Abstract
Over the last three decades, accurate modeling and forecasting of electricity prices has become a key issue in competitive electricity markets. As electricity price series usually exhibit several complex features, such as high volatility, seasonality, calendar effect, non-stationarity, non-linearity and mean reversion, price forecasting is not a trivial task. However, participants of electricity market need price forecast to make decisions in their daily activity in the market, such as trading, risk management or future planning. In this study we consider linear and nonlinear models for one-day-ahead forecast of electricity prices using components estimation techniques. This approach requires to filter out the structural, deterministic components from the original time series and to model the residual component by means of some stochastic process. The final forecast is obtained by combining the predictions of both these components. In this work, linear and non-linear models are applied to both, deterministic and stochastic, components. In the case of stochastic component, AutoRegressive, Nonparametric AutoRegressive, Functional AutoRegressive, and Nonparametric Functional AutoRegressive have been considered. Furthermore, two naive benchmarks are applied directly to the price time series and their results are compared with our proposed models. An application of the proposed methodology is presented for the Italian electricity market (IPEX). Our analysis suggests that, in terms of Mean Absolute Error, Mean Absolute Percentage Error, and Pearson correlation coefficient, best results are obtained when deterministic component is estimated by using parametric approach. Further, Functional AutoRegressive model performs relatively better than the rest while Nonparametric AutoRegressive is highly competitive.
Highlights
With the liberalization of the electricity sector, electricity has become a tradable commodity
The results in the table confirms the superiority of NonParametric AutoRegressive (NPAR) and Functional AutoRegressive (FAR) models compared to the rest, especially when the deterministic component is estimated through parametric approach
The deterministic component consists of long-run dynamics, multiple periodicities and calendar effects whereas the stochastic component accounts for the short-run dynamics of the process
Summary
With the liberalization of the electricity sector, electricity has become a tradable commodity. Extensive studies have been made on the problem of electricity prices forecasting using different modeling techniques and procedures Statistical models such as time series models, regression models, and exponential smoothing methods are widely used to forecast electricity market variables. These models are extensively used to accommodate multiple periodicities in time series data In this method, the variable of interest is predicted as an exponentially weighted average of the sequenced past values. A three-layered feedforward neural network, trained by the Levenberg-Marquardt algorithm, is used by Catalão, et al [34] for forecasting next-week electricity prices from the electricity markets of mainland Spain and California Based on their results, the neural network approach outperforms the ARIMA technique and the naive procedure in all considered weeks.
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call