Monthly Archives: May 2016

Which companies will beat this quarter? A brand new quant approach

I’ve worked on earnings surprise prediction for about 17 years now.  In the late ‘90s and early 2000’s as the head of research at StarMine I built innovative models on top of detailed sell side forecasts, using data that was commercially available but had not yet been analyzed sufficiently. And over the last three years I’ve been an advisor to Estimize, whose groundbreaking dataset of crowdsourced forecasts turns out to be consistently more accurate than the pros and represents a logical next step in earnings prediction.

So I’m very excited to announce ExtractAlpha’s recent collaboration with alpha-DNA on the Digital Revenue Signal (DRS).  DRS gives us a completely new and unique way to get at the likelihood of a company beating expectations.  Rather than collecting explicit forecasts of company financials, DRS builds an implicit estimate of the likelihood of a company to beat or miss revenue expectations,based on trends in a company’s digital demand footprint.  That is, it measures changes in the demand for a company’s products and brands, collected from datasets covering Site, Search, and Social trends.   Integrity Research, a well regarded equity research-focused blog, recently wrote about this new product.

This isn’t just another Twitter sentiment feed.  alpha-DNA collects from a wide swath of different publically and commercially available data sets to create a comprehensive view of consumer demand trends, all with a view towards measuring errors in the market’s revenue expectations.  The underlying technology also includes the Digital Bureau, a proprietary mapping of companies, brands, products, and digital properties which allows us to associate these entities with a security.  Doing this is a lot harder than it might seem at first, as the Digital Bureau must be both comprehensive and regularly updated.

The hard work that goes into building DRS has payed off.  Historically, the top 5% of companies most likely to beat sell side revenue expectations according to DRS’s underlying algorithm did so 70.8% of the time, per the chart below, and the bottom ranked 5% did so only 36.6% of the time.  The most recent quarter was no exception, with 70% of the top 5% beating and only 35% of the bottom 5% beating expectations.  DRS recently correctly called big revenue beats for FOSL and KORS (both up about 30% on earnings day) and a big miss for Verint (VRNT), which took a 15% hit when they reported last week.

Not surprisingly, DRS’s ability to predict errors in the Street’s earnings expectations also translates into return performance.  Below is a chart of a market-neutral portfolio based on DRS deciles going back to 2012, for a liquid US universe.  DRS exhibits a Sharpe of nearly 2 and performs comparably across large, mid, and small cap names, with modest turnover compared to other earnings and revenue forecast signals.  As far as we know it’s also fairly uncorrelated to other commonly used factors, so it should be additive to most quant processes.

In earnings season a tool like DRS can be invaluable in capturing big consumer data in an innovative yet intuitive way.  Several of the top US quant managers are currently evaluating DRS as part of their systematic process.  Are you also looking for a way to figure out which companies are likely to beat or miss this quarter?  Contact us!

Announcing AlphaLetters: finding diamonds in the academic rough

Our clients often ask us where we get our research ideas.   And one of them – a head of research at a top quantitative firm – reflected our own attitude towards this question when describing his interest in our models: “we don’t have a monopoly on creativity.”  A forward-thinking quant needs to keep his eyes open for anything that might inspire a research project: a new analytical technique; a newly available dataset, especially one which comes from a provider outside of financial services; discussions with practitioners within and outside of the capital markets; and, of course, from academia.

To that end I’m pleased to announce ExtractAlpha’s new partnership with AlphaLetters, a service dedicated to finding great quantitative portfolio management ideas from academic research.  For more background, read on, but if you’d like to learn more about subscriptions, please contact us!

Quantitative finance has long been inspired by, and in the early days was largely pioneered by, academics.  Prominent academics often hold posts within large institutions, as personified by Gene Fama’s dual posts at my alma mater, the University of Chicago, and at Dimensional Fund Advisors.  And with the advent in the mid-1990’s of the Social Science Research Network (SSRN), where we post  some of our research, practitioners’ access to cutting edge academic working papers no longer required a university connection or subscriptions to expensive journals.

The wide availability of working papers hasn’t been an unqualified boon for practitioners.  A friend in academia once told me that some of his colleagues had largely held off from putting papers up on SSRN too early – perhaps in part for fear of being “scooped,” and in part because they might find greater benefit to implementing their best ideas at a fund management company rather than disseminating them, at least prior to acceptance in a peer-reviewed journal which would satisfy their need to publish rather than perish.

And then there is the concern that publishing an idea, especially a discovery of a stock selection anomaly, can kill off or dampen that same anomaly.  A widely known example is Sloan’s publication of the Accruals anomaly in 1996 and its subsequent implementation across many hedge funds, which at least according to some studies have let to its compression or even, to quote one academic paper on the subject, its demise.

All of this is not to discredit academic research as a source of ideas, but rather topromote more proactive use of academia as an idea generating tool.  This means actively searching academic journals for great ideas.  It also means being selective in which ideas to test, and being efficient about determining which of the many articles warrant attention.  Academics often find interesting results which are impractical to implement, because the investment horizon is too short or too long; because transaction costs and a practical investible universe aren’t considered; or because the data is proprietary or not available on an ongoing basis in something close to real time.  This makes sifting through academic journals a daunting task.

Since 2005, the team of Wall Street-trained professionals at AlphaLetters have done that task for investment managers.  They’ve scoured dozens of top journals, university sites, and conference proceedings, to select the most intriguing articles in their monthly newsletter.  A recent edition includes digests of interesting papers on persistent day-of-the-week effects across different stock types, and on momentum in various anomalies’ returns, in other words, timing quant strategies based on their recent performance.

Over many years I’ve consistently found AlphaLetters’ curation of this kind of academic research to be insightful, concise, and practitioner-friendly, and a great tool for idea generation, as do their clients who include 6 of the 10 largest quant hedge funds, and I’m sure you’ll find the same.

This is an unusual partnership for us – we generally partner with FinTech firms with unusual data sets – but we decided the offering was compelling enough to our clients to warrant a collaboration.  Contact us to learn about subscribing to AlphaLetters.