Predicting stock return and volatility with machine learning and econometric models— a comparative case study of the Baltic stock market
Date
2021
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Tartu Ülikool
Abstract
Predicting the stock market is a widely researched area of study that is a challenging
task. The nature of the problem lies in correctly forecasting the direction and the
magnitude of the stock market movement. The severity of the problem exists due to the
stock market being impacted by a multitude of factors. There are numerous ways to
analyse the stock market and make appropriate investment decisions, but it is challenging
to decide the best approach. Here we show which approach is more effective in predicting
the returns and volatility of the Baltic stock market: the machine learning or econometric
approach. There is a low amount of research on using machine learning or econometric
models to predict the Baltic stock market. However, there are no comparative researches
that offer a fair comparison between the different approaches for the Baltic stock market.
Regarding results, the lowest symmetric mean absolute percentage error for the support
vector regression model is 61.90%, and for the autoregressive moving average model,
it is 165.43%. The lowest symmetric mean absolute percentage error for GARCH is
51.05%, and for the GARCH-ANN model, it is 61.65%. Overall, the machine learning
models outperform the econometric models in most of the evaluated metrics. However,
the econometric models’ results are comparable to the machine learning models’ results
in most cases.
Description
Keywords
machine learning, neural networks, autoregressive moving average, generalized autoregressive conditional heteroskedasticity