The Natural Gas Announcement Day Puzzle.

• Date: 01 March 2021
• Author: Prokopczuk, Marcel

1. INTRODUCTION

   The natural gas market has undergone massive changes throughout the last decades, starting with its deregulation in the 1980s, the inception of the futures market in 1990, the inflow of financial investors at the beginning of the twenty-first century, and recent shifts in supply and demand due to the introduction of shale gas, a growing industry for liquefied natural gas (LNG) as well as increased attention related to climate change. Natural gas storage levels have always been an important indicator of changes due to their natural role as a buffer between supply and demand. As such, the release of the Weekly Natural Gas Storage Report by the Energy Information Administration (EIA), which contains information about the current storage level, draws attention from all market participants. When new information is released to an efficient market, participants adjust their expectations and prices accordingly. Figure 1 shows that more than 50% of the annual return of natural gas futures is generated on weekly EIA announcement days. Therefore, returns on natural gas futures are significantly different on EIA announcement days compared to non-announcement days. However, after controlling for the information of the announcement this difference should disappear.

   This article documents a significant difference between the average returns observed on EIA announcement days and non-announcement days. Puzzlingly, this difference in returns between announcement days and non-announcement days cannot be explained by the information content of the announcement. Indeed, we find a strong significant negative relationship between natural gas futures returns and the announcement surprise, but we cannot explain the return difference between announcement and non-announcement days. This result is robust after augmenting the model with supply and demand measures, spillover effects from options, energy, or equity markets, as well as commodity specific variables such as the slope of the futures curve, hedging pressure, liquidity, or volatility measures.

   At the intraday level, we decompose the return within a two-hour window surrounding the announcement into a pre- and post-announcement part. Curiously, the overall return divides equally into the pre-announcement part (49.4%) and the post-announcement part (50.6%). Albeit modest evidence for the leakage of information, this can only be a partial explanation as there is still a significant effect from the announcement. Lastly, we document that the pre-announcement return is entirely realized on days where the announcement surprise is positive, i.e., the published inventory exceeds analysts' expectations. The asymmetry of this result casts doubt on a simple explanation based on informed trading.

   From the perspective of an investor, this puzzling result raises the question whether the newly documented premium is economically large once transaction and funding costs are accounted for. Our results show that the simple strategy of opening a short position 90 minutes before the announcement and closing it 30 minutes afterwards yields a significant annual return of 12% (t-stat = 2.93) translating into a Sharpe ratio of 1.76 after transaction and funding costs. However, the time series of strategy returns and the accuracy of analysts' forecasts suggests that the anomaly has decreased in magnitude and efficiency has returned to natural gas markets, leaving open the possibility that our strategy was new to investors who are now arbitraging it away.

   Our work contributes to the literature on storage effects in energy markets. Linn and Zhu (2004) show that the intradaily volatility of natural gas futures is significantly higher in the hour surrounding the American Gas Association (AGA) report and this effect has carried on after the EIA took on the reporting. Gay et al. (2009) show that the announcement return is negatively related to the inventory surprise, i.e., when the reported inventory level is higher than analysts' expectations, futures returns tend to be negative and vice versa. Halova et al. (2014) find seasonal patterns relating to the withdrawal period from November to March and the injection period from April to October. During winter, when inventories are lower than on average, inventory shocks have a smaller effect on futures returns, while the effect is stronger in summer. They also find that the effect is weaker, when forecast dispersion is higher which in general is the case in winter where demand shocks due to weather are an important driver of energy prices. Chiou-Wei et al. (2014) show that the announcement effect is unique to the day of the announcement. Bu (2014), Ye and Karali (2016), and Miao et al. (2018) find similar results for oil and gasoline using the EIA Petroleum Report announcements. Ederington et al. (2019) revisit these studies, and find that analysts' natural gas forecasts efficiently impound the available time series information but crude oil forecasts do not. Demirer and Kutan (2010) and Schmidbauer and Rosch (2012) study the effect of OPEC announcements on crude oil markets. Wolfe and Rosenman (2014) show that announcements in oil and gas markets cause spillover effects to each other. Compared to studies for other energy markets and studies on the effect of crop reports on agricultural commodities (Adjemian, 2012; Mattos and Silveira, 2016), focussing on the EIA Weekly Natural Gas Storage Report provides a unique setting. The report only publishes storage information without any supplementary information on supply, demand, or future prospects of production, hence the effect can be clearly referred to the changes in inventory.

   Our work also relates to the broader literature on the effects of scheduled news on energy prices. Basistha and Kurov (2015) study the effect of Federal Open Market Committee (FOMC) announcements on energy prices. For crude oil, Kilian and Vega (2011) and Chatrath et al. (2012) find no evidence to suggest that energy prices respond to macroeconomic news. They conclude that crude oil prices are predetermined with respect to macro aggregates, confirming the view of Kilian (2009), that prices are determined by flow supply and flow demand.

   Moreover, our work adds to the growing literature on analyzing risk premia on announcement days. Savor and Wilson (2013) find that 60% of the annual equity risk premium can be earned by only investing when important macroeconomic news is released. They interpret this finding as the premium investors demand for bearing macroeconomic risk. Ai and Bansal (2018) develop a theoretical framework that explains the announcement premium with the generalised risk sensitivity of investors used as evidence for a class of non-expected utility models. Relative to these studies, we focus on announcements of natural gas inventories, which are presumably asset specific news. We find a sizeable premium on these days suggesting a risk premium for idiosyncratic news. The sign of the premium, however, is puzzling as it is negative, hence penalizing investors for holding the asset on the day of the event.

   The intraday analysis in this article is related to the work of Lucca and Moench (2015), who document the pre-FOMC announcement drift in the U.S. equity market. As Brusa et al. (2019) show, the Fed is unique in channelling such an effect compared to other central banks. The EIA report plays a similar role for the US natural gas market. Gu and Kurov (2018) find a pre-announcement drift, and link it to informed trading caused by superior forecasting abilities of certain participants. Rousse and Sevi (2019) find evidence of an asymmetric response of crude oil returns to the EIA Petroleum Report. Our study reveals that the documented pre-announcement effect in the natural gas market is asymmetric and only accounts for half of the entire return, and therefore casts doubt on an explanation based on informed trading.

   Lastly, our work relates to the literature on the pricing of commodity futures. Brown and Yucel (2008) show that natural gas markets are driven by weather, inventories, and spill-overs from crude oil markets. Besides these supply and demand driven factors, we relate to the growing literature on factor models for commodity futures that include hedging pressure (De Roon et al., 2000), open interest (Hong and Yogo, 2012), idiosyncratic volatility (Fernandez-Perez et al., 2016), or the slope of the futures curve (Szymanowska et al., 2014). We confirm earlier studies that the listed variables affect natural gas returns. However, they are not able to explain the EIA announcement effect.

   The remainder of this article is organised as follows. Section 2 describes the data and introduces the main variables. Section 3 documents the EIA announcement effect and explores possible explanations. Section 4 looks at the intraday frequency. Section 5 discusses robustness checks. Finally, Section 6 concludes.

2. DATA & VARIABLES

   From Bloomberg, we obtain the daily price, trading volume, and open interest series of 499 Henry Hub natural gas futures contracts (Ticker NG) from March 2003 to December 2018. (1) Since we are dealing with futures contracts, we need to construct an investable price series by rolling over contracts before expiry. (2) We follow Szymanowska et al. (2014) and roll over the entire curve at the end of the month preceding the month prior to delivery, i.e., the return on the futures price series is defined as

   [mathematical expression not reproducible] (1)

   where [F.sup.(n).sub.t] is the price of the nth nearby on day t. We provide summary statistics on the returns of the first six nearby contracts in Table 1, that confirm common characteristics of natural gas markets. We find a strongly negative average return of -31.65%, which is in line with other studies (de Groot et al., 2014; Paschke et al., 2020). Further, we see high volatility of up to 45% per annum and a decreasing pattern of volatility in...