Risk Management beyond Asset Class Diversification Sébastien Page, CFA Executive Executive Vice President PIMCO Newport Beach, CA Asset allocation is evolving into an approach based on forecasts forecasts driven by macroeconomics macroeconomics and risk factor diversification. The dynamic nature of markets requires both secular and cyclical investment horizons. In addition, investors should look beyond volatility as a measure of risk and explicitly estimate the risk of tail events.
T
hroughout my presentation, I will discuss four ways in which asset allocation is changing. These concepts are not new, but they are important and evolving trends that affect every aspect of asset allocation. First, investors have historically relied heavily on data and models to arrive at their optimal asset allocation. Increasingly, investors are recognizing the importance of formulating a forward-looking view. For example, historical asset class returns reflect the declining interest rate environment that has prevailed over the past 20–30 years. But going forward, the likelihood that interest rates will rise over the secular horizon should directly influence investors’ capital market expectations. Another example is the debt issued by emerging market countries: It used to be much riskier than developed market debt. Today, emerging market debt in countries with relatively clean balance sheets is often considered less risky than that of certain developed markets, such as the peripheral countries in Europe. Current conditions matter, and too often, datadriven models ignore the current state of the world. Data and models are extremely useful, but only to the extent that they help create a forward-looking view. Second, investors have traditionally diversified portfolios among asset classes. But investors are now realizing that asset classes are just “containers” for underlying risk factors. Therefore, diversifying among risk factors directly may be more efficient than diversifying across asset classes. Think of the macronutrients that food can be broken down into: protein, fat, and carbohydrate. Similarly, corporate bonds, venture capital, private equity, equity, hedge funds, This presentation comes from the Global Investment Risk Symposium held in Washington, DC, on 7–8 March 2013 in partnership with CFA Society Washington, DC.
52 • September 2013
and real estate can all be expressed as combinations of financial “macronutrients”: equity risk, interest rate risk, spread risk, and others. Third, asset allocation has historically been a static process for many institutional investors. They would specify a fixed time horizon of typically three or five years, formulate a strategic asset allocation, and then leave it alone until the end of the fixed time period. Increasingly, especially in the wake of the global financial crisis, investors are realizing that asset allocation must be a dynamic process. The passage of time—for example, the number of times the Earth goes around the sun— has very little to do with how abruptly the financial markets can experience regime shifts. Valuations— and thereby, long-term expectations about capital markets—can change significantly along a three- or five-year horizon (consider the changes in equity valuations in late 2008). As a solution, investors with rigid investment governance and no clear mandate to follow a dynamic process can outsource the process to external, multiasset mandates, such as a global tactical asset allocation strategy. Fourth, the investment industry has been using volatility, measured as the standard deviation of returns, as a convenient risk measure to compare investment choices. Investors like to use the Sharpe ratio, which is an investment’s excess return over cash divided by its volatility, volatility, to determine the optimal asset mix. But it is becoming evident that minimizing exposure to large losses, or “tail risk,” is what really matters. After all, large and permanent losses are ultimately far more relevant to the investor than transitory swings in valuations. The standard deviation of returns also fails to differentiate between downside and upside upsid e exposure e xposure to loss in asymmetrical probability distributions. As a result, there has been a gradual shift away from volatility ©2013 CFA CFA Institute • cfapubs.org
Risk Management beyond Asset Class Diversification
as the defining measure of risk toward different measures of tail risk. Andrew Lo provides a great example of why volatility as a risk measure can be misleading in “Risk Management for Hedge Funds: Introduction and Overview.”1 Lo created a hypothetical investment strategy that doubled the Sharpe ratio on the S&P 500 Index. The strategy used monthly data from 1992 to 1999 and did not require any skill or knowledge of the future. The simulation was simply to sell deeply out-of-themoney put options—essentially, sell insurance and collect cash premiums. The strategy reduced volatility as measured by standard deviation but significantly increased tail risk. This case is an extreme example of how the Sharpe ratio can be misleading. Because it uses volatility as the measure of risk, strategies characterized by potentially large but infrequent losses will often look unreasonably favorable in a Sharpe ratio context. This strategy is also an illustration of my first point: Tail risks might not be reflected in a historical data sample but can be present in a forward-looking perspective based on a fundamental understanding of the investment.
Diversifying Using Risk Factors One of the most important ways in which asset allocation is adapting to the current state of the world is the increased use of risk factor diversification. In the stock market, the key risk factor is broad equity beta. One example of an approach to risk factors for asset allocation is to think in terms of broad equity beta rather than in terms of domestic versus global or small-cap versus large-cap equity asset classes. Then, the approach defines risk factors on an incremental basis that are net of the broad equity beta, such as industry, country, value, momentum, size, and risk tilts. In bonds, the equivalent of the broad equity beta is duration, defined as sensitivity to interest rates. The duration risk factor drives most of the volatility among different bond asset classes. Bond market investors will also look at credit spread duration, which is the sensitivity of the portfolio to changes in the level of credit spreads. The corporate bond asset class provides a simple illustration of the risk factor approach to investing. In risk factor space, the corporate bond asset class becomes at least two key risk factors: (1) interest rate or duration exposure and (2) credit spread exposure. These two risk factors move in different 1Andrew
W. Lo, “Risk Management for Hedge Funds: Introduction and Overview,” Financial Analysts Journal, vol. 57, no. 6 (November/December 2001):16–33.
©2013 CFA Institute • cfapubs.org
directions during macroeconomic shocks, with spreads generally tightening when interest rates rise and vice versa. As such, credit spread exposure may act as a diversifier to a Treasury-centric bond portfolio. Also important to bond investors are exposures to the slope of the yield curve. Duration measures the sensitivity of a portfolio to a parallel shift in interest rates. But changes in the slope of the yield curve can be construed as a separate risk factor linked to the macroeconomy. For example, when upside economic growth surprises occur, the yield curve typically steepens because market forces push the long end of the curve up whereas central banks, which generally control the short end of the yield curve, are often slow to react. Private and illiquid asset classes (i.e., alternative assets), such as private equity, real estate, and hedge funds, often have significant exposure to the same risk factors that drive stock and bond volatility. Returns on alternative assets depend on changes in interest rates, as well as the way in which investors value risky cash flows, as reflected in equity market valuations and credit spreads. In addition, liquidity and some specialized factors can play a role.2 The lack of mark-to-market data often lures investors into the misconception that these asset classes represent something of a free lunch. Their relatively high returns appear to come with low risk and significant diversification to other asset classes in normal times. This misconception arises because return indices for privately held assets often are artificially smoothed, which is a result of the lack of a true mark-to-market mechanism in these asset classes and which biases both volatility and correlation with traditional asset classes downward. To address this problem, risk models for private asset classes should rely on public market proxies and should adjust for reporting biases in the series of illiquid return indices. Models that use transformed risk factor returns to account for the lag structure of the index can be particularly useful. Overall, keep in mind that the models used to estimate risk factor exposures for alternative assets are not as precise in their explanation of volatility as the models used for stocks and bonds. Often, the models for alternative assets include a significant amount of idiosyncratic risk (nonfactor-based risk). 2See
“Asset Allocation: Risk Models for Alternative Investments,” PIMCO Analytics, Quantitative Research (May 2013): www.pimco.com.
September 2013
• 53
CFA Institute Conference Proceedings Quarterly Table 1 shows risk factor versus asset class correlations based on monthly data from 31 March 1997 through 31 May 2011.3 I calculated the average cross-correlation for the risk factors I have described and compared them with the average cross-correlation for a typical list of asset classes used by an institutional investor. In addition, I conditioned my estimate of correlation between quiet and stressed markets. There are two insights to gain from this table. First, the average cross-correlation for risk factors is much lower than that for asset classes, and second, it is more stable. That said, note that taking the average typically masks correlation shifts for specific risk factor pairs. On the basis of the way I have defined risk factors (using market proxies), some factor pairs become more correlated during market stress (for example, equity and spread risk) whereas others become less correlated (for example, equity and duration or equity and momentum). So, even in risk factor space, correlations can be unstable. For asset classes, there is a jump in the average cross-correlation from 30% during quiet markets to 59% during turbulent markets. This rise in correlation comes primarily from the various equity and equity-driven asset classes. The well-known observation from this analysis is that asset class diversification tends to disappear when it is most needed: during periods of market stress. Table 2 shows the asset allocation and the risk allocation of an average endowment. From an asset class perspective, the endowment portfolio is well diversified among both traditional investments and alternative investments, such as hedge funds, private equity, real estate, and venture capital. Only 28% of the endowment is directly invested in public equities. But when the asset classes are decomposed by risk factor, 78% of the endowment’s overall risk stems from broad equity risk.4 The model can be used to calculate the risk contribution with the following equation: xi = βi σi ρi p , ,
where xi is the risk contribution for factor i as explained by three components: βi is the risk factor’s beta (exposure), σi is the risk factor’s volatility, and ρi,p is the correlation between the risk factor and the entire portfolio. Each of these three components contributes to the high percentage of equity risk:
Table 1.
Risk factor correlations Asset class correlations
analysis, see Sébastien Page, “The Myth of Diversification: Risk Factors vs. Asset Classes,” PIMCO Viewpoints (September 2010): www.pimco.com. 4Many
authors have proposed similar examples. See, for example, Vineer Bhansali, “Beyond Risk Parity,” Journal of Investing, vol. 20, no. 1 (Spring 2011):137–147.
54 • September 2013
Full Sample 4% 40
Quiet 3% 30
Stressed 4% 59
Notes: To calculate stress correlations, I scaled observations by the covariance matrix. The 15% outliers were categorized as stressed. All other months were categorized as quiet. Sources: Based on data from Windham Portfolio Advisors, PIMCO DataStream, Barra, MSCI, Barclays Capital, Dow Jones, and Standard & Poor’s.
Table 2.
Average Endowment’s Asset Class vs. Risk Allocation
Asset Class Cash Distressed debt Domestic bonds (investment grade/high yield) Domestic equity Emerging market equity Energy and natural resources Global equity Global/emerging market bonds Hedge funds Private equity Real estate Venture capital Asset Class Commodities Corporate spread Equity Emerging market currency High-yield spread Municipal spread Other
Market Value Allocation 3% 2 8 12 7 9 9 1 20 16 8 5 Risk Allocation 5% 4 78 3 3 3 5
Notes: The asset allocation percentages shown are based on the Average Endowment Portfolio as of 30 June 2011 for endowments > $1 billion. Percentages do not equal 100% because of rounding. Source: Based on data from the Average Endowment Portfolio from the 2011 NACUBO-Commonfund Study of Endowments (NCSE). This study is based on information collected as of 30 June 2011.
• •
3For an earlier version of this
Risk Factor vs. Asset Class Correlations, 31 March 1997–31 May 2011
•
Equity beta is often greater than investors assume because other asset classes, such as real estate, contain indirect equity risk. The equity risk factor is more volatile than most, if not all, other risk factors. Because of the risk-on/risk-off effect, the equity risk factor typically exhibits positive correlations with other sources of risk premiums in the portfolio.
©2013 CFA Institute • cfapubs.org
Risk Management beyond Asset Class Diversification
But important caveats apply to such analyses. First, the percentage of portfolio risk that is contributed by the equity risk factor does not relate directly to the actual riskiness of the portfolio. Imagine a portfolio with 95% cash and 5% equity. Almost all of this portfolio’s risk would come from equities, but its volatility would be lower than, say, a long government bond portfolio.5 Second, the model ignores nonlinearities: Correlations and volatilities change over time, and therefore, equity’s contribution to risk also changes. Last, the model decomposes volatility. A superior approach to risk decomposition would be to decompose tail risk directly. A tail-risk decomposition often shows similar, or perhaps more extreme, results in terms of the concentration of equity risk.
The Role of Macroeconomic Fundamentals In addition to improved portfolio diversification, one of the key advantages of the risk factor approach to investing is that it serves as a link between macroeconomic fundamentals and asset returns. To illustrate this concept, my colleagues Niels Pedersen and Helen Guo and I compiled historical forecasts and realizations of GDP and inflation based on quarterly data going back to the early 1970s. We estimated macro surprises, which we defined as the differences between what was forecasted at the beginning of the period and actual realized GDP and inflation during the period.6 Then, we tried to explain returns to key risk factors using the macro surprises based on the following equation: GDP
Factor returni ,t = αi + βi
GDPt
INFLt + β INFL i
+ β INT GDPt INFLt + εi,t , i
where αi is the intercept (approximately, the mean return over time because the surprise variables are approximately zero mean), βGDP GDPt is a GDP i
between growth and inflation. As an example of the interaction effect, the impact of an inflation surprise on yields may be more pronounced if it is accompanied by a positive shock to GDP growth because it may signal a more permanent increase in the path for future inflation. Lastly, εi,t is the usual error term used in regression analysis. Our initial analysis of the differences between macroeconomic forecasts and subsequent realizations yielded two observations. First, realizations are much more volatile than expectations; on average, investors do not make extreme forecasts, but economic fluctuations in the world tend to be extreme—at least more extreme than expected. Second, expectations tend to follow realizations. Investors have a tendency to extrapolate recent developments in the fundamentals of the economy. Our main idea was that fundamentals can be mapped to risk factors based on the differences between expectations and realizations—that is, the surprises in key macroeconomic drivers of risk factor returns. Overall, the power of our approach lies in estimating the deltas from general market consensus, which drive the sensitivities of risk factor returns to macroeconomic fundamentals. Risk factors can, in turn, help explain asset returns in general because different asset classes are exposed to those key risk factors. Note that this type of model is meant primarily for scenario analysis; it is not a model to implement systematically as a quantitative strategy. Table 3 shows the sensitivities of a few of the primary risk factors to GDP and inflation surprises. As an example, assume GDP growth at the end of the year is 1% higher than consensus expectation at the beginning of the year. The model predicts that the short interest rate will rise by 50 bps, or 0.5 times that 1% surprise; the 10-year yield will increase by 20 bps, or 0.2 times 1%; credit spreads will contract by 20 bps, or –0.2 times 1%; equities will return 3.9% above their expected mean return, which is 3.9 times 1%; and so on. Table 3.
beta times the growth surprise, β INFL INFLt is the i inflation beta times the inflation surprise, and β INT GDPt INFLt measures the interaction effect i 5Also
note that when a risk factor helps decrease portfolio risk, the correlation effect can generate negative risk contributions, which is not easily shown in pie charts but is important in understanding risk budgets. 6Forecast
data were obtained from the Survey of Professional Forecasters. For more details on this section on the role of macroeconomic fundamentals, see Sébastien Page, Niels K. Pedersen, and Helen Guo, “Asset Allocation: Does Macro Matter? Part II,” PIMCO In Depth (May 2012): www.pimco.com.
©2013 CFA Institute • cfapubs.org
Model Estimation: Sensitivities of Factor Returns to Surprises
Short rate 10-year yield Credit spreads (Baa) Equities Commodities
GDP Surplus
Inflation Surprise
0.5 0.2 –0.2
0.7 0.4
3.9 2.7
–2.2 6.3
Note: Surprises are defined as Expected (GDP and Inflation) – Realized (GDP and Inflation). Sources: Based on data from Page, Pedersen, and Guo, “Asset Allocation: Does Macro Matter? Part II,” PIMCO (2012); Haver Analytics; Survey of Professional Forecasters (conducted by the Federal Reserve Bank of Philadelphia).
September 2013
• 55
CFA Institute Conference Proceedings Quarterly
There are a few qualifying factors to mention. First, the short-term interest rate right now is almost entirely driven by Federal Reserve Board policy. The Fed is committed to holding rates low for the near term. Hence, in the current environment with the ongoing quantitative easing, the Fed is forcing that coefficient below 0.5. In addition, in our research, we were unable to find a meaningful relationship between credit spreads and inflation. The relationship varies significantly, for example, depending on which decade is observed. Lastly, we found a negative coefficient between equities and inflation surprises, which typically generates a lot of debate. Conventional wisdom has held that over the long term, equities will hedge against inflation. But some academic studies have shown that the equity beta is negative in the short term and that it can be negative in the long term. Empirical evidence reveals that equities are not a very good hedge for inflation, at least not historically or across countries and especially not relative to commodities.7 Commodities generate a response to inflation surprises that is “levered” six times the magnitude of the inflation surprise. Of course, commodities are also a component of the U.S. Consumer Price Index. Table 3 demonstrates the links between macro fundamentals and asset returns; portfolio managers can build similar tables to generate estimates of asset returns under different scenarios for macro fundamentals. Table 4 shows that, as expected, equities are much more sensitive to GDP surprises than to inflation surprises. A similar table for bonds would show that bonds are much more sensitive to
inflation surprises, all else being equal. The bottom line is that the risk factor framework goes beyond serving as a template for thinking about diversification; it helps form views on returns as well.
Tail-Risk Hedging Tail-risk hedging is another avenue through which investors are adapting their portfolios to the current market environment. Figure 1 shows two probability distributions built using the risk factor model. I mapped a simple 60/40 portfolio to its underlying risk factors, and given the portfolio’s current exposures, I created probability distributions by simulating risk factor returns. I used a simulation that captures fat tails in the data. Specifying regime probabilities and then sampling according to those probabilities is one way of creating a forward-looking probability distribution. Although it was not done in this case, risk factor returns can be oversampled during periods of rising rates or during periods of high market volatility if tails are expected to be fatter. The dark-gray area of Figure 1 shows the traditional portfolio distribution, and the lighter-gray area shows the portfolio hedged with an S&P 500 put option that costs 100 bps with an assumed 22% implied volatility. The unhedged traditional portfolio has significant negative skewness, or large losses, that the normal distribution—illustrated by the line—fails to predict. Hedging the portfolio, although it can be costly, eliminates this fat-tail risk. When they use volatility as their risk measure— for example, when comparing Sharpe ratios— investors implicitly assume a normal probability distribution. Understanding this assumption is critical. Under a normal probability assumption, the likelihood of a large gain is exactly the same as the
7See
Nicholas J. Johnson and Sébastien Page, “Inflation Regime Shifts: Implications for Asset Allocation,” PIMCO In Depth (October 2012): www.pimco.com.
Table 4.
Scenario Analysis, Equities Real GDP Growth Surprise
Inflation Surprise
–4%
–2%
–1%
Equities 4% (2σ) 2% (σ) 1% (σ/2)
–2σ –17% –13 –11
–σ –9% –5 –3
–σ/2 –6% –1 1
0% –1% (–σ/2) –2% (–σ) –4% (–2σ)
–9 –6 –4 0
–1 2 4 8
3 5 8 12
1% 0% –2% 3 5 7 9 12 16
σ/2
2% 7 9 11 13 16 20
2%
4%
6% 11 13
2σ 14% 19 21
15 17 20 24
23 25 27 32
σ
Notes: This is a hypothetical example for illustrative purposes only. Equities are proxied by the S&P 500. Surprises are defined as the differences between the scenarios and the market expectations from the Survey of Professional Forecasters at the beginning of the year. The sample period is 1970–2011, subject to data availability. The risk-free rate is assumed to be 0.1%. σ represents the estimated standard deviation of the respective surprise. Risk factor returns are assumed to be linear functions of surprises in the regression for simplicity, although other nonlinear relationships could be present. For example, although low inflation surprises have positive implications on equity returns in general, extremely low inflation/deflation surprises could be correlated with low equity returns. Sources: Based on data from PIMCO, Haver Analytics, and the Survey of Professional Forecasters.
56 • September 2013
©2013 CFA Institute • cfapubs.org
Risk Management beyond Asset Class Diversification Figure 1.
Probability Distribution of a Traditional Portfolio vs. a Hedged Portfolio Using the Risk Factor Model, 31 January 2000–30 June 2012
Probability Density
–50
–40
–30
–20
–10
0
10
20
30
40
50
Annual Return (%) Traditional Portfolio
Traditional Portfolio Hedged
Normal Distribution Notes: This is a hypothetical example for illustrative purposes only. The portfolio is 60% MSCI World Index, 40% Barclays U.S. Aggregate Index. Source: Based on data from PIMCO.
likelihood of a large loss. But empirically, negative skewness is present in most markets and, in particular, in strategies that involve selling volatility. Because it is difficult to diversify across underlying risk factors and because key macro factors drive most of the asset returns, it can be very difficult to diversify away fat tails. Often, the only way to eliminate tail risk through diversification is to de-risk the portfolio by, for example, loading up on Treasuries (although it can be argued that in the current environment, Treasuries are closer to becoming a risk asset than they have been historically). Another way to eliminate tail risk is to directly protect the portfolio using nonlinear instruments— in this case, a put option. My colleague Vineer Bhansali, who oversees PIMCO’s quantitative investment portfolios, calls this “tail-risk hedging.” Tail-risk hedging is becoming part of the asset allocation decision for many institutional investors as they begin to think about asset allocation in nonlinear terms. For example, if an investor is able to hedge tail risk, through direct or indirect hedges, and protect part of the portfolio’s downside risk, this protection can directly increase the investor’s risk tolerance. As a simplistic example, perhaps the allocation to equities can be increased to 70% if losses of 15% or more can be hedged. Hedging tail risk comes with a cost; it is not a free lunch. Hence, hedging is ultimately a matter of investor preference. An investor can choose to de-risk the entire portfolio or hedge the portfolio, pay the costs, and then perhaps finance that cost ©2013 CFA Institute • cfapubs.org
with a larger exposure to high-risk assets. These are examples of rethinking asset allocation in nonlinear terms, not just in terms of asset class or even risk factor diversification. I like to use seatbelts in cars as an analogy. Seatbelts were not mandated by law until the 1970s in the United States. Tail-risk hedging in asset allocation is evolving in the same way. At some point, most investors will think of downside risk protection in their portfolios as a natural part of asset allocation, just as using seatbelts for safety protection has become a routine part of driving.
Conclusion In the wake of the financial crisis, investors have become skeptical of financial engineering and risk management in general. Today, I have presented four trends in asset allocation and top-down risk management that I encourage all investors to pay attention to. First, rather than relying on a backward-driven view generated by historical statistics, investors should formulate a forwardlooking view driven by macroeconomics. Second, investors should focus on risk factor–based diversification in addition to asset class–based diversification. Third, investors must recognize the dynamic nature of markets and make asset allocation decisions on a cyclical and secular basis rather than a calendar-year basis. Finally, risk should not be defined solely as volatility; investors should seek to explicitly measure and manage tail-risk exposures. This article qualifies for 0.5 CE credit.
September 2013
• 57
CFA Institute Conference Proceedings Quarterly
Question and Answer Session Sébastien Page, CFA Question: Is the pricing of
that are further out of the money hedging. One strategy to address are more expensive. both the cost question and the Investors have a few options question of how to effectively when it comes to hedging tail risk. construct a protection portfolio They can decide hedging is too is to diversify the hedges by expensive and de-risk instead. including not just an equity put They can also decide to implement but also CDS tranches, puts on Page: The put is approximately tail-risk hedging when the cost is high interest rate currencies, 15% out of the money. If I low or when they think they will derivatives on interest rates, and assume that the implied volatilneed it as an active investor. so on. The outcome is that taility is 22% and that I am spending These last two scenarios are risk hedging is an active manage100 bps, I can back out the strike akin to saying, “I am going to ment process. price. These factors are the movput my seatbelt on right before ing pieces in pricing the put. An Question: What is the average the car crashes.” The alternainvestor can have a fixed budget length of a turbulent regime, and tive is, of course, “I am going to for protection of, for example, how does that compare with the wear my seatbelt at all times”; 100 bps a year. The expected average length of a quiet regime? the investor will bear the costs return might be better than but reap the benefits and harvest Page: There is a wide variation –100 bps because the option is more risk premium in other parts in the range of durations for both expected to pay off occasionally. of the portfolio. turbulent and quiet episodes. So, the total cost of the put In general, tail-risk events One turbulent market might last might be less than 100 bps, ex tend to be different from one for four or six months, whereas post. If the budget for protection another. The tech bubble was another lasts for a year and a half is fixed, then the strike price will very different from the 2008 or two years. The duration of move based on the cost (implied crisis. Both of these events the regime also depends on how volatility). Or the strike price were very different from the the regime is defined. Defining can be fixed if the investor is Long-Term Capital Management the regime based on the VIX concerned with limiting losses event, and September 2011 was and defining it based on another to, for example, 10%; in that case, its own outlier. Each crisis is risk index will result in different the cost of the put will vary over different. If investors do not want regime lengths. time. Either the strike price will to fight the last war, so to speak, All I can say with a high be fixed and the cost of the put they will need to be careful when degree of confidence is that risk will vary or the cost of the put defining a tail event ex ante and tends to cluster. To use a facile will be fixed and the strike price constructing the protection portexample, if Lehman Brothers will vary. folio from that perspective. defaults today, the probability But most crises do share that the world will be volatile for Question: Are investors overpaying for tail-risk insurance— some common characteristics. the next few months is higher and overestimating the need for One key characteristic of finanthan it would be if Lehman does it—based on the events of 2008 cial markets is that correlations not. and 2009, similar to the way GDP rise as equity markets sell off. For Question: Can you comment estimates are based on recent example, an investor can plot the on the inclusion of hedge funds growth figures? correlation between equities and in a portfolio diversified by risk credit spreads, which over time factors? Page: The cost of tail-risk is roughly 20%. But when the protection changes over time. monthly data are partitioned for Page: If I combine a lot of hedge Currently, with the short end 10% or 20% losses in the equity funds into a broad fund of funds, of the Chicago Board Options market, the correlation between alphas and “idiosyncratic risk” Exchange Market Volatility Index the two can go all the way to the may get diversified away. Risk (VIX) at 14, hedging tail risk is upper 90% range. factor analysis often reveals that not very expensive by historical This increase in correla broad hedge fund portfolios standards. But the VIX curve is tion in the tails facilitates proxy essentially provide a collection steep in the sense that options the S&P 500 put used in your example of tail-risk hedging at the money or out of the money, and does the degree to which it is out of the money vary over time?
58 • September 2013
©2013 CFA Institute • cfapubs.org
Q&A: Page
of betas: equity risk, interest rate stocks and commodities to inflaHistorically, there is no statistically risk, spread risk, and maybe tion? Would commodities still be significant positive beta. some liquidity exposure. as sensitive to inflation if another To be sure, the relationship Hedge funds are not an asset time period had been chosen? between equities and inflation is class per se but strategies that complicated because inflation is Page: In our research, my colinvest in a combination of risk often linked to growth concerns. leagues and I used quarterly factors present in other asset During inflationary peridata and rolled it over one-year classes. Investors should look for periods, which does create issues ods that are purely monetary nonbeta returns in hedge funds: phenomena, stocks may indeed with the econometrics, but we a manager that engages in a provide a hedge. After all, if corrected for the autocorrelalong–short strategy, a specialized twice the supply of money is tion bias. Each beta should be strategy, or in general, trades and thought of as a one-year horizon, printed, then stock prices should exposures that are uncorrelated beginning in the 1970s. This simply double in nominal price. with the key risk premium in analysis is time-period specific. For example, during the 1921–23 public markets. Much of the high-inflation tail period of hyperinflation in The bottom line is that risk risk in our data is from the 1970s, Germany, stocks protected invesfactor analysis reveals that when when inflation was driven by oil tors against hyperinflation. allocating to hedge funds, for shocks and, overall, the world But during many other perithe same expected return, it is was quite different. ods in history, the beta between preferable to seek volatility that is Nonetheless, although it will stocks and inflation was negauncorrelated with the key betas in not be exactly the same, I expect tive. One of the hypotheses that a portfolio. the coefficient for commodities in At the very least, if an inveshas been suggested to explain the future to continue to be high tor has a collection of hedge this negative beta is that poor and positive, given their explicit funds that produce only an array economic growth leads to an economic links to inflation. of beta returns and no net alpha, increase in government spendMany investors assume the risk factor framework will ing, which leads to unsustainable that equities will hedge inflahighlight that situation more effidebt-to-GDP ratios, which, in tion over the long run. But the ciently than a nice, multicolored turn, lead to monetization of the negative equity beta to inflation asset allocation chart. debt through inflation. There are is prevalent in the literature, even other explanations, and the topic at longer-term horizons. There are Question: Will you clarify the remains controversial in both time horizon over which you decades during which equities measured the sensitivity of failed to keep up with inflation. academic and practitioner circles.
©2013 CFA Institute • cfapubs.org
September 2013
• 59