Long-Term Rewarded Equity Factors What Can Investors Learn from Academic Research? Felix Goltz

Transcription

1

2 Long-Term Rewarded Equity Factors What Can Investors Learn from Academic Research? Felix Goltz

3 Outline The venerable academic grounding Three Lessons from academic research What academic grounding does and does not mean 3

4 The venerable academic grounding Providers emphasize the academic grounding of their factor indices MSCI currently identifies six equity risk premia factors. They are grounded in academic research In developing the Russell High Efficiency Factor Index series we ensured that all of our factor specifications were consistent with academic research findings The FTSE Global Factor Index Series is. designed to represent factor characteristics for which there is a broad academic consensus ERI Scientific Beta: factor indices are meant to be investable proxies for rewarded factors that have been analysed in the academic literature This presentation: So what does academic research have to say on equity factors? 4

5 Lesson One: Be serious with data Academic research is better than a product backtest We struggle to estimate expected returns reliably (Merton 1980). This is also true for factor returns When testing whether a factor carries a positive premium, academic research conducts a thorough assessment Long term data Across different regions and asset classes Corrects for multiple testing These studies are open to criticism. Numerous papers are written to question previous empirical results see e.g the debate on the low volatility puzzle Factors which have undergone academic validation constitute a much stronger empirical justification than a mere product backtest. 5

6 Lesson One: Be serious with data Long Term Assessment Studies on US stocks span at least 40 years of data. For some factors we have data as far back as the 1920s Factor Factor Definition Period Premium t-stat Source Market Low Risk Size Value Momentum Profitability Investment Excess returns of cap-weighted equity index Stocks with low versus high risk (beta, volatility or idiosyncratic volatility Stocks with low versus high market cap Stocks with high versus low bookto-market Stocks with high vs. low returns over past 12 months (omitting last month) Stocks with high vs. low profitability (e.g. return on equity or gross profitability) Stocks low vs. high investment (change in total assets) % (annual) 0.70% (monthly) 2.28% (annual) 6.87% (annual) 9.34% (annual) 0.17% (monthly) 0.22% (monthly) Ang et al. (2009) Frazzini- Pedersen (2013) Ang et al. (2009) Ang et al. (2009) Ang et al. (2009) Fama-French (2014) Fama-French (2014) 6

7 Cumulative Returns (Log arithmic Scale) 1000 Lesson One: Be serious with data Long Term Assessment Cumulative Returns for US Equity Factors (Nov-1926 to Dec-2013) Market Factor (Mkt-RF) Size Factor (SMB) Value Factor (HML) Momentum Factor (UMD) 0,1 Factor returns from Kenneth R. French s data library The analysis is based on daily total returns from 03/11/1926 to 31/12/2013. The cumulative returns are scaled to logarithmic scale

8 Lesson One: Be serious with data Assessment across regions and asset classes (cont d) Value Momentum US Equities Int l Equities FCC Fama and French (2012) Basu (1977); Rosenberg, Reid, Lahnstein (1985); Fama and French (1993) Jegadeesh and Titman (1993); Carhart (1997) Rouwenhorst (1998) Empirical Evidence for Selected Factor Premia Low Risk Size Profitability Investment Ang, Hodrick, Xing, Zhang (2006); Frazzini and Pedersen (2014) Banz (1981); Fama and French (1993) Novy Marx (2013); Hou, Zhang, Xue (2014); Fama and French (2014) Cooper, Gulen, Schill (2008); Hou, Zhang, Xue (2014); Fama and French (2014) Ang, Hodrick, Xing, Zhang (2009); Frazzini, Pedersen (2014) Heston, Wessels, Rouwenhorst (1999); Fama and French (2012) Ammann, Odoni, Oesch (2012) Watanabe, Xu, Yao, Yu (2013) Asness, Moskowitz, Pedersen (2013) Asness, Moskowitz, Pedersen (2013) Frazzini and Pedersen (2014) N.A. N.A. N.A. 8

9 From Asness, Frazzini, Pedersen (2013) Lesson One: Be serious with data Example: Tests across regions (Value & Momentum) Factor premia are generally consistent across regions. Momentum in Japan is the exception.

10 From Asness, Frazzini, Pedersen (2013) Lesson One: Be serious with data Example: Tests across asset classes (Value & Momentum) Factor premia are generally consistent across asset classes. Momentum in Fixed income is the exception.

11 Lesson One: Be serious with data Adjusting for Multiple Testing Standard statistical tests only valid when we test a given hypotheses. But researchers may run several tests in practice to snoop the data. Multiple Testing: Simulate data for 100 variables that have zero mean. We would expect to find about five variables with mean significantly different from zero. Researchers have come up with tighter requirements for t- statistics to take into account the possibilities of multiple testing Applying these methods to standard equity risk factors, researchers find that many discovered factors remain significant. 11

12 Lesson One: Be serious with data Adjusting for multiple testing (continued) Harvey, Liu, Zhu (2013): Adjusted t-ratios for 5% significance level for factor premia MRT: Market Beta, EP: Earnings-to-price, SMB: small cap, HML: value, MOM: Momentum, DCG: durable consumption good expenditure, SRV, Short Run Volatility, LRV: long run volatility, LIQ: liquidity, IVOL: idiosyncratic volatility, DEF: default risk, CVOL:consumption volatility 12

13 Lesson One: Be serious with data Disappearing factors and newly arising factors Research often debates whether a factor has disappeared or a new factor has appeared. Measurement of a risk premium is highly sensitive to the chosen sample (Merton 1980). Given the uncertainty around estimates of factor premia, investors should be prudent before changing their set of factors Conclusions based on empirical evidence should only be drawn from studying very long time periods Economic arguments suggesting the disappearance or appearance of new factors should be investigated (see Lesson Two) 13

14 Lesson One: Be serious with data Disappearing factors and newly arising factors Consensual factors frequently disappear during.. particular time periods 20% 15% 10 Year Annualised Trailing Returns for US Equity Factors ( ) 10% 5% 0% -5% -10% Market Factor (Mkt-RF) Size Factor (SMB) Value Factor (HML) Momentum Factor (UMD) USA Equity Factor returns are from Kenneth R. French s data library The analysis is based on daily total returns from 03/11/1926 to 31/12/2013. Starting from year average annual returns are calculated by rolling each year forward till 2013.

15 t-statistic 1936 Lesson One: Be serious with data Disappearing factors and newly arising factors Consensual factors frequently disappear during particular time periods t-statistic of 10 Year Average Trailing Reurns of US Equity Factors ( ) Market Factor (Mkt-RF) Value Factor (HML) Size Factor (SMB) Momentum Factor (UMD) USA Equity Factor returns are from Kenneth R. French s data library The analysis is based on daily total returns from 03/11/1926 to 31/12/2013. Starting from year average annual returns are calculated by rolling each year forward till 2013 and the corresponding t-statistics are calculated.

16 Lesson Two: Being serious with data is not enough Looking for a compelling economic rationale We should place less weight on the data, the models are able to match, and instead closely scrutinize the theoretical plausibility and empirical evidence in favour or against their main economic mechanisms. [Kogan and Tian (2013)] Equity risk premium as an example: The equity risk premium can be statistically indistinguishable from zero even for relatively long sample periods Economic reasoning suggests that stocks have higher reward than bonds: Investors are reluctant to hold too much equity due to its risks. Investors due diligence should look at the. economic rationale for a risk premium 16

17 Lesson Two: Being serious with data is not enough Risk-based explanations for factor rewards CAPM Lessons of Rational Asset Pricing Theory A risk premium exists for the market factor in equilibrium Stock-specific risk (unrelated to market) is not rewarded Rewarded risk of an asset is measured by its beta with the market portfolio Assets paying off in bad times (low market returns) are attractive and thus require lower returns Multi Factor Models Risk premia exist for each factor in equilibrium (ICAPM) or with no arbitrage opportunities (APT) Stock-specific risk (unrelated to factors) is not rewarded Rewarded risk of an asset is measured by its exposures to multiple factors Assets paying off in bad times are attractive and thus require lower returns (see Ang 2013, pp ) 17

18 Lesson Two: Being serious with data is not enough Behavioural explanations for factor rewards Investors, analysts, fund managers make systematic errors Disposition effect: Are investors reluctant to realize their losses? Odean (1998) Overreaction and Underreaction to news: Hong and Stein (2002) Investor Sentiment affects decisions: Baker and Wurgler (2006) But arbitrage activity by rational agents removes mispricing Limits to arbitrage cause persistent mispricing: Shleifer and Vishny (1997) arbitrageurs cannot act (short selling and borrowing constraints) arbitrageurs cannot hold on and wait (e.g. funding liquidity constraints) 18

19 Lesson Two: Being serious with data is not enough Economic mechanisms behind main factors Risk-based explanation Behavioral Explanation Value Momentum Low Risk Size Profitability Investment Costly reversibility of assets in place: high sensitivity to economic shocks in bad times High expected growth firms are more sensitive to shocks to expected growth Liquidity-constrained investors have to sell leveraged positions in low risk assets in bad times when liquidity constraints become binding Low liquidity, high distress and downside risk is compensated by higher returns. Firms facing high cost of capital will invest only in the most profitable projects Low investment reflects firms limited scope for projects given high cost of capital Overreaction to bad news and extrapolation of the recent past leads to under-pricing Investor overconfidence and selfattribution bias leads to returns continuation in the short term Disagreement of investors about high risk stocks leads to overpricing due to short sales constraints Limited investor attention to smaller cap stocks Investors do not discern high and low profitability in growth firms Investors under-price low investment firms due to expectation errors 19

20 Lesson Three: Being serious isn t enough. Be practical Considering Transaction Costs Product providers often justify deviations from academic factors by implementation needs But while early studies indeed abstract from implementation issues, recent academic research addresses this shortcoming Do the premia to common equity risk factors survive net of transaction costs? Can we use mitigation strategies to ease implementation when harvesting these premia? Investors should be careful to not throw out the baby (academic grounding) with the bathwater (unrealistic assumptions on implementation issues) 20

21 Lesson Three: Being serious isn t enough. Be practical Impact of Transaction Costs Net-of-cost factor premia mostly remain significant (Monthly) Gross premium Turnover T-costs Net premium Avg. [t-stat] Avg. [t-stat] Size 0.33% [1.66] 1.23% 0.04% 0.28% [1.44] Profitability 0.40% [2.94] 1.96% 0.03% 0.51% [3.77] Value 0.47% [2.68] 2.91% 0.05% 0.42% [2.39] Investment 0.56% [4.44] 6.40% 0.10% 0.46% [3.60] Low Vol 0.63% [2.13] 24.59% 0.52% 0.11% [0.37] Momentum 1.33% [4.80] 34.52% 0.65% 0.68% [2.45] Extracted from Novy Marx and Velikov (2014). See their Table 3. All values are monthly. Factors based on cap-weighted decile portfolios. Portfolios are rebalanced annually for most factors but monthly for low idiosyncratic vol and momentum. Factors are return differences between two extreme decile portfolios (cap-weighted). Time period is July 1963 to December

22 Lesson Three: Being serious isn t enough. Be practical Mitigating Transaction Costs Using mitigation strategies : The case of momentum Gross premium Turn- T- Net premium over costs Mitigation strategy Avg. [t-stat] Avg. [t-stat] None 1.33% [4.80] 34.52% 0.65% 0.68% [2.45] Restrict trading to stocks in the lowest cost tertile Staggered rebalancing of 1/3rd of portfolio per month 1.44% [6.66] 38.17% 0.62% 0.82% [2.29] 1.25% [4.85] 16.66% 0.34% 0.91% [3.53] Use buffer rules 1.20% [4.71] 18.82% 0.35% 0.85% [3.35] Extracted from Novy Marx and Velikov (2014). See their Tables 6, 7 and 8. Monthly values. Factors are return differences between two extreme decile portfolios (cap-weighted). Time period is July 1963 to December

23 Lesson Three: Being serious is not enough. Be practical Considering Transaction Costs Frazzini, Israel and Moskowitz (2012): We measure the real-world transactions costs and price impact function and apply them to size, value, momentum, and short-term reversal strategies. [ ] Strategies designed to reduce transactions costs can increase net returns and capacity substantially, without incurring significant style drift. Results vary across styles, [ ] short-term reversals being the most constrained by trading costs. We conclude that the main anomalies are robust, implementable and sizeable. 23

24 Conclusion: What academic grounding does not mean Index Design Process Providers often conduct ad-hoc variable picking exercises without reference to academic research e.g. picking value variables (for Developed markets stocks during the period March 2000 to January 2013) "For each composite value index, factors are selected on the basis of the most significant t-stat values" We base the choice on the following considerations: Book to Price and Dividend Yield historically have a detrimental effect on the performance e.g. picking momentum variables (for Developed markets stocks during the period January 2001 to December 2013) Our preferred measure of momentum is the Residual Sharpe Ratio, which displays relatively high risk-adjusted performance outcomes, and relatively low levels of volatility 24

25 Conclusion: What academic grounding does not mean Mismatch with academic factor definitions: Examples Provider Value Momentum Quality Fama-French (2012, 2014) Goldman Sachs Equity Factor Index World Price to Book Value score from proprietary risk model (Axioma), relative to stock s regional industry group Past 12 Months return (omitting last month) Residuals from cross sectional regression of twelve month return (omitting last month) on stock volatility ROE (operating profits divided by book equity) Composite based on asset turnover, liquidity, ROA, operating CF to assets, accruals, gross margin, leverage MSCI Multi Factor Indices Sector-relative Composite based on Enterprise Value / Operating CF, Forward P/E, Price to Book Composite score based on excess return divided by ann. volatility over past 12 months and past six months Composite based on return on equity, standard deviation of earnings, debt-to-equity FTSE Global Factor Index Series Composite based on cash flow to price, net income to price, and countryrelative sales to price Mean/Std.dev. of avg. residual from 11 rolling window regressions of past 36 months returns on country and industry index Composite based on operating CF to debt, net income to assets, annual change in (sales over assets), accruals Deutsche Bank Equity Factor Indices Composite based on inverse of Enterprise Value to EBITDA and dividend yield Twelve month return (omitting last month) minus risk adjustment times idiosyncratic volatility score Composite based on Return on invested capital and net operating assets growth

26 Conclusion: What academic grounding means Minimum requirements for good practice Avoid mismatch with academic factors Refer to indicators for which academic research has provided thorough tests and economic explanation Refrain from proprietary tweaks Alternatively, when using novel or proprietary factors, check the items on the following list Thoroughly tested Tested in very long term data Tested across regions Tested across asset classes Tested for robustness to datamining Tested for robustness to transaction costs Linked to economic mechanisms 26 26

27 Conclusions The value of parsimony Parsimony is a prerequisite for robustness Explain a lot with a little Avoid picking up noise Be sceptical before adding new factors New factors may be explained by existing factors New factors may be statistical artefacts New factors may be based on storytelling rather than well-documented economic mechanisms 27 27

28 Conclusions Practical factor index design Factor indices do not need to deviate from indicators used to capture factors in academic research Improved weighting schemes (such as diversified multi-strategy) allow designing well diversified factor indices based on standard academic indicators and widely used index universes US Long Term (Dec-73 to Dec-13) US Broad CW The analysis is based on daily total return data from 31 December 1973 to 31 December 2013 (40 years). The benchmark used for relative analytics is based on the 500 largest market cap US stocks. Mid Cap, High Momentum, Low Volatility, and Value selections all represent 50% stocks of said characteristics in USA universe of 500 stocks. The risk free rate is the return of 3 months US Treasury Bill. Maximum relative drawdown is the maximum drawdown of the long-short index whose return is given by the fractional change in the ratio of strategy index to the benchmark index. Probability of outperformance is the probability of getting positive excess return returns if one invests in the strategy for a period of 1 (or 3) years at any point during the history of the strategy. Rolling window of length 1 (or 3) years and a step size of 1 week is used. Source: scientificbeta.com. Size Factor Momentum Factor Low Vol Factor Value Factor SciBeta SciBeta SciBeta SciBeta CW Mid Cap Mom. Div. Low Vol. Value Div. CW CW CW Div. Multi Multi Div. Multi Multi Strategy Strategy Strategy Strategy Ann Returns 10.95% 14.28% 15.67% 11.95% 14.57% 11.19% 13.90% 12.78% 15.70% Ann Volatility 17.38% 17.75% 16.69% 17.52% 16.26% 15.79% 14.34% 17.97% 16.51% Sharpe Ratio Max Drawdown 54.53% 60.13% 58.11% 48.91% 49.00% 50.50% 50.13% 61.20% 58.41% Ann Excess Returns 3.32% 4.72% 1.00% 3.62% 0.24% 2.95% 1.83% 4.75% Ann Tracking Error 5.90% 6.65% 3.50% 4.83% 4.46% 6.13% 4.69% 5.74% Information Ratio Outperf. prob (3Y) 69.62% 74.69% 78.99% 84.52% 52.54% 76.45% 67.34% 78.83% 28

29 Conclusions What can we learn from academic research? Perhaps it s time that product providers actually used academic research, rather than merely advertise academic grounding Investors should hold providers to high standards and conduct thorough due diligence Lesson 1: Be serious with data. Beware of datamining Lesson 2: Assess the economic rationale Lesson 3: Consider implementation issues 29