Modeling Non-Monotonic Volatility in Electricity Futures

Summary

The document describes a volatility pattern in monthly electricity futures that changes as the delivery month progresses. Because electricity is delivered throughout the month, the contract price converges toward the monthly average as delivery unfolds. The author observes that volatility begins low when trading activity is limited, rises as the contract becomes more active, and then falls as the delivery month elapses. This pattern is non-monotonic and linked both to market activity and convergence toward the delivery average.

The author asks how to model this behavior and wonders whether ARIMA-GARCH is suitable, given advice to avoid higher-order model specifications. The document does not include an answer, proposed model, empirical comparison, or evidence about forecast performance. It is useful as a concise statement of the modeling challenge and its market-specific motivation, but it does not establish that ARIMA-GARCH is unsuitable or recommend an alternative. Any model choice would require examining the data and testing whether it captures the changing volatility pattern.

Key ideas

• Monthly electricity futures converge toward the average price over the delivery month.
• The described volatility first rises with trading activity and later declines as delivery progresses.
• This non-monotonic pattern raises a modeling challenge for standard time-series volatility approaches.
• The document poses, but does not resolve, whether ARIMA-GARCH can represent the observed behavior.
• No empirical model comparison or forecasting evidence is provided.

Tags

# Non-Linear Time-Dependent Volatility


# Non-Linear Time-Dependent Volatility












My data consist of monthly electricity futures contracts. Unlike other commodities, electricity is delivered throughout a month (rather than on a specific date), which means that, as the active month elapses, the price of the electricity contract converges to the monthly average. Unfortunately, however, volatility is not monotonic: it is initially low, due to a lack of trading volume, accelerates as a contract becomes increasingly active (much like with other commodities), and finally decreases as the active month elapses.

My question is: how can one model such a market? Sources I've encountered advise against (p, d, q) > 2, which would be necessary for modeling non-monotonic volatility, which leads me to think that ARIMA-GARCH is unsuitable here.

Thank you for your patience and assistance!

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.