Predicting asset price movements with a high degree of certainty is harder than ever in today’s dynamic environment which includes a greater focus on quantitative trading. The finance profession is no longer confined to well connected MBA candidates from Wharton claiming to have a knack for stock picking. Many of the best and brightest in physics and mathematics now hold high profile positions on Wall Street and are responsible for defining the mathematical models to unlock potential mispricing opportunities in the market. But given the increasing complexities and intricacies of the market, building an accurate model requires choosing a set of complementing factors which accurately explain and understand the pattern of security returns.
This forecasting technique predates contemporary trading strategies. In the early 90s, Kenneth French and Nobel prize winning economist Eugene Fama examined the differences in returns of a diversified portfolio, now known as the Fama-French 3 Factor Model. The model takes principles of the capital asset pricing model (CAPM), which states that expected returns of a portfolio equal the risk free rate plus a risk premium, two steps further. Decades of empirical research established that a single beta factor inadequately describes an asset’s true movement, only explaining 70% of actual returns. In other words, if a stock increases 10%, about 70% of that movement can be explained by a single risk factor with the remaining 30% due to other factors. In response to CAPM’s shortcomings, Fama and French added size and value components to the original risk factor for a more astute description of asset price movements. Extensive backtesting concluded that the combination of size, value and risk explained between 90 and 95% of total returns, doing a better job of clarifying market anomalies.
As factor analysis becomes more commonplace, a greater number of financial professionals are applying enhanced tools to explain market performance. Most funds now employ a multi factor discipline, often kept from the public eye, to discover hidden gems or impoverished stocks that can be leveraged into profitable trades. In general, many models use some form of fundamental analysis to understand the relationship between price movement and underlying financials, such as earnings. But over time, the ubiquity of fundamental and sell side earnings estimates data means the alpha previously found in quantitative models now turns to beta. Building an effective earnings model that generates alpha in today’s market requires an alternative data set.
With this in mind, we built the Estimize Signal. It’s a multifactor model based on multiple layers of our proprietary crowdsourced earnings estimates. We leveraged research from an earlier paper, “Generating Abnormal Returns Using CrowdSourced Earnings Forecasts from Estimize”, first published by our CEO, Leigh Drogen, and Head of Quant, Vinesh Jha, in 2014.
The biggest factors used in the construction of the Signal comprise a pre earnings component measuring the difference between Estimize and Wall Street as well as post earnings factors such as recent earnings surprises benchmarked against Estimize forecasts. It’s possible to trade both factors independently and generate above average returns, but together it forms a highly powerful tool to generate superior alpha. Ultimately, though, the signal aides funds in position sizing and execution timing across the 2000+ U.S. stocks covered in the Estimize universe and not predicting the exact results of an earnings report or any price movement through print.
The Signal’s inputs include the following factors:
- The difference between Wall Street expected earnings and the Estimize Select Consensus, which is a re-weighted composite forecast of Estimize EPS forecasts, with weights based on the contributor’s track record and the timeliness of their forecasts. The idea is that the Estimize community provides forecasts which are leading indicators of future Wall Street revisions
- The difference between the average Estimize contributor’s EPS estimate and the Estimize Select Consensus. The idea is that top Estimize contributors lead other Estimize contributors and provide an early indication of revisions generally
- The historical likelihood of a company, and others in its industry, to beat or miss EPS and Revenue targets. Wall Street estimates, unlike Estimize estimates, tend to exhibit predictable biases, possibly relating to companies’ ability to manage Wall Street estimates more effectively than they can manage the Estimize community
- The degree to which earnings beat or missed the Estimize Select Consensus, which is a more accurate representation of the market’s true expectations than is the Wall Street consensus – and therefore is a better benchmark for earnings surprises and the subsequent price drift
- The Signal also overweights cases in which the company beat (missed) relative to Estimize, but missed (beat) relative to Wall Street; these are even stronger signals
Delivery of the Signal occurs on a daily basis before the market opens and covers 400 companies reporting in a 13 day window. We specifically designed it this way to harness the most value out of the Estimize data set which peaks in the 10 days leading up and 3 days following an earnings announcement. Meanwhile, daily cuts of data provide managers with sufficient time to exit positions immediately ahead of a company’s report (i.e. same day for reports after market closes and prior day for before opening bell) or enter a new trades in those that recently reported. The Signal, itself, takes on a uniform distribution with scores ranging between -100 and +100.
Although every investor analyzes the data in their own unique way, we found using a dollar neutral strategy generates hefty returns. A basic market neutral strategy by definition seeks to limit systemic risk by taking simultaneous long and short positions either in individual instruments or at the portfolio level. Because one position occurs in conjunction with another the strategy effectively reduces directional exposure and shifts in market sentiment. Profiting using a market neutral strategy depends on the delta in price between two stocks, or in this case the long and short components.
One of the earliest forms of pairs trading (long/short) dates back to the 1960s when Ed Thorp launched the first market neutral derivatives hedge fund, Princeton Newport Partners. The fund compiled a track record of 227 winning months and only 3 losing months, touting annualized returns of about 20% before fees. But even before that Thorp successfully used a “delta hedging strategy”, another term for the same thing, to identify trading opportunities in options. Track records such as Thorp’s prove it’s possible to beat the market with a highly disciplined and mathematical approach.
In essence, the Estimize Signal trading strategy incorporates many of the fundamental principles pioneered by Thorp nearly 50 years ago. A score of positive 100 represents a buying opportunity while a negative 100 screams a short with each subsequent cut of data representing another chance to rebalance or make additional trades.
In our hypothetical portfolio, we created an equally weighted long/short portfolio of stocks in the top and bottom deciles according to a given day’s Signal score. The dollar neutral returns equal the difference between the long and the short portfolio’s return. Over the past 5 years, a dollar neutral strategy using the Signal produced average annualized returns of 23.1% with a 1.75 Sharpe. Returns peaked at 37.7% with a 2.75 Sharpe in 2015 and bottomed in the first 9 months of 2016, but more importantly the Signal produced returns that far exceed the S&P 500 in 4 of the 5 years of testing (Note: In 2013 the S&P 500 jumped 26% whereas the Signal produced returns of about 21%). In other words, investing with the Signal at the start of 2012 generated compounded annualized returns of 173% through Q3 2016 compared to 70% collected by the S&P 500.
Funds employ some of the greatest minds in the world to answer a simple question that escaped the finance profession for so long, “What really drives performance? The answer can vary depending on who you ask and what factors they use to model market behavior. However, many investors recognize that the most accurate models trade on some form fundamental data, namely earnings. To help meet this growing need we developed the Estimize Signal, a multifactor model that uses our proprietary pre and post earnings data to aide traders during earnings season. And in doing so, the model yielded a rate of return that far exceeds the broad market with each passing quarter.
Photo Credit: Robert Scarth