You are on page 1of 14

The Journal of Investment Strategies (119131)

Volume 1/Number 2, Spring 2012

Understanding risk-based portfolios


Ryan Taliaferro
Acadian Asset Management LLC, One Post Office Square, Boston, MA 02109, USA; email: information@acadian-asset.com

Equity portfolio managers traditionally form portfolios using forecasts of returns. They seek to identify and hold stocks that will have high returns and to avoid stocks that will have low returns. In contrast, a number of equity strategies that have received attention recently do not employ return forecasts at all. Instead, investors form portfolios based on stocks risk characteristics. This paper reviews three non-returns-based (ie, risk-based) alternative equity strategies: minimum variance, equal risk contribution and maximum diversication. Minimum variance strategies build intuitive, low-risk portfolios. Equal risk contribution strategies are hybrids of risk minimizing and equal weighting strategies, and they, too, have intuitive characteristics and can have lower risk. In contrast, maximum diversication strategies are puzzling. Maximum diversication strategies prefer stocks with low correlations, even if they are risky, and since they do not target risk reductions, return increases or Sharpe ratio increases per se, maximum diversication strategies can produce portfolios with desirable characteristics only by accident.

The views and opinions expressed in this paper are those of the author and do not necessarily reect the views ofAcadianAsset Management LLC or any of its afliates or employees. The data presented in this paper has been obtained from sources believed to be reliable, but the author can make no guarantee as to its accuracy and completeness, and no such guarantee is expressed or implied. In particular, results generated by the MercerInsight PST application are for illustrative purposes only and are not intended to project future value of actual investments or holdings. In no event shall Mercer or any of its service providers be liable to any persons or entities for reliance on any of this data, and results in this paper should be read in conjunction with, and are subject to, MercerInsightMPA: Important Notices and Third-Party Data Attributions. The data is copyright 2011 Mercer LLC and all rights are reserved. The benchmark is proprietary to MSCI, copyright 2011, and all rights are reserved. The author does not recommend that this research paper serve as the basis for an investment decision; rather it is made available for informational purposes and is not an offer, a solicitation of an offer, or a recommendation or advice to purchase nancial instruments and should not be so construed.
119

120

R. Taliaferro

1 THREE STRATEGIES
Alternative equity strategies use risk-based measures to determine allocations within a stock portfolio.1 Three frequently cited strategies are minimum variance (MV), equal risk contribution (ERC) and maximum diversication (MD).

1.1 Minimum variance


One of the most startling facts in modern nance is the observation that risk has been unrelated to average return in the cross-section of stock returns. That is, over many decades, risky stocks have, on average, not realized higher returns than safer stocks. Finance economists attribute this robust empirical fact to two causes: investors irrational willingness to overpay for risky stocks, and the unwillingness of institutions to purchase low-risk stocks and take on the resulting tracking error (Baker et al (2011)). The immediate implication for investors who are unconstrained by tracking error, or who can apply leverage,2 is to build stock portfolios that minimize risk: if risk is not compensated, why hold it? The resulting minimum variance strategy forms the lowest-risk portfolio from available stocks (see Appendix B). The strategy is rmly grounded in empirical observation and economic theory, and it results in intuitive tilts toward sectors such as utilities and consumer staples.

1.2 Equal risk contribution


A portfolio for which each constituent stock (or class of stocks) has the same weighted marginal contribution to risk is an ERC portfolio,3 sometimes said to exhibit risk parity. For example, one implementation for long-only, equity-only portfolios sets each industrys weighted marginal risk contribution equal to that of the others. A stocks or industrys marginal contribution to the risk of a portfolio is proportional to the stocks or industrys beta to that portfolio. Consequently, in an industry ERC portfolio, the weight of each industry multiplied by its beta to the portfolio is equal across industries. The resulting ERC portfolio can be thought of as a blend of the
One family of equity strategies that is often labeled alternative but that is unrelated to the riskbased strategies discussed in this paper is fundamental indexing. Fundamental indexing uses nonrisk-based stock-level characteristics, such as measures of relative value, to form portfolios. Since value characteristics are known to forecast returns, fundamental indexing is a form of traditional investing in which return forecasts, rather than risk forecasts, are the primary variables that determine portfolios. In contrast, the risk-based strategies discussed in this paper are unrelated to traditional returns-based investing. 2 Investors inability or unwillingness to take on leverage has also been suggested as an explanation for the high returns of low-risk stocks (see Black (1972) and Frazzini and Pedersen (2010)). 3 Other denitions also have been proposed, such as setting portfolio weights according to the inverse of each assets or asset classs volatility (see Maillard et al (2009) and Clarke et al (2011)).
The Journal of Investment Strategies Volume 1/Number 2, Spring 2012
1

Understanding risk-based portfolios

121

MV portfolio and an equally weighted portfolio.4 In practice, the difference between an ERC portfolio and an equally weighted portfolio can be slight, so that the ERC portfolio can be closer to an equally weighted portfolio than to a risk-minimizing portfolio. If the equally weighted portfolio overweights lower-risk assets relative to the value(cap-) weighted benchmark, an ERC portfolio will likely be lower risk than the benchmark. In addition, the ERC portfolio could be designed to minimize risk, as in an MV portfolio, subject to the constraint that industries in the resulting portfolio have equal marginal risk contributions. Therefore, ERC portfolios may take advantage of the failure of equity markets to link risk to return (as described in the previous section), but the strategy cannot exploit the mispricing of risk as fully as an MV approach.5

1.3 Maximum diversication


A recently introduced and novel methodology searches for the portfolio that exhibits the greatest difference between the risk of constituent stocks on their own and the risk of those constituent stocks in combination (Choueifaty and Coignard (2008) and Choueifaty (2011); see Appendix B). The construction of these portfolios is governed by the maximization of a ratio. The numerator of the ratio is the average risk of the portfolios constituent stocks, while the denominator is the risk of the portfolio itself. Consequently, there is a tension between the numerator and denominator: the numerator pushes the optimizer toward risky stocks, while the denominator pushes the optimizer toward stocks that have low risk in combination. The optimizer resolves this tension by looking for stocks with low correlations, regardless of their individual risk or their combined portfolio risk, so that the resulting portfolio exhibits the greatest difference between the risk of constituent stocks on their own and their total risk in combination. Because the portfolio that has the greatest change in pre- and post-formation risk displays the greatest effect of diversication, this strategy is called a maximum diversication portfolio. However, the name can be deceptive, because other portfolios may be less risky. In fact, there is no reason to think that the portfolio that has the greatest difference in pre- and post-formation risk has any desirable properties, such
equally weighted portfolio sets the portfolio weights of constituent stocks to be equal, eg, wi D wj for all stocks i and j . An MV portfolio sets i 0 w D j 0 w for all stocks i and j , and an ERC portfolio sets wi i 0 w D wj j 0 w for all stocks i and j (see Appendix B). In this sense, the ERC portfolio is a hybrid of equally weighted and MV portfolios (see also Maillard et al (2009)). 5 That is, the ERC portfolio will have higher risk than the MV portfolio, except in rare circumstances in which it has the same risk as the MV portfolio. This discussion of the ERC methodology as applied to equity-only portfolios is distinct from its application in multi-asset-class settings where risk parity approaches have been used to build portfolios with high Sharpe ratios.
Forum Paper www.thejournalonvestmentstrategies.com
4 An

122

R. Taliaferro TABLE 1 Example 1: return standard deviations and correlations for three stocks. Return standard deviation (%) Stock 1 Stock 2 Stock 3 1

Stock 1 1 0 0

Correlations Stock 2 Stock 3 0 1 0 0 0 1

TABLE 2 Example 2: return standard deviations and correlations for three stocks. Return standard deviation (%) Stock 1 Stock 2 Stock 3 1

Stock 1 1 0.5 0.5

Correlations Stock 2 Stock 3 0.5 1 0.2 0.5 0.2 1

as being the lowest risk, highest return or highest Sharpe ratio portfolio. (Indeed, as will be shown below, the risk of the MD portfolio can be high.)

2 SIMPLE, THREE-ASSET EXAMPLES


This section uses two three-asset cases to highlight the similarities and differences between the three portfolio-construction methodologies. In the examples, consider two of the assets, labeled stock 2 and stock 3, as particular stocks, and the remaining asset, labeled stock 1, as a blend of all other available stocks. Table 1 presents summary statistics for the rst three-asset example. The rst data column displays return standard deviations for three hypothetical stocks: stock 1 has a return standard deviation of 1, while the other stocks, stock 2 and stock 3, have unspecied but equal standard deviations . (The following analysis studies what happens to portfolio allocations as changes.) The remaining columns present correlations among the three stocks. In this rst example, the structure is as simple as possible: all inter-stock correlations are assumed to be zero. In contrast, Table 2 presents summary statistics for the second three-asset example. In this example, which we analyze below, correlations with stock 1 are 0.5 for both stock 2 and stock 3, while stocks 2 and 3 have negative correlation 0:2 with each other.
The Journal of Investment Strategies Volume 1/Number 2, Spring 2012

Understanding risk-based portfolios

123

Figure 1 on the next page plots total allocations to stock 2 and stock 3, measured on the vertical axis, against the risks of these two stocks, measured on the horizontal axis, for all three strategies under the assumptions of example 1. (The ERC and MD portfolios have identical allocations to stocks 2 and 3 in this example, so there are only two lines plotted in the gure.) When D 1 so that all stocks are equally risky, all portfolio methodologies allocate equal weights to the three assets: consequently, each methodology allocates a combined weight of 0.67 to stocks 2 and 3. However, as stocks 2 and 3 become riskier (ie, as increases), the MV portfolio, depicted by the dashed black line, quickly reduces exposure to the increasingly risky assets. On the other hand, the ERC and MD portfolios, depicted by the solid black line, are slower to reduce their exposures to stocks 2 and 3, though their exposures do fall somewhat. As a consequence, the MV portfolio has lower risk than the other portfolios whenever > 1. (For example, when D 4, the ERC and MD portfolios have standard deviations of returns higher than the MV portfolio by 22%.) Figure 2 on the next page plots total allocations to stocks 2 and 3, measured on the vertical axis, against the risks of these two stocks, measured on the horizontal axis, for all three strategies under the example 2 assumptions. In example 2, stocks 2 and 3 are negatively correlated, so a portfolio formed from those two stocks may have a considerably lower return standard deviation than either stock on its own. Therefore, the MD methodology, depicted by the gray line, focuses exclusively on these two stocks and allocates 100% of portfolio weight to them. The MV portfolio, depicted by the solid black line in Figure 2 on the next page, also recognizes the benet of combining stocks with low correlations, and, for low levels of , the portfolio allocates 100% of portfolio weight to stocks 2 and 3 as well. However, as the risks ( ) of stocks 2 and 3 rise, the MV portfolio responds by lowering weight to the increasingly risky stocks 2 and 3. For high values of these stocks become too risky, especially in light of their high correlation with stock 1, and the MV portfolio sheds all exposure to them and places all weight on stock 1. The ERC portfolio, depicted by the dashed black line, is a hybrid of an equally weighted portfolio and an MV portfolio (see footnote 4). Consequently, it is not as responsive as the MV portfolio. From a risk perspective, the ERC portfolio underallocates to stocks 2 and 3 when their risk is low, and it overallocates when is high. Crucially and pathologically, in Figure 2 on the next page the MD portfolio always allocates 100% of its portfolio weight to stocks 2 and 3, no matter how risky they become: as rises, the gray line stays locked at a full 1.0 allocation to stocks 2 and 3. The MD portfolio has a strong afnity for stocks that have low correlation and that
Forum Paper www.thejournalonvestmentstrategies.com

124

R. Taliaferro FIGURE 1 Total allocation to stocks 2 and 3 in example 1 (see Table 1) as a function of the return standard deviation (risk) of stocks 2 and 3, for three equity strategies: MV, ERC and MD.

1.0 Total allocations to stocks 2 and 3 0.8 0.6 0.4 0.2 0 1 2 Risk ( ) 3 4

ERC and MD MV

FIGURE 2 Total allocation to stocks 2 and 3 in example 2 (see Table 2) as a function of the return standard deviation (risk) of stocks 2 and 3, for three equity strategies: MV, ERC and MD.

1.0 Total allocations to stocks 2 and 3 0.8 0.6 0.4 0.2 0 1 2 Risk ( ) 3 MV ERC MD 4

therefore show lower risk in combination than when taken separately. Consequently, the MD portfolio can be a high-risk portfolio, depending on the risks and correlations of the available stocks.
The Journal of Investment Strategies Volume 1/Number 2, Spring 2012

Understanding risk-based portfolios FIGURE 3 Cumulative monthly returns for an MV composite, a comparable MD composite and the MSCI world index (August 2006 to September 2011).

125

1.5

Cumulative returns

1.0

0.5

MV composite MD composite MSCI world index 0 2006 2007 2008 2009 2010 2011

Gross of management fees. Source: Mercer database.

Further details supporting the examples in this section can be found in Appendix A, which focuses on the portfolios marked by diamonds (ie, D 2) in Figure 1 on the facing page and Figure 2 on the facing page.

3 LIVE PERFORMANCE TRACK RECORDS: COMPOSITE PERFORMANCE


The differences between the MV methodology (which targets risk reduction explicitly) and the MD methodology (which has no explicit or implicit goals regarding risk reductions or return increases) are stark. These stark differences play out in the live performance of portfolios that are built using these methodologies. To demonstrate this point, Figure 3 displays live cumulative gross-of-fee returns for two managers: one that builds portfolios based on the MV methodology, and another that builds portfolios based on the MD methodology. The cumulative returns to the MV composite favor that strategy during the period 200611, the period for which data for both strategies is available.
Forum Paper www.thejournalonvestmentstrategies.com

126

R. Taliaferro FIGURE 4 Risk (annualized standard deviation of returns, horizontal axis) and returns (arithmetic average returns, vertical axis) for an MV composite, a comparable MD composite and the MSCI world index (August 2006 to September 2011).

Return

1 MV MD composite MSCI world index 0 0 10 Risk 20

Gross of management fees. Source: Mercer database.

Figure 4 plots each strategy according to its risk (horizontal axis) and return (vertical axis) over the period. On a risk-adjusted (ie, Sharpe ratio) basis, the MV composite is a dramatic improvement over the MSCI benchmark, while the MD composite only improves modestly on the benchmark. Table 3 on the facing page summarizes important statistical moments for the two strategies. In the rst data column, the MV composite has an average annualized return of 4.0% and an annualized standard deviation of returns of 13.1%, for a Sharpe ratio of 0.2. In the second data column, the MD composite has an annualized return of 1.4% and an annualized standard deviation of returns of 16.8%, for a Sharpe ratio of about 0. (Note that Sharpe ratios adjust for the risk-free rate, which the MD composite did not exceed on average.) While the MD composite achieved a risk reduction relative to the MSCI index (third data column), it did not reduce risk as much as the MV composite, which targets risk reduction explicitly. The MD composites modest reduction in risk was accompanied by a modest increase in average return, while the MV composite achieved a dramatic decrease in risk while increasing returns markedly. The MD composite also had more dramatic drawdowns over the period than did the MV composite.
The Journal of Investment Strategies Volume 1/Number 2, Spring 2012

Understanding risk-based portfolios TABLE 3 Annualized arithmetic average monthly returns, standard deviation of monthly returns, Sharpe ratio and maximum twelve-month drawdown for MV composite, a comparable MD composite and the MSCI world index (January 31, 2007 to July 31, 2011). MV composite Average return (%, annualized) Return standard deviation (%, annualized) Sharpe ratio (annualized) Maximum twelve-month drawdown (%)
Source: Mercer database.

127

MD composite 1.4 16.8 0.0 43.5

MSCI world 0.5 19.7 0.1 47.1

4.0 13.1 0.2 33.1

4 CONCLUSION
Alternative equity strategies may be benecial when they are economically wellgrounded and well-executed. The goal of minimizing risk is grounded in a vast academic literature that nds weak evidence of any relationship between risk and return in the cross-section of stock returns (Fama and French (1992)). If risk is not rewarded, a rational investor will minimize risk. Therefore, the MV approach is a reasoned, coherent portfolio-execution response to the empirical evidence. Other alternative portfolio-construction methodologies have weaker rationales. The ERC methodology is a hybrid of equal weighting and risk minimization, and it inherits some favorable attributes. More curious is the MD methodology, which does not seek risk minimization, return maximization, Sharpe ratio maximization or the optimization of any other economically grounded measurement of investment performance. Instead, the MD methodology seeks portfolios with the greatest difference between pre- and post-formation risk, without regard for the risk or return prole of the resulting portfolio. Consequently, an MD portfolio only has desirable properties by accident. Investors are wise to articulate an investment thesis clearly and then to identify the portfolio-construction methodology that best expresses the thesis. Of the major alternative equity strategies, the MV methodology ts most comfortably within this framework.

APPENDIX A
This appendix provides additional details for the three-asset examples when D 2. Table A.1 on the next page presents portfolio weights from example 1, setting D 2, and statistics related to rst-order conditions, for each of the three strategies
Forum Paper www.thejournalonvestmentstrategies.com

128

R. Taliaferro TABLE A.1 Market betas for stocks in example 1, together with portfolio weights and statistics for the MV, ERC and MD portfolios, setting D 2. MD MV ERC Marginal Market Portfolio Beta to Portfolio Beta to Portfolio contribution beta weights portfolio weights portfolio weights to D Stock 1 Stock 2 Stock 3 0.33 1.33 1.33 0.67 0.17 0.17 1.00 1.00 1.00 0.50 0.25 0.25 0.67 1.33 1.33 0.50 0.25 0.25 0.00 0.00 0.00

(MV, ERC and MD). For comparison purposes, the rst data column presents market betas, assuming that the market portfolio is an equally weighted combination of the three stocks. In Table A.1, the MV portfolio allocates two-thirds weight (0.67) to the least-risky stock 1, and divides the remaining allocation budget among the risky stocks 2 and 3. Each stock has a beta to the resulting portfolio of 1.0, so that a marginal shift in portfolio weights does not change the portfolios risk. This relationship is a necessary condition for an (interior) optimum. The ERC portfolio in Table A.1 allocates more weight to the risky stocks 2 and 3 and less weight to the least-risky stock 1 than does the MV portfolio. As a consequence of having less weight on stock 1, the resulting ERC portfolio has lower correlation with stock 1, so stock 1 has a lower beta (0.67) with the ERC portfolio. Similarly, as a consequence of having higher weight on stocks 2 and 3, those stocks have stronger correlation, and higher beta, with the ERC portfolio. For each stock, the portfolio weight multiplied by the portfolio beta is the same: 0.33. This condition is the identifying condition for the ERC portfolio. As noted in the main text, the ERC portfolio is a hybrid of a minimum variance and an equally weighted portfolio: it shrinks MV portfolio weights toward an equally weighted allocation. The MD allocation can also be seen in Table A.1. The MD methodology maximizes the diversication ratio D , the ratio of pre-formation average risk of constituent stocks to post-formation total risk of the resulting portfolio (see Section 1.3). The last column in Table A.1 presents the marginal contribution to D of each of the stocks. The allocation in the penultimate column is at an (interior) optimum, so the marginal contributions in the nal column are each zero. Table A.2 on the facing page presents portfolio weights from example 2, setting D 2, and statistics related to rst-order conditions, for each of the three strategies (MV, ERC and MD). For comparison purposes, the rst data column presents market
The Journal of Investment Strategies Volume 1/Number 2, Spring 2012

Understanding risk-based portfolios TABLE A.2 Market betas for stocks in example 2, together with portfolio weights and statistics for the MV, ERC and MD portfolios, setting D 2. MD MV ERC Marginal Market Portfolio Beta to Portfolio Beta to Portfolio contribution beta weights portfolio weights portfolio weights to D Stock 1 Stock 2 Stock 3 0.79 1.11 1.11 1.00 0.00 0.00 1.00 1.00 1.00 0.40 0.30 0.30 0.82 1.12 1.12 0.00 0.50 0.50 0.32 0.00 0.00

129

betas, assuming that the market portfolio is an equally weighted combination of the three stocks. In Table A.2, the MV portfolio allocates full weight to stock 1 and zero weight to stocks 2 and 3, as a consequence of their high risk ( D 2) and high correlations with stock 1 (0.5, see Table 2 on page 122). However, the solution is interior in the sense that the rst-order conditions are satised and the long-only constraint (weights greater than or equal to zero) does not bind. Consequently, all stocks have a beta of 1.0 with the MV portfolio, as in Table A.1 on the facing page. The ERC portfolio in Table A.2 is very close to an equally weighted portfolio. As in Table A.1 on the facing page, the dening condition of having equal products of portfolio weight and portfolio beta for all stocks is satised. The MD portfolio in Table A.2 encounters a binding long-only constraint. It allocates full weight to the risky stocks 2 and 3, but wishes it could allocate more. Given the opportunity to take a short position in stock 1 in order to take a leveraged position in stocks 2 and 3, the MD methodology would do so. This result adds emphasis to the observation that the MD portfolio does not seek a low-risk portfolio, nor does it seek a high-return portfolio or a high Sharpe-ratio portfolio. Its achievement of any of these standard goals would be purely by accident, and, indeed, the MD methodology could create a portfolio far from anything ordinarily considered prudent or optimal.

APPENDIX B
Specications for MV, ERC and MD portfolios are well-known and widely available, but for completeness they are included in this appendix. An MV portfolio is constructed from a set of N stocks with covariances of returns specied by the N N covariance matrix . The N 1 vector of portfolio weights w that minimizes risk satises the rst-order condition i 0 w D j 0 w for all stocks i and j , where i and j are N 1 column vectors of zeros, except for a one in the
Forum Paper www.thejournalonvestmentstrategies.com

130

R. Taliaferro

i th or j th position, respectively. (Under a long-only constraint, if stock i is included in the portfolio and stock j is excluded such that wj D 0, then i 0 w 6 j 0 w.) Intuitively, each securitys beta to the resulting MV portfolio is equal (and, therefore, equal to 1). For an ERC portfolio, the N 1 vector of portfolio weights w that sets stocks weighted marginal risk contributions to be equal satises the rst-order condition wi i 0 w D wj j 0 w for all stocks i and j , where i and j are N 1 column vectors of zeros, except for a one in the i th or j th position, respectively. The MD portfolio solves: w0 1=2 max p D w w0 w such that w0 1 D 1

where w is an N 1 vector of portfolio weights, is an N N return covariance matrix, and D 1=2 is an N 1 column vector of return standard deviations. The rst-order condition for a maximum is:
i

w0 D 1=2 ip D

w0 D 1=2 jp

for all stocks i and j , where i and j are return standard deviations for stocks i and j , respectively, and where ip and jp are the betas of stocks i and j to the portfolio, respectively. At the optimum, all included stocks have marginal conditions of zero, since the marginal effect on the objective function of changing portfolio weights is zero. Under a long-only constraint, excluded stocks have negative marginal contributions, so that: 0D
i

w0 D 1=2 ip >

w0 D 1=2 jp

where stock i is included in the portfolio and stock j is excluded such that wj D 0.

REFERENCES
Baker, M., Bradley, B., and Wurgler, J. (2011). Benchmarks as limits to arbitrage: understanding the low-volatility anomaly. Financial Analysts Journal 67(1), 4054. Black, F. (1972). Capital market equilibrium with restricted borrowing. Journal of Business 45(3), 444455. Choueifaty, Y. (2011). Methods and systems for providing an anti-benchmark portfolio. US Patent No. 7958038, United States Patent and Trademark Ofce. Choueifaty,Y., and Coignard,Y. (2008).Toward maximum diversication. Journal of Portfolio Management 34(4), 4051. Clarke, R., De Silva, H., and Thorley, S. (2011). Minimum variance, maximum diversication, and risk parity: an analytic perspective. Working Paper, SSRN. URL: http:// ssrn.com/abstract=1977577.
The Journal of Investment Strategies Volume 1/Number 2, Spring 2012

Understanding risk-based portfolios Fama, E. F., and French, K. (1992). The cross-section of expected stock returns. Journal of Finance 47, 427465. Frazzini, A., and Pedersen, L. H. (2010). Betting against beta. Working Paper, AQR Capital Management, New York University (Working Paper 16601, NBER). Maillard, S., Roncalli, T., and Teiletche, J. (2009). On the properties of equally-weighted risk contributions portfolios. Working Paper, SGAM Alternative Investments, Lombard Odier and University of Paris Dauphine. URL: http://ssrn.com/abstract=1271972.

131

Forum Paper

www.thejournalonvestmentstrategies.com

You might also like