Enhancing predictive capabilities of ARIMA models by hybridization : a case study on OMXH25 index
Le, Trang (2024)
Pro gradu -tutkielma
2024
School of Business and Management, Kauppatieteet
Kaikki oikeudet pidätetään.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe2024061753328
https://urn.fi/URN:NBN:fi-fe2024061753328
Tiivistelmä
This thesis endeavors to adopt hybrid forecasting models and scrutinize their efficacy in financial markets relative to individual models. Specifically, ARIMA, SVM, RF, and the hybrid ARIMA-SVM and ARIMA-RF models are selected for implementation, leveraging their established predictive prowess in prior research. Focused on the Finnish stock index OMXH25, a benchmark of the Helsinki Stock Exchange, this study conducts a comprehensive case analysis.
Daily data spanning 2015-2023, encompassing opening, high, low, and closing prices of the index, forms the basis of examination. Through a partitioning approach allocating 80% to training and 20% to testing data, model parameter optimization is undertaken. This optimization, based on predictive outcomes from the training set, informs the selection of optimal parameter combinations for subsequent forecasting of the testing set. Assessment of predictive performance hinges on statistical metrics including MAE, MAPE, RMSE, and R-squared.
Notably, the hybrid ARIMA-SVM model markedly enhances predictive accuracy in financial time series one-step-ahead forecasting compared to individual models, underscoring the synergistic benefits of integrating linear and non-linear modeling techniques. Conversely, while the ARIMA-RF model exhibits improvements over standalone RF, it fails to decisively outperform the ARIMA(2,1,2) model, emphasizing the nuanced considerations in model combinations. These findings underscore the promising potential of hybrid models in financial forecasting.
Daily data spanning 2015-2023, encompassing opening, high, low, and closing prices of the index, forms the basis of examination. Through a partitioning approach allocating 80% to training and 20% to testing data, model parameter optimization is undertaken. This optimization, based on predictive outcomes from the training set, informs the selection of optimal parameter combinations for subsequent forecasting of the testing set. Assessment of predictive performance hinges on statistical metrics including MAE, MAPE, RMSE, and R-squared.
Notably, the hybrid ARIMA-SVM model markedly enhances predictive accuracy in financial time series one-step-ahead forecasting compared to individual models, underscoring the synergistic benefits of integrating linear and non-linear modeling techniques. Conversely, while the ARIMA-RF model exhibits improvements over standalone RF, it fails to decisively outperform the ARIMA(2,1,2) model, emphasizing the nuanced considerations in model combinations. These findings underscore the promising potential of hybrid models in financial forecasting.