The Financial Effect of the Electricity Price Forecasts’ Inaccuracy on a Hydro-Based Generation Company

Abstract

Electricity price forecasting has a paramount effect on generation companies (GenCos) due to the scheduling of the electricity generation scheme according to electricity price forecasts. Inaccurate electricity price forecasts could cause important loss of profits to the suppliers. In this paper, the financial effect of inaccurate electricity price forecasts on a hydro-based GenCo is examined. Electricity price forecasts of five individual and four hybrid forecast models and the ex-post actual prices are used to schedule the hydro-based GenCo using Mixed Integer Linear Programming (MILP). The financial effect measures of profit loss, Economic Loss Index (ELI) and Price Forecast Disadvantage Index (PFDI), as well as Mean Absolute Error (MAE) of the models are used for comparison of the data from 24 weeks of the year. According to the results, a hybrid model, 50% Artificial Neural Network (ANN)–50% Long Short Term Memory (LSTM), has the best performance in terms of financial effect. Furthermore, the forecast performance evaluation methods, such as Mean Absolute Error (MAE), are not necessarily coherent with inaccurate electricity price forecasts’ financial effect measures.

Multi-Day-Ahead Electricity Price Forecasting: A Comparison of Fundamental, Econometric and Hybrid Models

“…The Diebold and Mariano (2002) (DM) test is widely applied in the EPF literature (e.g. , Ziel and Weron (2018) and Ugurlu et al (2018)). It considers whether the loss differential series of two models exhibits an expected value of zero and is usually considered in its multivariate one-sided version (e.g.…”

Forecasting day-ahead electricity prices: A review of state-of-the-art algorithms, best practices and an open-access benchmark

“…The studies that followed in 2018 focused on one of three topics: 1) evaluating the performance of different deep recurrent networks [13,23,37,78]; 2) proposing new hybrid methods based on CNNs and LSTMs [14,44,79,80]; or 3) employing regular DNN models [23]. Independently of the focus, they were all more limited than the first and the third studies [12,59] as they failed to compare the new DL models with state-of-the-art statistical methods and/or to employ long enough datasets to derive strong conclusions.…”

Economic Forecasting With Deep Learning: Crude Oil

“…This can be achieved because of the existing memory in the gates. LSTM can pass information that has been captured in the early stages and keep that information for a long time, which enable the LSTM to generate the long-distance dependencies (Ugurlu, Tas, Kaya, & Oksuz, 2018), (Chung, Gulcehre, Cho, & Bengio, 2014).…”