Forecasting electricity price in Colombia: A comparison between Neural Network, ARMA process and Hybrid Models

Jorge Barrientos Marin, Elkin Tabares Orozco, Esteban Velilla


This study aims to predict electricity prices in the Colombian electricity market. To achieve this goal, conventional time series econometrics analysis and one alternative technique based on artificial intelligence algorithms have been implemented. We use autoregressive-moving-average models (ARMAX) and non-linear autoregressive neural networks (NARX). After estimating a hybrid model that combines ARMAX and ARNX models, including exogenous inputs, we forecasted an electricity price time series in a horizon of 12 months ahead (May 2017). Results show that NARX model's performance is not significantly better than ARMAX's. After applying a Diebold-Mariano test for forecasting accuracy, the null hypothesis is not rejected. This suggests no significant difference in predictive accuracy between the competing methodologies.

Keywords: stochastic process, ARMAX, NARX, random walk, predictive accuracy, electricity spot price.

JEL Classifications: C01, C12, C22, C45, C53, L11, L94.

Full Text:



  • There are currently no refbacks.