International Finance. Investment Styles. Campbell R. Harvey. Duke University, NBER and Investment Strategy Advisor, Man Group, plc.

Transcription

1 International Finance Investment Styles Campbell R. Harvey Duke University, NBER and Investment Strategy Advisor, Man Group, plc February 12, 2017

2 2

3 1. Passive Follow the advice of the CAPM Most influential paper in the history of finance The implication is to hold the market portfolio (i.e., market cap weighting) There are issues as to what the market portfolio is Anything that deviates from market portfolio is often called active 3

4 2. Active Investment Three sources of active returns Misvaluation of assets Premiums for taking various types of risk Luck 4

5 2. Active Investment A positive average active return should not be necessarily be confused with alpha (or risk adjusted outperformance) Premiums for taking various types of risk are generally not alpha ; they are just rewards for taking on risk Misvaluation of assets, if successfully identified and then harvested, is alpha 5

6 2. Active Investment Time diversification is an important qualification Various market segments may have differing tolerances for risk which depend on time horizon Illiquidity is a great example. An asset could be priced because the majority of the market has low tolerance for illiquidity risk whereas the another segment of the market, such as long term institutional investors, has more tolerance for this type of risk 6

7 2. Active Investment Important to separate the factor premiums into those due to risk exposure, time diversification and misvaluation For example, the shortvol is simply a premium for taking on negative skew risk (no alpha) Illiquidity (possible alpha due to time horizons) Value (largely misvaluation, positive alpha) 7

8 2. Active Investment Misvaluation opportunities decreasing through time Information more readily available (and lower cost) Computing costs dramatically lower Decrease in noise traders (see Stambaugh, 2014) 8

9 2. Active Investment Stambaugh (2014) 9

10 2. Active Investment Stambaugh (2014) 10

11 2. Active Investment While share of alpha opportunities decreasing, it is still economically very important. It is also unlikely that alpha will disappear in the near future though some markets are more efficient than others. 11

12 2. Active Investment Traffic dynamics:* Consider a busy highway. Each lane moves approximately the same speed because lane switchers ensure a relatively even number of cars in each lane However, they don t move exactly the same speed because of the cost of switching lanes The relative speed of the lanes reaches an equilibrium level where there is not much benefit to switching. However, switching still makes sense for those with a comparative advantage in lane switching (though switching and higher speeds increase risk). However, machine learning/self driving cars may defeat the active benefit *Traffic example from Lasse Pedersen s Efficiently Inefficient. Also see: 12

13 2. Active Investment: Active vs. Passive Passive investing assumes: No one has any particular skill to identify mispricing or There is no mispricing I strongly believe these assumptions are false How do we deal with the fact that there is not much difference in performance between passive and active strategy returns? 13

14 2. Active Investment: Active vs. Passive It is no surprise to me that active managers, on average, don t outperform The key words are average There are a lot of unskilled active managers. However, there is a skilled group of managers 14

15 3. Active Styles Main styles Value: Invest in securities or markets that have low price to book ratios Momentum: Invest in securities that have done well Carry: Long high yielding assets short low yielding assets Size: Invest in smaller market capitalization securities Reversal: Invest in securities that have done poorly 15

16 3. Active Styles Many other styles Example 1: Profitability factor Example 2: Low volatility/beta A portfolio of low volatility stocks has been documented to outperform high volatility 16

17 3. Active Styles Explanation of low vol investing: Leverage constrained hypothesis: Investors have trouble getting leverage, as a result they will bid up the price of high beta stocks, lowering the expected returns Behavioral hypothesis: Investors flock to high volatility stocks and treat them as lottery bets Skew 17

18 3. Active Styles: Systematic and Discretionary Systematic means a machine makes the trading decisions e.g., Man AHL Discretionary means individual managers make the trading decisions e.g., Man GLG Note: Both approaches (at least at Man) are quantitative. That is, there is a lot of quantitative analysis that underlies the Man GLG investment process. However, in the end, the manager trades. 18

19 3. Active Styles: Systematic and Discretionary Interestingly, not much difference in performance: 19

20 3. Active Styles: Systematic and Discretionary Interestingly, not much difference in performance: 20

21 3. Active Styles: Systematic and Discretionary There are differences in the types of risks that are taken Preliminary results in Harvey, Rattray, Sinclair, and Van Hemert (2017) suggest that discretionary funds, on average, take more risk 21

22 3. Active Styles: Systematic and Discretionary However, alpha and risk analysis has challenges: What are the relevant risk factors? How do we model how these risk factors change through time 22

23 3. Active Styles: Time series vs. Cross sectional momentum Time series For each market, we buy when prices going up and sell when prices go down This is similar to replicating a long call option and a long put option Long straddle (positive convexity) 23

24 3. Active Styles: Time series vs. Cross sectional momentum Time series There are many subtle aspects. For example, sophisticated trend followers employ a response function which reduces the buying if the price goes up too much (or reduces the selling if prices plunge too far) Like any long option position, you benefit from big moves in volatility (sometimes called long volatility or long gamma) In crisis periods, we do well as evidenced by trend following performance during the global financial crisis 24

25 3. Active Styles: Time series vs. Cross sectional momentum Time series Key insight. With time series momentum, you might be long in many, many markets at the same time. Hence, your profitability will depend on what happens in the market. 25

26 3. Active Styles: Time series vs. Cross sectional momentum Cross sectional momentum Example of a U.S. cross sectional program. Sort all stocks based on past 12 month returns (time series momentum applied to individual equities) Form a portfolio buying the best performing stocks and selling the worst performing stocks 26

27 3. Active Styles: Time series vs. Cross sectional momentum Cross sectional Notice this is a hedge portfolio. Half the portfolio is always long and the other half is always short. If constructed to be market neutral, this portfolio is immune to movements in the overall market. Key insight: Cross sectional momentum is not dependent on overall market moves. 27

28 3. Active Styles: Time series vs. Cross sectional momentum Which one too choose? Both should be in the portfolio but there are reasons to believe that they will perform differently in different market environments. For example, suppose the overall market is trending. You definitely want to be in time series momentum because you will profit from the trend (in cross sectional, the market is hedged out) 28

29 3. Active Styles: Time series vs. Cross sectional momentum Which one too choose? If the market is flat, then there might be more profit opportunities by capitalizing on trends in individual stocks rather than the overall market. 29

30 3. Active Styles: Time series vs. Cross sectional momentum Which one too choose? Time series has the important property of positive convexity It provides insurance Like any insurance policy, some times you pay the premium without the insurance payoff Hence, it is expected that some years time series momentum will generate small negative returns. 30

31 3. Active Styles: Time series vs. Cross sectional momentum Which one too choose? Cross sectional momentum does not have the positive convexity property Cross sectional momentum is not exposed when there are large market draw downs (if market neutrality is enforced) Notice that cross sectional momentum is simply one of many factors that needs to be considered in a cash equity strategy 31

32 3. Active Styles: Measuring Risk in TS Momentum Investors care about alpha which is the performance over and above the risk exposure So if a long only fund moved, on average, 1 for 1 with the market as a whole (beta=1) and the fund produced a 15% excess return when the market delivered 12%, the alpha is 3%. However, time series momentum is not a long only style there are many longs and shorts and it evolves through time. 32

33 3. Active Styles: Measuring Risk in TS Momentum TS momentum has a type of risk that the long only equity fund does not necessarily have: Convexity risk Investors do not like negative convexity (an asset that can crash) and for negative convexity assets, prices are bid down (expected returns are high) However, TS momentum induces positive convexity not negative. 33

34 3. Active Styles: Measuring Risk in TS Momentum As a result of having this anti risk like performance profile, the alpha of TS momentum is almost always greater than its reported performance. Indeed, in the years of small negative returns that are realized, the alpha may be positive. What does this mean? 34

35 3. Active Styles: Measuring Risk in TS Momentum What does this mean? To get the type of protection that TS momentum is positioned to deliver (for example outright buying puts and calls), it would be expensive for the investor. Suppose TS momentum loses 3% in one year. However, the portfolio with the same type of protection, loses 8%. As a result, the TS momentum alpha is +5%. This is a very important insight that many investors don t understand. 35

36 4. Allocation Traditional method is mean variance optimization. However, problems: Assumes exact knowledge of means, variances and covariances (i.e., no uncertainty in any of the estimates) As a result, weights extremely sensitive to small changes in inputs Challenge when asset returns are non normal Ignores higher moments like skew Generally not used in raw form in asset management 36

37 4. Allocation Solutions: Deal with Estimation Error Resample or bootstrap the mean variance frontier (see my note on resampling). This explicitly allows for uncertainty in the means, variances and covariances Alternative is a Bayesian approach detailed in Harvey, Liechty, Liechty and Muller (2010) 37

38 4. Allocation Solutions: Minimum variance Use the mean variance framework but solve for the minimum variance portfolio (which does not depend on the means) 38

39 4. Allocation Solutions: 1/N or Risk Parity Abandon mean variance. Simplest strategy is just invest equally to each asset (1/N) Variant is risk parity where you assign equal risk allocations based on the volatility of the assets (e.g. suppose 2 assets, one with 50% volatility and the other with 25% volatility, you would invest half as much in the high volatility asset) 39

40 4. Allocation Solutions: 1/N or Risk Parity There are many variants of risk parity Minimum Variance, Maximum Diversification, Risk Efficient, Risk Cluster, Maximum Entropy (PCA), etc. 40

41 4. Allocation Solutions: 1/N or Risk Parity All depend on an estimate of the variance and we know that volatility changes through time All have some type of rebalancing which could lead to skew 41

42 4. Allocation Solutions: 1/N or Risk Parity Case study 1. One asset S&P

43 4. Allocation 43

44 4. Allocation 44

45 4. Allocation Solutions: 1/N or Risk Parity Case study 2. S&P 500 and Treasury bond 45

46 4. Allocation 46

47 4. Allocation 47

48 4. Allocation Many other long only and long short styles Global macro Event driven Directional Relative value 48

49 5. Evaluation Beware Sharpes could be misleading Peso problems Good and bad performance could be the result of good and bad luck 49

50 6. Summary Here is the way I look at it. Strategy has a true risk premium that is observable historically and maintains in the future (e.g., stocks have higher return that T bills, illiquidity rewarded, negative tail risk rewarded). Strategy has a true risk premium that is unobservable in the past that will be realized in the future Strategy has a true risk premium which may be observable in the past or may not but so many investors pile into the strategy that the crowded premium goes to zero or negative. Strategy produces significant excess returns in the past which is the result of a mispricing. This is not a risk premium but a mispricing. The excess return can repeat in the future but only until the mispricing goes away. Strategy produces significant excess returns in the past but it is a fluke (not mispricing, not a risk premium). This may be repeated in the future temporarily due to: more good luck or investors creating momentum. 50