Why are sell-side stock market analysts so wrong? Is there a better way to gather earnings estimates?
It's earnings season again. So you will hear umpteen statements like this from the financial press: "XYZ stock reported earnings of 37 cents, beating estimates by 2 cents."
A few weeks ago, an old friend, Christine Short, emailed me to say she was leaving her post covering earnings for S&P Capital IQ and moving to a new company, Estimize.
I had known Christine for years, and we had spent dozens of hours dissecting stock and sector earnings in that time. She said she was leaving because she thought Estimize had a better way of developing earnings estimates.
My ears immediately perked up. Christine and I both shared an intense dissatisfaction with sell-side analyst estimates.
Here's my beef: The quality of research has declined as many of the brightest sell-side analysts have left for more lucrative jobs with hedge funds and other buy-side firms. Many of those remaining do little if any original research. They often just parrot what the company has told them.
The decline started twelve years ago, when then-New York Attorney General Eliot Spitzer sued Merrill Lynch alleging that its analysts were touting firms as a shill for its investment bankers. That suit forced Merrill and other Wall Street firms to sever the link between compensation for analysts and investment banking. The SEC later concluded a Global Settlement with 10 of the largest investment firms to address these conflicts of interest.
Since then, analyst pay has declined and talent has fled.
But my biggest complaint is this: As less original research is done, companies have increasingly learned how to play the "game" of lowering estimates going into the quarter, then beating estimates when the earnings are announced, which usually causes a pop in the price.
About 20% of companies provide guidance every quarter, and the majority guide lower. So about a fifth of the S&P play this game, which is statistically enough to swing estimates for the as a whole.
Estimize does not rely solely on sell-side analysts; rather, they crowd-source company earnings estimates. They established an online community that allows many different types of investors to contribute forecasts: Buyside, individual traders, independent researchers, even students add their projections. They have been gathering data since 2011 but the number of stocks covered has expanded rapidly just in the past year.
The estimates are contributed anonymously, however each contributor has a unique ID which makes it possible to track the accuracy of every contributor.
Sound crazy? On its surface, it does. Why let in anonymous contributors, even if they are reasonably well-informed? Leigh Drogen, the CEO, says that harnessing "the wisdom of the crowd" can more accurately estimate earnings than competitors.
This is not an academic issue. Improving the accuracy of earnings forecast, especially just prior to earnings announcements, would enable traders to capture the post-earnings drift more accurately.
Is it effective? A few months ago Deutsche Bank back-tested their model. Their conclusion: "We find that the timelier Estimize forecasts can more accurately identify earnings surprise which results in a greater capture of the post earnings drift."
Sounds promising, but there are a couple of caveats. First, the database is still fairly shallow—Deutsche Bank's analysis only starts with 2012—so it's not clear how accurate the forecasts are over many years.
Also, most of the Estimize estimates are made within one week of the earnings announcement, as opposed to sell-side analysts', which are often made a month or more in advance. Estimize data is more accurate within one week of earnings announcement, whereas sell-side analysts are more accurate farther out.
I know what you're thinking: Who's the crowd? What if someone tries to come in with a wild number to skew the results?
There is a reliability algorithm they use every time an estimate is sent in. Say you go onto the site and you are looking at XYZ stock, which currently has a consensus of, say, $1.10. If you put in an estimate at, say, $1.80 and it's the first time you are on the site, it will likely flag you and say, you're estimate is too far outside the trusted range, and is being reviewed manually.
Deutsche Bank concluded: "We found that some of the value-added in the Estimize dataset was due to the 'wisdom of crowds' effect as more predictions give way to greater accuracy."
This makes sense.Think about it: Apple may have almost 50 analysts (Estimize will have about 250 in the final week leading up to the earnings report), but once you get past the S&P 100 most stocks have a dozen at most, with many at half a dozen or so. That makes it more likely you will get a consensus number that is far from the reported number.
But when you have, say, 30 or 40 reasonably well-informed people contributing estimates instead of 10 or 12, the probability that you can get closer to the reported number increases. It is a simple fact of the law of larger numbers.
Is this enough to get me to use Estimize consensus numbers over, say Thomson Reuters or Factset or S&P Capital IQ, all of whom aggregate sell-side analyst data? Not yet, but I will be watching. I'd like to see more follow-up studies using a larger data base.
A lot of money is at stake for anyone who can get the forecasts right! It certainly merits a closer analysis as the database gets larger.
By the way, Wall Street seems to have discovered the value of data mining in general. Recently, another company—TipRanks—was started to gather sell-side analyst estimates, but instead of going to the analyst company's directly, the information comes from sweeping the internet for publicly available information! They have end-run the analysts and their companies, but are still using their numbers!