USDA Announcement Effects in Real Time

Transcription

1 USDA Announcement Effects in Real Time Michael K. Adjemian 1 and Scott H. Irwin 2 1 Economic Research Service, United States Department of Agriculture, Washington, DC 2 Department of Agricultural and Consumer Economics, University of Illinois at Urbana-Champaign Abstract This version: 6/8/2016 In 2012, the Chicago Board of Trade (CBOT) eliminated a morning trading halt that coincided with the normal publication time for important U.S. Department of Agriculture (USDA) commodity reports. Previously, market participants had hours to review the information in the reports and adjust their trading strategies in advance of market re-opening. We use intraday grain futures market price and volume data to show that, without a trading halt, ensuing real-time trading on USDA WASDE and Grain Stocks announcements exhibits volatility spikes in the corn, soybean, and wheat markets, but that this heightened volatility dissipates quite rapidly. Our results demonstrate that, although real-time trading exhibits larger immediate USDA announcement shocks via heightened volatility relative to same-regime non-announcement days, crop markets take nearly the same time to fully absorb these shocks, following a very similar time path. When we account for the fact that announcement times in the real-time format no longer coincided with the opening of pit trading on non-release days which themselves drew larger spells of price volatility real-time trading exhibits far faster absorption times despite the higher initial shock. Because markets incorporate news more quickly under a real-time format, re-imposing a timeout would carry a substantial cost to price discovery. Keywords: announcement effects, CBOT, commodity futures, intraday, trading hours, price discovery, price volatility, USDA This material is based upon work supported by Cooperative Agreement with the Office of the Chief Economist, U.S. Department of Agriculture, under Project Number Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the authors and do not necessarily reflect the view of the Office of the Chief Economist or the U.S. Department of Agriculture. This article may not be cited without the permission of the authors. 1

2 Until May 2012, U.S. Department of Agriculture (USDA) crop and inventory reports were released outside of the regular trading hours for the Chicago Board of Trade (CBOT) grain futures markets. 1 This decadeslong convention afforded market participants a lengthy timeout to read and digest the information provided by these important USDA reports and then place informed trading orders in advance of the market opening, including the commencement of live floor (pit) trading at 9:30 Central Time (CT). 2 After the Intercontinental Exchange (ICE) announced that it would introduce competing, look-alike grains contracts that allowed trading during the timeout (Arasu and Saphir, 2012), the CBOT eliminated the longstanding trading halt that had occurred each morning from 7:15-9:30. 3 As shown in figure 1, this policy of trading on USDA announcements in real-time marked a sweeping departure from the established, historical practice. The possibility of active trading during the publication of important USDA reports drew strong concerns from some market participants. Of the 147 comments about these developments received by the National Agricultural Statistics Service (NASS) and the World Agricultural Outlook Board (WAOB) of the USDA, and published in the Federal Register, 64% were in favor of some sort of timeout including the CBOT itself and prominent industry associations like the American Farm Bureau Federation (USDA, 2012). Most conspicuous among the concerns expressed by survey respondents were worries over access equity and the increased price volatility that could result from real-time trading around USDA announcements. Several large commercial firms and individual farmers involved in the actual production, merchandising, and processing of agricultural commodities worried that the new announcement format favored nontraditional, speculative traders who have better access to sophisticated, algorithmic trading 1 The CBOT merged with the Chicago Mercantile Exchange (CME) in 2007 and became known as the CME Group. We refer to the corn, soybean and wheat contracts as CBOT contracts because they originated on the Chicago Board of Trade and this is still the official market designation for regulatory purposes. 2 For consistency, and to save the reader from unnecessary acronyms, all times in this article are reported in Central Time and a 24-hour format. 3 USDA crop and stocks reports were published at 15:00 after the close of trading until 1994, when the release time was changed to 8:30, before markets opened. USDA instituted this change due to concerns that U.S. futures exchanges might lose trading volume after the release of USDA reports to non-u.s. exchanges, particularly in Japan, that were open at 15:00. See Colling (1993) for interesting background on this change. 2

3 methods. Institutional trading firms have been known to spend hundreds of millions of dollars to install transcontinental fiber-optic cables and microwave systems that shave a few milliseconds off the arrival of information (Troianovski, 2012), which is then used to auto-execute trades according to pre-programmed instructions. 4 In contrast, participants in rural areas face limited internet bandwidth (Schmitz, 2012). Without the trading pause afforded by the timeout, the fastest to access new government data that routinely affects market prices has a clear trading advantage. In addition, critics worried that under the new trading format agricultural futures markets whose price volatility was already ordinarily heightened around USDA reports due to uncertainty about the implications of government expectations about fundamentals could be even more volatile or susceptible to overreaction as market participants attempted to sift through sometimes complex reports to settle on a consensus price expectation in real-time. In such an environment, trading strategies like stop-loss orders are more liable to be prematurely filled during more frequent large, post-announcement swings, complicating certain hedging practices. 5 Large enough reactions could even make price-limited trading periods more frequent, preventing normal futures market operation for an extended period of time. In theory, trading halts like timeouts allow markets time to resolve transitory information asymmetries that arise after the announcement of important information, permitting traders time to revise buy and sell orders once they ve considered a report s implications. Halts are routinely observed in equity markets when a firm publishes information about itself that may impact its share price (Hauser, et al., 2006). In an active trading format, with little time to review such information, traders may hesitate to initiate, or submit smaller-sized trades if they believe that counterparties have faster access to information, leading to a delay in the market s adjustment to news (Joseph, 2014). Longer adjustment times following halts have 4 From November 2012 October 2014, automated trading accounted for 39% of the volume in CBOT grain and oilseed markets (Hayes and Roberts, 2015a). Hayes and Roberts (2015b) also show that in the 10-Year Treasury and E-Mini S&P 500 markets, automated trading dominated aggressive (or liquidity taking) trading from the 2 nd -11 th trading seconds following government employment announcements, from November 2012-November Just after macroeconomic news announcement time, the resting time of passive orders particularly for manual traders spikes in the markets studied by Hayes and Roberts (2015b). 3

4 been demonstrated empirically in equity markets (e.g., Moshirian, et al., 2012), and for CBOT corn (Kauffman, 2013) and soybeans (Joseph, 2014). Even though timeouts may lead to more efficient post-announcement re-opening prices, by definition hey also introduce a substantial, structural lag into the own-market price discovery process at the CBOT during the many decades when the trading halt was observed, the lag lasted at least two hours potentially prompting some participants to trade on that information elsewhere. 6 An open market at least offers the opportunity for traders to discover prices quickly and transparently; under the timeout, only traders in open foreign or over-the-counter markets are permitted to trade actively on USDA information. That is, price discovery is occurring in response to government news, it is just not observable in a halted market. Indeed, planned trading halts around the release of public information are far from universal. Nonagricultural commodity futures markets have long traded actively during the release of important fundamental information; for example, the New York Mercantile Exchange (NYMEX) natural gas market remains open during the publication of weekly storage figures that regularly generate sizeable announcement effects. Traders are also free to act in all major domestic commodity markets during the publication of important U.S. macroeconomic announcements, such as the Bureau of Labor Statistics Employment Situation Summary (see, e.g., Hayes and Roberts, 2015b). Beyond lifting the traditional trading halt, the CBOT and USDA have made several subsequent changes that also impact trading on report days. 7 In June 2012, the CBOT began opening grains pit trading at 7:20 on some announcement days, ten minutes in advance of USDA's report time (Berry, 2012); it also extended pit trading by 45-minutes to match the later close of its electronic platform (Polansek, 2012). With the aim of allowing markets the best chance to absorb news at a time of ordinarily high liquidity, beginning in January 2013 USDA moved its report publication time to 11, three-and-a-half hours later than 6 Concern over its market share of agricultural commodities trading prompted the CBOT to drop the timeout convention in the first place. 7 For sample size purposes, we divide the data according to two regimes: Halt era announcements (following a timeout), and No-halt announcements. 4

5 it had been for nearly twenty years (Abbott, 2012). 8 Finally, in April 2013, the CBOT reduced grains trading hours and re-established a morning trading halt, from 7:45 8:30, although the new timeout did not coincide with release times and real-time trading on important USDA reports continued. We focus on the intraday price volatility and trading volume patterns around the announcement of the two most important USDA reports: monthly World Agricultural Supply and Demand Estimates (WASDE) and quarterly Grain Stocks estimates. 9 Due to their sensitive nature, both reports are prepared in a quarantined, high-security environment known as lockup ; once they enter the preparation area, analysts involved in the composition of these reports are unable to communicate with the outside world until official publication. As a balance sheet of consensus fundamental estimates for major domestic agricultural commodities, WASDE s preparation is administered by the World Agricultural Outlook Board, and draws on the expertise of four USDA agencies: the Agricultural Marketing Service, the Economic Research Service, the Farm Services Agency, and the Foreign Agricultural Service. Any WASDE can include changes to foreign supply and demand figures, domestic consumption, imports, exports, and ending stocks. In contrast, domestic crop production figures follow a fixed schedule. In certain months, survey acreage and production estimates prepared by the National Agricultural Statistics Service (NASS) are incorporated into the WASDE. NASS surveys also serve as the foundation for data published in the Grain Stocks reports. Only the January WASDE and Grain Stocks reports are released contemporaneously. Numerous researchers have found that commodity futures prices react to USDA news (Adjemian, 2012, Dorfman and Karali, 2015, Fortenberry and Sumner, 1993, Isengildina-Massa, et al., 2008, McKenzie, 2008, Sumner and Mueller, 1989). Most of the extant research, however, has focused on the daily change in prices resulting from report publication. Although end-of-day data can reveal important information about the existence of announcement effects, intraday price volatility and trading patterns offer several advantages. First, they can reveal the efficiency of the price discovery process, documenting how 8 Many of the survey respondents to the Federal Register notice mentioned on the prior page called for just such a timing change. 9 We consider the average announcement effect of a generic report, be it a WASDE or Grain Stocks release. 5

6 traders respond to information shocks and how long it takes for the market to reach a new equilibrium; market response to news generally occurs at a far shorter time scale than can be detected by measuring the difference in consecutive closing prices (Lehecka, et al., 2014). Second, intraday data can also detect changes in the price discovery process that would be far more difficult to uncover with daily data alone. Up to now, inattention to agricultural markets at the intraday level is probably due to the market structure at the major agricultural exchanges: an important stream of the intraday trading literature is concerned with how markets respond in real-time to scheduled government news releases, since these are predictable. Before the CBOT abolished the USDA report timeout, such a study was impossible to perform for corn, soybeans, and wheat. Two recent studies focus on the elimination of the timeout in agricultural futures markets. Using a limited number of No-halt period reports, Kauffman (2013) finds that the CBOT corn market displays higher volatility, over a longer duration, during actively traded USDA announcements. With a larger sample size, Joseph (2014) identifies a similar finding using soybean prices: a larger initial shock, and longer persistence of heightened volatility. We use over five years of electronic-platform tick data from to explore the reaction to WASDE and Grain Stocks reports in the CBOT corn, soybean, and wheat futures markets, and determine the effect of the elimination of CBOT timeouts. We focus on the magnitude of the initial shock resulting from government news, the duration of heightened volatility, and the path of news absorption, comparing the results across release regimes. We also identify whether hitting price limits or market over/underreaction is any more likely now that traders are free to act on USDA data at its announcement time. Because trading hours and report publication times have changed several times over the period of observation, we avoid comparisons based on absolute time and instead introduce a timescale based on the trading-time distance to the moment of USDA publication. This method permits a direct comparison of price volatility and trading volume in the minutes and hours surrounding announcements. 6

7 DATA We use intraday price and trading volume (termed time & sales ) data for CBOT corn, soybeans, and wheat electronic futures markets from July 20, 2009 through July 22, Over this interval, electronic corn, soybean, and wheat markets accounted for 80,339,922, 87,766,365, and 36,950,578 trades, respectively. The CBOT time-stamps all trades in the time & sales data to the second; in most of our analyses, we aggregate these to the trading minute for tractability. The data show that although the grain markets were open for nearly 1.35 million trading minutes, not all of these include executed trades, with most of the unutilized trading time occurred in low-liquidity overnight hours. For corn, 947,440 trading minutes saw positive trading volume, and executed trades occurred in 1,003,766 and 810,205 trading minutes for soybean and wheat markets, respectively. We focus on the release of two of the most important USDA reports: the monthly World Agricultural Supply and Demand Estimates (WASDE), published by the World Agricultural Outlook Board (WAOB), and the quarterly Grain Stocks report, released by the National Agricultural Statistics Service (NASS). 10 Release dates and times for historical USDA reports are published on the USDA calendar, at the department s website. 11 The period of observation included 1273 trading days. During that time, USDA published 60 WASDE reports and 20 Grain Stocks reports. 12 Because the January WASDE and Grain Stocks reports are released on the same day (and at the same time), there are a total of 75 announcement or release days in the data. We term the 1198 trading days without one of these reports nonannouncement or non-release days. In the data, 43 release days and 685 non-release days occurred 10 Several USDA reports are published concurrently with the WASDE, such as Crop Production reports. Because separate effects cannot be identified, we summarize their collective impact under the banner WASDE report. See Isengildina-Massa et al. (2008) for further discussion of this point. 11 At this writing, the report calendar can be accessed at: 12 USDA did not publish an October WASDE during the government shutdown of

8 before the transition to real-time trading on USDA announcements, while 32 release days and 513 nonrelease days occurred afterwards. 13 Previous research on intraday grains announcement effects has focused on measuring volatility and trading volume in either the nearby or harvest contract series (Joseph, 2014, Kauffman, 2013, Lehecka, et al., 2014). 14 Trading interest in these contracts is highly seasonal, i.e., the nearby series generally dominates the harvest contract s trading volume from December through early-june. Conversely, the harvest contract is more liquid from late-june until December. From the time that the September contract delivers until the December contract matures, the harvest and nearby contract series are one and the same. Because low trading volumes and liquidity widen the bid-ask spread (Wang, et al., 2014), they can make intraday trading appear more volatile as traders must cross a wider price divide to secure a transaction. Rather than simply dropping potentially useful data in low-trading months or leaving our results susceptible to confounding via sparse trading, we instead generate a composite contract series that chooses each day s trading data from the nearby and harvest contract series, based on the contract that has the highest trading volume. 15 METHODOLOGY We study the effect of shifting trading hours on intraday volatility and trading volume patterns using the following steps. First, we determine whether CBOT grain markets were any likelier to hit exchange-set price limits after the elimination of the traditional trading halt. Next, we use non-parametric methods to identify whether price volatility or trading volume has changed now that active trading is allowed at 13 The no-halt era includes several different release regimes, differing by trading hours. To increase our sample size for valid statistical comparisons, we pool the observations from the no-halt era, even though they have different publication and market closing times. 14 The nearby series is spliced from sequential futures contracts as they become nearest without yet reaching their delivery period. For corn, the harvest contract series splices together sequential December delivery futures contracts, again avoiding their delivery period. 15 Prices for grains futures contracts are tied together by storage across delivery horizons, particularly within a given crop year (Working, 1948), so USDA announcements should affect price reactions to the constellation of futures expirations about equally. Adjemian (2012) empirically verified this for several commodities and expirations. 8

9 announcement time. To accomplish this, we (1) measure the size of the announcement shock, and then (2) test for how long it takes the market to return to a normal pattern of volatility and trading volume or absorb the announcement shock as evidenced by non-announcement-day market behavior, and then (3) attempt to correct (2) for the fact that halt-era announcements regularly coincided with higher nonannouncement day volatility. Finally, we search for systematic under- and overreaction to the news contained in the reports, to determine if active trading makes the market any more likely to temporarily under- or overshoot. Approach to Tranquil Markets and Price-limited Trading Many times throughout the sample, no trading occurs for an extended period of time even though grain markets are open. Rather than treat these as missing observations, as some prior research has done, we instead assign these periods a price volatility of zero. Trader inaction is thus considered valuable information for the purposes of our analyses. Likewise, we identify in the data likely instances when exchange-set daily price limits had been reached. These spells are characterized by extended periods of trading (or non-trading) at a locked price equal to the maximum allowable difference from the market closed observed on the previous trading day. 16 Price-limited spells would otherwise appear to be zero-volatility, tranquil trading periods, while they in fact conceal market volatility from observation. Including them in the analysis would bias our volatility and volume estimates. We create dummy variables to account for these price-limited trading periods, as well as trading days when price limits had been reached at least once. Identifying the Basis of Comparison: Trading Distance from Announcement Time Shifting trading and announcement formats from complicate a comparison of intraday trading patterns based on absolute time for two reasons. First, timeouts obviously prevent the observation of price 16 CBOT price limits changed over the period of observation, so we identified in the data necessary conditions for a limit, and then investigated each in detail to confirm it. 9

10 volatility and volume, even though these would clearly be positive if unconstrained. Second, the change in announcement time means that both price volatility and trading volume concentrate at different times of the day over the sample. Figure 2a demonstrates the relationship between changing announcement regime and trading periods, and average intra-minute high-low price volatility. The largest volatility spikes occur around the trading time nearest USDA s scheduled announcement, i.e., just before 7:15 and at 9:30 in the trading-halt era, at 7:15 from May-December 2012, and at 11:00 afterwards. Consequently, we adopt a timescale based on the trading-time distance to the normal moment of report publication, by announcement regime. This methodology affords a direct comparison of price volatility and trading volume in the minutes and hours surrounding announcements. Figure 2b depicts the same information in figure 2a, but reorganized so that all announcements occur at time 0, and trading halts or breaks are removed. Using this approach, we can observe how prices and volumes behave around an event of interest, and compare them across different trading and publication-time regimes on a consistent basis. We split the trading day into trading minutes, indexed to the report announcement time. For each trading day, every trading minute, is indexed according to its trading date d, and distance in trading minutes, t, after the first trading minute following the regular USDA report announcement A time during that trading regime. To summarize all the information represented by a given trading minute, we write. Until May 2012, 7:14:00 7:14:59 was the last trading minute before the report came out, or. During that month, the first trading minute post trading-halt, or, was 9:30:00 9:30:59, and the next trading minute, or, was 9:31:00 9:31:59. We term this period as the No-halt era announcement regime, or Regime 1. In June 2012, after the transition to real-time trading, and and were 7:14:00 7:14:59, 7:15:00 7:15:59, and 7:16:00 7:15:59, respectively; this is referred to as announcement Regime 2. After USDA moved the report announcement time in January 2013, corresponded to 10:59:00 10:59:59, to 11:00:00 11:00:59, and to 11:01:00 11:01:59. This period represents announcement Regime 3. Under Regime 4, which commences in April 2013, the release time was the 10

11 same as in Regime 3, but the CBOT introduced a biscuit break timeout for grain futures contracts between 7:45 8:30, otherwise active trading on USDA reports was still permitted. To increase our sample size and permit more statistically robust comparisons, we pool all the active trading observations, Regimes 2-4, into a single No-Halt era announcement regime. Calculating Price Volatility and Trading Volume The set of prices and volumes traded in each minute is given by the vectors, and,, respectively. For robustness, we calculate each return, based on, using two different methods. Two recent studies focus on the period-to-period in our case, minute-to-minute return (Joseph, 2014, Lehecka, et al., 2014). The minute-to-minute return is calculated as the difference between the last traded price in consecutive trading minutes:,,, 100 In the expression above, the operator chooses the last recorded log price in each trading minute. In cases where the last traded price occurred in non-consecutive minutes, we identify the last traded minute and substitute it for, ; all interim minutes when the market is open but without trades are assigned a volatility of zero. An alternative measure of return is the hi-low difference, representing the maximum difference that futures prices achieved during each trading minute:,,, 100 As the difference in log prices, the high-low value has a lower bound at zero, and does not express the directionality of price changes. However, as a range estimator hi-low return has the advantage of capturing price variation within a trading minute. It also draws on nearly twice the amount of information than the minute-to-minute volatility does, and arguably presents a fuller portrait of volatility dynamics Across n trading minutes with positive activity, the minute-to-minute volatility series is calculated using n-1 price observations. Over the same time period, the hi-low volatility series is calculated using 2n price observations. 11

12 Once the return measure is selected, we calculate volatility by averaging over the period of observation for each trading minute. 18 For example, a simple mean of the minute hi-low observations during the halt-era, with d=1, 2,... H daily observations, is given by: 1, To capture trading volume patterns, we use the average trading volume on a trading minute basis, by taking the simple average of the number of contracts, traded in all minutes. For example, during the no-halt era, the average number of contracts that changed hands in minute over days d=1, 2, NH is calculated as: 1, Comparing Intraday Patterns in Price Volatility and Trading Volume Following Lehecka, Wang, and Garcia (2014), we use nonparametric tests to search for differences in the distributions of volatility and trading volume levels on a per-minute basis following a policy shock. We compare the average difference (and the significance of that difference) between per-minute volatility (and volume) on announcement and non-announcement days, within each trading regime. Then, we compare the patterns in these average differences before and after the cessation of the timeout, to determine if the magnitude and/or duration of elevated market volatility and trading generated by USDA announcements changed as a result. One complication to our identification strategy is that the policy shift to active trading at USDA announcement times also disconnected trading on non-announcement days from the potentially higher market volatility that can be experienced at the commencement of open outcry trading hours. During the 18 We also considered expressing volatility using the coefficient of variation applied to intraminute returns, i.e., the mean return divided by the standard deviation of returns for all trades the occur in a given m d,t. We choose to present mean hi-low volatility for much of this article because it representative of all the candidates. 12

13 halt-era in our sample period, USDA announcements were released during the timeout and both electronic and pit trading would commence simultaneously two hours later. After the elimination of the timeout, electronic and open outcry markets were open during USDA announcements (Polansek and Nelson, 2012), but only electronic markets were open at that time on non-announcement days. Differences in the baseline level of market activity due to pit trading at the normal announcement time on non-release days could confound our estimate of elevated volatility and trading volume due to the policy change. 19 We attempt to account for this problem by performing a counterfactual analysis, accomplished by differencing the average per-minute volatility and volume levels on release days in the no-halt era, from the values observed on non-release days in the halt era. That is, the goal of the counterfactual analysis is to judge the size of the no-halt era announcement shock and duration, assuming that non-announcement day volatility remained unchanged from the halt era. We report our results in the form of an absorption path to USDA information shocks, which reveals the time it takes for elevated release-day levels of volatility and trading volume to dissipate, i.e., so that they are statistically indistinguishable from non-release day observations. To estimate the average absorption path, we first calculate the mean hi-low return and the average trading volume on a per-minute basis for release and non-release days in both regimes (e.g., columns 1 and 2 in table 3a). Within each regime, we then identify statistical differences between these distributions per-trading minute using nonparametric tests, 20 and calculate the cumulative significant differences between the distributions over the first 100 minutes of trading after the normal USDA release time (e.g., columns 3 and 4 of table 3a). We report the average percent of USDA announcement absorbed as the percentage of the cumulative difference between release and non-release days realized by that trading minute, within each regime (e.g., column 5 of table 3a, which is also depicted graphically in figure 6). The same approach is used to 19 Baseline corn market volatility levels at and after the normal USDA announcement time were much higher before the timeout was removed when the resumption of trading coincided with the opening of the pit. 20 Partitioning the trading dates in the sample into the two announcement regimes, we perform Kruskal-Wallis tests on the hi-low and absolute minute-to-minute returns for each m t. The null hypothesis is that return distribution is the same on announcement and non-announcement days. Similarly, we perform Wilcoxon tests on the average volume to search for significantly different trading patterns due to USDA announcements, within each regime. 13

14 perform the counterfactual analysis. To search for differences in an even tighter window about the announcement time, we also isolate and compare the volatility and volume generated in the few hundred trades pre- and post-announcement, before and after the elimination of the timeout. Testing for Market Under- and Overreaction We also assess the grain returns data for a tendency to under- or overshoot based on the news in USDA reports, to determine whether this is any more likely now that markets stay open at publication time. These instances are characterized by a correlation between the returns at and for 1 or the cumulative return measured from to : as the market attempts to adjust to new information. Specifically, under-reaction implies that, 0 or, 0, as traders methodically adapt to the new equilibrium in the same direction, while overreaction implies that, 0 or, 0 as the market bounces around before settling on the consensus price expectation based on the information in the report. We test for these systematic inefficiencies using Pearson s correlation and Spearman s rank correlation, and use minute-to-minute returns since they are capable of expressing directionality. RESULTS AND DISCUSSION Descriptive Statistics Accounting for tranquil trading periods by assigning inactive trading minutes a return value of zero reduces all average returns and their standard deviations, and increases the population size. For all grains, returns are significantly skewed and leptokurtic, and therefore non-normally distributed. This is consistent with the findings of other research on agricultural futures returns (Isengildina-Massa, et al., 2008, Joseph, 2014, Lehecka, et al., 2014, Yang and Brorsen, 1994). Table 1 displays trading minute level descriptive statistics for the grain market returns and trading volume data that we used in the analysis, segregated by the announcement regimes over the period of interest. For all commodities under study, Kruskal-Wallis tests 14

15 reject equality of both the volatility and volume distributions between announcement regimes at the 5% level, for trading days with or without USDA releases. In the table, grain markets are less volatile with a lower mean and standard deviation of hi-low returns on non-release days in the No-halt era than in the Halt era. For release days, the results are mixed: average per-minute wheat volatility (mean and standard deviation) fell; the reverse is true for CBOT corn and soybeans. Occurrence of Price-limited Trading By preventing prices from straying too far from the closing price of the previous trading day, price limits can serve to censor announcement effects in terms of both volatility and volume, making markets appear calmer than they actually are. But even though latent market sentiment during limits cannot be observed using futures transaction data, the likelihood of achieving a price limit helps illustrate the ability of a market to convey its most explosive fundamental information. Table 2 describes the occurrence of price limits in each grain market s composite contract series for the five-year trading period from July 2009 through July 2014, segregated by the elimination of the trading halt near the end of May 2012, and reported for both release and non-release days. For every commodity we studied, fewer price limits were hit on release-days after the transition to active trading at the normal USDA announcement time, in both nominal and percentage terms. For CBOT corn, the results are particularly dramatic: its market reached a limit on 21% of the possible release days prior to the shift to real-time trading at announcement time, and only on 6% of release days afterwards. Likewise, CBOT soybean and wheat trading dropped from 2% and 5% to zero price limited release days, respectively. It is important to note that the system CBOT uses to define price limits is not constant, but occasionally changes to accommodate market conditions; this may serve to reduce the probability to hit a trading limit in the latter part of the sample. For example, the CBOT expanded price limits for corn futures in 1993, 2000, 2008, and 2011 (CME Group, 2011, Polansek, 2014). And in May 2014, the CBOT grain 15

16 markets moved to a variable rather than a fixed price limit scheme; variable price limits re-set biannually based on the average recent price of the July-delivery contract (Burgdorfer, 2014). 21 Comparing Market Reactions to USDA Announcements By comparing the reaction to USDA reports between announcement regimes, we can identify periods of heightened (or reduced) volatility and trading. First, we judge the report reaction within each regime relative to normal market conditions, or those that prevail on same-regime non-release days. The size and duration of elevated volatility or trading periods reveals the additional effort expended by the active market to discover a new consensus price based on USDA s updated demand and supply expectations. If the reaction pattern is identical across regimes, the traditional two-hour trading halt afforded no advantages to align potentially asymmetric expectations. As an additional step, we calculate a counterfactual comparison, between the No-halt era release day activity, and Halt era normal, non-release day conditions, to account for potential confounding due to reduction in non-release day volatility after the disconnection of the pit opening from announcement-time trading once the timeout was eliminated. Table 3 summarizes the volatility comparison for each commodity market we studied. 22 In the table, columns 2-3, and 6-7 display a selected portion of the mean hi-low volatility under each trading regime for both announcement and non-announcement days, on a per-trading minute basis. 23 To set up the counterfactual comparison, column 11 equals column 6, while column 12 equals column 2. We identify significant periods of elevated volatility by conducting non-parametric Kruskal-Wallis tests on the equality of the hi-low return distributions; the p-values associated with the χ 2 statistic produced by those tests are 21 The first update under the variable limit system actually decreased the maximum daily price limit for corn and wheat, but increased the limit for soybeans. 22 Although not shown to conserve space, the same procedure was used to compare mean trading volume across trading regimes. A Wilcoxon test was performed verify whether the mean ranks of the volume distributions were equivalent on announcement and non-announcement days. 23 We settled on the mean to convey the per-minute average volatility, because it is the most natural way to express an interval value like the hi-low return. We also considered using a measure that is more sensitive to outliers such as the average absolute deviation from the mean or median return, as used by other researchers (see e.g., Lehecka, Wang, and Garcia, 2014). That approach generated very similar results. 16

17 shown in columns 3, 8, and 13. If these distributions are significantly different at the 5% level, we differenced the mean hi-low values, and displayed those results in columns 4, 9, and 14. These results are extended graphically in panels a and b of figures 2-4, representing the corn, soybean, and wheat markets, respectively, in order to span a larger amount of trading time than can be easily shared in table format. Panel a in these figures plots the mean hi-low values for both release (solid line) and non-release (dotted line) days for both trading regimes (Halt era is red; No-halt era is blue), and panel b shows the differences of each regime s mean hi-low series if they were judged significantly different at the 5% level (same colors as in panel a, with the addition of the counterfactual difference shown in black). The results of our mean trading volume analysis conducted in identical fashion are shown in Panels c and d of the same figures. Table 3 and Figures 2-4 demonstrate several consistent findings across all three commodity markets. First, USDA announcements are immediately followed by a period of heightened volatility and trading just after the USDA publishes sensitive crop and stocks reports, during both the Halt and No-halt eras. Panels a and c in every figure exhibit a distinct spike in volatility and volume, respectively, beginning at the first trading minute following the USDA announcement. This is as-expected, since government crop and stocks reports are known to be informationally valuable, and are eagerly awaited by traders who adjust their supply and demand expectations in response. Although these spikes dissipate fairly rapidly, they are not L-shaped, indicating that it takes each market some trading time to adjust to news (Lehecka, et al., 2014). Second, the magnitude and duration of announcement shocks is more substantial after the elimination of the timeout. This is most pronounced during the first minute after release, when average hilow returns for corn, soybeans, and wheat during the No-Halt era are 2.15, 1.07, and 1.50 percentage points higher than during comparable non-release periods. The same differences during the Halt era are only 0.43, 0.31, and 0.33 percentage points for corn, soybeans, and wheat, respectively. So, in the immediate aftermath of WASDE releases during the No-Halt era, grain futures prices are roughly three to five times more volatile. It is equally interesting to observe how quickly the elevated volatility dissipates during the No- 17

18 Halt era. By the fifth minute after release the differences decline to 0.40, 0.29, and 0.36 percentage points for corn, soybeans, and wheat, respectively. Panel b in figures 2-4 highlights the degree of difference between the average hi-low volatility on release versus non-release days during the No-Halt era and the persistence of significant differences. The bulk of elevated volatility in the Halt-era corn market disappears by about the 40 th trading minute following the USDA report publication, while in the No-halt era elevated volatility levels persist (albeit at a relatively low level) for an additional hour of trading. In sum, these results indicate that without the time afforded by the traditional timeout to digest USDA news, announcement shocks are indeed bigger and it takes more time for each market to return to normal conditions on same-regime non-release days. Panel d of figures 2-4 reports quite similar results in terms of trading volume. For soybeans and wheat, though, mean trading volume (panel b) just following the announcement is slightly lower in the Halt era compared to the No-halt era; that could be due to slightly more reticence to trade in the immediate aftermath of the report absent the timeout, whether due to uncertainty about the exact implications of the report information on prices or a strategic concern about asymmetric information, i.e., the identity, information access, and processing speed of a potential trading counterparty. Finally, although the announcement spike is noticeably higher particularly for volatility after the elimination of the timeout, the path of the shock and its decay is strikingly similar across announcement regimes. Panels a and c of figures 2-4 show that, after about trading minutes following the USDA announcement, the average volatility and trading volume values are about the same regardless of announcement regime. In contrast, the baseline, non-release day volatility and volume levels are quite different in each market, they are far larger in the Halt era. This is likely due to the fact that, prior to the elimination of the timeout, baseline trading captures the opening of pit trading on non-release days. To address the potential problem of confounding by non-constant baseline trading conditions, we displayed the results of our counterfactual analysis in panels b and d. For every commodity, the counterfactual path shows that, although the magnitude of announcement shocks to volatility were larger after the abolition of the timeout, they dissipated quite rapidly, and returned to Halt era baseline market conditions even more 18

19 quickly than Halt era shocks. Compared to the same baseline, No-halt era announcement-driven volume shocks were actually smaller for soybeans and wheat, and about the same size for the corn market. Absent a timeout, our volatility results indicate that news shocks were bigger in the first few trading minutes after an announcement. However, average volatility and trading volume in the major CBOT commodity markets followed a very similar path after a USDA report, regardless of the announcement regime. Moreover, when compared to the same non-release day volatility benchmark, No-halt era announcement shocks actually subside to non-release day volatility levels even more quickly than before the elimination of the timeout. Comparing Absorption Paths between Announcement Regimes Another way to compare announcement structures is to consider the rate at which a market fully absorbs news shocks. Regardless of the size of a USDA announcement shock, such an approach explores the market s adjustment process. We accomplish that by measuring the total divergence between average perminute price volatility and trading volume on release and non-release days within each regime, then calculating the cumulative share of divergence resolved after each post-announcement trading minute. 24 In table 3, this divergence is displayed as the sum of average excess hi-low per-minute volatility at the bottom of columns 4, 9, and 14; values are summed only if they are judged to be significantly different. The cumulative share of the difference observed each minute is displayed in columns 5, 10, and Figure 5 documents the average path of announcement absorption over the first hundred postannouncement trading minutes graphically, using the same hi-low volatility and trading volume values used to construct figures 2-4. Volatilities are represented by solid lines, while volumes are dashed. Red, blue, and black lines represent the average Halt era, No-Halt era, and counterfactual paths, respectively. In the 24 Because the counterfactual difference was sometimes upside-down average volatility and volume on No-halt era release days was lower than those values for Halt era non-release days during the same m t the maximum difference, rather than total difference, was used to calculate the absorption rate for the counterfactual scenarios. Statistically significant upside-down situations were not observed for the same-regime calculations. 25 Again, although not shown in the table, a similar procedure is used to calculate the absorption path of excess trading volume generated after USDA announcements. 19

20 panels of figure 5, representing each commodity market under study, the vertical distance attained by each line represents the percent of the average announcement shock absorbed by that post-announcement trading minute; the time to reach normal, non-release day trading conditions is represented by the horizontal distance each line travels from 0% to 100% absorption. The absorption path for every commodity market in figure 8 is concave in nature, rather than a level shift. That is, although the bulk of the shock absorption occurs fairly quickly, crop markets don t adjust immediately to USDA news, even under the timeout format, when traders had hours to review USDA reports. By about trading minutes after a USDA crop or stocks report, all of the commodity markets had absorbed over half the elevated volatility and trading volume generated by the announcement, on average. 26 Our results are consistent across all commodities: in each case, crop markets absorbed news shocks at about the same rate regardless of announcement regime. For CBOT corn, the average No-halt shock was absorbed slightly faster over the first 25 post-announcement minutes, and then at a slightly slower pace over the next hour compared to the average Halt-era shocks but the differences are not very large. The corn and soybean markets both achieved full absorption a little faster in the Halt era (~90 versus ~98 minutes, in terms of volatility). But overall, the similarity in adjustment paths suggests that the additional digestion time afforded by the timeout did not meaningfully affect the rate at which conditions return to normal. Moreover, when we account for the pit trading discrepancy between regimes, the counterfactual analysis shows that, for every commodity market, the news shock decayed at a faster relative rate in the No-halt era, compared to the non-release day during the Halt era. That is, once we adjust for the fact that baseline, non-release day volatility and trading volume were significantly lower in the No-halt era likely due to the connection with pit opening on announcement days every commodity market s absorption rate accelerated. Every panel in figure 8 shows that announcement shocks resolved at a more rapid pace under the counterfactual scenario; that is, they approach normal, non-release Halt-era conditions quicker. Table 26 In reality, given that markets were closed during the timeout, Halt era grain markets could only absorb half the announcement shock after the passage of two hours and two minutes on a given trader s watch. 20

21 3a and panel a of figure 8 show that, when the non-release day baseline was fixed for both regimes, 81% of the elevated corn market volatility attached to USDA news announcements had resolved in the No-halt era by the first 10 post-announcement trading minutes; for the Halt era, that figure was about 49%. In the soybean and wheat markets, the same values by the 10 th post-announcement trading minute were 72% and 45%, and 94% and 45%, respectively. In addition, the time to fully absorb the average USDA news shock was far shorter in every commodity market under the counterfactual scenario. For example, in panel c, excess No-halt era wheat market volatility and volume resolved to Halt era non-release levels by the 12 th trading minute after the announcement; excess values on Halt era release days took well over an hour. First Five Hundred Trades Figure 6 explores these release-day differences at an even finer scale. Here, we report on a per-trade basis the average tick change and volume, respectively. Ticks in each market represent the smallest change allowed under exchange rules; for CBOT corn, soybeans, and wheat, each tick equals ¼ cent per bushel, or $12.50 per contract at 5000 bushels/contract. 27 In the figure, which shows the last 100 trades before and first 500 trades following the USDA announcement, it s clear that outside of the very first trade volatility is higher and trading volume is lower in the No-halt era. Although pre-announcement volatility is comparable, after the announcement, the average tick change per trade is over two times higher during the active trading format. During the Halt era, when corn traders had two hours to review and digest USDA reports, the average price change at market re-open was 45 ticks, and almost 1200 contracts changed hands; subsequent trades in the chart averaged about 0.6 ticks and 5.4 contracts. After the CBOT move to realtime announcement trading, the average price change immediately following USDA publication fell to just 8 ticks, with a volume of 1.2 contracts; subsequent trades averaged about 1.5 ticks but just 3.6 contracts. These results are consistent with Joseph (2014), who finds that soybean market trading intensity at announcement-time or the average number of contracts per trade declined after the abolition of the 27 Because soybean and wheat market patterns are similar, they are suppressed to conserve space. 21

22 timeout. It should also be noted that the time to reach 500 post-announcement trades fell considerably, from a median of 22.5 seconds in the Halt era, to just 3.9 seconds in the No-halt era. Market Under- or Overreaction to USDA News Table 4 shows the results of Pearson and Spearman correlation tests between and the minute-to-minute and cumulative returns over the next fifteen trading minutes, by announcement regime. 28 These results provide scant evidence for systematic under- or overreaction to USDA news, for any commodity market. In the Halt era, CBOT wheat returns in the first minute correlated positively to the return at the announcement minute, indicating a temporary under-reaction that was not detected in subsequent trading minutes. Two more tests in the table were found to be significant: one for corn, and one for soybeans. The nonparametric Spearman and parametric Pearson tests did agree about the significant negative correlation (overreaction) between the CBOT corn minute-to-minute returns in the announcement minute and the 7 th post-announcement minute, during the No-halt era. But such limited findings given the number of tests we performed in the table do not amount to strong evidence. CONCLUSIONS The elimination of the morning Chicago Board of Trade (CBOT) timeout for grain futures markets and the introduction of real-time trading on U.S. Department of Agriculture (USDA) crop and inventory announcements was a major departure from longstanding practice. Until May 2012, traders had always benefitted from having several hours to review important USDA crop and stocks reports and adjust their strategies in response before CBOT markets re-opened. Many stakeholders, from traders, to industry associations, to the CBOT itself, expressed concern about adverse market conditions that could arise were trading permitted on USDA announcements without some time to adjust to the news. Nevertheless, real- 28 Minute-to-minute returns are used because they express directional changes. Cumulative returns are calculated in a similar way, but take the difference between prices at more distant trading minutes. 22

23 time release of USDA reports has continued since May 2012 up until the present day. From an academic perspective, this new announcement structure offers the ability to for the first time view the complete process by which agricultural commodity markets adjust to important government news. Our results show that when agricultural futures markets are permitted to discover prices freely in response to new USDA reports, the adjustment process is not instantaneous; active crop futures markets experience heightened volatility and trading volume. But we show that the adjustment was not instant in the days of the timeout, either, consistent with the findings of related work (Joseph, 2014, Lehecka, et al., 2014). Compared to non-release days in the same announcement regime, we find that real-time trading on USDA WASDE and Grain Stocks announcements exhibits much larger average news shocks and longer sustained periods of elevated volatility and trading. This is most pronounced during the first minute after release, when average hi-low returns for corn, soybeans, and wheat are 2.15, 1.07, and 1.50 percentage points higher than during comparable non-release periods. The same differences with a timeout are only 0.43, 0.31, and 0.33 percentage points for corn, soybeans, and wheat, respectively. But the rate at which corn and wheat markets absorbs news is actually quite similar, with or without the benefit of a timeout. And when we account for the potential confounding due to non-constant baseline, non-release day volatility and volume levels across announcement regimes, we find that absorption rates were actually faster in the real-time release era. This is due to the fact that, despite their larger shocks, elevated announcement day volatility and trading volume dissipated quite rapidly, and reached timeout era non-release levels even faster. Also, price-limited trading was observed less frequently after the shift in announcement regimes, and we found no strong evidence of market under- or overreaction before or after the timeout was eliminated. Although crop markets experience higher announcement-time volatility today, USDA news enters markets more transparently and faster in calendar terms, than it did in the days of the timeout. Under the timeout format, futures markets could only begin publicly absorbing USDA news after a lengthy trading halt. We show that full absorption occurred after more than an additional hour of trading time, fully three hours after report publication in the years before In the current, real-trading trading format, nearly 23

24 two hours of actual calendar time are shaved off that total. Re-imposition of a timeout around USDA announcement would therefore carry a substantial price discovery cost. On the other hand, although agricultural markets incorporate news more quickly under the current system, our data do not permit the consideration of the impacts of different announcement regimes on trader equity. The change in announcement format may well have affected trading strategies for different types of trader classes. We do find that the average size of trades just after announcement time has fallen in the No-halt era, given heightened uncertainty about the implications of USDA news and the identity or information awareness of potential counterparties. At the same time, the average size of price changes and overall announcement-induced volatility has increased, making resting orders, like stop-loss strategies, riskier. Adapting recent efforts focused on the prevalence of automated traders around government macroeconomic news announcements to agricultural markets and expanding that work to consider different sorts of traders, like hedgers and speculators would help inform the discussion about the welfare effects of the transition to real-time trading on USDA news. 24

25 REFERENCES Abbott, C "USDA Sets New Release Time For Big Agricultural Reports." Chicago Tribune. Adjemian, M.K "Quantifying the WASDE Announcement Effect." American Journal of Agricultural Economics 94: Arasu, K.T., and A. Saphir "Exclusive: CME Readies Round-the-clock Grains Trade: Sources." Reuters. Berry, I "Early Report Day Trading Approved." Dow Jones Newswires. Burgdorfer, B. "Variable Price Limits on CME Grains, Oilseed Futures Start May 1." Farm Futures. CME Group "Corn Price Limits & Margin Requirements". Chicago Mercantile Exchange Group. Dorfman, J.H., and B. Karali "A Nonparametric Search for Information Effects from USDA Reports." Journal of Agricultural and Resource Economics 40: Fortenberry, T.R., and D.A. Sumner "The Effects of USDA Reports in Futures and Options Markets." The Journal of Futures Markets 13: Hauser, S., et al "The Effect of Trading Halts on the Speed of Price Discovery." Journal of Financial Services Research 29: Hayes, R., and J.S. Roberts. 2015a. "Automated Trading in Futures Markets". White paper. U.S. Commodity Futures Trading Commission b. "Macro News Announcements and Automated Trading". White paper. U.S. Commodity Futures Trading Commission. Isengildina-Massa, O., et al "The Impact of Situation and Outlook Information in Corn and Soybean Futures Markets: Evidence from WASDE Reports." Journal of Agricultural and Applied Economics 40: Joseph, K "Intraday Market Effects in Electronic Soybean Futures MarketDuring Non- Trading and Trading Hour Announcements." Ph.D. dissertation, University of Illinois Department of Agricultural and Consumer Economics. Kauffman, N "Have Extended Trading Hours Made Agricultural Commodity Markets Riskier?" Economic Review of the Kansas City Federal Reserve Bank: Lehecka, G., X. Wang, and P. Garcia "Gone in Ten Minutes: Intraday Evidence of Announcement Effects in the Electronic Corn Futures Market." Applied Economic Perspectives and Policy:1-23. McKenzie, A.M "Pre-Harvest Price Expectations for Corn: The Information Content of USDA Reports and New Crop Futures." American Journal of Agricultural Economics 90: Moshirian, F., L.H.G. Nguyen, and P.K. Pham "Overnight Public Information, Order Placement and Price Discovery During the Pre-Opening Period." Journal of Banking and Finance 35: Polansek, T "CME extends open-outcry grain trading day." Reuters "CME Group Weighs Overhaul for Daily Grain Price Limits." Reuters. Polansek, T., and S. Nelson "CBOT Grain Pits to Open Earlier on USDA Report Days." Reuters. Schmitz, D "USDA may revise reporting schedule." Farm World. Sumner, D.A., and R.A.E. Mueller "Are Harvest Forecasts News? USDA Announcements and Futures Market Reactions." American Journal of Agricultural Economics 71:1-8. Troianovski, A "Networks Built on Microseconds." Wall Street Journal. 25

26 USDA "Public Comments Received in Respnse to Federal Register Notice, June 8". The National Agricultural Statistics Service and the World Agricultural Outlook Board. Wang, X., P. Garcia, and S.H. Irwin "The Behavior of Bid-Ask Spreads in the Electronically Traded Corn Futures Market." American Journal of Agricultural Economics 96: Working, H "Theory of Inverse Carrying Charge in Futures Markets." Journal of Farm Economics 30:1-28. Yang, S.R., and B.W. Brorsen "Daily Futures Price Changes and Non-linear Dynamics." Structural Change and Economic Dynamics 5:

27 Figure 1. Trading Hours (Central Time) for CBOT Grain Futures Contracts, July 2009 July 2014 Note: Red bars indicate periods when the electronic market was open for trading; blue bars represent the open outcry, or pit, trading hours. The dashed blue bar in Regime 2 represents the extra open outcry trading hours on days when sensitive USDA reports were published. Green arrows indicate the scheduled USDA announcement time. 27

28 Table 1. Summary Statistics for Mean Returns and Trading Volume Per Trading Minute for CBOT Grain Futures Contracts, by Announcement Regime, July 2009 July 2014 Panel 1a. CBOT Corn Halt Era: No halt Era: Jul 2009 May 2012 Jun 2012 Jul 2014 Release Days Hi Low Returns 0.046% 0.053% (0.09%) (0.12%) Trading Volume* (447) (526) # of Days Non release Days Hi Low Returns 0.043% 0.037% (0.07%) (0.06%) Trading Volume* (292) (247) # of Days Panel 1b. CBOT Soybeans Jul 2009 May 2012 Jun 2012 Jul 2014 Release Days Hi Low Returns 0.034% 0.041% (0.064%) (0.076%) Trading Volume* (268) (310) # of Days Non release Days Hi Low Returns 0.031% 0.03% (0.053%) (0.04%) Trading Volume* (170) (163) # of Days

29 Panel 1c. CBOT Wheat Jul 2009 May 2012 Jun 2012 Jul 2014 Release Days Hi Low Returns 0.047% 0.043% (0.102%) (0.099%) Trading Volume* (177) (181) # of Days Non release Days Hi Low Returns 0.042% 0.03% (0.086%) (0.055%) Trading Volume* (124) (109) # of Days *Trading volume is reported as the number of traded contracts. Notes: Standard deviations appear in parentheses. Kruskal-Wallis tests reject equality of underlying returns and volume distributions between the regimes at the 5% level, for all commodities release and non-release days. 29

30 Table 2. Occurrence of Price Limited Trading Periods in CBOT Grain Futures Contracts, July 2009 July 2014 Panel 2a. CBOT Corn Release Days Non Release Days Days that Hit the Limit % of Possible Days Locked Through Close Days that Hit the Limit % of Possible Days Locked Through Close Halt Era: Jul 2009 May % % 5 No halt Era: Jun 2012 Jul % 2 3 1% 1 Table 2b. CBOT Soybeans Release Days Non Release Days Days that Hit the Limit % of Possible Days Locked Through Close Days that Hit the Limit % of Possible Days Locked Through Close Halt Era: Jul 2009 May % % 2 No halt Era: Jun 2012 Jul % % 1 Table 2c. CBOT Wheat Release Days Non Release Days Days that Hit the Limit % of Possible Days Locked Through Close Days that Hit the Limit % of Possible Days Locked Through Close Halt Era: Jul 2009 May % % 3 No halt Era: Jun 2012 Jul % 0 0 0% 0 Note: The CBOT changed the conditions required for price-limited trading several times over the period of observation. 30

31 Table 3a. Magnitude and Path of USDA Announcement Shock Absorption in CBOT Corn for the Halt era and No-halt era, by Selected Trading Minute Using Hi-Low Returns, July 2009 July 2014 Minutes after Announcement (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) Halt Era No halt Era No halt Counterfactual Release Nonrelease p(χ 2 ) Sig. Difference % Absorbed Release Nonrelease p(χ2) Sig. Difference % Absorbed Release Nonrelease p(χ 2 ) Sig. Difference % Absorbed % 0.06% < % 0.11% 0.07% < % 0.11% 0.06% < % % 0.06% < % 0.14% 0.07% < % 0.14% 0.06% < % % 0.06% < % 0.16% 0.07% < % 0.16% 0.06% < % % 0.07% < % 0.17% 0.07% < % 0.17% 0.07% < % % 0.10% < % 0.29% 0.07% < % 0.29% 0.10% < % % 0.46% < % 11% 2.23% 0.08% < % 15% 2.23% 0.46% < % 30% % 0.31% < % 16% 1.07% 0.08% < % 22% 1.07% 0.31% < % 42% % 0.27% < % 21% 0.74% 0.08% < % 27% 0.74% 0.27% < % 50% % 0.25% < % 25% 0.74% 0.08% < % 32% 0.74% 0.25% < % 58% % 0.25% < % 27% 0.60% 0.07% < % 36% 0.60% 0.25% < % 64% % 0.24% < % 30% 0.47% 0.07% < % 38% 0.47% 0.24% < % 68% % 0.23% < % 33% 0.42% 0.07% < % 41% 0.42% 0.23% < % 71% % 0.22% < % 36% 0.39% 0.07% < % 43% 0.39% 0.22% < % 74% % 0.21% < % 38% 0.33% 0.07% < % 45% 0.33% 0.21% < % 76% % 0.20% < % 39% 0.39% 0.07% < % 47% 0.39% 0.20% < % 79% % 0.21% < % 42% 0.33% 0.07% < % 49% 0.33% 0.21% < % 81% % 0.20% < % 44% 0.32% 0.07% < % 51% 0.32% 0.20% < % 83% % 0.19% < % 46% 0.38% 0.07% < % 53% 0.38% 0.19% < % 86% % 0.19% % 48% 0.32% 0.07% < % 55% 0.32% 0.19% < % 89% % 0.19% < % 49% 0.31% 0.07% < % 57% 0.31% 0.19% < % 91% % 0.20% < % 51% 0.26% 0.07% < % 58% 0.26% 0.20% < % 92% % 0.18% < % 53% 0.24% 0.07% < % 59% 0.24% 0.18% % 93% % 0.18% < % 55% 0.28% 0.07% < % 61% 0.28% 0.18% < % 94% % 0.18% < % 57% 0.22% 0.07% < % 62% 0.22% 0.18% % % 0.17% < % 60% 0.23% 0.07% < % 63% 0.23% 0.17% < % 95% % 0.18% < % 63% 0.21% 0.07% < % 64% 0.21% 0.18% % % 0.12% % 0.08% 0.08% % 100% 0.08% 0.12% < % 100% Total = 3.90% Total = 14.05% Max. Diff.= 5.97% 31

32 Table 3b. Magnitude and Path of USDA Announcement Shock Absorption in CBOT Soybeans for the Halt era and No-halt era, by Selected Trading Minute Using Hi-Low Returns, July 2009 July 2014 Minutes after Announcement (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) Halt Era No halt Era No halt Counterfactual Release Nonrelease p(χ 2 ) Sig. Difference % Absorbed Release Nonrelease p(χ2) Sig. Difference % Absorbed Release Nonrelease p(χ 2 ) Sig. Difference % Absorbed % 0.03% < % 0.10% 0.05% < % 0.10% 0.03% < % % 0.04% < % 0.09% 0.05% < % 0.09% 0.04% < % % 0.04% < % 0.11% 0.05% < % 0.11% 0.04% < % % 0.04% < % 0.11% 0.05% < % 0.11% 0.04% < % % 0.07% < % 0.18% 0.05% < % 0.18% 0.07% < % % 0.37% < % 11% 1.14% 0.07% < % 11% 1.14% 0.37% < % 19% % 0.25% < % 17% 0.69% 0.06% < % 17% 0.69% 0.25% < % 30% % 0.21% < % 24% 0.52% 0.06% < % 22% 0.52% 0.21% < % 37% % 0.19% < % 27% 0.54% 0.06% < % 27% 0.54% 0.19% < % 46% % 0.18% < % 30% 0.43% 0.05% < % 31% 0.43% 0.18% < % 52% % 0.17% < % 33% 0.34% 0.05% < % 34% 0.34% 0.17% < % 56% % 0.16% < % 36% 0.30% 0.05% < % 36% 0.30% 0.16% < % 60% % 0.15% < % 39% 0.30% 0.05% < % 38% 0.30% 0.15% < % 63% % 0.14% < % 42% 0.27% 0.05% < % 41% 0.27% 0.14% < % 66% % 0.14% < % 44% 0.28% 0.05% < % 43% 0.28% 0.14% < % 70% % 0.15% < % 46% 0.27% 0.05% < % 45% 0.27% 0.15% < % 72% % 0.14% < % 48% 0.23% 0.05% < % 47% 0.23% 0.14% < % 75% % 0.13% < % 51% 0.24% 0.05% < % 49% 0.24% 0.13% < % 77% % 0.14% < % 53% 0.23% 0.05% < % 51% 0.23% 0.14% < % 80% % 0.13% < % 54% 0.24% 0.05% < % 53% 0.24% 0.13% < % 83% % 0.14% < % 56% 0.20% 0.05% < % 54% 0.20% 0.14% < % 84% % 0.13% % 57% 0.18% 0.05% < % 56% 0.18% 0.13% < % 85% % 0.12% % 58% 0.21% 0.05% < % 57% 0.21% 0.12% < % 87% % 0.12% < % 60% 0.17% 0.05% < % 58% 0.17% 0.12% < % 89% % 0.12% < % 62% 0.14% 0.05% < % 59% 0.14% 0.12% % 89% % 0.12% < % 64% 0.16% 0.05% < % 60% 0.16% 0.12% < % 90% % 0.08% % 0.08% 0.05% < % 100% 0.08% 0.08% % Total = 2.68% Total = 9.88% Max. Diff.= 4.06% 32

33 Table 3c. Magnitude and Path of USDA Announcement Shock Absorption in CBOT Wheat for the Halt era and No-halt era, by Selected Trading Minute Using Hi-Low Returns, July 2009 July 2014 Minutes after Announcement (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) Halt Era No halt Era No halt Counterfactual Release Nonrelease p(χ 2 ) Sig. Difference % Absorbed Release Nonrelease p(χ2) Sig. Difference % Absorbed Release Nonrelease p(χ 2 ) Sig. Difference % Absorbed % 0.04% < % 0.12% 0.06% < % 0.12% 0.04% < % % 0.04% < % 0.09% 0.06% < % 0.09% 0.04% < % % 0.05% < % 0.12% 0.06% < % 0.12% 0.05% < % % 0.06% % 0.18% 0.05% < % 0.18% 0.06% < % % 0.09% % 0.05% < % 0.28% 0.09% < % % 0.61% < % 9% 1.56% 0.06% < % 13% 1.56% 0.61% < % 36% % 0.37% < % 17% 0.89% 0.06% < % 20% 0.89% 0.37% < % 55% % 0.30% < % 21% 0.56% 0.06% < % 24% 0.56% 0.30% < % 65% % 0.29% < % 25% 0.55% 0.06% < % 28% 0.55% 0.29% < % 75% % 0.28% < % 28% 0.45% 0.06% < % 32% 0.45% 0.28% < % 81% % 0.29% < % 30% 0.41% 0.06% < % 35% 0.41% 0.29% < % 86% % 0.25% < % 33% 0.34% 0.06% < % 37% 0.34% 0.25% < % 89% % 0.24% < % 35% 0.30% 0.06% < % 39% 0.30% 0.24% < % 92% % 0.24% < % 37% 0.27% 0.06% < % 41% 0.27% 0.24% % 93% % 0.22% < % 38% 0.25% 0.05% < % 43% 0.25% 0.22% % 94% % 0.23% % 39% 0.33% 0.06% < % 45% 0.33% 0.23% % % 0.22% < % 41% 0.32% 0.06% < % 47% 0.32% 0.22% % % 0.22% < % 43% 0.31% 0.05% < % 50% 0.31% 0.22% % 98% % 0.20% < % 46% 0.27% 0.05% < % 51% 0.27% 0.20% % 100% % 0.20% < % 48% 0.22% 0.05% < % 53% 0.22% 0.20% % % 0.24% % 49% 0.19% 0.06% < % 54% 0.19% 0.24% % 100% % 0.20% % 50% 0.21% 0.06% < % 55% 0.21% 0.20% % % 0.20% < % 53% 0.23% 0.06% < % 57% 0.23% 0.20% % % 0.19% < % 56% 0.18% 0.06% < % 58% 0.18% 0.19% % % 0.18% < % 58% 0.17% 0.06% < % 59% 0.17% 0.18% % % 0.19% < % 59% 0.18% 0.06% < % 60% 0.18% 0.19% % % 0.11% % 100% 0.08% 0.06% % 0.08% 0.11% < % 100% Total = 3.69% Total = 11.70% Max. Diff.= 2.66% 33

34 Figure 2. Announcement Time Price Volatility and Trading Volume for CBOT Corn Futures, July 2009 July 2014 Panel 2a: Mean Hi-low Return 2.5% 2.0% 1.5% 1.0% 0.5% 0.0% Trading Minutes After USDA Announcement Halt Era Release Halt Era Nonrelease No halt Era Release No halt Era Nonrelease Panel 2b: Differenced Mean Hi-low Return (Significant at the 5% level) 2.5% 2.0% 1.5% 1.0% 0.5% 0.0% Trading Minutes After Announcement Halt Era No halt Era No halt Counterfactual 34

35 Panel 2c: Mean Trading Volume Trading Minutes After USDA Announcement Halt Era Release Halt Era Nonrelease No halt Era Release No halt Era Nonrelease Panel 2d: Differenced Mean Trading Volume (Significant at the 5% level) Trading Minutes After Announcement Halt Era No halt Era No halt Counterfactual 35

36 Figure 3. Announcement Time Price Volatility and Trading Volume for CBOT Soybean Futures, July 2009 July 2014 Panel 3a: Mean Hi-low Return 1.2% 1.0% 0.8% 0.6% 0.4% 0.2% 0.0% Trading Minutes After USDA Announcement Halt Era Release Halt Era Nonrelease No halt Era Release No halt Era Nonrelease Panel 3b: Differenced Mean Hi-low Return (Significant at the 5% level) 1.20% 1.00% 0.80% 0.60% 0.40% 0.20% 0.00% Trading Minutes After Announcement Halt Era No halt Era No halt Counterfactual 36

37 Panel 3c: Mean Trading Volume Trading Minutes After USDA Announcement Halt Era Release Halt Era Nonrelease No halt Era Release No halt Era Nonrelease Panel 3d: Differenced Mean Trading Volume (Significant at the 5% level) Trading Minutes After Announcement Halt Era No halt Era No halt Counterfactual 37

38 Figure 4. Announcement Time Price Volatility and Trading Volume for CBOT Wheat Futures, July 2009 July 2014 Panel 4a: Mean Hi-low Return 1.8% 1.6% 1.4% 1.2% 1.0% 0.8% 0.6% 0.4% 0.2% 0.0% Trading Minutes After USDA Announcement Halt Era Release Halt Era Nonrelease No halt Era Release No halt Era Nonrelease Panel 4b: Differenced Mean Hi-low Return (Significant at the 5% level) 1.6% 1.4% 1.2% 1.0% 0.8% 0.6% 0.4% 0.2% 0.0% Trading Minutes After Announcement Halt Era No halt Era No halt Counterfactual 38

39 Panel 4c: Mean Trading Volume Trading Minutes After USDA Announcement Halt Era Release Halt Era Nonrelease No halt Era Release No halt Era Nonrelease Panel 4d: Differenced Mean Trading Volume (Significant at the 5% level) Trading Minutes After Announcement Halt Era No halt Era No halt Counterfactual 39

40 Figure 5. Absorption Path of Announcement Shock for CBOT Grain Futures, July 2009 July 2014 Panel 5a: CBOT Corn 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Panel 5b: CBOT Soybeans 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Trading Minutes After Announcement Halt Era Volatility No halt Era Volatility No halt Counterfactual Volatility Halt Era Volume No Halt Era Volume No Halt Counterfactual Volume Trading Minutes After Announcement Halt Era Volatility No halt Era Volatility No halt Counterfactual Volatility Halt Era Volume No Halt Era Volume No Halt Counterfactual Volume 40

41 Panel 5c: CBOT Wheat 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Trading Minutes After Announcement Halt Era Volatility No halt Era Volatility No halt Counterfactual Volatility Halt Era Volume No Halt Era Volume No Halt Counterfactual Volume Notes: The figures use mean hi-low returns to represent volatility, as well as mean trading volume on a perminute basis. The counterfactual path represents the comparison between No-halt era announcement day values, and Halt era non-announcement day values. 41