TY - GEN
T1 - Two-factor and ARIMA-LS-SVR models for forecasting of EUA futures prices
AU - Han, Jun S.
AU - Kordzakhia, Nino
AU - Shevchenko, Pavel V.
AU - Trück, Stefan
PY - 2025
Y1 - 2025
N2 - We analyse the h-step forecasts of futures prices, using the extended two-factor model in Han et al. (On autoregressive measurement errors in a two-factor model. 2021-22 MATRIX Annals (2023)) based on the Schwartz-Smith two-factor model in Schwartz and Smith (Manag. Sci. 46(7), 893–911 (2000)). The two-factor model assumes that the short- and long-term components of the logarithm of futures prices are mean-reverting. Moreover, the short- and long-term factors represent correlated latent Ornstein-Uhlenbeck processes and will be jointly estimated, along with model parameters, using the Kalman filter through the maximum likelihood method. Furthermore, we assume that the error terms in the measurement equation system are inter-dependent and serially correlated. A comparative analysis has been carried out between three models: (a) the reduced-form model in Schwartz and Smith (Manag. Sci. 46(7), 893–911 (2000)), (b) the full model in Han et al. (On autoregressive measurement errors in a two-factor model. 2021-22 MATRIX Annals (2023)), and (c) a hybrid model, where futures prices follow the ARIMA process. In (c), we will use the least squares support vector regression (LS-SVR) method for the prediction of the residuals of the multivariate ARIMA process. Historical daily prices of European Union Allowance (EUA) futures contracts from January 30, 2017, to April 1, 2022, were used for illustration in this study.
AB - We analyse the h-step forecasts of futures prices, using the extended two-factor model in Han et al. (On autoregressive measurement errors in a two-factor model. 2021-22 MATRIX Annals (2023)) based on the Schwartz-Smith two-factor model in Schwartz and Smith (Manag. Sci. 46(7), 893–911 (2000)). The two-factor model assumes that the short- and long-term components of the logarithm of futures prices are mean-reverting. Moreover, the short- and long-term factors represent correlated latent Ornstein-Uhlenbeck processes and will be jointly estimated, along with model parameters, using the Kalman filter through the maximum likelihood method. Furthermore, we assume that the error terms in the measurement equation system are inter-dependent and serially correlated. A comparative analysis has been carried out between three models: (a) the reduced-form model in Schwartz and Smith (Manag. Sci. 46(7), 893–911 (2000)), (b) the full model in Han et al. (On autoregressive measurement errors in a two-factor model. 2021-22 MATRIX Annals (2023)), and (c) a hybrid model, where futures prices follow the ARIMA process. In (c), we will use the least squares support vector regression (LS-SVR) method for the prediction of the residuals of the multivariate ARIMA process. Historical daily prices of European Union Allowance (EUA) futures contracts from January 30, 2017, to April 1, 2022, were used for illustration in this study.
KW - Forecasting
KW - Futures prices
KW - Two-factor model
KW - LS-SVR
UR - https://doi.org/10.1007/978-3-031-69750-0
U2 - 10.1007/978-3-031-69750-0_18
DO - 10.1007/978-3-031-69750-0_18
M3 - Conference proceeding contribution
SN - 9783031697494
SN - 9783031697524
T3 - Contributions to Statistics
SP - 299
EP - 312
BT - Time series analysis and forecasting
A2 - Valenzuela, Olga
A2 - Rojas, Fernando
A2 - Herrera, Luis Javier
A2 - Pomares, Héctor
A2 - Rojas, Ignacio
PB - Springer, Springer Nature
CY - Cham
T2 - 9th International Conference on Time Series and Forecasting Time Series Analysis and Forecasting (ITISE 2023)
Y2 - 12 July 2023 through 14 July 2023
ER -