In Part 2, Peter Hafez, Director of Quantitative Research for the data provider RavenPack, describes a new indicator he has developed and researched called “news beta.” This is a measurement of the degree to which a security responds to market level sentiment as captured through news stories. Hafez reports that “news beta” is relatively uncorrelated with other traditional measurements.
In the video here, I interview Peter Hafez, Director of Quantitative Research for the data provider RavenPack. Hafez has published a number of empirical studies dealing with ways of measuring the impact of news stories on securities prices. He has developed a number of trading strategies that have shown consistent profitability over the past, turbulent decade.
The number of good news events reported in the business press is very similar to the number of bad news events. However, Green, Hand, and Penn (2012) – from Pennsylvania State University, the University of North Carolina, and Florida State University – documented that bad news events were more wildly disseminated after they were initially reported than were good news events.
They hypothesized that the bad news dissemination bias occurred because business journalists, like their popular press counterparts, responded to “if it bleeds, it leads” incentives to increase readership. Consistent with this hypothesis, they found the bad news dissemination bias was greatest for companies and events that were least likely to attract readers’ attention (i.e., small firms and less investor-relevant events), and for which competition between journalists was high (i.e., firms with greater amounts of recent press coverage).
Of course, the major trend over the past decade concerning trading the news is the focus on algorithms for analyzing the news and making trading decisions. Stephane Gagnon (2012), of the Université du Québec, Canada, developed a taxonomy with five different categories of news trading algorithms. In summary they are listed below:
Gagnon proposed a rules-based, semantic reasoning engine that would ultimately incorporate all of these categories into a single system. This trend toward greater complexity appears to be the direction in which advanced news-based trading is moving. The research studies summarized in Book Three of The Alpha Interface series have been selected because they offer empirical evidence suggesting viable strategies for traders and investors.
Groß-Klußmann and Hautsch (2011) confirmed the usefulness of the machine-indicated relevance of news items. Significant market responses to news were only observable for items which were identified as being relevant. Their results showed that the classification was crucial to filter out noise and to identify significant relations between market activity and the news flow.
The news sentiment indicator used by the researchers had predictability for future price trends. However, significantly increased bid–ask spreads around public news arrivals rendered simple, sentiment-based trading strategies rather unprofitable. The researchers noted that more sophisticated algorithms would be necessary to overcome this obstacle.
This chart shows that the abnormal returns were too low to overcompensate increased bid-ask spreads around news and to provide economic gains of the underlying trading strategies. Reprinted from Groß-Klußmann and Hautsch (2011) with permission from Elsevier.
to be continued ….
Prototype experiments for predicting market reaction to financial news.
The research literature has described a number of prototypes for predicting market’s reaction to news. The first recorded approach was by the trader Victor Niederhoffer in the early 1970’s. Niederhoffer organized stories from the day’s newspapers into 19 separate categories with a sliding polarity scale (positive to negative). Trends were inferred from the aggregation of the polarity information. This manual approach would have been slow – but it was a fresh approach. Since then, the advent of machine readable news has allowed a number of systems to automatically classify news stories, eliminating the inherent lag of Niederhoffer’s approach.
Recording and analyzing the overall news flow for a specific asset has been challenging since the amount of news, the number of news sources, and the speed of information dissemination has rapidly increased over time. Due to the huge amount of information published, and digitally stored, in modern media, news has been overlaid by substantial noise caused by irrelevant information. These effects have made it difficult to identify significant links between high-frequency trading activity and the intraday news flow.
To reduce the impact of noise, Groß-Klußmann and Hautsch (2011) – of Humboldt University, Berlin, Germany – were the first researchers to make use of unique data provided by an automated news analytics tool of the Reuters Company. Designed for use in algorithmic trading applications and employing linguistic pattern recognition techniques, these novel news data allowed researchers to disentangle relevant news from irrelevant ones and to identify the sign and the novelty of news items. Using this news engine their study explored the impact of news items on high-frequency returns, trading volume, volatility, depth of market, and bid–ask spreads for a cross-section of 39 stocks that were actively traded at the London Stock Exchange (LSE). They looked at intraday data regarding over 29,000 news stories from January 2007 through June 2008.
Analyzing the unconditional and conditional effects of news items on intraday trading activity, Groß-Klußmann and Hautsch summarized the following results. They found significant market responses in volatility, money value traded, average trade sizes and bid–ask spreads. Given the fact that earnings announcements were explicitly discarded from the analysis, these findings were remarkable and indicated that the news engine successfully filtered the news flow according to positive and negative impact.
Cumulated abnormal returns around relevant positive, negative and neutral news. One can see that the greatest returns occurred before the news was time-stamped by the Reuters system. Reprinted from Groß-Klußmann and Hautsch (2011) with permission from Elsevier.
to be continued …
Periodicities of Earnings Announcements
Earnings announcements have distinct characteristics. The following charts show that there are periodicities to news announcements, particularly earnings announcements:
Based on data from Hirshleifer, Lim, and Teoh (2009).
The day-of-week chart above, for example, shows that companies have a clear preference for making earning announcements on Tuesday, Wednesday, and Thursday. Friday is the day most avoided.
Based on data from Hirshleifer, Lim, and Teoh (2009).
As the chart, above, shows, there is a clustering effect regarding the months of the year during which earnings announcements are made. This is mostly due to fiscal years that end with the traditional calendar year. The lowest number of earnings announcements are distinctly in June, September and December – perhaps good months to take vacations!
To be continued …
Efficiency. It’s a simple word. In the world of modern finance, it means that stock market prices – rapidly and without prejudice – reflect what’s going on in the business world.
This notion was settled upon because it was believed by analysts that buyers and sellers revised their expectations about future firm performance. These revisions in expectations changed the risk-adjusted value of firms and impacted market prices. Of course, the notion of informational efficiency has never implied that markets were somehow omniscient. No supposition was made as to the degree of precision with which prices should respond to news. Because of the continual noise prevalent in markets, one should not be surprised to find indications of pricing error in many situations.
In the last two decades researchers from a variety of disciplines have challenged the notion of information efficiency. Research studies questioned the completeness of the immediate market reaction to corporate news events. An extensive body of empirical literature examined a wide-ranging set of specific news events and found that markets appeared to initially under-react. In other words, it took some time for traders and investors to digest the news and process its potential implications and ramifications for asset prices.
How does this economic information get to the public, or at least to traders? When information is released, news agencies may start by summarizing its content in a short version and instantly redistribute it to end-users. The news agency then gathers information from various sources, eliciting comments from industry experts and adding other contextual information. This results in a second distribution of relatively longer news items within a couple of hours of the first news blast.
Successive editing and distribution based on the original information release may continue depending on its level of materiality. Often, following large corporate events equity research and credit rating analysts publish a report. Following in line, news agencies and newswires distribute news items discussing or summarizing the contents of these reports.
In the meantime, newspaper journalists gather news for the next issue of their publication. Some news items included in the next daily publication will reflect information that has been processed and distributed through newswires the day prior to publication. Journalists working on these news items, will add further insight by gathering more contextual information and adding further synthesis and analysis.
In summary, news items are the result of the activities of media industry participants as they edit, aggregate, and distribute raw economic information. Media industry participants choose the degree to which items are edited and aggregated to fit the medium’s distribution frequency – i.e. continuously, daily, weekly, etc. – and distribution form.
While not true in all cases, positive news events generally are met with positive market reactions. In these cases, returns subsequent to announcements showed positive, drifts. Similarly, negative news events generally meet with negative market reactions and tend to be followed by negative drifts. On the other hand, traders and investors often over-react to price shocks, causing excess trading volume and volatility – and then leading to reversals.
The reaction of financial markets to news cannot be studied in isolation, as there can be important interdependencies. For example, one piece of news not only has direct effects on asset prices and market volatility, but it can also alter the relative importance of other pieces of news.
Companies historically published the larger share of news (about 64%) outside trading hours. However, in recent years in Germany and the United Kingdom, at least, this trend has shifted. The majority of news (about 55%) in Germany and the U.K. is now published intraday. The following chart tracks this trend.
Reprinted from Hagenau, Liebmann, and Neumann (2013) with permission.
Why is this trend important today? It means that there are more opportunities to react to the news as it appears in real-time during the trading day. What changes in the social landscape does it reflect? It reflects the burgeoning of the 24-hour investigation and reporting of news, concomitant with fast-paced lifestyle of modern societies.
To be continued …
The opposite was true for low-volume stocks. When past week returns were negative, a drop in volume led to a continuation of negative returns. When the past week’s returns were positive, a drop in volume led to a reversal in returns. In both instances, a drop in volume was associated with a drop in prices. This is in accord with much trading lore.
The chart above shows the momentum continuation effect for stocks with low volume that were price-losers during the prior week. The horizontal axis groups stocks according to the percentage loss during the previous week. The colors show the degree of the percentage increase in volume. The vertical axis shows the percentage gain or loss. Based on data from Alsubaie and Najand (2009).
The chart above shows the reversal effect for stocks with a drop in volume that were price-winners during the prior week. The horizontal axis groups stocks according to the percentage gain during the previous week. The colors show the degree of the percentage decrease in volume. The vertical axis shows the percentage gain or loss. Based on data from Alsubaie and Najand (2009).
One must be mindful, that these results were generated from a unique bull market phase within an emerging market. Nevertheless, the lore generated by traders who focus on volume (i.e., see Leibovit, 2011) in other markets largely agrees with the pattern documented in Saudi Arabia.
Trading strategy: In a bull market, buy stocks that have dropped sharply in price the previous week on high volume. Hold for one week. In a bull market, buy stocks that have risen strongly in price the previous week on high volume. Hold short positions for one week. Sell stocks that have risen sharply the previous week on decreasing volume. Hold short positions for one week.
However, the primary result (as presented in the previous blog post, Part 1) did not hold in the price-winner, high-volume and price-winner, low-volume portfolios. Overall, high volume led to continuation in weekly returns while low volume led to reversal in weekly returns. During the test period, high-volume stocks were more profitable than low-volume stocks. When the previous week’s returns were negative, high-volume stocks led to reversal. When previous week’s returns were positive, high-volume stocks led to continuation. This is, of course, what one expects in a bull market environment.
The chart above shows the momentum continuation effect for stocks with high volume that were price-winners during the prior week. The horizontal axis groups stocks according to the percentage gain during the previous week. The colors show the intensity of the percentage increase in volume. The vertical axis shows the percentage gain. Based on data from Alsubaie and Najand (2009).
To be continued…
Alsubaie and Najand (2009), from the Department of Public Administration, Saudi Arabia, and Old Dominion University, Norfolk, Virginia, examined the relationship between abnormal changes in trading volume and short-term price behavior in the Saudi stock market. Considering prior research and theory, they hypothesized that if uninformed traders providing liquidity dominated the Saudi market, then price changes accompanied by high volume would tend to reverse their previous trend. On the other hand, if informed investors dominated the market, stock returns follow the direction of trading volume.
(Liquidity traders are also called “noise traders.” These are investors that are buying or selling securities for a variety of unpredictable reasons rather than reasons based on informed decision making. Examples of this type of buying or selling would be deciding to retire early, needing the money suddenly, or getting a hot tip.)
Alsubaie and Najand collected the daily data for return, turnover, and market capitalization for all Saudi firms from 1993 through 2005. They divided this data into two, sub-sample periods: from 1993 through mid-1999, and from mid-1999 through 2005.
Their sample started with 41 firms and ended with 77 firms. The market capitalization of Saudi Arabia stocks had increased substantially in recent years. It had increased by more than 112% during the year ended December 2005. The market index gained over 40% in 2005, which followed six years of growth at an average annual rate of 38%. Market volumes had also increased significantly. On average, market volume was worth over $4 billion a day in 2005.
Every week each stock was placed by the researcher into one of the four portfolios: price-loser, high-volume; price-loser, low-volume; price-winner, high volume; and price-winner, low-volume.
They found a reversal in weekly stock returns following a change in volume occurring the previous week. Their results were consistent for the whole sample, the two sub-sample periods, and both the large-firm and small-firm portfolios. They also found that the reversal was most pronounced with the price-loser portfolio. When liquidity traders sold, prices dropped to induce market makers to assume the other side of the trade; consequently, prices tended to reverse the following week. This result was consistent with earlier research showing that the pressure of the liquidity trader was higher when stocks dropped in price.
The chart above shows the reversal effect for stocks with high volume that were price-losers during the prior week. The horizontal axis groups stocks according to the percentage loss during the previous week. The colors show the intensity of the percentage increase in volume. The vertical axis shows the percentage gain. Based on data from Alsubaie and Najand (2009).
To be continued….
