Alternative Data Integration, Analysis and Investment Research

Transcription

1 Alternative Data Integration, Analysis and Investment Research Yin Luo, CFA Vice Chairman Quantitative Research, Economics, and Portfolio Strategy QES Desk Phone: L2Q Conference June 20, 2017 DO NOT FORWARD DO NOT DISTRIBUTE DOCUMENT CAN ONLY BE PRINTED TWICE This report is limited solely for the use of clients of Wolfe Research. Please refer to the DISCLOSURE SECTION located at the end of this report for Analyst Certifications and Other Disclosures. For Important Disclosures, please go to or write to us at Wolfe Research, LLC, 420 Lexington Avenue, Suite 648, New York, NY 10170

2 Agenda Table of Contents 1. Introducing Luo s QES Research 2. Crowdsourcing Revenue and Earnings Estimates 3. Text Mining Unstructured Corporate Filing Data 2

3 1. Introducing Luo s QES Research Top Ranked Quantitative and Macro Research Team. The team has been ranked #1 in the Institutional Investor s II-All America, II-Europe, and II-Asia surveys in the Quantitative Research sector, and top ranked in the Portfolio Strategy and Accounting & Tax Policy categories. Big Data and Machine Learning. We fully incorporate Big Data (e.g., text mining, news sentiment, satellite imagery, securities lending, crowding sourcing) and machine learning in our research, as reflected in our LEAP global stock selection model. Systematic Global Macro Research. Our research on Nowcasting economic growth in >40 countries/regions has received tremendous feedback from clients. Style rotation and factor timing is a core component of our global stock selection models. Alternative data sources (e.g., news sentiment, real-time hiring, satellite imagery, Google trends) are also fully integrated in our macro research. Useful Tools for Fundamental Managers. In addition to research, we also provide a suite of useful tools, including online screening and factor performance tracking, industry-specific modeling, position sizing and portfolio construction, and portfolio analytics to discretionary managers. Please contact us at Luo.QES@wolferesearch.com, if you are interested in our research and services. 3

4 The Complex Web of Data Sources: Wolfe Research Luo s QES 4

5 2. Crowdsourcing Earnings and Revenue Estimates Alternative Big Data on Earnings and Revenue Estimate. In this research, we study an alternative data source based on the concept of crowdsourcing. Estimize is an online platform that allows individuals with different background to contribute their financial forecast. We find Estimize estimates to be not only more accurate and timelier than the sell-side, but also highly complementary to traditional factors. Estimize FES Model. Diving into detailed Estimize estimates, we find that accuracy can be further improved along three dimensions: the freshness of the estimates, analyst experience, and analyst skill. We then introduce a smart Estimize consensus called FES (Freshness, Experience, and Skill). Smart Strategies around Earnings Announcement. We explore three different type of trading strategies around earnings releases using the Estimize data. The pre-earnings announcement strategy buys stocks based on earnings revisions in the week before the earnings reporting date. PEAD (Post Earning Announcement Drift) strategy attempts to capture the drift alpha immediately after the earnings announcement, based on earnings surprise. Lastly, we propose a low risk longonly strategy by avoiding earnings risk and earnings uncertainty. Enhanced Value Strategies. Many fundamental and quantitative strategies explicitly or implicitly rely on earnings and revenue estimates. For long-term value investors, we show how Estimize data and our FES model can be used to boost performance. In the end, we also overlay our enhanced value strategy with a low risk tilt (by avoiding earnings uncertainty) to further improve return and reduce risk. 5

6 a) The Basics of Crowdsourcing Earnings and revenue estimates are probably the most important drivers of stock returns and risks. Unlike conventional sell-side consensus, Estimize crowdsources estimates from a wide range of contributors. Breakdown of Estimize data Finance Professionals Non-professionals Contributors to the Estimize database Non Professional Sell Side Financial Professional Buy Side Independent Telecommunication Services Utilities Energy Consumer Discretionary Consumer Staples Materials Industrials Financials Information Technology Health Care Student Academia Insurance Firm Wealth Manager Investment Bank Financial Advisor Broker Other Endowment Fund Pension Fund Private Equity Fund of Funds Mutual Fund Proprietary Trading Firm Venture Capital Hedge Fund Asset Manager Other Independent Research Other 46% 54% Breakdown of finance professional Independent Buy Side Sell Side 30% 45% 25% Sources: Estimize, Wolfe Research Luo s QES 6

7 Estimize Data Distribution # of companies (in the Russell 3000 index) with Estimize analyst coverage 1600 >=1 anlalyst >=3 anlalysts 1200 >=10 anlalysts >=30 anlalysts Number of analysts Estimize data sector distribution 25% Number of days before earnings report 20% 15% 10% 5% 0% % of stocks in the Estimize database % of stocks in the Russell 3000 index Sources: Estimize, S&P Capital IQ, FTSE Russell, Wolfe Research Luo s QES 7

8 b) The Accuracy of Crowdsourced Estimates EPS estimate accuracy Revenue estimate accuracy 100% 100% 90% 90% 80% 70% 57% 59% 60% 62% 64% 80% 70% 48% 49% 50% 51% 51% 60% 60% 50% 50% 40% 40% 30% 20% 10% 0% 43% 41% 40% 38% 36% >=1 analyst >=3 analysts >=5 analysts >=10 analysts >=30 analysts 30% 20% 10% 0% 52% 51% 50% 49% 49% >=1 analyst >=3 analysts >=5 analysts >=10 analysts >=30 analysts Sell-side more accurate Estimize more accurate Sell-side more accurate Estimize more accurate 100% 50% 0% Estimize EPS accuracy by sector Estimize accuracy, domestic versus multinational firms 100% 90% 80% 70% 59.1% 60.6% 60.8% 61.4% 60% 50% 40% 30% 20% 40.9% 39.4% 39.2% 38.6% 10% 0% <10% exus >10% exus >20% exus >50% exus Sellside more accurate Estimize more accurate Sellside more accurate Estimize more accurate Sources: Estimize, IBES, S&P Capital IQ, FTSE Russell, Wolfe Research Luo s QES 8

9 c) Factors Determining Analyst Accuracy Comparing the above and below median freshness Older estimates, more accurate Newer estimates, more accurate Comparing above and below median experience Below median experience analysts, more accurate Above median experience analysts, more accurate Comparing above and below median skill Below median skill analysts, more accurate Above median skill analysts, more accurate 39% 38% 41% 61% 62% 59% Comparing top and bottom decile freshness Oldest estimate, more accurate Newest estimates, more accurate Comparing top and bottom decile experience Bottom decile experience analysts, more accurate Top decile experience analysts, more accurate Comparing top and bottom decile skill Bottom decile skill analysts, more accurate Top decile skill analysts, more accurate 34% 35% 38% 66% 65% 62% Sources: Estimize, IBES, S&P Capital IQ, FTSE Russell, Wolfe Research Luo s QES 9

10 Estimize FES (Freshness, Experience, and Skill) Model Weighting the estimate by the freshness, analyst experience, and analyst skill Equal weight, more accurate Weighted by freshness, more accurate Equal weight, more accurate Weighted by experience, more accurate Equal weight, more accurate Weighted by skill, more accurate 54% 46% 53% 47% 53% 47% Estimize FED model versus single weighting scheme FES model, more accurate Weighted by freshness, more accurate FES model, more accurate Weighted by experience, more accurate FES model, more accurate Weighted by skill, more accurate 48% 52% 46% 54% 45% 55% Sources: Estimize, IBES, S&P Capital IQ, FTSE Russell, Wolfe Research Luo s QES 10

11 d) Trading around Earnings Announcements Earnings announcement is among the most significant market moving corporate events As earnings release date approaches, the Estimate data becomes more and more accurate. Estimize provides timelier estimates, while sell-side analysts are better at predicting earnings over a longer horizon. If the consensus moves up right before earnings announcement, a company is more likely to beat the consensus. Estimize EPS accuracy as a function of the # of days prior to earnings announcement Sell-side more accurate Estimize more accurate 100% 50% 0% # of days prior to earnings release An example of Pandora Media before earnings announcement CIQ EPS IBES EPS Estimize EPS Actual EPS Price # of estimates for Pandora Media before earnings announcement 60 CIQ IBES Estimize EPS Stock price # of estimates Sources: Estimize, IBES, S&P Capital IQ, FTSE Russell, Wolfe Research Luo s QES dates dates 11

12 Earnings revisions prior to the announcement date leads announcement day return EarningsRevision,,. = EPS,,. EPS,,.34 ENRP,,. = EPS,,. EPS,,.34 EPS,,.34 Price,,. Percentage of EPS changes in the week before earnings announcement A) Most positive EPS revisions B) Most negative EPS revisions 120% 100% 80% 60% 40% 20% 0% Sell-side Estimize FES top 1% top 2% top 5% top 10% Average excess earning announcement day return 0% -10% -20% -30% -40% -50% -60% -70% -80% Sell-side Estimize FES bottom 1% bottom 2% bottom 5% bottom 10% Sources: Estimize, IBES, S&P Capital IQ, FTSE Russell, Wolfe Research Luo s QES 12

13 Post Earnings Announcement Drift (PEAD) PEAD is more significant, if a company beats (or misses) the Estimize estimates. PEAD decays away after two days. Positive earnings surprise, first day PEAD 0.6% 0.5% 0.4% 0.3% 0.2% 0.1% 0.0% 10% surprise 20% surprise 30% surprise 40% surprise Estimize Sell-side Negative earnings surprise, first day PEAD 0.0% -0.1% -0.2% -0.3% -0.4% -0.5% -0.6% -0.7% 10% surprise 20% surprise 30% surprise 40% surprise Estimize Sell-side Excess return when EPS beats over 40% Excess return when EPS misses over 40% 0.8% 0.2% 0.6% 0.0% 0.4% -0.2% 0.2% -0.4% 0.0% -0.6% -0.2% day 1 day 2 day 3 Estimize Sell-side Sources: Estimize, IBES, S&P Capital IQ, FTSE Russell, Wolfe Research Luo s QES -0.8% day 1 day 2 day 3 Estimize Sell-side 13

14 e) Long-term Investment Strategies Benchmark earnings yield. For the benchmark factor, we use the consensus sell-side EPS, by taking a simple average of CIQ and IBES. Standard Estimize earnings yield. In this case, we replace the sell-side consensus with the Estimize estimates when we have at least three contributors in the Estimize database. Estimize FES earnings yield. Instead of replacing with the standard Estimize estimates, we use the FES model introduced in the previous section. EarningsYield = Monthly turnover FQ1 EPS Price Cumulative performance, long/short quintile portfolio on S&P Benchmark earnings yield Estimize Avg earnings yield Estimize FES earnings yield Sharpe ratio, different transaction cost assumptions 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Benchmark earnings yield Estimize Avg earnings yield Estimize FES earnings yield Sources: Estimize, IBES, S&P Capital IQ, FTSE Russell, Wolfe Research Luo s QES No cost 2 bps 5 bps Benchmark earnings yield Estimize Avg earnings yield Estimize FES earnings yield 14

15 Enhanced Value Strategies Enhanced value strategies produce decent performance in a long-only context. Sharpe ratio, different universe Sharpe ratio, monthly versus daily rebalance Enhanced earnings yield using Estimize data performs equally well in both large- and small-cap universes. S&P 500 Russell 3000 Russell 3000 sector neutral Baseline Estimize Benchmark earnings yield Estimize Avg earnings yield Russell 3000 universe, monthly rebalance Russell 3000 universe, daily rebalance Estimize FES earnings yield Long only portfolio performance, Russell 3000 universe Sharpe Ratio, Long only portfolio Russell 3000 universe Equally weighted Russell 3000 Top quintile Estimize earning yield Sources: Estimize, IBES, S&P Capital IQ, FTSE Russell, Wolfe Research Luo s QES 0.0 Equally weighted Russell 3000 Top quintile Estimize earning yield 15

16 4. Text Mining Unstructured Corporate Filing Data # of EDGAR Filings (Daily) Average # of Words in the 10-K Filings Sources: EDGAR, Wolfe Research Luo s QES 16

17 Lazy Prizes Firms in general use a well-defined format for their annual and quarterly filings. Almost identical language is simply repeated year after year, until someone actively intervenes and makes changes. When firms do break away from their tradition of textual descriptions in their filings, they typically foresee significant changes in their business, risk, or corporate strategy. Our NLP algorithms are able to identify potential issues from textual information among US transportation companies. An Example: American Express Price Substantial and increasingly intense competition for partner relationships; risk surrounding Airline industry. Ongoing legal proceedings can cause substantial monetary and reputational damages. US Transportation Industry, Cumulative Performance US Transportation Industry, Returns by Merciless Sources: Bloomberg Finance LLP, EDGAR, FTSE Russell, S&P Capital IQ, Thomson Reuters, Wolfe Research Luo s QES 17

18 Systematic Profiling EDGAR Composite (SPEC) Model SPEC model rank IC SPEC model Long/short portfolio performance Quintile portfolio returns Portfolio Sharpe ratio Long/short monthly turnover Rank IC decay Sources: Bloomberg Finance LLP, FTSE Russell, S&P Capital IQ, Thomson Reuters, Wolfe Research Luo s QES 18

19 Interaction with traditional factors Lastly, it is interesting to note that our factor has almost no exposure to classic risk factors such as size, beta or volatility. One of the most important reasons of relying on alternative Big Data sources rests on the diversification benefit, in that they are expected to have minimal correlation with traditional factors. As expected, our signals based on EDGAR text mining are almost uncorrelated to any of our traditional factors. Finally, as a unique alpha source, the SPEC model should complement and add value to traditional factors. The performance of all traditional factors improves remarkably, when the SPEC is added. Growth factors witness the largest improvement. Improvement in risk-adjusted performance is even more significant. The Sharpe ratio improves by 50% for the majority of the conventional factors. Sources: Bloomberg Finance LLP, FTSE Russell, S&P Capital IQ, Thomson Reuters, Wolfe Research Luo s QES Factor correlations with EDGAR composite models Earnings Yield Earnings Revision Momentum 1M Reversal ROE EPS growth Market cap Volatility Composite 10K 11% 6% 4% 1% 10% -3% 11% -14% Composite 10Q 8% 3% 3% 1% 7% -2% 8% -10% Composite All 11% 5% 5% 2% 11% -2% 11% -13% CAGR (%) Annualized returns of Style composites vs base factors Earnings Revision Earnings Revision composite Earnings Yield Earnings Yield composite EPS growth Sharpe ratio of Style composites vs base factors Sharpe ratio Earnings Revision Earnings Revision composite Earnings Yield Earnings Yield composite EPS growth composite EPS growth EPS growth composite Momentum Momentum composite Momentum Momentum composite ROE ROE ROE composite ROE composite 19

20 An Interactive Web Portal 20

21 DISCLOSURE SECTION Analyst Certification: The analyst of Wolfe Research primarily responsible for this research report whose name appears first on the front page of this research report hereby certifies that (i) the recommendations and opinions expressed in this research report accurately reflect the research analysts personal views about the subject securities or issuers and (ii) no part of the research analysts compensation was, is or will be directly or indirectly related to the specific recommendations or views contained in this report. Other Disclosures: Wolfe Research, LLC does not assign ratings of Buy, Hold or Sell to the stocks it covers. Outperform, Peer Perform and Underperform are not the respective equivalents of Buy, Hold and Sell but represent relative weightings as defined above. To satisfy regulatory requirements, Outperform has been designated to correspond with Buy, Peer Perform has been designated to correspond with Hold and Underperform has been designated to correspond with Sell. Wolfe Research Securities and Wolfe Research, LLC have adopted the use of Wolfe Research as brand names. Wolfe Research Securities, a member of FINRA ( is the broker-dealer affiliate of Wolfe Research, LLC and is responsible for the contents of this material. Any analysts publishing these reports are dually employed by Wolfe Research, LLC and Wolfe Research Securities. The content of this report is to be used solely for informational purposes and should not be regarded as an offer, or a solicitation of an offer, to buy or sell a security, financial instrument or service discussed herein. Opinions in this communication constitute the current judgment of the author as of the date and time of this report and are subject to change without notice. Information herein is believed to be reliable but Wolfe Research and its affiliates, including but not limited to Wolfe Research Securities, makes no representation that it is complete or accurate. The information provided in this communication is not designed to replace a recipient's own decision-making processes for assessing a proposed transaction or investment involving a financial instrument discussed herein. Recipients are encouraged to seek financial advice from their financial advisor regarding the appropriateness of investing in a security or financial instrument referred to in this report and should understand that statements regarding the future performance of the financial instruments or the securities referenced herein may not be realized. Past performance is not indicative of future results. This report is not intended for distribution to, or use by, any person or entity in any location where such distribution or use would be contrary to applicable law, or which would subject Wolfe Research, LLC or any affiliate to any registration requirement within such location. For additional important disclosures, please see The views expressed in Wolfe Research, LLC research reports with regards to sectors and/or specific companies may from time to time be inconsistent with the views implied by inclusion of those sectors and companies in other Wolfe Research, LLC analysts research reports and modeling screens. Wolfe Research communicates with clients across a variety of mediums of the clients choosing including s, voice blasts and electronic publication to our proprietary website. Copyright Wolfe Research, LLC All rights reserved. All material presented in this document, unless specifically indicated otherwise, is under copyright to Wolfe Research, LLC. None of the material, nor its content, nor any copy of it, may be altered in any way, or transmitted to or distributed to any other party, without the prior express written permission of Wolfe Research, LLC. This report is limited for the sole use of clients of Wolfe Research. Authorized users have received an encryption decoder which legislates and monitors the access to Wolfe Research, LLC content. Any distribution of the content produced by Wolfe Research, LLC will violate the understanding of the terms of our relationship. 21