Compare calendar months without confusing observation weights

Define the monthly statistic exactly

Assign each retained weekly stock observation to the calendar month containing its reference date. Add the levels and divide by the number of observations in that month. Each observation receives equal weight. The result is a mean of weekly snapshots, not a daily time-weighted average or the inventory on the last calendar day.

Months can contain different numbers of weekly reference dates. This is not automatically an error, but it changes the relationship between snapshot weighting and calendar-time weighting. The lab preserves observation counts so a reviewer can see how each monthly mean was formed instead of assuming every month represents four equal weeks.

Construct the five-year reference in two stages

For a synthetic current July, suppose the five preceding annual July means are 100, 110, 120, 130 and 140 units. Their equal-weight average is 120. A current July mean of 126 is six units above that reference, or five percent when divided by the positive baseline.

Now suppose the first four historical Julys contain four observations each and the final July contains five. Pooling all observations would yield 2,540 divided by twenty-one, approximately 120.952381, if each year’s observations equal its stated mean. That is a different weighting rule. The lab uses the mean of annual month means so the five-observation year does not receive extra weight.

Require all five years to contribute valid evidence

The lab requires exactly five prior calendar years, with at least four observations in each corresponding month, before calculating a baseline comparison. The selected current month also needs at least four observations. This is a minimum-count eligibility rule, not proof that every scheduled weekly report is present; source coverage still needs inspection.

If only four annual means are available, their average may be a useful separately labeled statistic, but it is not the promised five-year reference. Keeping the distinction prevents a chart from appearing comparable across months while its history length and selection rule change invisibly. With retained data limited to 2019–2024, earlier selected years cannot obtain all five preceding years from that file, so their five-year comparison is withheld.

Distinguish the agency’s benchmark

EIA’s weekly natural-gas storage methodology describes daily interpolation for its official five-year comparison. The publication’s calendar-month snapshot method is deliberately different. It should therefore be named as a publication calculation rather than displayed as though it were the official WNGSR five-year average.

A numerical difference between the two methods need not indicate an error. They can use different dates, weighting and revision vintages. Reconcile definitions before trying to force agreement. If the purpose is to reproduce the agency figure exactly, use the agency’s specified methodology and appropriate source vintage instead.

Interpret a gap as a description of the chosen reference

A five-percent gap does not state whether inventories are economically adequate, whether demand is stronger or whether prices will rise. The reference window may include unusual disruptions or structural changes. Show its constituent annual means so readers can assess whether one year materially shapes the comparison.

The lab supports a transparent seasonal description of retained historical observations. It supplies no predictive calibration or trading rule. A research dossier should preserve the month assignment, five-year membership, units, counts and retrieval vintage so another reader can reproduce the gap and understand the narrower question it answers.

Keep in the decision record

• Average within each year-month, then average the five annual means.
• Do not pool weeks when equal year weights are intended.
• This method differs from EIA’s official daily-interpolated gas reference.

Source notes

1. EIA — Weekly Underground Natural Gas Storage methodology, updated 10 September 2018
2. EIA — Weekly Petroleum Status Report explanatory notes

Source dates and scope matter. Historical inventory observations are retained snapshots; forecasts and synthetic curve assumptions are labelled separately. No investment outcome or product coverage is implied.

Continue the analysis