The long-term trends on Russian electricity market: comparison of empirical mode and wavelet decompositions

The problem of trend-cyclic component filtering from price time-series arises in many commodity market studies, including those of wholesale electricity market. The long-term component filtering is an important part of price analysis since incorrect determination of this component may result in substantial risk underestimation, distorted expectations of both consumers and power generating companies, as well as financial losses. A great strand of literature on this topic proposes quite a lot of approaches and procedures for solving this problem, but all of them suffer from two principal flaws: (1) inability to deal with non-stationary and nonlinear processes; (2) assumption of an "a priori", knowledge of the phenomenon being studied. The complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) allows to effectively overcome these flaws and is expected to produce more adequate results as compared to other methods. In order to check this, we compare the performance of CEEMDAN with the ordinary EMD and yet another well-known approach - the wavelet-decomposition, with an example of the Russian day-ahead electricity market (price zones Europe-Ural and Siberia). Our results shows that the CEEMDAN is much more effective than the standard EMD and is comparable with the wavelet-decomposition (in terms of trend estimation error). At the same time, we found that there are some real data problems with the criterion of the number of low-frequency modes that are included into trend.