Finding Alpha in Ownership Data StarMine Smart Holdings Model Dirk Renick, David Sargent

Transcription

1 Finding Alpha in Ownership Data StarMine Smart Holdings Model Dirk Renick, David Sargent July 2011

2 AGENDA Background Model formulation Performance Trading Strategies Final Thoughts

3 Smart Holdings predicts what stocks are going to be bought by institutions and mutual funds - the smart money Avg IC Bottom Decile Top Decile Spread Max Drawdown Sharpe Ratio Global % 22.0% 23.5% -19.0% 1.70 North America % 15.3% 8.0% -42.2% 0.44 Developed Europe % 14.8% 21.2% -32.4% 1.33 Developed Asia % 24.3% 34.0% -13.2% 2.43 Emerging Markets % 24.3% 23.6% -15.4% 1.53 Japan % 4.5% 9.1% -17.1% 0.81 Monthly rebalance, , top 98.5% by market cap Our model is based on predicting which stocks are going to be bought by institutional managers by identifying the unique investment profile (value, growth, momentum, etc.) of each institutional investor, and then determining which stocks best fit each investor s unique profile 3

4 The typical methods used to follow the smart money using regulatory holdings data don t work out of the box. Traditional approaches and shortfalls Best Ideas (Cohen) Identify high conviction holdings as holdings that differ significantly from index weight or average portfolio weight Invest in high conviction names Works well until ~2002 and then goes dead Hedge Fund Holdings Buy names held by significant number of hedge funds Only works if you knew what they held at the end of the qtr, falls apart with 45 day lag in 13-F filings Buy what worked last month Worked for a while in the old days, then fell apart in more recent times of market turbulence Cohen, Randolph B., Polk, Christopher K. and Silli, Bernhard, Best Ideas (March 15, 2010). Available at SSRN: 4

5 Ownership-based models are profitable only to the extent that they can be predictive of institutional behavior. Holdings data is reported to SEC at a substantial lag to the quarter end (typically 45 days). Globally, reported holdings are lagged as well. Comparison of returns to a naïve ownership model assuming three horizons of data availability: Data available (45 days after quarter) Data at end of quarter Look-ahead predictive model Data Available Perfect Data Perfect Foresight Annualized Return -9.97% -0.62% 46.60% Sharpe Ratio Max Drawdown -69% -25% 0% 5

6 Quarterly Return Larger changes in institutional ownership lead to more buying pressure and higher prices. 8% Forward 1Q Return by Change in Institutional Ownership Decile 7.6% 6% 5.5% 4% 2% 0.4% 1.3% 1.9% 2.8% 3.5% 4.0% 0% -2% -4% -2.8% -0.9% Smallest (most negative) Changes in Institutional Ownership Largest Positive Changes in Institutional Ownership Quarterly rebalance, perfect foresight model Source: Thomson Reuters & StarMine research 6

7 By reverse-engineering fund manager decisions, we have built a predictive model of buys and sells. 15% Net Securities with Positive Change in Institutional Ownership 0.12 Rank Correlation with Institutional Ownership Change at Different Horizons 10% % 0% -5% q0 q1 q2 q3-10% Decile of Aggregate Screens Aggregate Screens 7

8 AGENDA Background Model formulation Performance Trading Strategies Final Thoughts

9 StarMine Smart Holdings Model Fund Preferences Peer Holdings Data Sets used: Thomson Reuters Ownership Feed (OP) Reuters Fundamentals Datastream pricing Screening Component Change Component StarMine SmartHoldings Model 9

10 Fundamental Factors Screening Component: Track Fund Preferences over time For each owner-factor pair, we not only determine if the owner is using the factor in their process, but the relative importance of each factor. Cells highlighted in blue indicate an owner-factor where the owner shows preference towards that factor. Each owner occupies a column, and each factor a row. This allows us to determine exactly what is important to each owner, and ultimately, what the market will be purchasing next. Thousands of Owners in the Thomson Reuters OP Datafeed Vanguard Growth Equity Fund (VGEQX) Top 5 Factors Basic PriceMo Long Term PriceMo StarMine ARM EPS Long Term Growth FY1 Revenue Growth Bottom 5 Factors F12M Dividend Yield Price / Intrinsic Value EV / Sales Relative Value F12M Earnings Yield 10

11 The 25 Factors used to generate fund profiles Value Growth Leverage Momentum Profitability Sentiment F12m E/P G1 EPS Debt / Book LT PriceMo ROE FY1 EPS Chg (30 day) T12m CF/P G1 REV Debt / Assets Basic PriceMo ROA FY2 EPS Chg (30 day) T12m B/P F5Y CAGR Int Coverage Volume (t30 day avg) Net Margin Recc Chg (30 day) T12m EBITDA/P T5Y EPS CAGR F12M EV / Sales Basic Accruals T12m Div/P LTG F12m Div/P 11

12 The most popular fundamental ratios (across all funds) change over time. Price momentum and growth factors dominate ROE, Earnings Quality and value factors dominate StarMine PriceMo EPS_CAGR3 ROE ROE ROE ROE LTG ROE Profit Margin Interest Coverage F12m E/P F12m E/P G5 EPS Profit Margin Interest Coverage F12m E/P Interest Coverage Interest Coverage Debt/Assets Debt/Assets LTG Profit Margin Profit Margin Profit Margin Interest Coverage LTG F12m E/P LTG StarMine EQ StarMine PriceMo in-sample years The shift from growth to value matches intuition around how investors approached these different market regimes. 12

13 Average Monthly Spread Screening Component: Peer Holdings We utilize analyst and industry peer information to determine if a fund is already concentrated in a given peer group We found that incorporating peer ownership into our model added value by determining which securities have high peer ownership levels at funds. 1.8% 1.6% 1.4% 1.2% 1.0% 0.8% 0.6% 0.4% 0.2% 0.0% Performance for Ownership model by peer quintile Peer Ownership Quintile Low peer ownership High peer ownership 13

14 Change Component track securities We monitor securities across all fundamental factors and also look at changes to capture stocks becoming more (or less) aligned with the Street Securities that are beginning to show up on manager s screens are of special interest. These names are potentially new targets for funds and are increasing the alignment of their fundamentals with the preferences of funds. Example screen result: 14

15 AGENDA Background Model formulation Performance Trading Strategies Final Thoughts

16 Each component produces good risk-adjusted returns with low drawdowns and adds to good performance of the overall global model. Annualized Return Annualized Sharpe Ratio Worst Drawdown IC Screening Component 25.62% % Change Component 8.1% % Final Combination 23.5% % Top 98.5% global securities 16

17 Global performance of Smart Holdings model 17

18 The model works well globally across different market cap ranges Global securities, top 98.5% of market cap, , no reranking within each cap tranche 18

19 Model correlations to existing StarMine factors are relatively low and quantcentration confirms different names coming into the portfolio. ARM ValMo RV IV EQ PriceMo Final Model Portfolio Quant -centration: What fraction of the top/bottom deciles selected by the SmartHolidngs model overlap with the top/bottom deciles of other StarMine factors? Top Decile Bottom Decile ARM 17.6% 23.7% ValMo 26.3% 34.7% RV 16.4% 32.3% IV 10.5% 21.5% EQ 21.5% 30.8% PriceMo 16.3% 20.6% 19

20 Smart Holdings turnover is driven by the change component, which we previously showed reduces the overall drawdowns. Bottom Decile (Short) Top Decile (Long ) Global 50.7% 55.8% Dev Asia 50.0% 57.3% Dev Europe 55.5% 63.1% North Amer. 61.6% 75.0% Emerg. Mkts 48.4% 53.5% Japan 54.9% 64.5% Monthly turnover for Smart Holdings is about the same as an analyst revision or price mo model. Turnover can be reduced by de-emphasizing the change component. 20

21 AGENDA Background Model formulation Performance Trading Strategies Final Thoughts

22 AA institutional ownership went from 81% in Aug 2008 to 70% in July 2009 while both Smart Holdings scores decreased measurably. Smart Holdings Screening Component Rank Smart Holdings Change Component Rank 22

23 The top decile of Smart Holdings global rank outperforms the MSCI world index which shows efficacy in a long-only portfolio. Annualized Return Sharpe Ratio Max Drawdown Smart Holdings 14.50% % MSCI World 2.07% % Smart Holdings > 90 and Market Cap > $1B USD Monthly rebalance, zero transaction costs 23

24 Combining Smart Holdings with the StarMine Analyst Revision Model improves the longonly performance of the ARM model. Return Sharpe Ratio Max Drawdown ARM > 80 17% % ARM > 80 & SH > 50 20% % Improvement 17% 35% 6% Global securities, top 98.5% of market cap by region 24

25 ARM Quintile ARM Quintile ARM Quintile ARM Quintile The combination of Smart Holdings and StarMine s Analyst Revision Model generates superior returns across all regions. Global Portfolios: Average Monthly Return Japan: Average Monthly Return Smart Holdings Quintile Smart Holdings Quintile % 0.28% 0.34% 0.51% 0.68% 1-0.5% 0.0% 0.0% 0.0% -0.1% % 0.59% 0.75% 0.77% 1.02% 2 0.0% -0.1% 0.2% 0.3% 0.2% % 0.67% 0.88% 0.98% 1.13% 3 0.1% 0.5% 0.4% 0.2% 0.4% % 0.90% 0.93% 1.06% 1.41% 4 0.4% 0.6% 0.5% 0.3% 0.5% % 1.29% 1.50% 1.58% 2.03% 5 0.7% 0.8% 0.8% 1.0% 0.8% US-only Portfolios: Average Monthly Return Smart Holdings Quintile % 0.4% 0.8% 0.8% 1.1% 2 0.6% 0.9% 1.0% 0.9% 1.1% 3 0.5% 0.8% 1.0% 1.0% 1.0% 4 0.6% 0.7% 0.9% 1.0% 1.1% 5 0.7% 1.1% 1.3% 1.3% 1.5% Dev Asia: Average Monthly Return Smart Holdings Quintile % 0.5% 0.3% 0.8% 0.9% 2 0.7% 0.7% 0.8% 1.0% 1.5% 3 0.7% 0.9% 0.8% 1.2% 1.4% 4 0.8% 1.3% 1.2% 1.4% 2.0% 5 1.2% 2.4% 2.3% 2.2% 2.8% We see the large differences between the alignment buckets: when both analyst sentiment and buyside sentiment is positive (negative) 25

26 A simple linear combination of Smart Holdings and StarMine s Analyst Revision model improves long-short spread returns and IC s. All weight on Smart Holdings All weight on ARM 26

27 Using Smart Holdings peer concentration metric to identify overbought securities improves performance of StarMine s Intrinsic value model Remove securities that are not widely held but whose peers are widely held according to the Smart Holdings peer concentration metric (implies that the security is the un-popular stock amongst a popular peer group) to uncover securities that are relatively unknown. 27

28 Removing low scoring Smart Holdings securities from a universe of stocks removes higher volatility names not of interest to the buyside and results in improved Sharpe Ratios and reduced drawdowns in StarMine s multifactor Val-Mo model 28

29 AGENDA Background Model formulation Performance Trading Strategies Final Thoughts

30 StarMine Smart Holdings uses Ownership data to follow the Smart Money Some Smart Holdings lessons (and debunking): Don t chase factor returns Lags in reporting kill performance of buying backward-looking holdings Don t buy names others are buying, buy the factors Leverages a relatively under-utilised data set No good academic references on this approach, but our white paper is a good start. 30

31 StarMine Smart Holdings is available today White paper summarizing model formulation Historical daily data back to 2000 for testing Daily ftp datafeed us: 31

32 QUESTIONS? 32