• —Why are trading floors quiet?
• —All the robots are thinking.

News surrounds us. It shouts at us from the television. It pops up at us when we use our computers. It streams across monitors. Now, it tweets at us, too. Nothing is more real time than the news, and nothing carries word of calamities, scandal, failures, and disappointments as fast as the news.

Exposure to real‐time risk from the news comes in many forms. The value of understanding news is not only in general event awareness but also in understanding the impact of those events. There is risk in being late to hearing about the news, and there is also the challenge of digesting the news across all of the aspects that matter to your investments. In this chapter, we review news from a risk management point of view to understand how events that carry risk are communicated and analyzed today.

Investors, market makers, traders, and risk managers are watching for news that affects their portfolios of investments. Information is gleaned from earnings announcements, government indexes, and the growing volume of streaming data from the Internet of things (known as IoT). This IoT, discussed in the next chapter, is a vast array of numbers that comes from all of the wireless devices that are collecting and sending information about the movement of goods, the conditions of agriculture, and the weather, to name a few.

But what is news? News is defined as newly received information. In the strictest sense, anything that has arrived prior to here and now is historical data, and no longer information, even though you personally may not have seen it until now.

In the past, financial firms invested only into market data platforms such as Bloomberg, Thomson Reuters, Factset, and others. These platforms collected news and streamed it through their computer systems. Over time, these platforms offered broader services and an increasing numbers of feeds, which we will describe below. Viewing news only from formal news services changed with the Internet and intelligent devices. The introduction of email and social media created a vast amount of information generated by consumers about themselves and their views about everything. A marketplace of alternative, often more timely, interesting and predictive data sources has sprung up and is blooming. In fact, the proliferation of alternative data sources is so vast that many financial practitioners are now shunning traditional data providers like Bloomberg altogether and are strictly relying on direct data feeds from data sources. Sourcing news directly from its origin bypasses delays and extra markup costs that inevitably occur whenever the news travels through a news aggregator.

THE DELIVERY OF NEWS

News arrives in a continuous flow, spaced at random intervals. News origins are not always real‐time. Some news is delayed or embargoed until a certain pre‐determined release time. News aggregators further slow things down, even if only by seconds, when they collect articles, government announcements, and press releases and provide them to the public through mainstream media and Internet‐based news services.

News Is Written So Computers Can Read It

Algorithms have been developed over time to recognize specific words and, more recently, to read and understand entire articles. Traditional approaches to machine reading and comprehension have been based on either manually specified grammar sequences stored in databases or by using a method that transforms language into probabilistic sequences, in the method known as Natural Language Processing (NLP) and discussed in Chapter 8.

The deluge of data makes it really hard for buy‐side and sell‐side analysts to stay on top of the news affecting their sectors and portfolios. Many traders, however, rely on their internal analyses and are making their trading moves before the announcements. Wouldn't it be great to learn from how they are making their bets?

Understanding the behavior of the markets around announcements is important, and this section shows why. The previous chapter discussed how using the power of analytics can bring clarity on the influence of aggressive HFT on the price of WMT on October 14, 2015, the day when Wal‐Mart released news about its upcoming underperformance. While aggressive HFT algorithms were prominent during sell‐off of WMT immediately following the announcement, as shown in Figure 7.1, further analysis shows they were not the only force behind the sell‐off.

What else was happening at this time? What were institutions doing? When institutional trading activity is considered as a percentage of buy and sell volume recorded in 30‐minute intervals, as Figure 7.2 depicts, one can see that the institutional trading in Wal‐Mart stock has actually peaked between 10:00 AM and 10:30 AM, at the very onset of the sell‐off. At that time, the institutional activity comprised 100 percent of trading volume, making the 30 percent of activity previously marked as aggressive HFT activity clearly driven by institutions deploying HFT algorithms in their trading. As Figure 7.3 shows, when considered as a percentage of the total daily volume, institutional trading activity in WMT on October 14, 2015, peaked from 11:00 AM to 11:30 AM, way after the majority of sell‐off was completed. This observation may make market participants believe that it was, in fact, aggressive HFT that was most responsible for the sell‐off. However, closer examination of institutional activity paints a different picture.

The Wal‐Mart case study followed a very well‐documented surprise, but what evidence is there that someone was trading the news early across all stocks?

Analysis appears to imply that people are trading on macroeconomic news announcements well in advance of their formal announcement times. Traders seem to be in the know or are placing their bets about the upcoming news announcements and profit on the news at least ½ hour (30 minutes) ahead of the news announcement, significantly moving the markets and rendering post‐announcement trading a true gamble.

According to classical finance, markets are not supposed to move in response to an announcement, as most of the news is priced in prior to the news announcement. Specifically, the news is incorporated into prices through a trading process, with trades carrying information to the market. Traders are acting on their beliefs and “putting their money where their mouths are.” It would seem that according to the rational expectations hypothesis, therefore, the markets are operating normally, except that the rational expectations hypothesis was created to work over the days and weeks preceding the event announcement, not minutes.

According to yet another pillar of classical finance, the efficient markets hypothesis (Fama, 1970), the incorporation of news depends on the “universe” of the news: whether the news is public or private. News that is public and is, therefore, known to a large number of traders is incorporated into the markets nearly instantly, while the news that is not widely known tends to seep into the markets slowly. In theory, public announcements, like the ISM Manufacturing Index, are not available for distribution until their precise release time. After the embargo ends, related stocks should experience a clean “step” in price action similar to the one shown in Figure 7.4 for positive news and Figure 7.5 for negative news. Under the efficient markets hypothesis, the price neatly follows the rational expectations hypothesis, fully incorporating all the publicly available news just before and immediately following the news release. With these situations, there are no gradual price transitions.

The efficient market hypothesis, however, allows for selected public news, like the ISM Manufacturing Index value, to be estimated by economists. The economists' thinking, in turn, could gradually filter into pricing through trading, but would not result in a concerted price action. Instead, copious research shows that, whenever public news is released, the price undershoots just before the news announcement, and overshoots temporarily just after the announcement, as shown in Figures 7.6 and 7.7.

Furthermore, the theory of rational expectations suggests that upcoming news is priced in the markets long before the announcement. One way it may be priced in is by trading on the summary forecasts of a cohort of economists polled by Bloomberg or other forecasting services ahead of the announcement. The average of these estimates is considered to be a “consensus forecast.” Although the consensus forecast might not be perfect, it can be fairly informative. For the ISM Manufacturing Index, for example, the consensus forecast predicted the correct direction of the index (increase or decrease from the prior month) 79 percent of the time from 2010 through 2015, and 83 percent of the time from 2013 through 2015.

Figure 7.8 shows the impact of the ISM Manufacturing Index news releases on the Russell 3000 equities, the most commonly held 3,000 stocks in the US markets. The ISM Manufacturing Index is released once a month on preannounced dates at 10:00 AM by the Institute of Supply Management (ISM). ISM asks over 300 manufacturing firms about their employment, production inventories, new orders, and supplier deliveries, and then creates a composite index reflecting their current manufacturing conditions. An improvement in this index tends to signal better manufacturing conditions, translating into a pick‐up in economic growth. Conversely, a decrease in the index potentially signals a flagging economy. Bloomberg considers the ISM Manufacturing Index an important leading indicator.

As Figure 7.8 illustrates, average equity prices on the 3,000 stocks are already moving at the market open, 9:30 AM, a half hour ahead of the upcoming news announcement. By the time the general public receives the news at 10:00 AM, the markets have, on average, moved up by $0.01 per share of the Russell 3000 stocks when the announcement exceeds expectations and down by $0.07 per share of the 3,000 stocks when the news is worse than expected. On average, if you are in the know, you can capture $0.04 per share per news announcement, or $1,440,000 per year if you trade only 1,000 shares of each of the Russell 3000 stocks just once a month. With hundreds of distinct macro announcements released every month (see The Quant Investor Almanac 2011 for details), it becomes easy to build a billion‐dollar global macro hedge fund that is consistently profitable.

The profits, however, come directly from the pockets of the general investing public who trade after the announcements. As Figure 7.8 shows, the content of the news is completely priced in ahead of news announcements. The markets barely move in response to the news after the news.

Is it possible that the preannouncement trading is not done on the embargoed news, but, instead, is based on superior economic powers of prediction of the upcoming values of the ISM Manufacturing Index? While some economists try to intelligently guess the upcoming ISM Manufacturing Index number, few succeed. Bloomberg compiles and publishes consensus forecasts for the ISM Manufacturing Index by polling well‐respected economists and then averaging the respondents' opinions. The reported consensus only has a 42 percent correlation with the realized value.

How does the news leak out ahead of time? The answer likely lies in the antiquated methodology by which the government and other macroeconomic news sources release their updates. Typically agencies distribute the news to journalists and others parties an hour ahead of the news release under a news embargo. The embargo is an honor system whereby the recipient of the news is expected to keep the news confidential until the scheduled news release time.

How can market behavior and that of aggressive HFT be explained in a sensible manner? Some market participants have blamed aggressive HFTs for obtaining news and acting on it ahead of its release to the public. While the assumption of advanced knowledge is tenuous, it is not impossible due to an outdated concept of news embargo. News embargo emerged in the 1960s as a solution to the issue of news fairness raised by market participants. To ensure wide access to news, the figures were to undergo the fullest possible distribution, which at that time equaled television, radio, and print. To provide adequate time for television and radio broadcast preparation, the news sources embargoed the news content for one hour, allowing all the news outlets to broadcast in unison, ensuring equal access to all investors. The embargo system, however, has always been voluntary, and no government penalties of any sort exist for cases where a reporter decides to inappropriately email the news to a hedge fund or an HFT friend. Given all the financial incentives, the often‐starving reporters may violate the embargo and share the news with a hedge fund or trading desk that can trade on the embargoed news.

There is the risk that at least some parties receiving news appear not to care to observe the embargo. When the rewards of ignoring the embargo become excessively attractive, all the incentives are there to throw the honor code out the window.

In the past, placing a single trade took hours or even days, and the Internet as we know it today did not exist. At that time, news embargoes served an important function: they gave journalists a chance to prepare articles in order to achieve the widest news coverage possible, enabling all individuals to benefit from the news releases at the same time. Today, however, the embargoed news distribution not only makes no sense, it is actually disadvantaging regular investors and enabling a clever few to make outsized profits at everyone else's expense.

Are high‐frequency traders (HFTs) involved in this news arbitrage? According to AbleMarkets Aggressive HFT Index, yes, roughly three‐quarters of the preannouncement trading is due to trades initiated by aggressive HFTs. However, given the timing of trading (half hour, not microseconds), the observed aggressive HFT strategies are likely deployed as order execution/footprint minimization strategies by the entities that receive the advanced news.

Examining the trading patterns ahead of the ISM Manufacturing Index and Construction Spending announcement, we find that trading on the not‐yet‐publicly released embargoed news or at least placing educated bets consistently takes place as long as 30 minutes ahead of the news announcement times. The aggressive HFTs appear to be involved in the pre‐announcement trading. In fact, as much as three quarters of the pre‐announcement price move appears to be driven by aggressive HFTs.

Of course news moves the markets as discussed previously. Furthermore, according to the rational expectations hypothesis, news is the only thing that moves the markets; everything else is noise. Companies like Bridgewater Associates have built small empires with annual revenues exceeding the GDPs of small countries combined just following, interpreting, and acting upon the news. Not surprisingly, the question of whether news announcements are released in a fair manner remains a hot topic. Even less surprising, the fairness of news releases has surfaced as one of the key concerns associated with high‐frequency trading. Specifically, some market participants have accused high‐frequency traders of using fast technology to front‐run lower‐speed traders following major news announcements. This research considers market activity surrounding news events.

Event studies are a classic way to measure the happenings surrounding news announcements. The event study methodology is as established as the science of finance, dating back to the 1930s. An event study compares the impact of the news on market conditions before and after the event, in what's known as an event window. The window can be as large or as small as one may like it to be, provided that there are enough data points in the selected window and that the distribution of the dependent variable matches the selected analysis model. Given the short‐term behavior of aggressive HFT, we focus on smaller time windows to consider the behavior of aggressive HFTs around a news announcement.

Choosing the event is another matter, no less important than the selection of the event window. One factor to consider is that many news announcements are scheduled outside of regular trading hours. To examine the short‐term HFT activity around news announcements, however, we consider news that was (1) released during common market hours, and (2) likely to generate a similar reaction across many financial instruments at once.

One such news is construction spending, computed by the U.S. Census Bureau. It estimates the total value of construction performed in the United States during the previous month, including labor, materials, architecture and engineering costs, overhead, interest, and even taxes. The index covers construction in both public and private sectors. Construction spending is an indicator of economic optimism. The higher the construction spending, the reasoning goes, the more people are investing into long‐term projects, the higher is the optimism about the economy's future.

Construction spending announcements often coincide with ISM Manufacturing Index survey figures. The index, now computed by IHS Markit in collaboration with the Institute of Supply Management, is based on the responses to the questionnaires sent out to managers in selected companies. ISM asks over 300 manufacturing firms about their employment, production inventories, new orders, and supplier deliveries, and creates a composite reflecting the current manufacturing conditions. An index increase tends to signal better manufacturing conditions, translating into a pick‐up in the economic growth. Conversely, a decrease in the index potentially signals a flagging economy.

Current research analyzes the two events in tandem and uses the latest event study methodology to separate the impacts of the two announcements and the aggressive HFT behavior on the returns.

PREANNOUNCEMENT RISK

On July 1, 2015, the month‐to‐month change in Construction Spending and ISM Manufacturing Index were reported at 10:00 AM. According to Bloomberg, Construction Spending had increased by 0.8, beating analysts' consensus forecast of a 0.5 increase by 0.3. The simultaneously reported ISM Manufacturing Index value was 53.5, an increase of 0.8 from the value reported in June and a 0.3 improvement over the “consensus forecast,” a composite figure aggregating opinions of a range of economists polled by Bloomberg on the matter. The news was good. The economy was observed to be growing, and stocks were expected to go up.

The 10:00 AM announcements were preceded by a 9:45 AM value of Institute for Supply Chain Management (ISM) Manufacturing Index (ISM Manufacturing Index). The 9:45 AM figures were worse than expected and worse than the prior month's figures.

Figure 7.9 shows the minute‐by‐minute cumulative price response to the news by Agilent Technologies (NYSE:A) recorded on BATS‐Z exchange. As Figure 7.9 shows, in the 60‐minute time interval prior and immediately following the news announcement, the biggest growth in price occurred nearly a half hour ahead of the news release, just after the market open. The stock price of Agilent appears to be unaffected by the ISM Manufacturing Index values made public at 9:45 AM, 15 minutes prior to the 10:00 AM event.

Figure 7.10 shows the proportion of aggressive HFT activity by volume traded in Agilent around the news announcement, as estimated by AbleMarkets. As Figure 7.10 shows, aggressive HFT buying activity peaked at market open, and again nearly a half hour following the news announcement. Aggressive HFT selling activity was elevated around 10:13 AM, potentially explaining some of the observed post‐announcement sell‐off in NYSE:A. The aggressive HFT numbers immediately surrounding the event announcement may have been dampened by the influx of trading volume brought on by other market participants: institutions and retail looking to capitalize on the news.

Is the behavior of the price of A an anomaly? Is the preannouncement gain a random occurrence? To answer this question, we looked at the price response of the entire Russell 3000 index to the same announcement, ISM Manufacturing Index report on July 1, 2015. Using Bats‐BYX data, and averaging the cumulative dollar gains and losses of each of the Russell 3000 stocks each minute, we arrive at an even more pronounced preannouncement market movement pattern shown in Figure 7.11. In fact, across all the Russell 3000 stocks, the preannouncement price movement is so pronounced and precise that very little volatility can be observed after the announcement, as Figure 7.12 shows. Figure 7.12 quantifies volatility by measuring the cross‐sectional dispersion of returns each minute across all Russell 3000 stocks. As Figure 7.12 shows, volatility indeed declines dramatically following the news announcement.

The minute‐by‐minute average of the aggressive HFT activity for the entire set of stocks comprising the Russell 3000 surrounding the events of 10:00 AM on July 1, 2015, is equally interesting: the aggressive HFT buyers dominated sellers from 9:45 until the 10:00 AM news announcement, at which point relative proportion of aggressive HFT dropped dramatically from the 20 to 30 percent range to single digits, as shown in Figure 7.13.

Putting aside aggressive HFT behavior for a moment, let's consider what is wrong with the pictures of Figure 7.11 and Figure 7.12.

As shown in Figures 7.11 and 7.12, market behavior of the Russell 3000 stocks had little to do with the market responses expected under the efficient markets and rational expectations hypotheses. The average price of the Russell 3000 stocks began to rise 15 minutes prior to the news announcement and barely moved after the news is released at 10:00 AM.

What can be done to ensure fairness in the financial markets? Perhaps catching up with the times and distributing news using social media may do the trick—after all, most traders today are capable of making fundamental stock pricing calls on the basis of the released news figures alone and do not require a reporter's interpretation of the figures. Why not release the news via Twitter or other social media and eliminate the now‐ancient embargo process?

Decades‐old changes to the news distribution process, however, may take years to complete. In the meantime, investors of all stripes may choose to follow the markets dynamics and observe aggressive HFT behavior in an effort to extract the information about upcoming events directly from the markets. Thus, for example, observing elevated levels of aggressive HFT buyers prior to the 10:00 AM news on July 1, 2015, would suggest that the about‐to‐be‐formally‐released news is likely to be positive. With a 15‐minute lag prior to the news announcement, such observations do not require high‐speed technology, yet deliver powerful predictability and, as a result, profitability.

DATA, METHODOLOGY, AND HYPOTHESES

Deploying an event‐study methodology on ISM Manufacturing Index announcements from January 2013 through October 2015, we analyze movements of price, volatility, and aggressive HFT activity around the news release. The results are interesting and surprising, or not so much, depending on whom you ask:

1. The news is “leaking” into the markets well prior to the news announcement.
2. Aggressive HFTs do appear to be trading on the news preannouncements. However,
   1. The aggressive HFTs comprise only a portion of observed trading activity; and
   2. The aggressive HFT activity can be due to institutions using aggressive HFT strategies to trade. While aggressive high‐frequency trading activity appears to contribute to preannouncement news incorporation in the markets, it is not the overwhelming factor in the preannouncement trading activity.

How do we know that the news is leaking into the markets well ahead of the proper news announcement time? Consider once again Figure 7.8, which shows the cumulative price move in minutes for the ISM Manufacturing Index before and after the actual time the news is released, averaged over the following two dimensions:

1. All the ISM Manufacturing Index announcements from January 2013 through October 2015
2. All the stocks in the Russell 3000

Figure 7.8 shows three lines:

1. The average cumulative price across all the event announcements and all the Russell 3000 stocks dips mildly ahead of the announcement time (Time 0)
2. The average cumulative price across only those ISM Manufacturing Index news release dates where the announced ISM Manufacturing Index values were higher than those announced in the immediately preceding month (Avg Cum+). This line rises sharply as early as 15 minutes ahead of the announcement (from Time −15).
3. The average cumulative price across only those ISM Manufacturing Index news release dates where the announced ISM Manufacturing Index values were strictly lower than those announced in the immediately preceding month (Avg Cum−). This line begins its steep descent full 30 minutes ahead of the announcement time (at Time −30).

At the time and shortly following the announcement, the markets move little. The lack of movement underscores the dearth of information at the actual news release time (Time 0).

How persistent are the observed price responses across various announcements? What if a single announcement dominates this entire dynamic? Do the rest of the announcements generate a consistent response? To answer this question, we look at the ratio of averages shown in Figure 7.8 to understand the standard deviation of minute‐by‐minute price responses across different announcements across all the Russell 3000 stocks. Figure 7.14 shows the standard deviations, and Figure 7.15 presents the t‐ratios: the averages of Figure 7.8 divided by the standard deviations of Figure 7.14.

As Figure 7.14 shows, the variation in price responses is the highest just before the scheduled news announcements, and the lowest following news announcements. However, the variation in the price response also happens to be low about 21 and 8 to 5 minutes prior to the proper news announcement time, 10:00 AM. The consistency of correct “guesses” of the impending news direction is so high that it is highly unlikely to be purely accidental.

Figure 7.15 displays the t‐statistics of the cumulative price responses (averages of Figure 7.8 divided by the standard deviations of Figure 7.14). While, as shown in Figure 7.15, the response is much more statistically significant after the news announcement, it still reaches 99.9 percent significance at least 10 minutes prior to the official news announcement time.

What about the Construction Spending announcements that occur at the same times as the ISM Manufacturing Index? Figure 7.16 shows the average price to the realized vs. prior month change in the Construction Spending value. As Figure 7.16 shows, the response to Construction Spending is much more convoluted than it is to the ISM Manufacturing Index, shown in Figure 7.17 and 7.18.

How can anyone possibly trade on the news announcements prior to the news announcements? How would one know what the news value is going to be? An intelligent forecast may certainly be one answer to this question. Bloomberg compiles one set of such forecasts.

How good is Bloomberg's consensus forecast for ISM Manufacturing Index? Research shows that it is not particularly good. From January 2010 to October 2015, the direction of the forecast coincided with the direction of the realized value just 32 out of 73 times. In other words, over 56 percent of time when the forecast said that the ISM Manufacturing Index was going to go up (down) the following month, the released figures actually went in the opposite direction: down (up). From January 2013 through October 2015, that directionally incorrect proportion of forecasts has decreased to 52 percent, with not‐even‐close forecasts outnumbering somewhat useful ones.

What about the forecasts for Construction Spending announcements? The latter are much better: since January 2010 through October 2015, over 71 percent of time when the forecast said that the Construction Spending was going to go up (down) the following month, the released figures actually went up (down). Since January 2013 through October 2015, that number has actually increased to 74 percent.

Aside from the directional successes and failures, both Construction Spending and Manufacturing indexes exhibit high correlation between differences in realized values and prior values and realized values and forecasts, as Table 7.1 shows. The difference between the realized construction values and their prior month values, for example, exhibits 72 percent correlation with the realized index value less its economic consensus forecast. Between the Construction Spending Index and the ISM Manufacturing Index, however, correlations are quite low, as is also shown in Table 7.1. As a result, the news announcements, while overlapping, leave distinct marks on prices at different times.

Is the consensus forecast of Construction Spending driving prices? This does not appear to be the case. Since the consensus forecast is typically released several days ahead of the announcement, the price change would have occurred at that time. Furthermore, no significant changes in prices would be observed in the 30 minutes immediately preceding the news announcement. The latter is not at all the case. Figure 7.19 shows the average price response across all of the Russell 3000 stocks preceding and following positive and negative announcement values as compared with the consensus forecast values reported by Bloomberg. As Figure 7.19 shows, when realized Construction Spending is above the forecasted values, the average stock price across all the Russell 3000 stocks actually happens to fall ahead of the news release! On the other hand, when the announced Construction Spending figures are below the consensus forecast, the prices tend to rise ahead of the announcement. The prices stabilize immediately after the news is publicly announced.

Figure 7.20 shows the t‐ratios of the averages documented in Figure 7.19. As Figure 7.20 illustrates, while the response is much more pronounced after the announcement time, the trading behavior consistent with the realized news release is prevalent up to 20 minutes before the scheduled news release time. The evidence suggests that news is indeed leaked or bets are being placed before being made available to all.

How much is this information worth? Suppose one is only trading on the ISM Manufacturing Index. As shown in Figure 7.8, if one were to receive the realized ISM Manufacturing Index value 30 minutes ahead of the announcement time and trade on that information, one would, on average, make 1.5 cents per share ($0.015) on good news—that is, the announcement is better than last month—and 6.5 cent ($0.065) per share on bad news. Trading just 100 shares in each of the Russell 3000 stocks 30 minutes ahead of the news would thus produce, on average, in excess of $12,000 per announcement, not accounting for transaction costs. Trading the same on the Interactive Brokers, where each round‐trip trade costs $2.00, our embargoed‐information trader would still clear $6,000 per announcement, trading just 100 shares a half hour prior to the news scheduled release. Given the half‐hour window allowed for trading, the strategy can be easily scaled, to, say, at least 1,000 shares, easily becoming a meaningful incentive to any journalist!

Fast forward to 2016, and the embargo system is no longer cutting it. One can trade faster than one generates a trading idea, computers can process news as soon as it hits the news wires, and the “level‐playing‐field” idea behind the original news embargo system no longer makes sense. If anything, the system allows trading on the news to the chosen few, chosen by their ability to procure the embargoed news ahead of the masses. How is this fair?

Who is trading on the embargoed news? While it is hard for a bystander to point the finger at the exact trader in the anonymous markets, we can separate categories of traders active before and after the announcements.

One observation from using the AbleMarkets Aggressive HFT Index is the ability to track the behavior of aggressive HFT around the news releases. Aggressive HFT is of particular interest as it has often been associated with advanced trading on news in the popular press.

Figures 7.21 and 7.22 document the behavior of aggressive HFT buyers and sellers, respectively, averaged across the Russell 3000 stocks and all ISM Manufacturing Index announcements from January 2013 through October 2015. As the figures show, behavior of aggressive HFT buyers and sellers is the same whether the realized figures are higher or lower than those of the prior month. The balanced nature of the aggressive HFT activity and higher volumes ahead of higher‐than‐previous announcements may explain this phenomenon. Aggressive HFTs hold positions for a very short term, and faced with the potentially higher‐than‐normal flow ahead of positive announcements, the aggressive HFT activity goes up. In Construction Spending announcements, the separation of aggressive HFT buyers and sellers is much clearer. When the soon‐to‐be‐released value of Construction Spending is higher (lower) than the forecast, the aggressive HFT Buyers (Sellers) are more prominent than when the realized value is lower (higher) than the forecast, as shown Figures 7.23. In other words, selected aggressive HFTs appear to receive and act on advanced Construction Spending news, if not the ISM Manufacturing Index values. However, given that the cumulative price of Russell 3000 stocks moves opposite to the realized value ahead of the news release, the “in‐the‐know” aggressive HFTs appear to be trading to their disadvantage.

Figure 7.23 shows aggressive HFT buyer participation around events where the realized value was higher than the forecast and lower than the forecast. As Figure 7.23 shows, in cases like this the aggressive HFT buyers are seen to account for a larger proportion of trading activity after the proper event announcement time. As seen in Figure 7.23, aggressive HFT accounts for about 1 percent more of trading activity after a higher‐than‐forecasted figures release than lower‐than‐forecasted figures release. Aggressive HFT buyers, however, account for a considerably higher participation before the announcement time when the announced news is higher than the forecast. This finding is consistent with the price movement ahead of news whereby the released figures are higher than the forecasted ones. It appears that aggressive HFTs are at least partially responsible for the consistent price drop/rise ahead of news whereby the realized numbers differ from the forecast.

How can this be the case? One hypothesis can be that the entities deploying aggressive HFT around the announcements prepare to maximize their profitability and volume traded on a given macro trade well ahead of the announcement. In the process, they assume that the forecast will come short of the realized value and overaccumulate stocks. Then, when the news is revealed to them or when they make their bets about the announcement, the aggressive HFT strategies sell off excess inventory to align their holdings with the expected post‐announcement price, now easily quantifiable under the rational expectations hypothesis. The resulting strategy benefits the participating entities in two ways: (1) maximizes traded capital, and (2) helps avoid detection as the direction of preannouncement trading is reversed vis‐à‐vis the expected price direction given the announcement.

What are the implications of the aggressive HFT activity for market makers? Aggressive HFT flow is toxic and is best avoided from the market‐making perspective. As a result, market makers may significantly improve their profitability around news announcements by explicitly tracking aggressive HFT behavior.

CONCLUSIONS

Whether news is being leaked or traders place their bets in advance, there is predictive value in studying market data in the 30 minutes before a macroeconomic news announcement.

What should investors do to minimize the impact of the news on their portfolios without the same access to the preferentially distributed embargoed information to? Tracking aggressive HFT may help make informed portfolio and market‐making decisions when trading around macroeconomic news.

END OF CHAPTER QUESTIONS

1. What is news?
2. How is news distributed?
3. How does news impact markets in theory?
4. How does news impact markets in practice?
5. What are the key risk implications for investors?