Day Ahead Price Forecasting Models in Thin Electricity Market

TL;DR

This paper develops and tests advanced day-ahead electricity price forecasting models for India's thin and complex market, incorporating demand-supply drivers like weather and outages to improve accuracy.

Contribution

It introduces models that account for demand-supply drivers specific to India's market and evaluates their performance using the Model Confidence Set approach in a real-world setting.

Findings

Models outperform standard benchmarks in forecasting accuracy.

Inclusion of weather and outage data improves model performance.

Models are validated in a live two-year business environment.

Abstract

Day Ahead Electricity Markets (DAMs) in India are thin but growing. Consistent price forecasts are important for their utilization in portfolio optimization models. Univariate or multivariate models with standard exogenous variables such as special day effects etc. are not always useful. Drivers of demand and supply include weather variations over large geographic areas, outages of power system elements and sudden changes in contracts which lead the players to access power exchanges. These needs to be considered in forecasting models. Such models are observed to considerably reduce forecasting errors by outperforming other models under conditions, which are neither infrequent nor recur at defined intervals. This paper develops models for India and tests the utility of these models using Model Confidence Set (MCS) approach which picks up the best models. The approach has been developed for a power utility in India over a period of two years in live business environment.