Most electricity spot price series exhibit price spikes. These extreme observations may significantly impact the obtained model estimates and hence reduce efficiency of the employed predictive algorithms. For markets with only positive prices, the logarithmic transform is the single most commonly used technique to reduce spike severity and consequently stabilize the variance. However, for datasets with very close to zero (like the Spanish) or negative (like the German) prices the log-transform is not feasible. What reasonable choices do we have then? To address this issue, we evaluate 16 variance stabilizing transformations within a comprehensive forecasting study involving two model classes (regression models, neural networks) and 12 datasets from diverse power markets. We show that the probability integral transform combined with the standard Gaussian distribution yields the best approach, significantly better than many of the considered alternatives.

In day-ahead electricity price forecasting (EPF) variable selection is a crucial issue. Conducting an empirical study involving state-of-the-art parsimonious expert models as benchmarks, datasets from three major power markets and five classes of automated selection and shrinkage procedures (single-step elimination, stepwise regression, ridge regression, lasso and elastic nets), we show that using the latter two classes can bring significant accuracy gains compared to commonly-used EPF models. In particular, one of the elastic nets, a class that has not been considered in EPF before, stands out as the best performing model overall.

Recent electricity price forecasting (EPF) studies suggest that the least absolute shrinkage and selection operator (LASSO) leads to well performing models that are generally better than those obtained from other variable selection schemes. By conducting an empirical study involving datasets from two major power markets (Nord Pool and PJM Interconnection), three expert models, two multi-parameter regression (called baseline) models and four variance stabilizing transformations combined with the seasonal component approach, we discuss the optimal way of implementing the LASSO. We show that using a complex baseline model with nearly 400 explanatory variables, a well chosen variance stabilizing transformation (asinh or N-PIT), and a procedure that recalibrates the LASSO regularization parameter once or twice a day indeed leads to significant accuracy gains compared to the typically considered EPF models. Moreover, by analyzing the structures of the best LASSO-estimated models, we identify the most important explanatory variables and thus provide guidelines to structuring better performing models.

The recent developments in combining point forecasts of day-ahead electricity prices across calibration windows have provided an extremely simple, yet a very efficient tool for improving predictive accuracy. Here, we consider two novel extensions of this concept to probabilistic forecasting: one based on Quantile Regression Averaging (QRA) applied to a set of point forecasts obtained for different calibration windows, the other on a technique dubbed Quantile Regression Machine (QRM), which first averages these point predictions, then applies quantile regression to the combined forecast. Once computed, we combine the probabilistic forecasts across calibration windows by averaging probabilities of the corresponding predictive distributions. Our results show that QRM is not only computationally more efficient, but also yields significantly more accurate distributional predictions, as measured by the aggregate pinball score and the test of conditional predictive ability. Moreover, combining probabilistic forecasts brings further significant accuracy gains.