[FREE] Quick Context: US Gasoline Demand Data
Concerns re: weak US gasoline demand data are likely overstated and distorted by especially-pronounced pandemic-era data estimation issues
I’ve decided to open this post outside my normal paywall because this issue is causing a lot of confusion and it’s a pretty interesting exercise in data quirks.
If you value this free research, hit the LIKE button.
Fears, and at worst conspiracy theories, about weekly [unadjusted] US gasoline product supplied (PS) data appearing lower today than the summer of 2020 are misplaced; instead, two compounding factors have tragically muddied the water.
First, a simple but powerful statistical mirage: the Energy Information Administration’s (EIA) Weekly Petroleum Status Report (WPSR) estimates aren’t corrected once more accurate estimates are released after the fact, which distorts comparisons across multiple years (vs. 2020, for instance).
Second, a data definitional issue: the EIA is measuring PS as a proxy for consumer demand, but key differences between the behaviour of gas stations and their customers are likely distorting the magnitude and timing of a downturn here.
Overall, US gasoline demand probably isn’t especially strong at the moment and has likely been weakening counter-seasonally for more than a month; the PS data is very likely overstating the extent of the downturn—although it’ll take a month or two for the data to confirm this one way or the other.
The last few weekly releases of US gasoline product supplied (hereafter “PS”) data, a rough proxy for demand, have been notably and unseasonably weak—so much so, in fact, that the weekly series appears to show US gasoline PS dipping even below this period in 2020 (chart above, left). This has people understandably alarmed, and confused, given that much of the US economy was still under various stages of early-pandemic lockdown through the summer of 2020. This confusion is then mixed with the heightened stakes of a violent negative oil price correction, which now has various parties claiming that the data is manipulated (it isn’t) for political purposes—a deeply regrettable state of affairs.
To address this alarm and confusion, I propose that we break the matter down into two core questions: Has gasoline demand weakened recently? Very likely, though the magnitude remains an open question. And are we really back down to below summer 2020 levels? Almost certainly not.
Upfront, it’s very likely that US gasoline consumption has fallen back over the past few weeks, an early warning sign regarding both the strength of consumer balance sheets and the resilience of end-user fuel demand. Demand is likely both lower than usual for this time of year as well as headed in a counter-seasonal direction, effectively—if this trend persists—ending the summer driving season a month and a half early. What’s more, this slowdown appears to be global in scope, with gasoline inventories building counter-seasonally around the world and the gasoline crack spread collapsing precipitously over the past month and a half.
But ultimately, I feel that these two core questions have been muddied considerably by both:
a simple but powerful statistical mirage that is distorting comparisons across multiple years (vs, 2020, for instance), and
a data definitional issue (“product supplied vs. “consumer demand”) that has a high likelihood of sending lumpier mixed signals, potentially distorting the magnitude and timing of the downturn
So, let’s dig into the funny data.
Statistical Mirage: Stop Using Unadjusted Weekly Data Histories
The high-frequency PS data—effectively a measure of “demand” but with important differences discussed below—is released regularly in the Energy Information Administration’s (EIA) Weekly Petroleum Status Report (WPSR). This data appears to show that recent US gasoline “demand” just ticked below the same period of the year in 2020—which, admittedly, feels very “off” given that much of society was still facing various degrees of pandemic restriction through to summer of 2020.
Except, I don’t actually think that’s what it shows at all; let me explain how the statistical mirage works.
The WPSR data release is an early best estimate of gasoline product supplied during a given period of time, while that same period is covered and estimated far more accurately by the monthly Petroleum Supply Monthly (PSM), which comes out a couple months later. Typically, the WPSR estimate is pretty close to the more-accurate PSM estimate, but there are always discrepancies (by which I mean the residual vs. the ultimate PSM estimate)—and sometimes those errors can be large and biased to one direction, especially in these bizarre times (because, you know, pandemic and all). Most importantly, the WPSR estimates aren’t corrected once the PSM data is released, even though the errors may inform tweaks to future WPSR estimates in the hopes of improving the metric over time.
So, when it comes to comparing current WPSR estimates and estimates from the summer of 2020, things run amuck. Back in the summer of 2020, the WPSR error was persistently positive (i.e., the weekly estimate was overestimating product supplied) to the tune of 200-300 kbpd vs. the ultimate PSM estimate. And now, the most recent WPSR errors have been negative with the WPSR underestimating product supplied with the PSM as of May 2022).
So, even barring any other considerations, the known sampling error alone could result in the July 2022 PS figure being, in truth, almost half a million barrels per day above July 2020 levels, even though the unadjusted WPSR figures show roughly the same observation for those two dates. While we won’t really know until the PSM releases more accurate data in a couple months, we know, based on historical estimation errors, that it is extremely unlikely that current demand is anywhere near as low as the summer of 2020—even if demand is indeed easing off counter seasonally, and even with this adjusted PS series is comfortably below both the seasonal average and last year-levels.
(Special thanks to a thread from Makai Marine for prompting my digging on this subject and whose work on this kind of stuff is consistently excellent)
Definitional Issues: Wholesale Demand is Lumpier, Especially Amidst Volatile Prices
The second issue with some interpretations of the PS data is around definitions. The data that the EIA releases is called “product supplied” (PS) and is a rough proxy for domestic demand; but, in practice, it’s a residual function that measures the barrels that have “disappeared” from the visible distribution system. In the EIA’s words, “Approximately represents consumption of petroleum products because it measures the disappearance of these products from primary sources, i.e., refineries, natural gas-processing plants, blending plants, pipelines, and bulk terminals.”
Essentially, the difference is this: the PS data isn’t actually estimating consumers filling their car at the station; but, rather the gas stations filling their tanks by purchasing fuel in the wholesale market. This difference matters because businesses and consumers don’t always act the same; gas stations have some discretion about how frequently and how fully to fill their tanks and also generally have a better sense of what’s happening in the market than the typical consumer. (Another thread credit here goes to Mason Hamilton, who explained this phenomenon in further depth.)
So, if you’re a gas station owner watching gasoline prices plummet over the past two months, it’s not unreasonable to think that you, the business, would delay your purchase if given the choice between today or next week at potentially 10% lower prices. Stations typically hold a substantial volume of effective gasoline storage capacity, which means there is a material difference to the market between running those tanks full in fear of rising prices (say when they were hitting all-time highs in early June) and running those tanks a little lower in hopes of lower prices next week.
This type of delayed purchasing could be exacerbating any preexisting weakness in consumer demand, and would help explain the divergence between deteriorating weekly PS data and more explicit retail demand data, like that collected by GasBuddy, that shows relatively steady consumption. Suffice to say, this dislocation between wholesale and retail demand has the potential to be large enough as to materially distort the state of end-consumer demand. While PS and consumer demand can’t deviate for long and will likely normalize over the next couple WPSR releases, the effect is large enough that we can’t yet know for sure how low retail demand has actually fallen.
Conclusion
Fears, and at worst conspiracies, about PS figures being lower today than the summer of 2020 when substantial COVID lockdown measures were still in place are misplaced. Yes, US gasoline demand probably isn’t especially strong at the moment and has likely been weakening counter-seasonally for more than a month; but, it isn’t shocking coming out of a fresh all-time high price spike and the extent of this slowdown is likely being exacerbated by the commercial decisions of wholesale market participants. Wider than usual errors in the WPSR estimates vs. the more-accurate PSM estimates through much of the pandemic era have also materially distorted comparisons between current data and data from 2020. Despite temptation, it is especially important to delay settling on firm interpretations of PS until better data (aka the PSM) has been released.
Clicking the LIKE button is one of the best ways to support my research.
Disclaimer: These materials incorporate third-party data, are provided for informational purposes only, and do not constitute advice or opinion of any kind. Commodity Context does not warrant or guarantee the accuracy or completeness of these materials.
well done Rory. good reminder that weekly stats are not revised.
Top Shelf content Rory, as always.