The CFO is often thought of as a company’s calm, steady voice of reason amid all the noise generated by industry and market dynamics. And, it turns out, the more visible a finance chief is, the more likely it is that such a reliable presence will be reflected in the company’s stock price.
That’s one interpretation of data provided by Prattle, a small, young company looking to make a big splash in the investment research arena.
In late 2016, Prattle launched its first product, an advanced, quantified sentiment analysis of speeches and other communications by the U.S. Federal Reserve Bank and 19 other central banks worldwide. Its clients — asset managers, hedge funds, investment banks — use the data to help guide their investment decisions.
The algorithms that calculate positive or negative scores for each speech, in terms of their predicted impact on subsequent short-term market movements of various asset classes, are based on speech patterns. The analysis looks at words, sentences, phrases, and the relationships between them to generate a score indicating the speech’s relative dovish or hawkish quality.
That’s not all. Using machine learning and natural language processing, Prattle analyzes how the language patterns in a speech differ from those the speaker used in previous addresses. The analysis tracks how markets moved after the earlier speeches and, if language patterns are different in the new speech, adjusts the expected market movements accordingly. (The analyses are typically available within minutes after a speech.)
In this way, with each new speech the algorithms grow more accurate, according to Brian Peterson, Prattle’s chief operating officer and de facto finance chief.
A convenient example of a language subtlety that Prattle takes into account involves usage of the words “moderate” and “modest.” The company’s founders discovered that it makes a huge difference in financial markets how people use those words, says Peterson.
“In a straight-up sentiment analysis, they could both be positive and they’d have the same weighting,” he says. “But in financial market settings, they can have significantly different meanings. And then, factoring in how individual people use those words in the context of all their speeches can further differentiate whether the words are positive and at what level.”
Prattle upped its game considerably last September, when it began selling access to sentiment analyses for the earnings calls of 3,000 publicly traded U.S. companies. The database includes only companies that have been public for four years, allowing a history of earnings calls to be established.
A clear trend Prattle noticed is that when a company’s CFO speaks for a larger percentage of the call, compared with the CEO, there is less short-term volatility in the company stock price (see the graph below).
The trend holds true for both positive and negative volatility. So, while a relatively verbose CFO is associated with a more predictable stock price, there is no association with the direction of price movement.
Prattle doesn’t point to any clear cause for the trend. However, one may surmise that investors simply react with more volatility to CEOs’ comments, no matter what they are, than to the comments of finance chiefs.
More specific scenarios underscore the point. For example, some companies that typically experience relatively high stock-price volatility may naturally allocate more air time to the CEO, giving greater weight to any explanations for the volatility.
Also, it’s fairly common for CFOs to spend much of their air time on earnings calls simply reciting the newly published financial results, which may not be expected to trigger investor behavior.
Meanwhile, how is Prattle’s methodology working out? A third-party research firm, Lucena Quantitative Analytics, performed portfolio backtests using Prattle data to guide trading strategies. The backtests, which factored in realistic trading conditions such as transaction costs, involved “buying” stocks of companies that Prattle scored as positive and holding them for one month before “selling.”
Applying that strategy to stocks included in the Dow Jones Industrial Average from the beginning of 2013 through mid-September 2017 resulted in a 112% return premium over the Dow’s actual performance. The same strategy, applied to the Russell 1000 stocks from the beginning of 2012 through mid-September 2017, scored a 44% return premium.
The Sharpe ratios for the two portfolios, representing risk-adjusted return, were 1.33 and 1.46 — “huge returns over the benchmarks,” says Peterson.
“At its core, the typical sentiment analysis takes all the positive and negative words in a communication, and the net difference is the score,” Peterson says. “Our version is much more mature.”
Prattle may not need many clients to profit from its groundbreaking products. The company’s main asset is the algorithms, and they are developed and refined by a small number of people. Total headcount is just 18. Startup funds were negligible, although last year the company landed $3.3 million in seed funding for its expansion into equities.
The company is currently working toward expanding its analyses to company press releases and stocks listed internationally.