock prices, showing the viability of this model for forecasting on the MSE.