Diebold-Mariano test results for paper "Forecasting spot electricity prices: Deep learning approaches and empirical comparison of traditional algorithms"
datasetposted on 30.04.2020, 00:00 by Jesus Lago, F. (Fjo) De Ridder, B. (Bart) De Schutter
This dataset comprise the results of the Diebold-Mariano (DM) tests when comparing several models for predicting day-ahead electricity prices in Belgium in the years 2015-2016. These results are part of the research paper: Forecasting spot electricity prices: deep learning approaches and empirical comparison of traditional algorithms The results are provided for each of the 27 models evaluated in the original paper. In particular, there are 27 files representing the p-values of the statistical tests, each one of the corresponding to one model. The name of the file represents the model evaluated. The results of the DM tests are given defining the model under evaluation as M1 (as defined in the original research paper), and then considering the remaining 26 models as M2. The DM test considers the null hypothesis of the forecasts of M1 being equal or worse than those of M2 against the alternative hypothesis of the forecasts of M1 being better. The p-values represent the probability of observing the obtained experimental data if the null hypothesis is true. Thus, very low p-values represent the cases where the model evaluated (M1) is significantly better than its counterparts. For each model pair M1, M2, the results are also given for the 24 predictive horizons as well as for the full loss differential with serial correlation (as defined in the original paper). In addition, figures representing the DM results are also included. Instead of plotting the p-values, the test statistics are employed; this is done to obtain figures that are easier to read. A threshold line is also depicted to indicate the test statistic value that represents a p-value of 0.05.