EIA Storage Announcements, Analyst Storage Forecasts, and Energy Prices.

• Author: Ederington, Louis H.

The energy sector has and continues to undergo revolutionary, what some have called epic, changes over the last 15 years, including the introduction of abundant, lower-cost energy sources like shale oil and natural gas owed in large part to significant and ongoing technological developments. These developments have been accompanied by a shift in investor focus regarding portfolio holdings of oil and natural gas futures as well as the development of new information providers. (1) The result has been a significant increase in open interest and volume in both oil and natural gas futures. (2) Notwithstanding, and perhaps because of, these developments, market participants pay close attention to fundamentals--in particular the weekly announcements of the change in oil and natural gas in storage by the Energy Information Administration (EIA) and how they relate to analyst forecasts of those changes. In particular, it is well documented that energy prices react to the surprise in weekly EIA storage announcements, where the "surprise" is generally measured as the storage figure announced by the EIA minus the median analyst forecast as compiled by Bloomberg, Reuters or some other service. (3) Left largely unexplored are: (1) how accurate and efficient these analyst forecasts are, (2) how they (as opposed to the EIA announcements per se) contribute to price discovery and informational efficiency in energy markets, and (3) how the energy market reaction to the subsequent EIA release depends on characteristics of these analyst forecasts. A comprehensive up-to-date study of these issues is both warranted and crucial to interpreting and predicting the impact of storage surprises on oil and natural gas futures prices.

In this paper, we provide a fresh look at these questions, building on and going beyond the existing studies documenting the price reaction to the EIA storage announcements. We first examine the accuracy and efficiency of the analyst storage forecasts and explore what drives these forecasts. We then investigate how efficiently natural gas and crude oil prices react to both the analyst forecasts and the subsequent EIA announcements and how the reaction to the EIA announcements depends on analyst forecast uncertainty.

The data show that analysts do not always agree when it comes to inventory change forecasts for the oil and natural gas markets. By-and-large an in-depth examination of the determinants and properties of analyst forecasts has not received the attention one might expect; instead most of the literature has focused only on the relation between unexpected changes in inventories and unexpected changes in futures prices. Our analyses showing how analyst forecasts behave and have changed and how those forecasts contribute to price discovery and informational efficiency in the energy markets contribute to the literature the following new findings. One, the median natural gas analyst storage forecast reported by Bloomberg is highly accurate in anticipating the EIA storage figures, accounting for over 99% of the variation over time in natural gas storage levels. Analyst crude oil storage forecasts, on the other hand, are much less accurate. This difference in forecast accuracy is partially due to seasonal patterns and serial correlation being much stronger in the natural gas market, but both forecasts clearly bring information to the market beyond seasonal patterns and past storage flows. Two, while analyst natural gas storage forecasts are basically unbiased, analyst crude oil forecasts tend to underestimate upcoming changes in crude oil storage. Three, analyst natural gas forecasts efficiently and completely impound the available time-series information but analyst crude oil storage forecasts do not. Four, analyst storage forecast dispersion has decreased for both natural gas and crude oil. However, while forecast accuracy has improved considerably over time for natural gas, this has not occurred for crude oil. Five, natural gas and crude oil prices appear to react to the new information in the analyst forecasts prior to the EIA announcements. Six, we find that a storage flow higher or lower than analysts had expected one week tends to be partially reversed the following week, especially in the natural gas market, suggesting that storage forecast errors are partially due to transitory forces that tend to reverse in the future. In other words, in some cases, analysts correctly forecast the storage flow but miss the timing. Seven, natural gas price responses to EIA natural gas storage announcements are considerably greater when the dispersion of analyst forecasts around the mean forecast is small than when the forecast dispersion is large. Our interpretation of this finding is that when different analysts forecast roughly the same storage level change, traders have more confidence in the forecasts and thus are more surprised by a given size forecast error. When analyst forecasts widely disagree, the same size forecast error evinces less of a surprise and the market reaction is smaller consistent with traders being less confident about the forecasts. Eight, contemporaneously, absolute analyst natural gas forecast errors are positively correlated with the degree of disagreement among analyst natural gas forecasts. Inversely, a larger absolute forecast error one week is generally followed by greater disagreement among analysts in subsequent weeks, indicating that analysts tend to disagree on the cause and permanence of the unexpected storage flow. By providing a comprehensive investigation of how analyst forecasts contribute to price discovery and informational efficiency in both natural gas and crude oil markets, our paper joins the growing literature on energy market informational efficiency.

The remainder of the paper proceeds as follows. Section 1 reviews the existing evidence. Section 2 describes the data. Section 3 reports our empirical findings. Section 4 concludes.

(1.) RECENT EVIDENCE

While several studies of the oil and natural gas markets report that analysts disagree on how inventories will change in any given week, little evidence exists on the causes, consequences, and behavior of such disagreement. Further, most studies focus on the forecast error, measured usually as the difference between the forecasted change and the actual change announced by the EIA with no attention to the determinants of the former. We present new evidence on both the forecast error and the dispersion of analyst forecasts in the natural gas and oil markets.

Central to any such examination is whether the forecasts are unbiased. The literature on this point is limited. The exceptions include Ye and Karali (2016), who study oil forecasts from Reuters during the period August 26, 2012 to December 30, 2013 and find some evidence of bias. Similarly, Chiou-Wei et al. (2014) in a study of Bloomberg's natural gas inventory forecasts during the period from August 30, 2002 to August 18, 2011 also find evidence of bias. Bu (2014), on the other hand, finds results consistent with unbiasedness for the Reuters oil forecasts for the period May 10, 2006 to Oct. 31, 2011. We take a fresh look at this question for both the oil and natural gas markets taking into account additional as well as more recent data explicitly accounting for the significant changes in these markets during the last 15 years.

Chiou-Wei et al. (2014) show that forecast dispersion (standard deviation of analyst forecast changes) for Bloomberg forecasts of natural gas varies across the seasons of the year for 2002-2011 and Gay et al. (2009) find that forecast accuracy tends to be worse during the "withdrawal" season from November to March. Halova et al. (2014) show that for the period 2003-2012 Bloomberg analyst forecast dispersion tends to be larger during the withdrawal season and also report evidence that forecast dispersion has declined over time for natural gas. Similar evidence on oil storage forecasts is lacking. We present evidence on both the oil and natural gas markets expanding the analysis to encompass the last 15 years.

Previous studies have documented that natural gas and crude oil prices react negatively to the forecast error in EIA announcements and that the price reaction is stronger the larger the forecast error: Gay et al (2009), Chiou-Wei et al (2014), Halova et al (2014), Chang et al. (2009), Elder et al. (2013), Bu (2014), Ye and Karali (2016), Miao et al. (2018), Linn et al. (2017), and Bjursell et al. (2015). However, evidence on whether analyst forecast disagreement influences the price reaction is limited and mixed. Chiou-Wei et al. (2014) in a study of natural gas storage announcements find no evidence that analyst forecast dispersion influences the price reaction for the period they study (August 30, 2002-August 18, 2011). Halova et al. (2014) in contrast find that the response coefficient is greater in absolute value when forecast dispersion is above the median forecast dispersion for their sample period. We present new evidence on the price response for both oil and natural gas accounting for forecast dispersion.

Gay et al. (2009) report that the market appears to condition natural gas futures price responses on how accurate overall the forecasts of the contributing analysts have been. However, the response coefficients accounting for accuracy do not differ significantly in magnitude. Their study period ends in August 2005. Chang et al. (2016) examine the forecast accuracy of Bloomberg oil inventory forecasts for the period June 2003 through March 2005. In contrast, our study spans the 2003-2017 period and includes analyses of both the natural gas and crude oil storage forecasts produced by Bloomberg. Importantly and as emphasized earlier, our sample period spans an epic period of significant change in both oil and natural gas markets not heretofore examined.

(2.) DATA

Each Thursday, the EIA issues a natural gas storage...