Pertinent power price forecasting for a retailer
Categories: Hedging, Power Utilities
-
Situation
A Dutch power retailer hedges its sales commitments in the forwards market, based on its sales channels estimation.
-
Complication Volumetric risk due to deviation of actual power demand from forecast can cause under and over hedging situations. Wholesale power price spikes in under-hedging situations can cause huge losses to the energy retailer.
-
Solution We proposed a hybrid model using historical statistical properties of power prices, combined with supply and demand indicators to produce a reliable short term price forecast. Over the period of June 2013 to May 2014, the model predicted 76% of the price spikes that occurred, with a 0.55% error in suggesting a spike occurrence.
The predicted spikes allowed traders to turn risk into opportunity.
More details
What we know about Power Prices
- Prices are set by the intersection of supply and demand
- Prices are cyclical, seasonal and mean-reverting with price-dependent volatility
- Since power is not economically storable (at least before year 2018), demand or supply shocks cannot be smoothed out by inventory
- This creates extreme spikes
- Spikes can cause bankruptcy to energy retailers if not properly managed
- However, these huge losses can be turned into high profit if the spikes timing can be predicted
PRICE RISK
If an energy retailer is not hedged and has to source power at the spot market, this could cause huge costs, especially if spikes occur.
VOLUMETRIC RISK
Even if the retailer is hedged, volumetric risk still exists because the actual consumption cannot be predicted precisely, which results in over hedging and under hedging almost all of the time.
Therefore spikes constitute a huge risk even in hedged cases
Suppose that you know with high probability that a price spike will occur the next day.
You can monetize this information by taking forward positions (day-ahead futures or options) with an underlying volume that exceeds your forecasted volume (you over-hedge)
Statistical Part
- The statistical part of the model captures the dynamics of historical prices (volatility, cycles, seasonality and auto-correlation).
- Calibrating the parameters of the statistical part should differentiate between two states: “normal prices” and price spikes. This is a regime switching model!
Fundamental Part
- History does not necessarily repeat itself, especially with the introduction of new energy resources such as wind and solar.
- The fundamental part of the model captures the supply and demand factors responsible for the price formation: Reserve margins, Electricity load, Marginal cost, Fuel prices, etc.
Hybrid model
The hybrid model is a combination of both statistical and fundamental parts. On one hand, it is based on a regime-switching model for the statistical part. On the other hand, it utilizes the supply and demand factors that are responsible for the price formation.
We are using an iterative inference for parameter estimation, which is the Hamilton procedure:
1. Write the densities of the observations under the two regimes.
2. Compute the probabilities of being in each regime based on transition probabilities and densities.
3. Write the conditional likelihood function as a function of the above quantities.
4. Estimate the autoregressive parameters Ck maximizing the log likelihood.
5. The maximum-likelihood estimation optimal parameters are then used in a final pass through the filter to draw probabilistic inference about regime states St.
| Switching probability cutoff | 50% | 75% | 85% |
| Percentage of spikes detected | 91% | 86% | 81% |
| Percentageof spikes suggested incorrectly | 12% | 4.3% | 0.5% |
- The quality of the input data is of prime importance.
- Notably, demand forecast and the available capacity are critical inputs that need to be accurately estimated before being used in the forecast model.
- As shown in the table, in most cases, the regime switching times are correctly detecting price spike occurrences.
We have the skills and experience to support you in:
- The review and testing of your existing forecasting models
- The enhancement of your current models by adding other options
- The design and implementation of forecasting models in various programming languages (e.g. C++, Java, Python, R, Matlab) and interface with your trading system
- The back-testing of the models using both historical and forward looking (simulations) approaches
- An advisory in other quantitative risk management fields
What clients said aboutour projects
-
Bron Sharman Senior Oil Trader “Unlike many advisers I’ve worked with, Trading Integral Solutions don’t have to pretend to be interested in their clients: they focus on what really matters and strive to add value in an optimal way.” Bron Sharman, Senior Oil Trader at SOCAR Trading -
Olga Kutuzova IFRS Group Manager "Trading Integral Solutions applied high quality standards when they delivered their project to us. We benefited greatly from our interaction with their highly skilled team of experts!" Olga Kutuzova, IFRS Group Manager at Novatek Gas & Power -
Jaroslaw Wajer Partner "I have cooperated with Trading Integral Solutions several times. They supported us during the design and implementation of trading departments and company strategies for our power utilities clients. I appreciate their trading expertise and profound knowledge of risk management. I would truly recommend cooperation with Trading Integral Solutions." Jaroslaw Wajer, Partner Performance Improvement at Ernst & Young