Investing in Global Equity Markets with particular Emphasis on Chinese Stocks John B. Guerard, Jr., McKinley Capital Management, LLC Anchorage, AK 99503 JGuerard@McKinleyCapital.com May 20, 2016 1
Based on joint research with Professor Shijie Deng of the Quantitative and Computational Finance (QCF Program) at the Georgia Institute of Technology, Harry Markowitz and Ganlin Xu of the McKinley Capital Management (MCM) Scientific Advisory Board, and Rob Gillam, CIO, and Ziwei (Elaine) Wang, Quantitative Analysis, of MCM. 2
Research Conclusions: 1. Models Produce Statistically Significant Active Returns in Global, Non-US, and EM Markets using MVM59, MVTaR, and EAW Optimization Techniques! 2. The Public Form of Forecasted Earnings Acceleration, E’, CTEF, Produces Statistically Significant Asset Selection (Stock Selection) in Global, Non-US, R3, EM, and JP using the Three Methods of Markowitz Optimizations! 3. Models Pass Markowitz-Xu Data Mining Corrections Tests in all Markets except China A Shares, where the time frame is too Short! 3
Questions to be Answered 1. How is this analysis consistent with previous work in the literature? 2. What is the role of forecasted earnings in creating expected returns? 3. Can the models be implemented in the world of business? 4
Earnings Forecasting, Expected Returns, and Data Mining Early Research Includes § Cragg and Malkiel (JF 1968) § Elton, Gruber, and Gultekin (MS 1981) § Wheeler (1991) § Brown (IJF 1993) § Bloch, Guerard, Markowitz, Todd, and Xu (JWE 1993) § Markowitz and Xu (1994) § Blin, Bender, and Guerard ( IJF 1998) § Ramnath, Rock, and Shane (IJF 2008) § Guerard, Rachev, and Shao (IBMJoR&D, 2013) § Deng and Min(JOI, 2013) § Guerard, Markowitz, and Xu (IJF, 2015) 5
Portfolio Construction and Modeling Process Expected Returns (MQ, GLER, USER) Risk-Return Efficient Frontier (Varying Lambda, Risk Models (Axioma, Targeted Tracking Error) APT) Statistical versus Markowitz Portfolio Fundamental Risk Optimization Models Data Mining Corrections Constraints:Turnover; Test of Portfolios Stock Weights: Equal Active Weights (EAW) versus Stock Mean- Variance (MV) Weights Efficient Return Frontier Attribution Analysis (Specific Returns, Factor Exposures and Returns) Risk
Bloch et al. (1993) Stock Selection Model TR t+1 = a 0 + a 1 EP t + a 2 BP t + a 3 CP t + a 4 SP t + a 5 REP t + a 6 RBP t + a 7 RCP t +a 8 RSP t + e t (1) where: EP = [earnings per share]/[price per share] = earnings-price ratio; BP = [book value per share]/[price per share] = book-price ratio; CP = [cash flow per share]/[price per share] = cash flow-price ratio; SP = [net sales per share]/[price per share] = sales-price ratio; REP = [current EP ratio]/[average EP ratio over the past five years]; RBP = [current BP ratio]/[average BP ratio over the past five years]; RCP = [current CP ratio]/[average CP ratio over the past five years]; RSP = [current SP ratio]/[average SP ratio over the past five years]; and e = randomly distributed error term. 7
Public Form of Stock Selection Model TR t+1 = a 0 + a 1 EP t + a 2 BP t + a 3 CP t + a 4 SP t + a 5 REP t + a 6 RBP t + a 7 RCP t +a 8 RSP t + a 9 CTEF t + a 10 PM t + e t (2) where: EP = [earnings per share]/[price per share] = earnings-price ratio; BP = [book value per share]/[price per share] = book-price ratio; CP = [cash flow per share]/[price per share] = cash flow-price ratio; SP = [net sales per share]/[price per share] = sales-price ratio; REP = [current EP ratio]/[average EP ratio over the past five years]; RBP = [current BP ratio]/[average BP ratio over the past five years]; RCP = [current CP ratio]/[average CP ratio over the past five years]; RSP = [current SP ratio]/[average SP ratio over the past five years]; CTEF = consensus earnings-per-share I/B/E/S forecast, revisions and breadth, PM = Price Momentum; and e = randomly distributed error term. 8
Regression Issues and Analysis 1. Financial data has Outlier issues and we use Robust Regression to estimate Expected Returns, using the Beaton-Tukey (1974) Bisquare Criteria. Ongoing research finds that the MM-Methods of Robust Regression using the Tukey Optimal Influence Function (1999) offer enhancements. 2. Multicollinearity exists in Financial data and we estimate total stock returns models using the Gunst et al. (1974,1976) Latent Root Regression (LRR) procedure on Robust-Weighted data, hence WLRR. 9
Research in “Threes” Research in “Threes” Three Levels of Testing; Three Levels of Testing; Three Methods of Markowitz Optimizations; Three Methods of Markowitz Optimizations; Three Testing Universes; Three Testing Universes; Three Research Conclusions. Three Research Conclusions. 10
Levels of Testing Level 1. Information Coefficients, ICs; Level 2. Markowitz Efficient Frontiers with Transactions Costs; Level 3. Markowitz-Xu Data Mining Corrections testing 11
Markowitz Optimization Techniques § 1. Mean – Variance Model using Total Risk (MVM59); § 2. Mean – Variance Tracking Error at Risk (MVTaR); § 3. Equal – Active Weighting (EAW); § The Goal: Maximize the Geometric Mean (Latane, 1959; Markowitz, 1959 and 1976; and MacLean, Thorp, and Ziemba, 2011) and Sharpe Ratio. 12
We present evidence on three Modeling Universes: 1. In Guerard, Rachev, and Shao (2013) and Guerard, Markowitz, and Xu (2015), we used a Global Broad Universe, defined as all Companies on FactSet with Sales and Net Income, Two Analysts on I/B/E/S Database, Top 7500 stocks in terms of $USD, 1982-2011. 2. MSCI Index Constituents with FactSet Net Income and Sales Data and I/B/E/S coverage, 1/2003 – 5/2015. 3. Global Stocks with FactSet Net Income and Sales Data and I/B/E/S coverage, 1/2003 – 12/2015. China A Shares Stocks, 1/2009 – 12/2015. 13
Universe I: 1. In Guerard, Rachev, and Shao (2013) and Guerard, Markowitz, and Xu (2015), we used a Global Broad Universe, defined as all Companies on FactSet with Sales and Net Income, Two Analysts on I/B/E/S Database, Top 7500 stocks in terms of $USD, 1982-2011. 14
APT Optimization Techniques Test: Guerard, Markowitz, and Xu (2015) Efficient Frontier of the Global Stock Selection Model with Various Portfolio Optimization Techniques 1999 -2011 APT Risk Model Earnings Model or Mean-Variance Annualized Standard Sharpe Information Tracking Component Methodology Lambda Return Deviation Ratio Ratio Error GLER M59 1000 15.84 24.97 0.590 0.78 13.11 500 16.34 24.85 0.590 0.82 12.08 200 16.37 24.38 0.610 0.85 12.68 100 15.90 24.61 0.580 0.81 12.66 5 10.11 19.36 0.440 0.51 8.81 Benchmark 5.59 0.240 GLER TaR 1000 16.10 21.93 0.660 0.94 11.18 500 15.91 21.99 0.651 0.90 11.44 200 16.09 20.95 0.691 0.97 10.83 100 14.18 21.24 0.591 0.77 11.23 5 8.51 20.03 0.344 0.33 8.75 GLER EAWTaR2 1000 14.80 21.96 0.600 0.94 11.07 500 14.30 21.65 0.590 0.80 10.87 200 14.15 20.92 0.600 0.85 10.04 100 13.49 20.82 0.570 0.80 9.84 5 10.77 20.79 0.440 0.43 12.18 15
Axioma Attribution: WLRR Model in Guerard, Markowitz, and Xu (2015) Attribution of FSGLER APT-Created Portfolios using Axioma World Fundamental Risk Model Source of Return Contribution Avg Exposure Hit Rate Risk IR T-Stat Portfolio 14.52% 21.25% Benchmark 1.51% 20.38% Active 13.01% 10.81% 1.20 4.34 Factor Contribution 7.87% 8.28% 0.95 3.43 Style 4.44% 7.47% 0.59 2.14 Exchange Rate Sensitivity -0.07% 0.0281 51.28% 0.25% -0.27 -0.98 Growth 0.33% 0.1589 64.74% 0.25% 1.30 4.68 Leverage -0.59% 0.2732 41.67% 0.36% -1.63 -5.88 Liquidity 0.30% 0.1223 51.92% 0.81% 0.37 1.34 Medium-Term Momentum 5.14% 0.4534 72.44% 2.29% 2.25 8.10 Short-Term Momentum 0.82% 0.0371 44.23% 1.33% 0.61 2.20 Size 0.69% -1.0072 53.85% 6.28% 0.11 0.39 Value 2.67% 0.5142 66.03% 1.36% 1.96 7.05 Volatility -4.85% 0.5467 36.54% 4.58% -1.06 -3.82 Country 2.27% 2.59% 0.88 3.16 Industry 0.49% 2.38% 0.21 0.74 Currency 0.62% 1.35% 0.46 1.67 Local 0.08% 0.31% 0.24 0.88 Market -0.02% 2.23% -0.01 -0.03 Specific Return 5.13% 6.69% 0.74 2.66 16
Level II Test: Axioma Efficient Frontiers Test in Guerard, Markowitz, and Xu (2015) Table 4: Axioma FSGLER Efficient Frontiers 1999 - 2011 Axioma Fundamental Risk Model Tracking Annualized Standard Active Active Sharpe Information Number of Errors Return Deviation Return Risk Ratio Ratio Stocks 3 4.87 18.78 3.59 3.50 0.259 1.027 309 4 6.22 19.40 4.94 4.79 0.326 1.031 254 5 7.90 20.22 6.62 6.11 0.391 1.083 227 6 7.90 21.20 6.62 7.24 0.373 0.913 211 7 9.09 22.10 7.81 8.38 0.411 0.932 195 8 8.54 23.05 7.26 9.42 0.371 0.771 226 9 10.45 23.30 9.17 10.06 0.449 0.911 238 10 11.62 24.18 10.35 11.05 0.481 0.936 229 Axioma Statistical Risk Model 3 8.79 20.63 7.51 6.00 0.426 1.253 411 4 9.86 21.79 8.58 7.73 0.453 1.110 323 5 11.92 22.51 10.64 8.95 0.530 1.189 275 6 13.00 23.20 11.72 9.95 0.561 1.178 247 7 12.03 23.83 10.75 10.95 0.505 0.983 232 8 12.35 24.93 11.27 12.09 0.504 0.932 225 9 12.71 25.48 11.43 12.93 0.499 0.884 222 10 12.68 26.00 11.40 13.47 0.488 0.846 227 17
Universe II: MSCI Index Constituents and Broad Global Testing § Analysis is 12/2012 – 5/2015; § MSCI Index Constituents with FactSet Net Income and Sales Data and I/B/E/S coverage. 18
Recommend
More recommend
