How Fund Managers Trade on Sentiment

On the first trading day of this year Valentijn van Nieuwenhuijzen’s laptop pops open.  He sees right away that something special is going on. The MarketPsych Indices, the system by which NN Investment Partners [formerly ING] measure sentiment in the market, are coming in…words that are associated with stress and gloom course through digital media...

The main strategist of the Dutch asset manager, with € 187 billion under management, taps the brakes: he reduces investments in shares and raises more cash. Later that week the stress spreads: worldwide stock markets went down sharply...

~ Lenneke Arts & Jeroen Groot.  February 28, 2016.  “Beleggen met big data wordt langzaam gemeengoed.” Financial Daily.  (Translated from the Dutch).

Some of the largest global asset managers are actively predicting price swings using sentiment analysis.  They understand that staying in front of the herd is key to successful investing.  The language used by media, with its broad influence on financial risk takers (investors), appears predictive of stock prices.  Today's newsletter is a summary of our findings into how media, language, and herds of investors predictably move prices – explained in detail in our new book "Trading on Sentiment:  The Power of Minds over Markets."

Information Impact

Information sometimes hits market prices hard.  The non-farm payrolls number, released the first Friday of each month, has a significant impact on the value of the U.S. Dollar and Treasury bonds.

Elijah DePalma, Senior Quantitative Research Analyst at Thomson Reuters, analyzed the millisecond impact of the nonfarm payrolls (NFP) on the U.S. dollar future contract (DXZ4) on one day—December 5, 2014. A minutely chart of the December dollar index future contract below shows that the price impact is nearly instantaneous with the news release.


SOURCE:  Courtesy of Elijah DePalma, Senior Quantitative Research Analyst at Thomson Reuters

Dr. DePalma notes that on December 5, 2014, $5.7 million of USD contracts (DXZ4) were traded within 63 milliseconds of the NFP release, and $29 million was transacted within 100 milliseconds.

Information that is not numerical (as Nonfarm payrolls is) that is conveyed in text is more difficult to measure.  MarketPsych’s expertise in text analytics allows us to tackle the non-numerical side of information flow – the concepts that influence and bias investors.   Our sentiment-based data feed allows us to deeply understand how information causes herding, and when it doesn’t.  This feed is called the Thomson Reuters MarketPsych Indices, and it is consumed by the world’s largest quant funds and banks for trading and risk applications.

Trading on a Feeling

Daniel Kahneman’s research on the framing effect demonstrated that the words used to describe a financial risk - words such as “risk” “lose” or “opportunity” - could influence choices more than the stated odds themselves.  In Kahneman’s experiments the odds of various outcomes could be mentally calculated, yet the language used to describe the risk (or opportunity) changed the choices of particpants.  Unlike in Kahneman’s experiments, in financial markets the odds are unknown, and minor deviations in language may hold even greater influence over the investing herd.

Through text analysis we have tapped into a profoundly new way of understanding markets.  Most investors rely on lagging indicators such as past price action and fundamental numbers.  But if investors want to predict the future it helps to understand the information that drives investors and the markets they trade in. 

Media conveys information.  That information creates feelings in traders.  Those feelings are acted on immediately in some cases, and slowly in others.  The following image depicts the channels, and feedback loops, that dictate how information is incorporated into prices.


Some information drives intense feedback loops that result in an “echo-chamber” effect, in which fear is amplified over the following weeks (as described in the opening quote).

Understanding Information Flow

Statistical analyses demonstrate the existence of two broad sentiment-based patterns in financial prices: overreaction and underreaction. These concepts refer not only to patterns of prices but also to the collective investor reactions to information that fuel such patterns.  In the case of underreaction - boring, routine, or complex information may be initially overlooked by investors. As they wake up to its meaning, they gradually buy (or sell), slowly moving prices. Underreaction is what NN Investment Partners exploited in January.

The overall feeling tone—positive versus negative—expressed in the media about the S&P 500 and its constituent stocks—is called sentiment. When simple moving averages of the net sentiment of the combined news and social media commentary about the S&P 500 are plotted against the value of the S&P 500 itself, a simple underreaction relationship is evident, as depicted below from 1998 through mid-2015.


When short-term sentiment drops below longer-term sentiment, prices tend to follow it lower. When short-term sentiment rises above the long-term average, S&P 500 prices often rise higher. This simple strategy earned 314% from January 1, 2000 through July 31, 2015 with a maximum peak-to-trough drawdown of 33%. The S&P 500 itself, adjusted for dividends, earned 91% over that period with a maximum drawdown of 57%. Investors using this strategy would have been short stocks during the drawdowns associated with the 2000-2002 tech stock crash, the 2007-2009 financial crisis, and the August 2015 and January 2016 volatility.

The 200-day versus 500-day sentiment averages for the S&P 500 work well to demonstrate the sentiment effects over the business cycle, but over shorter time periods, shorter MACDs have similar profitability for timing turns in sentiment.  The following image depicts the 30-day average of sentiment versus the 90-day average for the S&P 500 from July 2015 through early April 2016.  Like the 200 versus 500 day averages, the relationship isn’t perfect, but it is significantly better that buy-and-hold.



While these sentiment MACDs are fairly good at predicting downturns in market prices, they are poorer when it comes to timing recoveries.  The recovery from a correction is not preceded by an upturn in sentiment.  Typically it is a bounce as a response to excessively low sentiment – bouncing off a base of extremely negative sentiment - a process called overreaction.

When Media Provokes Collective Panic

When the herd panics, and the consensus forecast becomes very dire, eventually all of the “weak hands” sell out of the market.  In the absence of sellers, there are only cautious buyers left, and prices rise.  Such bounces typically follow a panicked overreaction.  Such overreactions occur not only across the entire stock market, but also in individual stocks.

Value investors such as Warren Buffett search for stocks that are inexpensive, with high earnings-to-price ratios.  Similarly, value investors including Buffett often reference the value of buying on fear (buying when the herd has overreacted, typically to some negative news or expectations).  In order to test Warren Buffett’s advice to buy cheap stocks that have been scared lower, our team statistically examined the returns of stocks that are both value stocks and associated with high media fear. Using an annual holding period and starting at a notional $1, the returns since 1999 of the top 5 percent of S&P 500 stocks by value (earnings-to-price ratio) are plotted below in the light gray line, yielding a 7-fold return to July 31, 2015.  The equity curve of a subsection of stocks that are both value stocks and associated with high fear are represented in the darker line, yielding a 23x return, more than a threefold increase in returns over value alone.  Note that transaction costs were not included in this analysis.



Importantly, this is a conceptually simple strategy, and it has been described anecdotally by distressed asset investors, lending further support to its legitimacy.

Conclusion

The herd of investors tend to underreact to new information (useful in timing as seen in the opening quote), and then to overreact (taken advantage of by value strategies like those of Warren Buffett).

At times feelings overwhelm cold reality, and prices become driven by collective emotions of panic or greed. Prices often reverse after such events, leading to the label of overreaction. Conversely, information that is too complex, boring, or in disagreement with investors’ prejudices is ignored, and price trends form as investors underreact to it.

As seen in the MACDs of sentiment, investors underreact to gradual shifts in media sentiment, and as seen in the value strategy example they overreact - excessively discounting stock prices - when frightened. 

Learn more about improving your investment returns with insights from sentiment analysis of the herd in our new book, “Trading on Sentiment:  The Power of Minds Over Markets.”

We love to chat with our readers about their experience with psychology in the markets.  Please send us feedback on what you'd like to hear more about in this area.

If you represent an institution, please contact us if you'd like to see into the mind of the market using our Thomson Reuters MarketPsych Indices to monitor real-time market psychology and macroeconomic trends for 30 currencies, 50 commodities, 130 countries, 50 equity sectors and indexes, and 8,000 global equities extracted in real-time from millions of social and news media articles daily.

Good Reactions,
The MarketPsych Team