Day-ahead electricity price forecasting with high-dimensional structures: Univariate vs. multivariate modeling frameworks (1805.06649v1)

Abstract: We conduct an extensive empirical study on short-term electricity price forecasting (EPF) to address the long-standing question if the optimal model structure for EPF is univariate or multivariate. We provide evidence that despite a minor edge in predictive performance overall, the multivariate modeling framework does not uniformly outperform the univariate one across all 12 considered datasets, seasons of the year or hours of the day, and at times is outperformed by the latter. This is an indication that combining advanced structures or the corresponding forecasts from both modeling approaches can bring a further improvement in forecasting accuracy. We show that this indeed can be the case, even for a simple averaging scheme involving only two models. Finally, we also analyze variable selection for the best performing high-dimensional lasso-type models, thus provide guidelines to structuring better performing forecasting model designs.

• The paper empirically compares univariate and multivariate modeling frameworks for day-ahead electricity price forecasting using 12 different datasets.
• Multivariate models, particularly lasso-based ones, generally show a slight edge in predictive performance over univariate models, though no single model consistently outperforms.
• Combining predictions from both univariate and multivariate lasso models significantly improves forecasting accuracy across most markets.

An Empirical Analysis of Electricity Price Forecasting: Univariate vs. Multivariate Approaches

The paper "Day-ahead electricity price forecasting with high-dimensional structures: Univariate vs. multivariate modeling frameworks" provides a comprehensive empirical paper concerning the optimal model structure for short-term electricity price forecasting (EPF). Specifically, it explores whether univariate or multivariate modeling frameworks are preferable when predicting day-ahead electricity prices. The analysis is based on 12 different datasets spanning various European power markets and one dataset originating from the Global Energy Forecasting Competition 2014.

Overview of Methodologies

The researchers analyze the performance of 10 models representing eight different model classes. These models are divided into two primary categories: univariate and multivariate frameworks. The multivariate frameworks include the use of independent models for each hour of the day as well as fully multivariate models like VAR, which account for the interdependencies between different hours of the day. Univariate frameworks treat the price series as a single 'high-frequency' time series and forecast based on recursive schemes.

Of special interest are the least absolute shrinkage and selection operator (lasso) models, implemented within both multivariate and univariate frameworks. These models perform variable selection and shrinkage, which are crucial for managing high-dimensional data effectively and improving forecasting accuracy.

Key Results

The paper demonstrates that while multivariate models provide a slight edge in overall predictive performance, no single model consistently outperforms across all datasets, hours of the day, or seasons. Interestingly, the lasso-based models—both univariate and multivariate—show superior performance across several datasets, with multivariate lasso models being the best for most analyzed markets.

Importantly, the combination of predictions from multivariate and univariate lasso models—by simple arithmetic averaging—yielded improved forecasting accuracy, suggesting that each framework captures different aspects of the underlying price dynamics. This improvement was statistically significant for the majority of the markets.

Implications

The findings suggest that there is no definitive model structure that universally excels, highlighting the importance of adaptive and flexible modeling approaches. The application of lasso-type models, which allow for efficient variable selection, demonstrate significant promise in advancing electricity price forecasting methodologies.

Future Research Directions

Given the nuanced findings, future research could explore more sophisticated combination methods or hybrid models that dynamically select between univariate and multivariate predictions based on market conditions. Additionally, extending the analysis to incorporate and assess additional exogenous variables such as weather forecasts or economic indicators could provide further insights. Research leveraging advanced machine learning techniques or hierarchical modeling structures might offer further improvements in EPF, especially in dealing with high-frequency and high-dimensional data.

Overall, this paper contributes valuable empirical evidence and methodological guidance for academics and practitioners in the field of energy economics, emphasizing the growing complexity and sophistication needed in forecasting models to navigate the dynamic nature of electricity markets.