In continuously evolving power markets, with highly volatile electricity prices, market participants are interested in managing their price risk in addition to their volumes and other risks. Price information from future contracts traded on standard markets do not provide the flexibility or granularity required for different intervals, such as hours of the days or weekdays. With electricity consumption different at time of the day, week and seasons, and affected by weather conditions, this may lead to unanticipated price changes. Having access to accurate price expectations is therefore very relevant. This is how a forward price curve will help you Manage Price Risk with a Price Forecaster!
The emergence of less predictable renewable electricity generation has increased volatility in power markets, where positions may change rapidly. In addition, less liquid markets may not provide enough pricing information. Portfolio Managers who can forecast wholesale power prices position themselves to adjust their trading strategy or their generation and demand plans, thereby managing their price risk.
Where modern ETRM systems already provide real-time position information, what-if risk scenarios capabilities, and price curve management, these functions can be further leveraged when the system is equipped with and integrated ability to forecast short and long-term power prices.
Short term here is defined as next day to a few weeks out, whereas Long-term forecasting can be a month to several years out. As ingredients, the price forecaster will use historical price and demand data as well as forecasted data such as weather. Other input consists of calendar attributes, such as on-peak, off-peak hours, weekdays, weekend days, holidays etc.
For the Short Term forecasts a Neural Network method will be performed, while for Long-term forecasting various methods are available. Under the Neural Network method, historical input data are divided into train and test sets. By taking the train data set, the method performs optimizations in an iterative fashion and provides the optimized model parameters. Those model parameters are used in the test data for the model validation. Once the models are validated the optimized model parameters with new input data are fed in the model for the price forecasting.
For a long-term forecast, Multiple Linear Regression is one of the simplest and widely used methods. It requires at least one year of historical price data. The method generates two-dimensional grids (month vs. hours) and distributes averages of the historical price in the corresponding grids. Then it calculates regression coefficients and those coefficients are used in forecasting for the next year. The process is repeated sequentially up to the desired number of years in such a way that the forecast of year 1 will be the input of year 2 and so on. Other available methods for Long-term forecast that can be selected include the Kalman Filter, Fuzzy Logic and Econometric models. Working mechanism of these methods can be found elsewhere.
Whether a utility supporting industrial customers who need to minimize the risk of highly volatile power prices, or a generator who must cover his production cost, they are interested in hedging their price risk. A price forward curve generated by a price forecaster allows them to get reasonably accurate predictions of future prices to help manage their price risk and thereby ‘keep an eye on the prize’!