Regression tree model versus Markov regime switching: a comparison for electricity spot price modelling and forecasting

Abstract

This paper models electricity spot prices using a Markov regime switching (MRS) model and regression trees (RT). MRS models offer the possibility to divide the time series into different regimes with different underlying processes. RT is a data driven technique aiming in finding a classifier that performs an average guessing for the response variable in question, which is the short term electricity spot price. We use a dataset consisting of average day ahead spot electricity prices for the MRS model. Then, we use hourly data to build the RT model. The empirical evidence supports that the regression tree approach outperforms the MRS model. We also compare the forecasting accuracy of the regression tree model by incorporating different predictors sets for electricity prices and logarithmic electricity prices. We find that a model with 11 predictors, accounting for logarithmic prices fits best our data.