Comparative analysis of 3D-CNN models, GARCH-ANN, and VAR models for determining equity prices

TL;DR

This study compares 3D-CNN, GARCH-ANN, and VAR models in predicting S&P 500 prices, finding that complex models like GARCH-LSTM outperform simpler ones in accuracy, especially for long-term forecasts.

Contribution

It introduces a comparative analysis of three different financial models, highlighting the superior performance of the GARCH-LSTM model in stock price prediction.

Findings

GARCH-LSTM achieved the lowest RMSE among tested models.

Complex models outperform simpler models in long-term financial forecasting.

Forecast accuracy decreases as the forecast horizon extends.

Abstract

Financial models have increasingly become popular in recent times, and the focus of researchers has been to find the perfect model which fits all circumstances; however, this has not been thoroughly achieved, and as a result, many financial models have been created. Artificial Intelligence modelling has increasingly become more popular in the financial space as an answer to the weakness of the advanced mathematical models studied in Economics. This paper introduces three commonly used models and tests them on the S&P500 to give a strong projection as to the future values of the prices. It then introduces various error metrics like the root mean square error (RMSE) to ascertain the viability of the models. The results show that a longer-term forecast indeed has more arduous consequences as there is a veer between the actual and the forecasted readings. The models can produce a strong growth estimate for businesses and policymakers to plan more appropriately in the short term. Overall, the GARCH-LSTM model produced the least RMSE which also shows that complex models tend to perform better than simpler models