Estimating Seasonal Factors for Natural Gas Prices
Summary
The document outlines a way to estimate seasonal patterns in natural gas prices before analyzing or modeling the futures term structure. For each year, calculate the average daily spot price, then calculate the average for each month. Divide each monthly average by that year’s annual average to get monthly shaping factors, and average those factors across several years. The suggested history is at least three years, though the choice depends on judgment and data availability.
For daily adjustments, the method can also compare weekday prices with monthly averages. This produces a separate factor for each weekday and month; for gas, the answer suggests that distinguishing workdays from weekends may be enough. The factors can then be expressed as a seasonality-related drift term for the winter-to-summer term structure. The document provides a proposed procedure rather than empirical results, and it does not specify how to handle unusual years, changing market regimes, or differences between spot and futures prices.
Key ideas
- Estimate each month’s seasonal factor by comparing its average spot price with the annual average.
- Average monthly factors across multiple years to reduce reliance on one year’s pattern.
- Use weekday adjustments when daily seasonal shaping matters, with workday and weekend factors as a possible simplification.
- Seasonality factors can be translated into a drift term for modeling the winter-to-summer term structure.
- The suggested historical window is a judgment call and may depend on available data.
Tags
Full text
# How to de-seasonalize natural gas term structure data?
# How to de-seasonalize natural gas term structure data?
I need to de-seasonalize Nat Gas futures data for a project and am hoping to get good suggestions. As we all know natural gas futures are priced higher for the winter months and to analyze/model the term structure we need to de-seasonalize the data.
Any ideas how would one do it?
## Answer by ZRH (score 5)
https://quant.stackexchange.com/a/43867
As a starting point to this, determining seasonality for a given market is as follows:
i) Take several years of historical spot price time series, e.g. TTF spot prices. For year $i$ work out a yearly price $p_{yr,i}$ by taking the arithmetic average of daily spot prices. Do the same in respect of month number $j$ of the same year to get a monthly price $p_{mth,i}^{j}$. The monthly shaping factors $f_{i}$ are then $f_{i}=\frac{p_{mth,i}^{j}}{p_{yr,i}}$. Determine the $f_{i}$ for a number of years (where possible i use at least 3, but that is a judgement call), and use their average. As you say, winter will be more expensive, i.e. you expect $f_{i}>1$ for $i\in\{1,2,3,10,11,12\}$ and $f_{i}<1$ for $i\in\{4,5,6,7,8,9\}$
ii) if you need to use daily shaping, you can determine the ratio of weekday prices (numbered 1 to 7) to the monthly prices. This results in 7 weekday factors for each month, i.e. another 84 factors. This is how it is done in electricity, where intraweek shaping is very pronounced. I guess in gas you might find it sufficient to have only two factors per month, one for the workdays and one for the weekend.
Having determined the seasonality factors, one can turn them into a seasonality-related drift term $\mu(t)$ to describe W/S term structure.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.