Daniel Kahneman earned the Nobel prize in economics for his research on prospect theory. His scholarship helped demonstrate how behavioral finance — and by extension, sentiment analysis — can improve our understanding of market behavior.
Sentiment analysis applies algorithms to news articles, social media, and other data sources to gauge how people feel about the market, while behavioral economics identifies the cognitive biases that affect decision making. Sentiment analysis can help illuminate how these biases manifest in the financial markets. Of course, what people do is often more revealing than what they say, so sentiment analysis doesn’t always capture the complexity of human emotions in a domain as charged as the financial markets.
Nevertheless, it can help us interpret and anticipate market behavior. Here’s how.
Technical analysts tend to measure sentiment tangentially, by approximating when a turning point will occur. But their results are often inconsistent since their methods are associational and may not identify the “cause” behind market outcomes. Fundamental analysis takes a more causal approach, but its feedback loop is often longer than investor time frames and does not always differentiate value from a value trap.
The best investors intuitively understand that markets are not good at discounting future outcomes. During the subprime crisis, for example, the pricing of subprime securities indicated that the market valued 80% of the underlying loans at roughly zero. This made for a highly favorable risk–reward proposition for those investors who knew what to look for. Similarly, last year, market sentiment largely anticipated a recession this year.
“The best trades are the ones that will get you laughed off the set of CNBC.” — Jared Dillian
Jared Dillian is one of my favorite sentiment traders, and his point is an important one. While he believes in sentiment analysis, he acknowledges that it is a hard strategy to raise money around. After all, a trade that inspires laughter doesn’t necessarily inspire confidence or investment capital. Moreover, many doubt sentiment analysis’s scientific rigor and see it as akin to astrology.
But by reorganizing market data and applying the principles of auction theory, we can use sentiment analysis to classify market behavior. James F. Dalton has pioneered the application of the Market Profile technique, developed by J. Peter Steidlmayer, to recognize the behavior of different market participants. Specifically, Dalton’s technique observes the shape of a day and other “market-generated information.” For example, if the market is falling on a particular day and only a limited set of market participants is selling or the sales are driven by long liquidation and not new sellers, the shape of the day might resemble the letter “b.” At the other end of the spectrum, if speculation and short-covering activity are driving the buying, the shape of a day might resemble a letter “p.” These behaviors indicate weaker forms of buying and selling and may signal that the market may not be as strong or weak as price alone would make it seem.
How can we know if these shapes are conveying important and actionable information? By applying artificial intelligence (AI), we can test whether the shape of a day is due to a truly random process. How? By modeling such a process and comparing that with the actual shapes observed in the market. If market moves are arbitrary, the distribution of shapes from a random process would match the actual distribution of shapes. But they don’t.
Auction Process: Day Classification
The test shows with 99% confidence that these results do not conform to a truly random process. If they’re not arbitrary, then they must yield valuable information. Indeed, the largest deviations from the random distribution occur when the shapes indicate the market is too long and too short due to short covering or long liquidation. This supports the intuition that these behaviors are both unique and potentially actionable from an investment perspective.
In “Market Profile with Convolutional Neural Networks: Learning the Structure of Price Activities,” Chern-Bin Ju, Min-Chih Hung, and An-Pin Chen show that using similar image-recognition techniques can identify market patterns that may inform commodity producers’ hedging strategies. Such research could lead to a deeper understanding of the market’s price-setting process and help quantify investor sentiment. Investors tend to focus on price alone, and momentum strategies are widely followed. Such trades can get too crowded at times, leading to reversals. This is not random behavior, and now we have a way to objectively measure that behavior.
This research provides a heuristics-based technique for causality testing. Markus Schuller and Andreas Haberl laid out the forward-looking case for causality in “Causality Techniques in Investment Management: Five Key Findings.” They observe that financial markets are “complex, dynamic, and forward-looking” and are driven by “market participants with imperfect information and bounded rationality.” The ability to objectively observe and measure the behaviors of these market participants is “both appealing and potentially very lucrative.”
That’s how sentiment analysis can help uncover alpha opportunities and why it is worth including in our investment toolkits.
For more market commentary from Joshua J. Myers, CFA, subscribe to his Substack at Cedars Hill Group (CHG).
If you liked this post, don’t forget to subscribe to Enterprising Investor.
All posts are the opinion of the author. As such, they should not be construed as investment advice, nor do the opinions expressed necessarily reflect the views of CFA Institute or the author’s employer.
Image credit: ©Getty Images / vladystock
Professional Learning for CFA Institute Members
CFA Institute members are empowered to self-determine and self-report professional learning (PL) credits earned, including content on Enterprising Investor. Members can record credits easily using their online PL tracker.
Bitcoin 
Tether 
XRP 
Dogecoin 
USDC 
Lido Staked Ether 
LEO Token 
