Liquidity prediction in Tehran stock exchange using learning models

Liquidity in the stock market expresses how close the stock is to cash. Since the Tehran Stock Exchange is included among the world's non-cash stock exchanges and the issue of stock liquidity is one of the main concerns of investors, therefore, in this research, an attempt is made to predict the liquidity of the stock exchange using deep learning models. The statistical population includes companies active in the Tehran Stock Exchange in the years 1394-1400, and 23 companies were studied as a sample. The transaction volume and value, stock turnover ratio, Amihood, the difference between the bid and ask prices (spread) and the relative spread were measured as liquidity measure and a fully connected neural network based on multilayer perceptron (MLP), mixed deep learning (MDL) model and classical linear regression (LR) model was tested. To measure the predictive power of the models, the mean squared error (MSE) and mean absolute error (MAE) measures were calculated, and t-test was used to compare the accuracy of forecasting methods. According to the results, the prediction error rate of MDL model is lower than the other two models, and the statistical tests also confirm the significance difference in the prediction accuracy of the models at the 95% confidence level, which shows the proper performance of the proposed hybrid model compared to two other models.