The Energy Information Administration (EIA) is attempting to post warning signs to help prevent the gas industry from entering what might be considered the Bermuda Triangle of weekly gas storage levels. Ever since the EIA took over the weekly storage survey from the American Gas Association (AGA) in May, the industry has been lost in a sea of revisions, struggling to find its way through weekly changes that have strayed so far from the historical norm.

Now industry observers have to brace themselves for changes to EIA’s revisions policy, which could mean the agency will use more back months of actual data from its national storage survey to come up with a better estimation ratio (see EIA’s recent request for comments in the Federal Register on July 11). While EIA believes the changes will help observers better understand its report, some people in the industry already are so confused that these changes could have the undesired affect of leading them further astray. One part of the proposed new system, that of announcing any large revisions on any morning of the week, could conceivably run some traders out of business.

Everyone in the gas industry knows that the most important information about supply and demand is captured by the weekly storage report. However, some experts are willing to bet that most of the people in the industry really don’t know how different EIA’s report is from AGA’s survey and are making market forecasts and doing storage analysis based on mistaken assumptions.

“The industry still thinks they are doing it just like the AGA, even though they are not,” said Tom Saal of Pioneer Futures. “The technique that has been used to produce the sequence of numbers for years has been changed.” Saal said people are making market decisions and using forecasting analysis for storage that is based on the incorrect assumption that the sequence of numbers will change in the same way that AGA’s data changed over the course of the year.

EIA already has stated in a comparison of its monthly data and AGA’s weekly data that AGA’s series showed a more rapid seasonal storage build and a more rapid seasonal storage decline than did EIA’s data. That fact sheds some light on the enormous differences between the two surveys.

In order to produce weekly numbers, AGA did a one-time snapshot using a survey sample; EIA has sought to duplicate that quick snapshot process, but then attempts to make it conform to the time-lagged monthly survey it has always done of the entire industry. This is where the revisions come in.

“It is not like the AGA survey,” said EIA’s Roy Kass. And people probably should not react to it in the same way they reacted to AGA’s weekly report, he said. “Our estimation procedure is dramatically different from AGA’s. We use information from not only the weekly survey that is confidential but also the monthly survey that is confidential and the monthly survey goes to [every storage operator in the nation].”

One of the keys to understanding EIA’s recent revision (of data for the week ending April 26) is the differences in the behavior of storage operators. If all the storage operators behaved the same, EIA’s estimation ratio would stay the same and there would not be any revisions based on changing to a new set of monthly data in its estimation process.

EIA uses a complex estimation formula that is understood by few, but impacts many. EIA’s weekly report is based on a weekly survey of a portion of the total storage operators in the country and then on a calculation taken from complete national storage data that was gathered months earlier. When a new set of monthly data is complete — which is usually three months behind, e.g., for the weekly storage report on Aug. 8 EIA started using May’s monthly data in its estimation process — EIA breaks down the data between the storage operators who are participating in its weekly survey and the remaining storage operators across the country who are not participating in the weekly survey. EIA determines what percentage of the whole its weekly survey participants represented, and then uses an inverse of that percentage, along with the data from its current weekly survey, to extrapolate the current weekly inventory for each region of the country and for the country as a whole.

One important issue that can cause fluctuations in the sequence of EIA weekly storage changes is the difference in the way the storage operators in EIA’ weekly survey manage their storage, compared to the way the other storage operators in the country do it. There may be more marketers holding capacity in one group or there may be more electric generators in one group. What that can do is make the transition from one set of monthly base data to the next cause a shift that may require EIA to report a revision if the shift leads to a change of more than 7 Bcf.

“If every storage operation in a region behaved the same way, then no matter what we used as a reference we would get the same ratio,” said Kass. “We knew that they probably didn’t [behave the same], but we didn’t know when we started this, nor do we know now, how different one operator is from another on a week-to-week or month-to-month basis. When we tooled this up, we didn’t have an opportunity to do any investigation over several years or whatever it would have taken to see what the behavior pattern of the different operators was.”

EIA made an assumption that the behavior of the weekly sampled companies is approximately that of the unsampled companies and it develops a new estimation ratio each month. That way if there are dramatic differences as the year goes on, the ratio would not represent a very great deviation from reality.

When EIA changed to the May reference period for the Aug. 8 report, a big difference popped up in the East (11 Bcf compared to the 7 Bcf threshold for a revision) from the previous week’s number. EIA books the current week and the previous week using the new reference month, so that people will have the ability to do an apples-to-apples comparison on the two numbers.

Kass said if EIA had calculated the data for the week ending Aug. 2 using the estimation ratio based on May monthly data, but then retained the April data for the previous week’s calculation, that would have given a much smaller implied weekly injection number. “If we had used the ratios from the two different months, it would have given an incredibly small and not correct difference of 12 Bcf less than what was reported (33 Bcf). That 21 Bcf injection would not have represented reality.”

Weekly changes in storage drive prices like no other information except news of some unusual cataclysmic event, such as a hurricane or pipeline rupture. Even the slightest difference between the reported weekly change and what was expected by the market can cause a dramatic shift in natural gas futures prices. Revisions to previously reported storage data also have a major impact on gas prices.

The 12 Bcf revision to two-week-old data reported on Aug. 8 was mainly to blame for the sudden 13 cent increase in September gas futures prices in the five minutes following the report, according to Pioneer’s Saal. But the misunderstanding of EIA’s storage methodology could lead to unexpected sharp changes in the weeks to come, he added.

“I don’t think injections will return to above 60 Bcf,” said Saal prior to last week’s report of a 53 Bcf injection. “You can’t think too logically with this. We just had a revision to a sequence of numbers. We have a new base month of storage data to work from. The numbers that we’ve been seeing were based off of one set of monthly numbers (April), and now we’re going off the May number. From what little I know about statistics, that is telling me they just lowered the base because they revised the [report for the week ending July 26] downward. My guess is the next sequence of numbers is going to be lower than the sequence we just had.”

Saal said one of his clients expected the next weekly injection number to be higher than the 60 Bcf-plus injections of the past few weeks because May’s working storage total was higher. “That’s not how it works,” noted Saal. “They changed the base that they were projecting off of and they revised the number downward. It’s a calibration thing. The technique that they use is different than the technique the AGA used. The AGA didn’t have complete monthly data on the nation’s entire storage situation. That’s a big difference in projections.”

People are used to the AGA, which revised annually its estimation number called “percent full” and what “full” was. AGA’s operationally full level of 3,200 Bcf is another area where AGA and EIA differ substantially, despite what many observers think. According to AGA, “full” is an average of where working gas was at the end of the injection season over the past several years.

“It’s a completely different concept than what we use,” said Kass. “It’s completely accurate for what it is, but I don’t think people understand what it is. If every reservoir in the country were filled to its capacity, it would be more than that. There’s a concept, which other people use, of working gas capacity or the difference between developed capacity and base gas, and the total of that would be more than 4 Tcf. I don’t use that as a measure either, ” he said. “Conceptually you could say that could happen, but would every reservoir ever be filled to its developed capacity? That’s something that’s unlikely to ever happen.”

Unlike AGA, EIA also states that its weekly data has a standard error of 2% of the total working gas on any given week. Last week that was about 51 Bcf, a huge number in terms of weekly changes, and one that, if it ever were fully realized, would have dramatic market implications.

There’s little doubt that the proposed changes to EIA’s methodology also will have significant implications for the market, which nearly always has a knee-jerk reaction to storage data when it’s released. One potential outcome of the next set of changes could lead to a sudden revision of more than 35 Bcf that would be announced on any morning (the day after it is discovered) during the week. That certainly would trigger a chain reaction through the market. One market observer predicted there would be traders (who had gone broke) picketing EIA if that occurred.

“I think we arbitrarily said 35 Bcf, but that was a straw man,” said EIA’s Kass. “What should be a large revision? If we find out, how can we fairly and equitably get the word out?” Those are some of the questions EIA wants answered in the comments filed. “AGA’s policy was to wait until the next week no matter what. But if we hear at 11 a.m. on Thursday that someone made a mistake, and this has an impact of really changing dramatically the number we just put out, our current policy says we don’t do anything in terms of distribution of that information for a week. Should we continue to do that? If not — and I think overwhelmingly people are saying ‘not’ — what should we do in terms of fairly and equitably getting the information out?”

The proposed new revision policy will allow EIA’s current method for “scheduled” revisions to continue, but new “unscheduled” revisions will be added. Under EIA’s current scheduled revision method, changes that are made during the week are reported in the next regularly scheduled weekly survey and occur when the cumulative effect of all reported changes is at least 7 Bcf at either a regional or national level. Changes can have various causes: a respondent may need to file more accurate data because of new information; data previously submitted may have been incorrect; data may be submitted late; there may have been a reclassification of working and base gas; there may have been a change to a field’s operating status; or — perhaps the least understood of the six — EIA may have incorporated a new reference month in its estimation process.

EIA’s proposal for “unscheduled” revisions would include changes of significant magnitude and interest. They would occur only if EIA has received changes to the data from the most recent report or the report prior to the most recent report that amount to at least 35 Bcf or one standard error of the national estimate at the end of March 2001 when inventories were 742 Bcf. They also have to be received so they can be disseminated 24 hours prior to the next weekly report. EIA said it would disseminate the unscheduled revision at 10:30 a.m. EDT on the next business day.

EIA has requested comments on whether the new revision policy is appropriate and whether the threshold criteria for the revisions and the other components of the proposed revision methodology are appropriate. It also has requested comments on any additional actions it might take to help “ensure and maximize the quality, objectivity, utility and integrity” of the survey.

One perhaps unexpected change is what Kass called a “smoothing effect,” which was developed from unsolicited comments about EIA’s estimation procedure. “We’re looking at smoothing the change to a new base month, so that rather than a step function month-to-month there would be a smoothing mechanism. We’re still going to make use of the monthly survey, but instead of taking the ratio from a single month we may take a three-month average or an exponential smoothing in which the most recent month gets the most weight and they successively lose weight as you move back in time.”

Kass hopes these changes will help the industry gain a better understanding of the survey and what the weekly numbers actually represent. “Right now people don’t really understand what our number is. I talk to traders and ask them what the first thing is that they look for and some say they look at the stock number for the week and others say they look for the difference from the previous week. But in the first five minutes I don’t know how much thought can really go into it.”

©Copyright 2002 Intelligence Press Inc. Allrights reserved. The preceding news report may not be republishedor redistributed, in whole or in part, in any form, without priorwritten consent of Intelligence Press, Inc.