Crowdsourced Earnings Estimates Vinesh Jha CQA - 24 April 2014 Agenda • • • • • Background: crowdsourcing financial forecasts Data Accuracy of a crowdsourced consensus Returns analysis Future directions 2 Forecasting • • • • • • Crowdsourced forecasts have mostly focused on stock price performance (e.g., Motley Fool CAPS) or the outcomes of economic events (e.g., prediction markets) – There are a lot of moving parts in stock prices By focusing on EPS forecasts, we can isolate a particular aspect of forecasting skill Replaces phone calls and buy side huddles And we have a ready-made benchmark in the form of sell side estimates – Sell side biases are well documented. Herding, banking, risk aversion Hope is that crowdsourced forecasts better represent the market’s expectations Improve valuation, revisions and surprise models, research 3 Estimize • • • • Founded in 2011 by Leigh Drogen Platform is free and open for contributors and consumers Pseudonymous Contributor base – Buy side, independent, individuals, and students – Diversity of backgrounds and forecasting methodologies – Users can contribute biographical information 4 Estimize • • • • • 25,000 registered users 75,000 unique viewers of data last quarter 4,000 contributors 17,000 estimates made last quarter Coverage (3+ estimates) on 900+ stocks last quarter 5 Agenda • • • • • Background: crowdsourcing financial forecasts Data Accuracy of a crowdsourced consensus Returns analysis Future directions 6 Data • US listed stocks, Nov 2011 – Mar 2014 • Universe, updated monthly – # Estimize contributors ≥ 3 – Market cap ≥ $100mm – ADV ≥ $1mm – Price ≥ $4 • Potentially erroneous estimates flagged for review or removal • In sample analysis restricted to quarters reporting during 2012 • Returns residualized to industry, yield, volatility, momentum, size, value, growth, leverage 7 Coverage 8 Seasonality 9 Agenda • • • • • Background: crowdsourcing financial forecasts Data Accuracy of a crowdsourced consensus Returns analysis Future directions 10 More accurate For what % of EPS reports is the Estimize consensus closer to actual EPS than is the sell side? >= 1 analyst >= 3 analysts >= 10 analysts >= 20 analysts n % more accurate Estimize Wall Street error error 8971 53% 17.3% 17.4% 4916 58% 13.7% 14.5% 1438 62% 11.7% 12.6% 487 62% 12.6% 13.3% 11 What makes for an accurate estimate? • Regress estimate-level accuracy (% error) against – Track record + • how good has the analyst been in this sector in the past? • accuracy is persistent: better forecasters remain better – Difficulty of forecasting • condition track record on the overall accuracy of the Estimize community • Expect less accuracy if everyone’s been inaccurate – Amount of coverage + • more is better, to a point – Days to report • more recent forecasts contain more information – Bias + • higher estimates tend to be more accurate 12 What makes for an accurate estimate? N 19,796 Factor Parameter Track record 0.09 Difficulty (0.04) Coverage 0.03 Days to report (0.10) Bias 0.15 T 10.90 (3.95) 5.30 (12.58) 25.85 p <.0001 <.0001 <.0001 <.0001 <.0001 13 Agenda • • • • • Background: crowdsourcing financial forecasts Data Accuracy of a crowdsourced consensus Returns analysis Future directions 14 After earnings Estimize N IC 4614 Mean return All surprises 4548 > 1% surprises 4059 > 5% surprises 2521 > 10% surprises 1654 1 day 0.010 0.14% 0.14% 0.20% 0.20% 2 day 0.016 0.14% 0.13% 0.20% 0.25% 5 day 0.024 0.19% 0.16% 0.21% 0.27% Wall Street N 4614 4417 4107 2755 1849 1 day (0.018) 0.08% 0.07% 0.13% 0.10% 2 day (0.012) 0.03% 0.02% 0.06% 0.05% 5 day (0.001) 0.00% -0.01% 0.01% -0.09% 15 After earnings (2) Holding period 1 day 5 day Ann ret 25.7% 10.7% Ann SD 19.8% 14.5% Sharpe 1.30 0.73 % days invested 29% 77% 16 Before earnings • Estimize Delta = % diff between Estimize and Wall St consensus • Delta predicts earnings surprises 17 Before earnings (2) 18 Before earnings (3) Ann ret Ann SD Sharpe % days invested 21.0% 5.8% 3.61 96% 19 Agenda • • • • • Background: crowdsourcing financial forecasts Data Accuracy of a crowdsourced consensus Returns analysis Future directions 20 Improve forecast accuracy • Earlier contributions during the quarter • Forecasts farther out than one quarter • Leverage biographical data, estimate commentary, historical surprise 21 Forecast more things • • • • • Mergers & acquisitions (www.mergerize.com) Macroeconomics Growth & valuation Industry aggregates Industry specific (same store sales, iPods/iPads, FDA approvals, etc) • Other structured data 22 Thanks! vinesh@estimize.com 23