Estimize, the first crowdsourced earnings estimate service, has launched a weighted consensus based on measuring the accuracy of its analysts' forecasts.
Known as the Estimize Select Consensus, it looks at the attributes of an analysts’ track record, biographical data and the estimate itself, to give the highest confidence that the individual estimate is more or less accurate than the other analysts in the distribution.
“It’s a significantly more accurate than even our flat consensus which beats Wall Street 67% of the time,” said Leigh Drogen, founder and CEO of New York-based Estimize. What’s more, the Select Consensus if 70 to 80% more accurate than the sell side consensus, added Drogen.
To lead the project and develop the algorithm underpinning the Select Consensus, Estimize has hired Vinesh Jha, who is the former head of quantitative research at StarMine, the only estimate service to develop intelligent weightings for the accuracy of each of its analyst’s forecasts, according to Drogen.
“We basically built StarMine 2.0, and Vinesh built that,” says Drogen. Most recently, Jha worked from 2007 to 2013 on the buy side as executive director at Morgan Stanley’s internal quant trading unit PDT Partners. The unit – whose acronym stands for Process Driven Trading — and earned about 20 percent per year over the past decade — was recently spun out as an independent hedge fund PDT Partners.
“During my tenure at PDT I came to realize that unique and innovative data sources were a key driver for outperformance among quants; to outperform, you need to take bets that are based on intuitive and robust concepts, and you need to do so in a way that’s different from the competition,” commented Jha in a backgrounder about the research at Estimize.
[For more on Estimize Looks to Tap the Wisdom of the Buy Side Crowd, see Ivy Schmerken's related story.]
Hiring Jha was a big step since he brings considerable experience on both the vendor and quant sides of the aisle. “He’s created the tools to assess analysts and he’s used those tools to generate alpha and he’s used different innovative data sets and understands how to generate alpha form those data sets,” says Drogen.
Launched in December of 2011, Estimize has grown into a community of 3,000 contributing analysts from the buy side and hedge funds along with 16,000 registered members.
To create the Select Consensus, Jha’s team has looked the different attributes of analysts contributing to Estimize and run linear regressions to figure out which attributes are highly correlated to predictive analysis, says Drogen.
Each estimate is given a confidence score, which is used to weight the consensus more or less heavily in favor of that individual analysts’ projection. The confidence score is based on the attributes of the estimate itself and the attribute of the analyst.
Inside the distribution of estimates for a given earnings release, Estimize determines whether a particular estimate is aggressive or conservative.
“For certain analysts who are aggressive, they are more accurate, and for some that are more conservative, they are more accurate, and we know this and are able to weight it accordingly,” said Drogen.
A good example would be the time between when the analyst made the estimate and when the company reported its earnings. Estimize’s view was that the most recent estimates are more accurate than older estimates. It turns out that is the case, says Drogen, adding that if an analyst took more time to submit an estimate it was because they waited to get the latest information. Thus, as part of that algorithm, more recent estimates are higher weighted, he says. They end up giving the analyst a higher confidence score.
There are many pieces of data to weight. Another discovery is that analysts who provide unstructured text notes on the platform, such as on what a company is doing, are often more accurate, says Drogen. Estimize is collecting data from a community of 3,000 contributors from buy side and independent research analysts. It offers coverage on 930 stocks with at least 3 three estimates per quarter.
“We are in the process of using these and other insights to build better measures of analyst accuracy, a more accurate composite estimate, and analytics which will help institutional investors make the best use of data in a rigorous way,” stated Jha in background material about the new data strategy.
Jha said, initially the focus of the firm’s quant research was about gaining insight into the characteristics of the Estimize community, such as when and how do analysts make forecasts and who are good analysts, and are they consistently good, stated Jha in background material. “Next, we wanted to get a handle on how much alpha is in the data, and how one might capture it in a trading strategy,” wrote Jha.
Looking ahead, Estimize’s focus is on building an analytics layer and rolling out alerts and notifications based on this Select Consensus data, for use in filtering and sorting. Clients have also asked the firm to put the Select Consensus estimates into the API.
Drogen says the firm has focused heavily on bringing more institutional analysts onto the platform and has been doing direct outreach to hedge funds an asset management firms.
While the service is free on the front-end and open to consumers and contributors who register to participate, Estimate will offer a paid service to quantitative hedge funds and others that want to capture the data through the API.
[Check out this case study: Major Online Brokerage Combines People, Process, and Technology to Improve End-User Experience, which will be discussed at the upcoming Interop event in NYC.]