Commodity Seasonality Analysis: Weather, Production Calendars and Comparisons
An average that rises every month on a chart is not yet a tradable seasonal rule. Harvest, heating, cooling, refinery maintenance, river levels and holidays create physical calendars, but markets anticipate them and the baseline, stocks, policy and climate also change. This guide separates a calendar pattern into causal hypothesis, data definition, normal, outliers, forecast vintage and out-of-sample test.
Who this guide is for: Readers who want to test seasonal charts with physical evidence and reproducible statistics
Key points to understand first
- Start seasonality with a repeating physical process and transmission hypothesis, not a monthly average.
- Use returns for price and normal deviations for stocks, fixing the transformation to the question.
- Start month, holidays, week number, leap year, futures roll, missing data and outliers can change the result.
- Separate full sample, recent regime and out-of-sample evidence and audit structural change and forecast vintages.
Six audits before saying "every year"
- 01Physical mechanism
Which harvest, demand, maintenance or transport process repeats?
Primary calendar - 02Series definition
Cash, future, continuous, inventory or spread?
Specification - 03Time alignment
Week, month, holiday, crop year and leap year?
Date rule - 04Baseline
Normal, full period, recent regime or post-change?
Sample window - 05Robustness
Median, quantiles, outliers and sign frequency?
Distribution table - 06Unseen period
Does the fixed rule survive data not used to design it?
Out-of-sample
Seasonality is a repeating process that changes quantities
The calendar does not move a commodity; production, consumption and logistics that recur with the calendar do. Crops have planting, flowering, harvest and export windows. Gas has heating and injection seasons. Refineries have maintenance and driving-demand cycles. Write the path from process to supply, demand or stocks and then to a price or spread.
Prices anticipate. Harvest month need not produce a decline if the crop was already discounted during development. Treat quantity seasonality, return seasonality and volatility seasonality as separate hypotheses.
| Target | Useful transformation | Question |
|---|---|---|
| Price | Return or calendar spread | Directional change by window |
| Inventory | Year-on-year or normal deviation | Position versus seasonal band |
| Supply/use | Daily rate or four-week average | Physical peak and trough |
| Volatility | Absolute/squared return or range | When uncertainty rises |
Do not mix calendar year, crop year, contract month and report week
Crop years differ across countries and southern and northern harvests reverse. Weekly data contain week 53, holidays and release shifts. Futures liquidity and maturities change each year, while a front-month series embeds a roll rule.
Choose calendar month, ISO week, days to harvest or days to expiry and keep it fixed. Normalize agriculture to the production calendar and energy to heating, cooling or maintenance seasons. Put the date rule at the top of the analysis.
- Boundary: calendar, crop or marketing year.
- Week: ISO week, report week, week-end date and week 53.
- Contract: fixed maturity, front month or constant maturity.
- Holiday: missing observations, shortened sessions and release delays.
Use median, quantiles and sign frequency beside the mean
Monthly average return can be dominated by a few extreme years. Display mean, median, interquartile range, positive-year share, standard deviation, maximum, minimum and observation count. Normalize holding length and record how many products and windows were tested.
Seasonal deviation = observed value − baseline for the same calendar positionSeasonal deviation % = (observed − baseline) ÷ baseline × 100Sign frequency = years matching the hypothesized direction ÷ valid yearsChoose mean, median or climate normal for the purpose and do not switch after seeing the result.Designing and evaluating on the same sample overfits. Freeze the rule on an early window and test a later out-of-sample period. Compare rolling subperiods and post-structure-change results so an obsolete effect is not preserved by a long average.
Separate climate normal, forecast, realization and production response
Climate normals are calculated over a reference period and change when that period updates. Use the same normal for comparisons. Climate, irrigation, crop genetics, building efficiency and the generation mix can change the mapping from temperature to yield or demand.
Forecasts update by issue time. Overlaying the final realized weather on historical prices creates look-ahead. Retain issue timestamp, horizon, model or ensemble, realization and error, then measure response around forecast changes.
Continuous contracts, selection and regime change manufacture patterns
Continuous futures can add or multiply roll gaps, changing long-run levels and returns. Front-month series also vary days to expiry, mixing curve shape with seasonality. Identify fixed maturity, constant maturity or index method.
Choosing a start date after viewing results, retaining only successful commodities or testing hundreds of windows creates false patterns. Add spread, commission, roll, liquidity and execution time, and retain the number of tests and failed examples.
- Survivorship: did the dataset omit discontinued contracts?
- Look-ahead: did later revisions, realized weather or future index membership enter?
- Multiple testing: how many months, starts and products were tried?
- Regime change: did policy, capacity, climate or specifications change?
Test mechanism, fixed rule, unseen data and execution in that order
- Write the mechanism
Explain what repeats, when quantities change and which spread should respond.
- Freeze series and dates
Lock transformation, week/month, maturity, roll and missing-data treatment.
- Show the distribution
Report mean, median, quantiles, sign frequency and sample size.
- Use unseen data
Evaluate a period not used to design the rule and the post-change regime.
- Add execution
Include spread, commission, roll, liquidity and release timing.
Backtest & Robustness Lab can inspect imported TradingView results through drawdown, Monte Carlo, stress and OOS views. Financial Templates Hub can preserve hypothesis, test count, date rule and data vintage. The tools do not prove seasonality or future return; they help expose overfit.
Seasonal-baseline deviation calculator
Enter an observation and a baseline for the same calendar position and definition.
Fix the normal period, week or month, unit and data vintage. This is not a directional forecast.
Frequently asked questions
Is a positive monthly average enough?
No. Review median, quantiles, sign frequency, outliers, unseen periods and trading cost.
Is a climate normal fixed?
No. It changes with the reference window. Use the same normal and assess structural change.
Can I use a continuous futures chart?
Yes with care. Roll adjustment and changing time to expiry can enter the apparent pattern, so compare fixed or constant maturities.
Can seasonality alone form a trading rule?
Future repetition is not guaranteed. Test mechanism, OOS, liquidity, cost and risk limits separately.
Primary sources and verification links
- WMO | Climatological NormalsClimate normal and reference-period principles
- NOAA NCEI | Climate at a GlanceOfficial temperature and precipitation data
- USDA FAS | Crop CalendarsCountry and crop planting and harvest calendars
- USDA NASS | Crop ProgressWeekly planting, development and harvest progress
- EIA | Weekly Petroleum Status ReportPrimary petroleum and refinery statistics for seasonal comparison
Edited and published by: SG Group · Editorial approach: We prioritize primary materials from EIA, USDA, CFTC, NOAA, international commodity bodies, exchanges and index providers. Data definitions, contracts, methodologies and release times can change; verify the current source before acting.
Important notice: This article provides general education about physical commodity markets, statistics, indicators and derivatives. It is not investment advice, a product recommendation, a trade signal, a price forecast or a promise of profit. Prices, quantities and ratios are fictional calculation examples unless an official statistic is expressly identified. Contract units, delivery terms, taxes, fees, margin, trading hours and data definitions vary by commodity, region, venue, provider and date. Verify primary sources and current provider terms before acting.