This action might not be possible to undo. Are you sure you want to continue?

3

Asset Allocation: The Mean–Variance Framework

Chapter Outline

3.1 3.2 3.3 Introduction: Motivation of the Mean–Variance Approach to Asset Allocation Theory: Outline of the Mean–Variance Framework Practice: Solution of Stylized Problems Using the Mean–Variance Framework Appendix 1: Returns, Compounding, and Sample Statistics Appendix 2: Optimization Appendix 3: Notation

3.1

**Introduction: Motivation of the Mean–Variance Approach to Asset Allocation
**

Asset allocation is the term used to describe the set of weights of broad classes of investments within a portfolio. For an individual investor, asset allocation can be represented by the proportional investment in bond mutual funds, stock mutual funds, and money market investments. For example, if an investor holds $1,500 in bond mutual funds, $3,000 in stock mutual funds, and $500 in money market funds, this asset allocation can be described as 30 percent bonds, 60 percent stocks, and 10 percent cash. This set of weights would provide an initial summary of the risk profile of the individual investor’s investments, prior to completing a more detailed description of the funds owned or an even more detailed description of the individual securities held in the funds. Why does this allocation represent the risk profile? Certain investments reflect greater volatility, and investing in multiple asset classes tends to be less risky than placing 100 percent in any one investment. Once an investor’s goals and objectives have been defined, setting the asset allocation target is the first step in developing an investment program. These weights will define the overall behavior of the portfolio and should be set to match the risk and return targets for the investor. For example, an investor concerned about total return risk would tend to have a higher weight in money market funds than would a more risk-tolerant investor.

51

52

Chapter 3

Asset Allocation: The Mean–Variance Framework

Asset allocation techniques represent tools that help professionals set the optimal mix between broad classes of investments. They are used to determine how much money should be placed in stocks versus bonds and so on. Different techniques may be used for short-term and long-term investment horizons, but they all have the same goal of setting proportional investments. Asset allocation tools can help solve many investment problems, including those faced by individual investors, defined benefit pension (DB) plans, defined contribution (401(k)) plans, endowments, and foundations. Although the techniques may be similar in concept, there are important differences between the tools applied to different problems. For example, DB plans are characterized by a detailed cash flow stream listing projected payments to retirees. This stream reflects the liability that must be met in the long term through funding and investment. The asset allocation of a DB plan must optimize the match between plan assets and this liability. On the other hand, 401(k) plans do not reflect a single, well-defined liability. Individual investors must decide their asset allocations utilizing proxies for the liability, such as expected retirement expenses and future sources of income, as well as risk preferences.

**Types of Asset Allocation
**

There are at least three types of asset allocation: strategic, tactical, and dynamic. Strategic allocation, as defined by Nobel Prize winner Bill Sharpe (1987), is set based on long-term goals. For example, if you need to set your asset allocation today for the next 30 years, you would be defining a strategic allocation. A fixed strategic allocation would be consistent with constant investment opportunities and risk tolerances. A fixed allocation may not be a good fit for an individual who plans to switch from saving to spending sometime in the future. In practice, strategic allocations are reviewed and revised at least every three to five years. Tactical asset allocation responds to short-term changes in investment opportunities. Investors who frequently adjust their exposure to stocks, bonds, and cash are called market timers and set their allocations tactically. They seek to profit from shortterm movements in the market, expect to change their asset weights in the near future, and may not worry much about the long-term implications of their average weights. Some active allocation managers may define a band around strategic weights, within which weights may be set in the short term but never deviate by so much so that tactical bets overwhelm the strategic allocation. For example, a long-term average target weight of 50 percent stocks with a 10 percent band yielding a 40 –60 percent range would incorporate both strategic and tactical allocations. Dynamic asset allocation is driven by changes in risk tolerance, typically induced by cumulative performance relative to investment goals or an approaching investment horizon. Portfolio insurance, popular in the mid-1980s, was a dynamic allocation strategy. It was designed to replicate the behavior of a put option by constantly adjusting the allocation to stocks based on the market level. This delta hedge was implemented using futures contracts and worked well until market liquidity collapsed in the 1987 crash. Other forms of dynamic allocation include constant proportion portfolio insurance, constant horizon portfolio insurance, and dynamic horizon asset allocation models.

Asset Classes

Asset allocation refers to setting the weights of asset classes. Asset classes are typically defined as groups of securities with similar characteristics. Statistically their constituents exhibit high correlations within each class, but low correlation between the

Chapter 3

Asset Allocation: The Mean–Variance Framework

53

**EXHIBIT 3.1 Historical Correlations: Monthly Gross Returns, 35 Years Ending 12/31/2008
**

US Small Stk US Small Stk S&P500 Int’l Stock US High Yield US Corp Bond US Govt Bond 30-Day Tbill US Inflation 1.00 S&P500 0.79 1.00 Int’l Stock 0.51 0.59 1.00 US High Yield US Corp Bond 0.57 0.57 0.46 1.00 0.27 0.36 0.26 0.62 1.00 US Govt Bond 0.04 0.14 0.09 0.34 0.87 1.00 30-Day Tbill 0.02 0.03 0.01 0.05 0.09 0.14 1.00 US Inflation 0.03 0.06 0.06 0.06 0.16 0.16 0.44 1.00

**EXHIBIT 3.2 Historical Returns: Annualized Gross, Ending 12/31/2008
**

Last 50 Years US Small Stk S&P500 Int’l Stock US High Yield US Corp Bond US Govt Bond 30-Day Tbill US Inflation 12.5% 10.0% N/A 7.1% 6.8% 6.9% 5.4% 4.1% Last 35 Years 12.6% 10.0% 9.8% 8.2% 8.3% 8.5% 5.9% 4.4% Last 10 Years 3.0% 1.4% 1.2% 2.2% 4.9% 6.3% 3.2% 2.6% Last 5 Years 0.9% 2.2% 2.1% 0.8% 2.6% 6.4% 2.9% 2.8%

classes. As mentioned previously, stocks, bonds, and cash are the most common forms of asset classes; they may be expanded to include international stocks and bonds, real estate, venture capital, hedge funds, high-yield (junk) bonds, and commodities. The return correlations between the classes are frequently assumed to be stable and offer attractive diversification, on average. However, under some circumstances such as periods of market crises, correlation levels can shift quickly and eliminate diversification benefits.1 This occurred in late 2008 when correlations among risky assets (such as equity classes and high-yield bonds) quickly approached 1. Long-term correlation estimates are reported in Exhibit 3.1, illustrating the high correlation between the domestic equity classes and relatively low correlation between equities and government bonds. Investment-grade corporate and lower-quality high-yield bonds exhibit somewhat stronger correlation with equities. Investment-grade corporate bonds are highly correlated with government bonds, but high-yield corporate bonds appear more closely related to equities. Inflation exhibits low correlation with investments, except for very short-maturity Treasury bills. Exhibit 3.2 reports historical returns across several asset classes for multiple time horizons. Over the long term, small-cap stocks have generated the strongest performance at the cost of higher volatility, as shown in Exhibit 3.3. Corporate bonds earned returns similar to those of government bonds, though the latter are of longer duration

1

This issue frequently arises in the context of global investing and is discussed in Chapter 10.

54

Chapter 3

Asset Allocation: The Mean–Variance Framework

**EXHIBIT 3.3 Historical Standard Deviation: Annual Gross Returns, Ending 12/31/2008
**

Last 50 Years US Small Stk S&P500 Int’l Stock US High Yield US Corp Bond US Govt Bond 30-Day Tbill US Inflation 25.2% 17.6% N/A 12.0% 8.8% 6.9% 2.8% 3.0% Last 35 Years 23.0% 18.9% 23.2% 13.6% 9.2% 6.5% 3.0% 3.2% Last 10 Years 21.7% 21.2% 25.5% 13.3% 4.7% 5.3% 1.8% 1.1% Last 5 Years 21.4% 21.2% 28.1% 15.5% 3.5% 4.8% 1.7% 1.6%

and the results were impacted by the events of 2008. Net of inflation, the real return of fixed-income investments has been much lower than that of stocks over longer periods. The reliability of sample statistics will be discussed in detail in Chapter 4. Asset allocation tools are motivated by economic theory and built using mathematics. They rely on assumptions about human behavior—most importantly that investors prefer higher returns and lower risk. There is a lot of flexibility in defining the parameters for the mathematics. For example, returns may be defined over different periods—one month, one year, or 30 years. Risk may be defined as short-term volatility, the chance of losing money, or the probability of meeting wealth goals. Combining investor objectives and preferences and the statistical behavior of asset classes within a mathematical model can provide a useful tool for setting optimal asset weights. However, oversimplification of investor goals or the behavior of asset returns can yield the wrong solution. For example, an economic utility function may be easy to apply mathematically, but it may not effectively capture individual or corporate preferences. Investors look at more than the standard deviation of returns when they evaluate risk. Moreover, the characteristics and behavior of asset classes are not stable and may need to be described by more than mean return and return standard deviation. Finally, highly sophisticated approaches that more accurately fit reality may be difficult for the final user to understand, creating the opportunity for a mismatch between the investor’s goals and the solution. The key to successfully using these tools is to balance sophistication with simplicity and apply healthy doses of intuition and skepticism.

**The Mean–Variance Framework
**

The mean–variance (M–V) framework, originally developed by Nobel Prize winner Harry Markowitz (1952) and others, is a popular model for computing optimal asset allocations. The math is easy and fairly well describes reality. The model is fair in the sense that risk, return, and preferences can all be included in some form; in addition, specific allocations can be computed. To apply this framework some assumptions about reality are required. Investors need to be risk averse and wealth maximizing; these assumptions reflect the real world. Another assumption is that either returns are normally distributed (or can be transformed to that distribution), or investors are interested only in mean and variance (or semivariance); this is only an approximation of reality. Statistical analyses indicate that historical returns exhibit fatter tails than are suggested by normal distributions. In other words, really bad and really good things happen more frequently than predicted by a bell-shaped curve. If you use a M–V model to set allocations, in practice it makes sense to examine setting allocations using risk and return data generated over different

Chapter 3

Asset Allocation: The Mean–Variance Framework

55

War Story

Dynamic allocation horizon funds were introduced in the mid-1990s as a better investment option for 401(k) investors than fixed allocation lifestyle funds. The first mutual fund product offered a linear path for the target allocation to equities, declining to a level close to zero as the fund’s horizon (the investor’s approximate retirement date) approached. On top of this structure was a tactical process that actively timed the equity allocation as market conditions allowed. A linear allocation path was simple, easy to understand and describe, but inappropriate. The investor’s perception of risk is not necessarily linear in time. Some portfolio managers recognized this and included nonlinear paths built on long-term downside risk targets. This technique is reviewed in Chapter 5. Many subsequently developed competing products avoided tactical allocation because the strategic levels were key to investment success, and these passive allocations were dynamic in their own right. Moreover, many funds included active security selection. During the development of these products in the 1990s (and later enhancements in the mid-2000s), historical testing alone could have suggested that higher equity allocations would always yield better results. This was because stocks had experienced a 10 + -year bull market, and any increase in equities would have generated higher historical returns. Left to their own devices, inexperienced financial modelers could demonstrate that 100 percent equity allocations would always dominate any other mix. Taken to its logical conclusion, this model predicted that there was no need for diversification and asset allocation! Only the understanding that things sometimes do not work out as expected, suspected through intuition and observed through experience, would lead to more conservative allocations for retirement savers. Retirees would all have benefited from this in late 1998, 2000–2002, and 2008.

periods. It is also helpful to confirm your results with historical simulations that include worst-case scenarios. Chapter 7 addresses these issues in greater detail. What can we do with the mean–variance framework? Given a model, we can estimate future market values of wealth. We can also calculate confidence intervals around these expectations. The probability of losing different levels of wealth over different periods can be estimated, given mean–variance assumptions. And a model user can calculate an optimal mix of investments reflecting an investor’s trade-off preferences between risk and return. Risk may be defined as return variance, semivariance, the probability of losing money, or the risk relative to a liability. Moreover, varying time horizons may be included as well as wealth, return, risk, and borrowing constraints. The following section provides the theoretical underpinnings of the mean–variance framework, followed by a section describing how to apply this theory in a computer program or spreadsheet. The theory begins by describing a world in which prices move randomly through time according to a lognormal probability distribution. Returns calculated from these prices move according to a normal distribution. Once expected returns, variances, and covariances of asset classes are estimated, we can calculate the expected returns and variances of portfolios reflecting different asset weights. For a given expected portfolio return, we can search for the weights associated with the portfolio offering the lowest overall variance; we call this the minimum variance portfolio (MVP), which we can use to plot the efficient frontier. In this world individuals display a utility function, which converts expected returns and variances into a single value that is meaningful to them and can be communicated to others. This function trades off mean and variance with a sensitivity parameter defined as the risk aversion coefficient. Utility values can be compared for portfolios with different asset weights, and the individual can select the asset weights with the highest value. This is the optimal asset allocation. Because the utility function dislikes variance, this portfolio will be on the efficient frontier.

the remainder being explained by style. reflecting a conservative investor. and security selection.Theory in Practice Asset allocation policy is important. mutual fund companies. a portfolio’s asset allocation policy will determine. This is called shortfall risk and is a useful tool for setting investment policy for individuals and institutional investors. with high probability. it is difficult to implement intraperiod cash flows (such as 401(k) contributions) within the model. Again. For example. by asset class manager performance. The 90 percent is the R-square from regressing quarterly plan returns on the returns of asset class indexes. What does this mean? A research study by Brinson. such as a pension plan’s benefit payment stream. but other factors are cumulatively more important. on average. The influence is determined by how different the asset allocation policies of the funds are. For example. for the long term. may be added to the model as a negative fixed-income asset class. Note. Hood. and Beebower (1986) reported that. asset allocation explained 40 percent of relative annual return variability. the risk aversion coefficient can be adjusted to make risk seem more unattractive. which requires all asset weights to be zero or positive. Relative returns are influenced less than total returns because asset class total returns have been removed to a large extent. that this will not necessarily be the case for all funds in all periods because the result is only an average. variance can be converted into probabilities of high or low returns so we can compare portfolios by. this will not be the case for all portfolios because the result is only an average. Each of these features will be either discussed in the following theory section or developed in the subsequent application section. If an investor hires index managers. Brokerage firms. But the decision to place 50 percent or 55 percent in stocks is less important than which equity manager is hired. It is easy to add asset classes—there is no need to be limited to two or three. however. trading costs. What can we take from this? In general. In practice. explained an even larger percentage (100 percent) of long-term (they used 5. As a result. In contrast. if an investor hires highly concentrated managers. With diversified equity and fixed-income investments. market timing. Security selection. trading costs. at least in the short run. asset allocation may be overwhelmed. the decision to place 0 percent or 50 percent in stocks is more important than which equity manager is hired.and 10-year) portfolio total returns. by definition asset allocation will determine future total returns. this can be represented within the model with a long-only constraint. on average. asset allocation is much more important than selecting the mix of asset class managers. With normality. over 90 percent of the variability in pension plan total returns could be explained by their asset allocation policies. changing investment opportunities (such as changing expected returns and variances through time) are difficult to include. for example. asset allocation policy is important in determining the performance of one portfolio versus another. Nonnormal distributions (those with fatter tails than reflected in the normal distribution) may not be included in the mean–variance model. much of the portfolio’s return. A liability. and financial pundits can all be found to state that 90 percent of portfolio returns are determined by asset allocation policy. and style factors should explain the rest. their chance of losing money over time. optimal portfolios often 56 . The relevant time frame can be changed by simply converting returns and variances into longer-period measures. In the study’s data set. For a given fund over a short period. Most investors do not wish to leverage their investments. In addition. The mean–variance framework is not fully customizable. Ibottson and Kaplan also determined that relative performance between two multi-asset class funds is influenced by asset allocation as well. A study by Ibbotson and Kaplan (2000) demonstrated that asset allocation. One of the nice things about the mean–variance framework is that it can be customized to more closely reflect a real investment problem. however. market timing.

Sophisticated consultants frequently supplement the process with studies of historical data. A review of Exhibits 2. 3. “paying it only lip service. Pension consultants in particular tend to be highly sophisticated—they need to be. They use this for educational purposes. For example. For example. for analyzing the trade-off between risk and return for portfolios containing several assets.” as well as the importance of setting today’s allocation with future changes in the liability in mind. 57 2 Utility Theory . They confirm the results of their mean–variance model with simulated confidence intervals. For pension consultants. Nonetheless it is by far the most common approach to practical asset allocation decisions. known as mean–variance (M–V) analysis. 100 percent allocated to a single asset. One simplifying assumption of the M–V model See the Appendix I for a review of the basic statistical concepts and common probability distributions used in this section. reflect corner solutions with 0 percent weights for most assets or. such as explaining risk–return trade-offs to their clients. Higher returns were not realized in the late 1980s and 1990s but were in the 2000s.2 Theory: Outline of the Mean–Variance Framework2 More than 50 years ago Markowitz (1952) developed a simple framework. and they will be addressed in Chapter 5. the decision to invest internationally was based on the expected benefit of both higher return and lower risk. mean–variance modeling was relied on heavily for making the case in the mid-1980s to invest internationally. As a result.’ Many don’t recognize that you need to consider that your liability will change again in another three years. And they use it for advising on asset allocation policy—both for education and for proposing targets. Less sophisticated consultants “apply an oversimplified 401(k) model. the asset classes. so we should set asset allocation policy based on that.10 and 2. they’ll say. in the extreme. “The good ones have a longer-term view of the problem and recognize the existence of the liability. results may be highly sensitive to small changes in assumptions. improper asset allocation will cost employers a lot of money. this means they ignore the liability side of the equation. who goes on to say. but their approach may be static. and the asset allocation.11 confirms the importance of these observations.Theory in Practice Investment advisors come in all shapes and sizes. ‘Looking out three years. Their clients are plan sponsors. Moreover. This section provides a formal introduction to the M–V model. making it difficult for the practitioner to set appropriate asset allocations. In practice an investor can make investment decisions at various intervals over the course of their investment horizon. Pension consultants know mean–variance modeling is a useful technique that must be applied with caution and supplemented with additional tools. As will become apparent later. One asset allocation tool they use is the mean–variance model.5 trillion as of December 2007. this is what your liability will be. For example. responsible for investing the money set aside to pay benefits to their companies’ employees after they retire. Pension consultants provide advice on how to invest pension assets—selecting the professional managers. These issues can wait—there are more advanced techniques to handle them. ignoring cash flow concerns. this framework has some important limitations. Pension assets are large—the largest 10 represented over $1.” in the opinion of one veteran investment professional. For now we will explore the theory behind mean–variance asset allocation.

Letting W denote wealth and U() the investor’s utility function. The utility associated with each level of wealth is reflected by the height of the utility function at that point. there are diminishing benefits to each increment of wealth. For most people.3 A simple example will illustrate the relationship among wealth. the same as the certain outcome from the riskless investment. one is riskless and the other entails risk. and marginal utility declines as wealth increases.1) where E represents taking the expected value of the expression in brackets. we can write this formally as max E [U (W )] (3. investors attempt to maximize the expected value of utility.4. and possibly even unpleasant. as a result. The utility function.58 Chapter 3 Asset Allocation: The Mean–Variance Framework EXHIBIT 3. The expected payoff on the risky investment is assumed to be W2—that is. and expected utility.4 Suppose an investor is offered a choice between two investments. in excess. has two key properties. utility rises with wealth because more wealth is assumed to be preferred to less. Second. however. The conditions required to validate this assumption will be discussed in Chapter 5. An investor who does not care about risk would be indifferent between these two investments because they provide the same expected wealth. and. third. The riskless investment results in a wealth level denoted by W2 in Exhibit 3. That is. the investor’s decision problem may be treated as if there were only a single period. . evaluates the investments on the basis of expected utility. utility rises at a decreasing rate as wealth increases. The slope of the utility function reflects the investor’s marginal utility: the change in utility due to a small change in wealth. or more cup of coffee in the morning. First.4 Risk Averse Utility Utility Wealth is that the investor will select the same portfolio in each of these subperiods. provided by this level of wealth. even something pleasurable becomes less pleasurable. 4 The following discussion is summarized in the Excel Outbox. The risky investment has two possible outcomes denoted by W1 and W3. second. risk. Our risk-averse investor. In economics it is standard to assume that investors base their decisions on a utility function that maps wealth to their subjective assessments of the utility. the level of welfare or satisfaction. 3 Consider the pleasure obtained from the first. illustrated in Exhibit 3. Because future wealth is uncertain.5.

The horizontal distance between the CEW and the expected level of wealth (W2) is a measure of the compensation—in expected wealth—that the investor requires for bearing the risk inherent in the risky investment. and has the same probability of outcomes. The most useful measure of risk aversion. The curvature of the utility function—the benefit of a gain versus the pain of a loss—is what matters. 5 . is defined by RRA % change in marginal utility % change in wealth (3. we might ask what level of certain wealth the investor would view as equivalent to the risky investment. The risky investment is preferred to any level of certain wealth less than the CEW. In the exhibit this is labeled as the “certainty equivalent wealth” (CEW).5 The point labeled “expected utility” lies on this line directly above the expected wealth level. known as the coefficient of relative risk aversion (RRA). Continuing the example. Note that this line lies below the utility function except at the endpoints. The utility of this wealth level is equal to the expected utility of the risky investment. W2. the negative sign in Equation (3. The absolute level of utility is not relevant to the investor’s decisions because more wealth is always better.5 Expected Utility and Risk Aversion Utility of Expected Wealth • • Utility • • Expected Utility • • Certainty Equivalent Wealth W1 W2 = Expected Wealth Wealth W3 The expected utility of the riskless investment is simply the utility of W2 because the outcome is known with certainty. The curvature of the utility function reflects the investor’s risk aversion or willingness to take risk. The upward-sloping dashed line in Exhibit 3. The vertical distance between this point and the point labeled “utility of expected wealth” is the penalty (in utility) that the investor assigns to the risk. In general the RRA depends on the The endpoints correspond to cases in which there is actually no uncertainty because one of the outcomes (W1 or W3) has zero probability while the other has probability 1.5 shows weighted averages corresponding to varying the probability of the two possible outcomes. the risk-averse investor prefers the riskless investment. Although the two investments give the same expected wealth (W2). The expected utility of the risky investment is a weighted average of the utilities at W1 and W3.2) Because marginal utility declines as wealth increases.Chapter 3 Asset Allocation: The Mean–Variance Framework 59 EXHIBIT 3.2) implies that RRA is (by convention) positive. while any level of certain wealth greater than the CEW is preferred to the risky investment.

assume that you are offered either $50.5 more concrete. Excel Outbox Utility Function To make Exhibit 3. the payoff is $50. To obtain concrete results it is usually necessary to limit attention to a specific class of utility functions. and in cell D6 enter =(1 D5). To capture more risk-averse preferences.000 payoff. U(W) = ln(W). Both options have an expected payoff of $50. Sure thing: In cell F4 enter =(D4*E4).6 ). U(W) = 2 W. where = 0. Use the worksheet <Utility – Template> in spreadsheet Chapter 03 Excel Outboxes. The expected utility of the options is calculated as i pi U(Wi). is an important special case of CRRA utility. is usually negative (as noted in the text).21.5. A > 0 is used here so that the utility function plots in the first quadrant where both wealth and utility are positive. In cells D5 enter 0. In terms of Exhibits 3.5. the CRRA exhibits constant relative risk aversion equal to (1 Typically is assumed to be negative.xls. there is a 50 percent chance of tails with a $0 payoff and a 50 percent chance of heads with a $100. Assume your utility function displays constant relative risk aversion and takes the form W / .000 (W3) with heads or nothing (W1) with tails. However. There is nothing wrong with letting the utility function take negative values. the higher the RRA. as it does when < 0. The sure thing has a utility of 447. and in cell D4 enter 1. positive values of utility are more intuitive.3) As its name implies.5 implies a relatively high tolerance for risk. The higher the RRA. That is. because it is still increasing in wealth. As a result. One of the simplest and most useful utility functions for investment analysis is the so-called power utility or constant relative risk aversion (CRRA) utility function.4 and 3. the more pronounced the curvature of the utility function. the more risk averse the investor.23.000 (W2) for sure or a coin flip where you win $100.5—that is. the sure thing is preferred to the coin flip. and the coin flip has an expected utility of 316. 6 Logarithmic utility.3) as goes to zero. and in cell C6 enter 100000. In cell B1 enter 0. Note: = 0.000 with a probability of 100 percent. It is obtained as the limit of Equation (3. A γ Option Sure thing Coin Flip Tails Heads B C D E F G 1 2 3 4 5 6 W Probability U(W) E[U(W)] CEW Define the options: • Sure thing: In cell C4 enter 50000. Coin flip: In cell F5 enter =(D5*E5) + (D6*E6). Calculate the utility of the options: • • • • U(W): In cell E4 enter =(C4^$B$1)/$B$1 and then copy E4 to E5:E6. This is the definition of risk aversion. defined by U (W ) W / 1 (3.60 Chapter 3 Asset Allocation: The Mean–Variance Framework level of wealth at which it is measured.000. • Coin flip: In cell C5 enter 0.5. That is. . but the lower-risk alternative is preferred.

and then copy C9 to cells C10 and C13 and cells C15:C36. . A γ Option Sure thing Coin Flip Tails Heads B 0. . In cell B13 enter =G5. This is the inverse function of the utility function.21 0.000 CEW C D E F G 1 2 3 4 5 6 Here are the inputs to the X–Y chart: • • • • • • In cell B9 enter =C4. . T. whereas the end of the investment horizon is denoted by t = T. In cell B12 enter =C6.000. the coin flip would be preferred to a sure thing with a payoff below $25. The time at which the investment decision is to be made is denoted by t = 0. Note that the CEW for the coin flip lies on the utility function directly to the left of the E[U(W)] while the E[U(W)] lies directly below U(W) for the sure thing. In cell B11 enter =$D5*B10 + $D6*B12.00 632.46 E[U(W)] 447. Suppose there are n assets (or asset classes) available to the investor and the investment horizon is T years.000 Probability 1. . The prices of the assets at the beginning of the period are known.23 25.Chapter 3 Asset Allocation: The Mean–Variance Framework 61 The question then becomes. at time t = 0.000 0 10. The result is $25. In cell C9 enter =(B9^$B$1)/$B$1. .5 W 50. . n. .5 U(W) 447.000. what level of wealth has a utility of 316. Let Pi(t) denote the price of asset i. 700 600 500 400 300 200 100 0 • • • E[U(W)] • • 0 CEW 20000 40000 60000 80000 100000 Return Behavior The next step in developing the theory is to define the behavior of asset prices and returns.5 0.0 0. i = 1. In cell B10 enter =C5. What sure payoff would make you indifferent to the coin flip? That is. . • In cell C11 enter = $D5*C10 + $D6*C12. The prices during and at the end of .23? • In cell G5 enter =($B$1*F5)^(1/$B$1).21 316. That is.

Equation (3. i2 .5) Because the continuously compounded return is equal to Xi(t). i.5) shows that the continuously compounded return equals the logarithm of the gross return. However. then the continuously compounded return is normal and vice versa.6) s 2 i To help distinguish between these two concepts of return. Our assumptions also imply that asset returns may be contemporaneously correlated across assets. A comparison of Equations (3. If the gross return has a lognormal distribution. The gross return is simply the ratio of prices at the beginning and end of the period.5. . This same theorem is responsible for the fact that the utility of expected wealth exceeds the expected utility of wealth in Exhibit 3.62 Chapter 3 Asset Allocation: The Mean–Variance Framework the investment horizon are uncertain and are assumed to have a lognormal distribution. and each has advantages in certain applications. In our model the gross return on the ith asset is exp[Xi(t)]. Under this assumption the expected values of these two measures of return are related by ln ( E[ exp( X i )]) E[ X i ] ai 1 2 1 2 var ( X i ) mi (3.4). Combining this with the assumption that means and variances are constants. Note that Equation (3. measure of return.4) where Xi(t) has a normal distribution with Mean Variance i 2 i . but they are not correlated across time periods. we need to clarify the relationship between continuously compounded returns as defined in Equation (3. perhaps more familiar. a theorem known as Jensen’s Inequality implies that the “log of the expected value” is greater than the “expectation of the log” for any probability distribution. The continuously compounded return on asset i in the t th period is defined as the natural logarithm of the ratio of prices at the beginning and end of the period. Before proceeding with the development of this model.5) and another. the gross return is the exponential of the continuously compounded return. According to Equation (3.4) gives the return as7 ri (t ) ln[ Pi (t ) /Pi (t 1)] X i (t ) (3. we have defined i to be the logarithm of the expected gross return on asset i.4) and (3. it has a normal distribution with mean and variance given for Equation (3.6) holds only when Xi has a normal distribution.6) this is equal to the expected continuously compounded return. plus one-half the variance of the continuously compounded return.8 Each of the return concepts introduced here is useful. So it matters whether we take the expected value first and then the logarithm or first take the logarithm and then the expected value. it follows that we are assuming investment opportunities are the same in every subperiod. Specifically it is assumed that the prices at times t and t 1 are related by Pi (t ) Pi (t 1) exp[ X i (t )] (3. The advantage of working with continuously compounded returns is that multiperiod returns are simply the sum of the returns in each 7 8 See the Appendix I for a review of compounding and multiperiod returns. The random variables Xi(t) and Xi(s) are assumed to be uncorrelated unless t = s. The advantage of gross returns is that the gross return of a portfolio is a simple weighted average of the gross returns of the underlying assets. Conversely. The difference between these two measures of expected return arises because the logarithm is a nonlinear function. This is not true for continuously compounded returns.

In cell E2 replace “COUNT” with “AVERAGE”. Then Ri (T ) ln [ Pi (T ) /Pi (0)] t ln [ Pi (t ) /Pi (t r ( t ). The standard deviation. we need to calculate the net return.Chapter 3 Asset Allocation: The Mean–Variance Framework 63 period. the cumulative (continuously compounded) return over the investment horizon is simply the sum of the (continuously compounded) returns in each of the underlying periods. we will work primarily with continuously compounded returns. Let Ri(T) denote the cumulative continuously compounded return on asset i over the entire T-year investment horizon. Note that the ending value. But we cannot forget about gross returns altogether. So the variance of the T-year return is t 2 i 2 i T (3. the expected return over the T-year horizon is t i i T (3. however.9) The standard deviation is the square root of the variance and is therefore equal to i T. In multiperiod contexts logarithmic returns are the natural choice. In cell E5 replace “COUNT” with “MIN”. Because we want to relate the M–V framework to more general multiperiod problems.xls. Copy cells E1:E5 to cells F1:F5. so we will be careful to specify “gross return” whenever that is the relevant concept. and then copy cell F8 to cells F9:F1004. Because these are index levels. all the covariances are zero. Note that the mean and variance of the continuously compounded return are proportional to the length of the investment horizon. In this case. The relationship among risk.8) Return Variance In general. . equals [P(t 1) + (P(t) – P(t 1))]. • In cell E8 enter = LN(C8/C9) and then copy cell E8 to cells E9:E1004. and investment horizon will be examined more fully later in this section. Use the worksheet <Returns Template> in the spreadsheet Chapter 03 Excel Outboxes. T. In cell E4 replace “COUNT” with “MAX”. 1)] (3. The calculation of continuously compounded returns is given by r(t) = ln[P(t)/P(t 1)]. there is no need to adjust for dividends and splits. P(t). is proportional to the square root of the horizon. The monthly index level for the S&P 500 is given in cells C8:C1005 in reverse chronological order. the gross return is given by = P(t)/P(t 1). the variance of a sum is not equal to the sum of the variances because the covariance between each pair of elements in the sum must be taken into account. however.7) t i So as indicated earlier. Because the mean of a sum is equal to the sum of the means. Therefore in single-period models it is usually best to work with gross returns. To isolate the change in value. • In cell F8 enter = C8/C9 • • • • • • 1. Summary statistics: In cell E1 enter = COUNT(E$8:E$1004) and then copy this to cells E2:E5. return. Excel Outbox Calculating Returns A spreadsheet provides an opportunity to explore these concepts. In cell E3 replace “COUNT” with “STDEV”. defined as [(P(t)/P(t 1) – 1]. As defined in the text. This is not true for gross returns.

the maximum (35.1% 0.] • Notice that the total continuous compound return from the sample [ LN(C$8/C$1004)] is the same as the sum of the monthly continuous compound returns [ SUM(E8:E1004)]. • The monthly = 5.28 percent) one-month returns are nearly identical in absolute value.0% 25.0% 0.$E$3. construct frequency distributions: H 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 45.46% −35.8% 12.3% 33. In cell I19 delete “-NORMDIST ($H19.5% 0.3% 29.0% −30.2% 0.Monthly Data D Obs Mean S.TRUE) and then copy this to cells I4:I19.TRUE)”.0% 0.0% 30. Max Min OBS 1 2 DATE Jan-09 Dec-08 SP500 842.88 percent.0% −25.56% −29.0% −35.$E$2.2% 0.$E$3.0% 100. • The monthly = 0.51 percent.4% 0.0% 0.25 E 997 0.0% 5.3% 28.78% 1 2 3 4 5 6 7 8 9 • There are 997 months in the sample.4% 2.3% 0.0% 0.28% r(t) −6. .2% 14.TRUE)-NORMDIST($H4.0% • In cell I3 enter =NORMDIST($H3. $E$3.0% 0.0% 0. which is the estimated monthly mean of the net return (gross 1) series.0% −20.88% 5.0% 100.4% 0.46 percent) and minimum ( 35.512 = 0.8% 45. Here x is the monthly average net return.0% 0.73% 5.0% 44.0% K Gross-1 0. which is expected for a symmetric distribution such as the normal.0% 0.62 903.4% 0.$E$2.0% 15.78% F 997 0.64 Chapter 3 Asset Allocation: The Mean–Variance Framework Here are the results: A B C S&P 500 Index .56 percent) is considerably higher than the absolute value of the minimum one-month return ( 29.5% 0.0% 35. For r(t).95% 0.0% 0.5% 4.54% 42. and (1 + x) is the monthly average gross return.73 percent. [Note: Strictly speaking we should compare to the log of the mean gross return.0% 20. This is not the case for the net return (gross 1): The maximum one-month return (42.73 percent). Notice that = 0.0% 0.71% 0.73% Gross-1 −6.0% 0.0% 0.7% 1.0% −45.0% I N 0.0% 0.2% 17.4% 0.0% −10.D.0% 12. To examine the shape of the return distributions.0% 0.3% 2.$E$2.5% 7.1% 0.8% 1.1% 0.73 + ½ 5. • The most noticeable difference is the maximum and minimum returns.0% −5.6% 0.1% 0.51% 35.2% 0.0% J r 0.0% 100.2% 7. But LN(1 + x) x for small values of x.7% 28.3% 2.0% 10.0% 0.1% 0.0% −15.

The gross return on the portfolio is then a weighted average of the gross returns on the individual assets. This adjustment is exactly analogous to the relationship shown in Equation (3. they both display positive skew (the empirical distributions are not symmetric) and kurtosis (there are more observations in the tails).10) holds exactly if portfolio weights are continuously rebalanced. that sum to unity (1). With rebalancing at discrete intervals it is only an approximation. Equation (3. Notice that while both the continuously compounded and net (gross 1) return distributions follow the general shape of the normal curve.10) is the weighted average of the underlying asset returns. So the continuously compounded portfolio return is not exactly a weighted average of the continuously compounded asset returns. the expected value of a weighted average is the same as the weighted average of the expected values. • In cells K3:K19 enter as an array =FREQUENCY(F8:F1004. However. Using a standard result from probability. The significance of these deviations from normality will be addressed in various contexts throughout the book.6) we are subtracting it from the left side.13) 9 This approximation is used extensively in Campbell and Viceira (2002).Chapter 3 Asset Allocation: The Mean–Variance Framework 65 • In cells J3:J19 enter =FREQUENCY(E8:E1004. Portfolio Return and Variance Suppose an investor allocates a portfolio according to a set of weights. Looking at Equation (3.11) where we have defined W i i ( i 1 2 2 i ) i i i (3. Together the first two terms capture a weighted average of underlying asset gross returns and are approximately equal to the log of the gross return on the portfolio.10) The first term in Equation (3.13). ( 2 T ). The third term then adjusts the mean of the portfolio return downward by one-half the variance of the portfolio return. Applying a standard result for the variance of a weighted average gives 2 W T [ i j i j ij ]T (3. it is approximately equal to that sum plus volatility adjustments. The continuously compounded portfolio return is approximately9 RW ( T) i i Ri (T ) 1 2 i i 2 i T 1 2 2 W T (3. let W (for wealth) denote the portfolio. which W will be defined in Equation (3. . Therefore. we see that this adjusts the mean of each return upward to the log of its expected gross return. Next we need to obtain the mean and variance of the portfolio return.H3:H19)/$E$1 and then <SHIFT> <CTRL><ENTER> to enter this as a range array formula. Given our assumption that gross asset returns are lognormal.H3:H19)/F$1. Unfortunately the logarithm of a sum is not equal to the sum of logarithms. i.6). To avoid confusion with asset prices.6) for individual asset returns except that instead of adding the variance term to the right side of Equation (3.10) the expected portfolio return is i i i T 1 2 i i 2 i T 1 2 2 W T ( W 1 2 2 W )T (3. The second term adjusts each of the underlying asset returns upward by one-half its variance.12) The variance of the portfolio return is more complex because it must take into account not only each asset’s own variance but also the covariance between each pair of assets. The second and third terms are volatility adjustments that adjust the level of the portfolio return for the difference in expected value between gross returns and continuously compounded returns. from Equation (3.

16) The exponent factor may be eliminated in the problem because maximizing Equation (3. . i. it follows that an investor with CRRA utility will select portfolio weights to maximize E{(1/ ) exp[ RW (T )]} (3. For i = j the covariance is equal to the variance of asset i.15) Because RW(T) displays a normal distribution.14) at the end of the investment horizon. reflects the penalty the investor assigns to an increase in the portfolio variance.17) where.11). Formally. Constraints The M–V model stipulates that the investor will trade off additional expected return against additional portfolio variance at a constant rate.16) is equivalent to maximizing ( 2 W W )T [ 1 W 2 (1 ) 2 W ]T (3. For example.3). the expression in brackets in Equation (3.10 To see this. the mean component would no longer be linear in the portfolio weights. The difference between these two measures has been subsumed into the variance component of the objective function. Substituting this for wealth in the constant relative risk aversion utility function (Equation (3. Also note that W is actually the log of the expected gross return on the portfolio rather than the expected continuously compounded return. Note that W is linear in the portfolio weights whereas the variance is quadratic because it involves squares and cross products of the weights.15) has a lognormal distribution. 10 1 2 1 2 . we have defined ½ (1 ). The risk aversion coefficient. The goal is to maximize the expected utility of final wealth. the investor’s decision problem is to select portfolio weights. to maximize Equation (3. look back at Equation (3. If the investor puts $1 of wealth into the portfolio at time t = 0. i i = 1. asset allocation. The mean of this distribution is given by (1/ ) exp[ ( W T 1 2 (1 ) 2 W T )] (3. but that has been translated to expected return less a weighted variance term. Objective Function Now we are ready to consider the growth of wealth.17) subject to the budget constraint that the portfolio weights sum to unity—that is. it will grow to exp[ RW (T )] (3. It is often desirable to add additional constraints to the basic M–V model.66 Chapter 3 Asset Allocation: The Mean–Variance Framework where i j denotes the covariance between the returns on assets i and j. But there would no longer be a clear distinction between the “mean” and “variance” components of the objective because the portfolio variance would appear in both components. This is the objective function for the basic M–V model. In addition. for simplicity. the investor may be prohibited from taking a short position in some or all ] so that the “mean 2 W ) 2 We could regroup the terms in the objective as [( W W term” is the expected log return (see Equation (3. and the setting of asset weights.11).

. the objective function is recast in terms of the mean and variance of return per period rather than of cumulative return (wealth). Therefore the probability of a return less than or equal to a threshold H is given by the cumulative normal distribution evaluated at H. Although the advice is probably sound—most practitioners take it for granted—the arguments advanced to support it are often not well grounded. . the investor may want to limit the probability of earning a return below some threshold level.17)) is likewise proportional to T.i T 1. 2 W T) K (3. the constraint would be given by N (H . Similarly. ( )T . does not change this conclusion. Imposing constraints on the portfolio weights. 2 W T) K for some H and K This problem is easily solved by standard optimization packages such as the “Solver” add-in in Microsoft Excel. Assuming the investor wants to limit this probably to at most K. This long-only constraint means that for each such asset the solution must satisfy i 0.18) where N denotes the cumulative normal function. we can write the more general problem compactly as Choose Maximize Subject to Budget constraint: Long-only constraint: Shortfall constraint: i i i i . It is therefore useful to examine this issue more closely. The optimal portfolio allocation must therefore be the same for all horizons. Under our assumptions. The investor’s objective function (Equation (3. return per period is simply the cumulative . we showed that both the mean and variance of portfolio returns are proportional to the investment horizon T. as a result. Bringing the pieces together. For continuously compounded returns. it could simply be eliminated without changing the problem.18) is usually referred to as a shortfall constraint. . ( 1 W 2 2 W ) T. Because the horizon is just a scale factor in the objective function. Therefore the basic M–V model does not support the notion that asset allocation should depend on the investment horizon. . such as the long-only constraint. Investment Horizon It is often argued that stocks are less risky over long investment horizons and. that investors should allocate a higher proportion of their portfolios to stocks the longer their investment horizon. In many applications of the M–V model. Under the assumption that (continuously compounded) asset returns have the same normal distribution in each subperiod and are uncorrelated over time. Equation (3. the continuously compounded return has a normal distribution.Chapter 3 Asset Allocation: The Mean–Variance Framework 67 of the available assets. n to 2 W W T 1 0 W 1 2 2 W N (H . We have just derived the basic M–V model from first principles: expected utility of wealth maximization and careful treatment of investment opportunities.

17) shows that the only difference is that the risk aversion parameter has been divided by the time horizon. Indeed it appears to drop quickly. A link between asset allocation and investment horizon can be introduced within the M–V framework by adding a shortfall constraint.20) is used—often in graphical form—as “proof ” that risk declines sharply as the investment horizon is extended. Given that assumption it should be no surprise that the investor selects a riskier portfolio over longer horizons—but this is not because the assets are less risky! It is not the risk that is declining. it gives a distorted view of the relationship between risk and horizon. Focusing on return per period is an expedient way to make asset allocation depend on the time horizon. . But what is really going on here? Comparing Equation (3. From this perspective risk does appear to decline with longer horizons. The secret is that the total risk is being spread across more and more periods. the variance of wealth—is actually rising but not fast enough to offset the effect of spreading the risk across more periods. but the variance of the return per period declines as the investment horizon increases.68 Chapter 3 Asset Allocation: The Mean–Variance Framework return divided by the investment horizon. it is the investor’s aversion to risk.13) give the mean return per period for the portfolio as (1/T )( while the variance is (1/T )2 ( 2 W W 1 2 2 W )T W 1 2 2 W (3.19) rather than mechanically replacing ( W T) with W. Notice that total risk (variance) increases at a rate of T while the per-period risk decreases at a rate 1/T due to the (1/T)2 scale factor. Unfortunately that is not how the story is usually told. Thus a more solid foundation can be built for the notion that risk.21) It should be clear that the optimal portfolio allocation will depend on the horizon if we maximize this objective instead of the original objective. the mean–variance objective function becomes11 W 1 2 ⎛1 ⎝ ⎞ T⎠ 2 W (3.11) and (3. Total risk—that is. If we replace the mean and variance of wealth with the mean and variance of return per period. however.21) with the right side of Equation (3. Instead Equation (3. As we have shown. and a 10-year investment looks only one-10th as risky. Thus using the mean and variance of return per period in the M–V framework is equivalent to assuming that risk aversion is lower with a longer investment horizon. In Chapter 5 we will relax the assumption that investment opportunities are the same in each period.19) T) 2 W /T (3.20) The expected return per period is constant. Equations (3. In doing so we are relaxing 11 Note that we are being careful here to replace the mean of the continuously compounded return with the corresponding per-period value from Equation (3. Without that assumption some assets may indeed be more or less risky over longer horizons. and asset allocation. for example. depends on the investor’s time horizon. a two-year investment appears to be only half as risky as a one-year investment.

Chapter 3 Asset Allocation: The Mean–Variance Framework 69 our assumption that the utility function fully describes the investor’s attitude toward wealth. and extending the horizon beyond that point will not affect the investor’s asset allocation. Moreover. . As a result. Consider the portfolio that the investor would select in the absence of the shortfall constraint. The Excel Outboxes will illustrate how to build and solve the problem within an Excel spreadsheet. There will be only two constraints: the long-only constraint and the budget constraint.22) where the left side is the probability of (log) wealth less than or equal to H according to the standard normal distribution. its strengths. This section begins with the development of a two–asset class problem. the constraint will no longer be binding. This simple application will illustrate the efficient frontier and the optimal weights for the investor. For example. As inputs are varied by estimating the mean–variance parameters over different periods. The next section explores the model with a series of progressively realistic applications. such as stocks and bonds. applying historical variance. The template of the spreadsheet is provided online. this probability declines as T increases. and mean returns within a two–asset class model with few constraints can provide varying optimal mixes as the modeler varies the risk aversion coefficient. as the number of assets and constraints are increased. It is your job to fill in the data and formulas. The shortfall constraint in Equation (3. where the investor seeks to diversify. In particular. The user will begin to understand that in the real world. or corner solutions. covariance. Beyond some horizon. applying the M–V framework to live examples. weaknesses. If it violates the constraint at short horizons.18) can be rewritten as ⎛H N⎜ ⎝ ( W W 1 2 2 W T )T ⎞ ⎟ ⎠ K (3. it will become clear that optimal mixes depend on the historical period chosen.3 Practice: Solution of Stylized Problems Using the Mean–Variance Framework Asset allocation modeling is best learned through doing—that is. optimal solutions may become unattainable. the investor is assumed to have strong feelings about falling below some threshold level of wealth. Holding everything else constant. At this point the modeler will be interested in the alternative methods for exploring asset allocation examined in Chapter 5. however. dependent entirely on how the constraints are formulated. While it is important to understand the conceptual foundations of the M–V framework. 3. and investment implications can be fully understood only by applying it. a portfolio that violates the constraint at a short horizon may satisfy it at a longer horizon. the investor will pick the best portfolio that does satisfy the constraint. unstable. Through this experience. the reader should understand the power and limitations of the model. results may be highly sensitive to forecasts.

70 Chapter 3 Asset Allocation: The Mean–Variance Framework Following the two–asset class example. However. This translates into the program (where n = number of risky assets and W = portfolio): Minimize by choosing subject to where W i i W W W i 1. then setting the weight of the second asset using the budget constraint. If there are only two risky assets.B ) 1 2 .n i j i j i. we will extend the model to more asset classes. The M-V frontier represents the portfolios available to and appealing to the investor. the goal is to identify the portfolio with the risk that best fits with their preferences. which is the set of portfolios with the lowest risk for a given level of return. Here we will explore a more general approach that may be applied to n > 2 assets. consider including a shortfall constraint. there are too many possible asset weight combinations to consider for the method to be efficient.n i j i 1. the goal is to: Minimize by choosing subject to the portfolio standard deviation the weights of the risky assets a fully invested portfolio (the budget constraint) for a given level of return. Second. we can easily identify this set of portfolios by examining a range of potential weights for one asset. Once the investor identifies the efficient frontier. The Efficient Frontier Consider an investor who seeks to determine the optimal balance between stocks and bonds. We will present the problem in words and then translate it into equations.j ( )½ For the two risky assets Stocks (S) and Bonds (B). it is not clear which asset weight combinations best satisfy the assumptions of investor preferences. and finish with a review of an asset–liability problem. we have the following: Minimize by choosing subject to where S W S and B B + 1 + W W W Target S S 2 S B 2 B B 2 B ( 2 S 2 S B S B S . explore the importance of time horizons. To trace the M–V frontier. the portfolio with the highest return for a given level of risk. this technique is not feasible with more than two assets. Given the assumptions about investor preferences.n i 1 Target 1. First. and finally illustrate it in the Excel Outboxes. From this set the efficient frontier. The first task is to create the M–V frontier. these portfolios have the lowest level of risk for a given return. 1 + 2 = 1.n i 1. is identified.

. all attractive portfolios are included on the efficient frontier. reflecting portfolios offering similar risk–return profiles. the portfolios with the highest return for a given level of risk. In other words. The frontier may be plotted as a line. the plot is typically a curved line. This is because the portfolio can take more advantage of the diversifying assets. There are several key observations regarding the frontier: .2500 0.1500 • 0.6 Efficient Frontier. • The ability to sell short will improve the risk–return characteristics of the efficient frontier. this is illustrated in Exhibit 3.3000 0. may lead to similar risk and return values.0500 Risk • Bond • Stock • MVP This program can be solved using an optimization tool such as Excel’s Solver (explained in the Appendix II). the portfolio with the lowest risk across all levels of return. The Optimal Portfolio Excel Outbox describes how to identify the particular portfolio the investor would select given his or her level of risk aversion.1000 0. without the long-only requirement. Two Asset Portfolios of Varying Weights Return 0. The idea is to illustrate the fact that many portfolios.0000 0. The Efficient Frontier Excel Outbox describes how to identify the portfolios the investor could select—that is.0500 0. the MVP lies at the leftmost point of the curve.0500 • • 0. Portfolios where MVP < MVP would never be held by a rational investor because there is another Target portfolio that offers a higher return for the same level of risk.1500 0. Note that eliminating the ability to sell short requires the addition of the long-only constraint ( i 0) for those assets that cannot be sold short. Templates are provided online. some with very different weights. • The efficient frontier consists of the portfolios where Target MVP • Rational investors will only hold a portfolio where Target . 12 Some practitioners present the efficient frontier as a fat line composed of many points.0000 0. The spreadsheets described in the following Excel Outboxes explore the creation and selection of portfolios of just two risky assets.1000 0.6. In the graph. For diversifying assets.3500 0. defined by the lowest level of risk for each level of return.2000 0. as a warning to the user to avoid assuming false precision in practicing mean–variance techniques.4000 −0.12 The frontier has essentially two regions defined by the portfolio’s expected return: Region one: Region two: Target MVP MVP < Target Here MVP is the return on the minimum variance portfolio (MVP).Chapter 3 Asset Allocation: The Mean–Variance Framework 71 EXHIBIT 3.

click on cell F25. For background on how Solver optimizes the problem. To enter the annual equivalent in cell F25. the annual log return for the SP500 is located in cell I6 of the “Data” worksheet. Use the following table to enter the remaining summary statistics: Column F SP500 =Data!l6 =Data!l7 =Data!l11 C Row 23 24 25 26 29 30 μ α σ ρ G Tbond =Data!L6 =Data!L7 =Data!I14 =Data!L14 = +F25+0. see Appendix II.xls. For example.Template> in the spreadsheet Chapter 03 Excel Outboxes. Use the worksheet <EF . Each construction step will be explained in the text. In this example the assets are the Treasury bond (Tbond) and S&P 500 (SP500) indexes as the bond and stock respectively. type “ = ”.72 Chapter 3 Asset Allocation: The Mean–Variance Framework Excel Outbox Efficient Frontier We will construct an Excel spreadsheet to demonstrate the application of the above equations.5∗F26^2 = +G25+0. click the “Data” tab.5∗G26^2 . The template appears as follows: A 1 2 3 4 5 6 7 8 Program 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 Data 24 25 26 27 28 29 30 B C D Mean-Variance Efficient Frontier Two risky assets No riskless asset Short sales E F G Objective MIN σω = Variables ω Constraints ∑ω μω = = Portfolio Stock Bond = = Where μω σω Statistics μ α σ ρ SP500 Tbond = = SP500 Tbond Data The first step is to select the two risky assets and enter their summary statistics into the Data range of the worksheet. You should see =Data!I6 appear in the cell. The summary statistics are available on the “Data” worksheet in the spreadsheet. and then click cell I6.

S B ($ Stocks $ Bonds)/($ Stocks $ Bonds) S $10.5 Constraints This program has two constraints: The portfolio has to be fully invested. entered in cells F12 and G12. and the target value. constraint: calculated value. Objective Because the goal is to minimize the portfolio standard deviation. and $G$29 reference the standard deviations and correlation of the SP500 and Tbond respectively. $G$26. 000/$10. the formulas for the portfolio return and standard deviation are entered into cells G20 and G21 respectively.Chapter 3 Asset Allocation: The Mean–Variance Framework 73 Where For formatting and presentation ease. so there is no formula to enter. if you had $10. you cannot leave these cells empty because Solver needs a starting point. the weights for the portfolio are S B $ Stocks/($ Stocks $ Bonds/($ Stocks $ Bonds) $ Bonds) $6.40 Notice that because all the money is invested in stocks and bonds. Start with an equal weighting for each asset by entering ½ in each cell: 0 . 000/$10. A constraint has three elements: the calculated value. 000 $4.B ) 12 and is entered into cell G21 as follows: SQRT (F12∧ 2 ∗$F$26∧ 2 G12∧ 2 ∗ $G$26∧ 2 2 ∗ F12 ∗$G$12 ∗ F26 ∗$G$26 ∗$G$29) Here cells $F$26. The “$” are used so we can copy this equation to other cells without changing the cell references. The portfolio return formula is W S S B B and is entered into cell G20 as (F12 ∗$F$24) (G12 ∗$G$24) where cells F12 and G12 are the weights and cells $F$24 and $G$24 are the returns on the SP500 and TBonds respectively. These are the solution. 000/$10. The sum of the weights. 000 . is entered into cell E15: + B = 1. or full investment. However. and we are interested in solving the program for a particular level of return.60 0. the SUM(F12 : G12) . 1. the objective cell simply references the standard deviation formula entered in cell G21: G21 Variables The decision variables. 000 0.000 in SP500 and $4. The portfolio standard deviation formula is W ( 2 S 2 S 2 B 2 B 2 S B S B S .000 to invest and put $6. are the portfolio weights for SP500 and Tbond. Budget Constraint Weights represent the proportion of the total investment held in each asset: i Investment in asset i /Total investment For example.000 in Tbond. the inequality or equality condition.00 This is the budget.

This is the parameter that will change as we repeatedly solve for different points along the frontier.0520 0.5000 = = 1.1184 0.0000 0.0966 Stock 0.0761 Statistics μ α σ ρ SP500 Tbond 1.0600 Where μω σω = = SP500 0.1128 Tbond 0. entered into cell G16. Start with a value of 6 percent: 0.1384 0.06 The worksheet is now complete and should appear as follows: A 1 2 3 4 5 6 7 8 Program 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 Data 24 25 26 27 28 29 30 B C D Mean-Variance Efficient Frontier Two risky assets No riskless asset Short sales E F G Objective MIN σω = 0.0549 0.1712 1.0000 .0000 0.74 Chapter 3 Asset Allocation: The Mean–Variance Framework The target value.1128 Variables ω Constraints Σω μω = = Portfolio 1. is the return of interest.1997 0.0966 0. or 1.0000 0. so enter this into E16: G20 The target value. The calculated value is the portfolio return for the given weights. Target Return The target return constraint determines the level of return examined. is entered into cell G15: 1.5000 Bond 0. This is the formula already entered into cell G20 and is referenced in cell E16.0000 The inequality is entered when the program is entered into Solver.

9212 1.2325 0.7: I 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 Target μ 0. Note that the MVP.0614 Portfolio σ 0. The results are listed in the following table.0000 0.0300 Stock Bond MRP Stock 1.1013 0.0162 0.1779 1.1393 0.0000 0.1384 0.0745 0. Copy the weights.0572 0.9212 μ 0.Chapter 3 Asset Allocation: The Mean–Variance Framework 75 We will calculate the efficient frontier by using Solver to minimize portfolio risk for a target expected return.4208 0.0400 0.0000 −0. portfolio return.xls.1200 0. and the M-V frontier is charted in Exhibit 3.0000 1.0000 −0.0300 0.0000 0. is computed by deleting the return target constraint.7800 1. and constraints should include the values in rows 15 and 16.1500 0. or the minimum variance portfolio.0000 1.0400 0.1393 0.1595 0.0549 0.0744 0.0761 0.1200 0.1997 0.0616 0.2259 0.6569 −1.0549 0.0000 0.0300 0.1062 0.5792 0.0906 0.0561 0.2977 −0.0162 1.1384 0.1656 0. The Solver window should include G8 as the objective to Min.0761 0.0900 0. as follows: This optimization should be repeated for the range of return values listed in the worksheet in column I.1384 0.0600 MRP 0.0788 0. The completed template is available online in the worksheet <EF-Completed> in the spreadsheet Chapter 03 Excel Outboxes.0744 J K L M .0300 0.0549 0.0000 −0.2977 1.6569 2.0900 0.0000 0.2200 0.1779 −0.0788 ω Bond −0. and risk into the columns to the right of column I.9384 0.1500 0.1997 0.

which portfolio should the investor select? Stated as an optimization program. given the assumptions about investor preferences.2000 0.1000 0.n i = 1.0800 Return 0.0600 0.7 Efficient Frontier for Excel Outbox 0.1500 0.n W W i =1 i i j = 1.76 Chapter 3 Asset Allocation: The Mean–Variance Framework EXHIBIT 3.0200 0.2500 MVP • SP500 • • Tbond The Optimal Portfolio Having identified the most efficient portfolios available to the investor. the goal becomes to identify the portfolio the investor should hold given his or her assumed risk preferences. we wish to: Maximize by choosing subject to Maximize by choosing subject to where the investor’s utility given his or her level of risk aversion the weights of the risky assets a fully invested portfolio.n )½ Notice that the target return constraint has been removed. Instead of identifying the portfolios the investor could hold.1400 0.0200 −0. We can simplify these equations for the two risky assets Stocks (S) and Bonds (B): Maximize by choosing subject to where W 2 W S . = B B S + S W W =1 S 2 S + 2 S B B 2 B 2 B ( 2 S B S B S .0000 −0. 2 W This translates into the equations: W i i = 1. the next question is.0400 0.B ) 1 2 The optimization solution represents the asset weights of the portfolio offering the highest utility for the given level of risk aversion.1600 0.j = =( i = 1.0400 Risk 0.1200 0.0500 0. The Optimal Portfolio Excel Outbox .n i j i j i.0000 0.1000 0.

Chapter 3 Asset Allocation: The Mean–Variance Framework 77 presents a template and provides steps for computing the optimal portfolio for a given level of risk aversion for two risky assets. The sections not covered can be completed using the instructions describing the <Efficient Frontier> worksheet. The utility function is entered into cell G8 as G19 D9 ∗ G20∧ 2 Note that the optimization should be set to maximize cell G8 Max. The template should appear as follows: A 1 2 3 4 5 6 7 8 Program 9 10 11 12 13 14 15 16 17 18 19 20 21 22 Data 23 24 25 26 27 28 29 30 B C D Mean-Variance Optimal Portfolio Two risky assets No riskless asset Short sales E F G Objective MAX λ E [μ] − λσ2 = Variables ω Constraints ∑ω = Portfolio Stock Bond = Where μω σω Statistics μ α σ ρ SP500 Tbond = = SP500 Tbond The objective is to maximize utility. Start with a value of 1.00: 1 Constraint Because we have already identified the portfolios that the investor could invest in. the target return constraint is no longer needed. xls. This is the parameter that will change as we solve for different portfolios along the frontier. The description covers the additions or changes made to the complete version of the previously presented “Efficient Frontier” program. which should appear as follows: . The assumed risk aversion coefficient is entered into cell D9. Delete it from the Solver window. Use the worksheet <OP – Template> in the spreadsheet Chapter 03 Excel Outboxes. Excel Outbox Optimal Portfolio This box reviews a simple spreadsheet that calculates the optimal weights for a portfolio of two risky assets.

the worksheet is complete and should appear as follows: A 1 2 3 4 5 6 7 8 Program 9 10 11 12 13 14 15 16 17 18 19 20 21 22 Data 23 24 25 26 27 28 29 30 B C D E F G Mean-Variance Optimal Portfolio Two risky assets No riskless asset Short sales Objective MAX λ E [m] .0839 Variables ω Constraints ∑ω = Portfolio 1.0761 Statistics μ α σ ρ SP500 Tbond 1. .0000 0.5000 = 1. you can compute optimal portfolios using different risk aversion parameters.0966 0.1997 0.1128 Tbond 0.0520 0.λσ2 1 = 0.0000 Stock 0.5000 Bond 0.1384 0.0000 Where μω σω = = SP500 0. Once you verify your answers.0000 The solution is included online in worksheet <OP _ Completed> in the spreadsheet Chapter 03 Excel Outboxes.5 for the weights and use the same formulas for the Where and Data sections as in the efficient frontier exercise.xls.78 Chapter 3 Asset Allocation: The Mean–Variance Framework After we enter 0.1184 0.1712 1.0549 0.

the 30-year return = 30 1 . the optimal portfolio allocation is the same for all investment horizons. that is. this gives a distorted view of the relationship between risk and horizon. we can calculate the annualized return . n # 0 T (T where W W i i j i j i.year return because it is a log-return. requiring some weights i to be greater than or equal to zero.Chapter 3 Asset Allocation: The Mean–Variance Framework 79 Investment Horizons The optimization program can be modified to include the possibility of extending the investment horizon to a longer period. The program is solved for any investment horizon T. and the solution represents the asset weights of the optimal portfolio (the highest utility for the given risk aversion coefficient over the investment horizon). will also be incorporated. • The asset weights are the same for all investment horizons. risk appears to decline with longer horizons. However. This translates into the equations: Maximize by choosing subject to W 2 W i i = 1. 2. that is. as illustrated here (the changes are shown in boldface): Maximize by choosing subject to where the investor’s utility given her or his level of risk aversion the weights of the n risky assets the budget constraint the long-only constraint the portfolio return is the weighted return over horizon T the portfolio variance is the weighted covariance over horizon T. There are two important results concerning the investment horizon in the basic M–V framework: 1.n ) 1 2 For this program the long-only constraint is applied only to designated assets—those that cannot be sold short (the symbol denotes “for all”). The key implications of this investment problem are: • The portfolio return increases in proportion to the investment horizon. we need to modify the program to include extended risk and return horizons. j i 1. • The portfolio standard deviation increases in proportion to the square root of the investment horizon. as we discussed in Section 3. If we include a variable investment horizon. If the program is recast in terms of the mean and variance of return per period.year standard deviation. n i j 1.n i i 1 designated i i 1. To illustrate how recasting the program into the mean and variance of return per period distorts the relationship between risk and horizon.2. The long-only constraint. the 30-year standard deviation = 30 1 . If investors maximize total holding period utility and the asset return distributions are the same in each subperiod and uncorrelated over time.

00% 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 −10.11 or $0. By definition.23) where W(1) is the one-year portfolio return. Suppose we start with $1 and our log return is either +1 percent per year or 1 percent per year for T years with equal probability. . W (1) ⎞ ⎟ T ⎠ (3.8 Horizon Return Confidence Interval 50.00% 40. As a result. As explained in the previous section.99. it appears that risk is declining with the horizon. which. is also the mean. That is. the per-period return will fall between these values with 90 percent confidence.00% 0. however.37. because the normal distribution is symmetric. the probability of realizing a return less than the target declines. A simple example should clarify the situation. after 10 years either $1.00% −20.20).5 2 W (1)B. W(1) is the one-year portfolio standard deviation.19) and (3.91. Exhibit 3.00% 20. and after 100 years either $2. but the 90 percent confidence interval narrows as the investment horizon increases. Note that the mean return is the same at every horizon. the mean return per period is zero and the standard deviation of perperiod return is 1 percent.8 illustrates the 5th and 95th percentiles of this distribution for horizons out to 31 years. The exhibit also shows the median (50th percentile) of the distribution.00% Return 10.00% −30.01 or $0. the standard deviation of final wealth increases with the horizon—but not fast enough to offset the effect of spreading the risk across more periods. The point is that 1 percent per period has a huge impact over a long horizon.00% 30. So a narrower distribution of per-period returns at longer horizons does not indicate low or even declining risk. the probability of exceeding a specified target return per period increases with the horizon if the target return is less than the expected return per period. For each horizon the per-period distribution of log return is given by13 ⎛ RW (T ) ∼ N⎜ T ⎝ W (1) .00% Horizon Mean 95% Quartile 5% Quartile Tbill for each investment horizon along with statistical confidence intervals. Conversely. On the other hand. 13 The mean and variance of the per-period log return were given in Equations (3.71 or $0. and T is the investment horizon.80 Chapter 3 Asset Allocation: The Mean–Variance Framework EXHIBIT 3. After 1 year we have either $1.

Excel Outbox Five-Asset Efficient Frontier The previous Excel Outboxes illustrated the application of mean–variance theory for two assets. Although the portfolio return formula could be entered directly as the sum of the weights times the return for each asset and the standard deviation formula as the sum of the weights times the covariance for . including Treasury bills.Chapter 3 Asset Allocation: The Mean–Variance Framework 81 For example. The template should appear as follows: E F G H I A 1 2 3 4 5 6 7 Program 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 Date 23 24 25 26 27 28 29 30 31 32 33 B C D Mean-Variance Efficient Frontier Two risky assets No riskless asset Short sales Objective MIN σω = Variables ω Corp SP500 SP600 Tbill Tbond Constraints ∑ω μω Where = = Portfolio = = μω σω Statistics μ α σ ρ Corp SP500 SP600 Tbill Tbond Corp SP500 SP600 = = Tbill Tbond Where The changes presented here extend the formulas for portfolio return and standard deviation to work easily with more than two asset classes. Treasury bonds.8 this probability is virtually zero at 30 years. xls. We will build on this idea soon when we introduce shortfall constraints. This box expands the number of asset classes to five. assuming the expected return on the portfolio exceeds the riskless rate. the S&P 500 index (large-cap stocks). In Exhibit 3. Use the worksheet <5AEF-Template> in the spreadsheet Chapter 03 Excel Outboxes. But first we need to allow for a broader set of asset classes. the probability of underperforming T-bills declines as the horizon increases. corporate bonds. and the S&P 600 index (small-cap stocks).

Otherwise the cell will display the error message #VALUE! The completed worksheet should appear as follows: A 7 Program 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 Date 23 24 25 26 27 28 29 30 31 32 33 B Objective C MIN D E F σω = G 0. SUMPRODUCT (F11: J11 F23 : J23) where cells F11:J11 are the cell range containing the weights.0761 Statistics μ α σ ρ Corp SP500 SP600 Tbill Tbond 1.1062 Tbond 0.1071 1. TRANSPOSE(F11 : J11)))} .1062 H I Variables ω Corp 0.2000 SP500 0.0995 Tbill 0.0568 0.0363 0. The resulting formula will appear bracketed: { SQRT (MMULT (MMULT (F11: J11 Data!H17 : L 21).0679 SP500 0.0520 0.2000 Tbond 0.1784 0. They may especially dislike losses of any kind. in addition to disliking overall return volatility.8364 0.82 Chapter 3 Asset Allocation: The Mean–Variance Framework each asset pairing. where cells Data!H17:L21 reference the cell range of the variance covariance matrix on the “Data” worksheet.2000 SP600 0.0256 1.0549 0.0000 0.) After entering the formula.0995 0.0591 0.2000 Constraints ∑ω μω ω = = > = = 1. may have a different view of risk than that assumed by the standard utility function. type <CTRL><SHIFT><ENTER> rather than simply <ENTER>.0000 0.0250 0 Where μω σω Corp 0. or the likelihood of negative returns.2000 Portfolio 1.8585 1.1184 0. TRANSPOSE(F11 : J11))) . particularly individuals.0000 0.1619 0.1712 0. Completed> in the The Shortfall Constraint Investors.0857 –0. this is clumsy and becomes impractical for large portfolios. (Note: This is an array formula.2090 0.1997 SP600 0. Instead we will use Excel matrix formulas.0000 0. The portfolio return formula is entered into cell J19 as .0000 The completed worksheet is provided online in worksheet <5AEF spreadsheet Chapter 03 Excel Outboxes. The portfolio standard deviation formula is entered into cell J20 as SQRT (MMULT (MMULT (F11 : J11 Data!H17 : L 21).3067 = = Tbill 0.1384 0.0000 0.0000 0.0093 0.2225 1.xls. One simple .1071 0.0161 –0.0363 0. and cells F23:J23 reference the cell range of the expected asset returns.

Asset–Liability Management In many instances the investor’s objective is to fund a set of cash liability payments.n i 1 designated i 1 W 0 N ( H . The probability of a return less than or equal to a threshold H is given by the cumulative normal distribution evaluated at H. by choosing the weights of the n risky assets This translates into the equations: Maximize by choosing subject to W 2 W i i i 1. 2 W T) K (3. the solution to the shortfall problem will provide the highest possible utility as long as the program’s constraints are met.2. then this will also be true at any longer horizon. The pension plan promises future payments to its beneficiaries based on salary. If the optimal portfolio is the same with or without the shortfall constraint for a given horizon. Defined benefit pension plans are a classic example of this type of problem. using the shortfall scenario is one way to make asset allocation depend on investment horizons. The key implications of this problem are: • The asset weights of the optimal portfolio change for the investment horizons where the shortfall constraint is binding.24) where N denotes the cumulative normal function. the probability of earning a return below H will be no higher than K. However. The program for computing the optimal portfolio weights now becomes: Maximize subject to the investor’s utility given her or his level of risk aversion the budget constraint the long-only constraint the probability of a return less than or equal to H being at most K where the portfolio return is the weighted return over horizon T the portfolio variance is the weighted covariance over horizon T.( where W W 2 2 sW ) T . The shortfall constraint can be used to construct a portfolio that limits the probability of earning a return below some threshold level. As a result.n i j i 1. 2 W T) K T (T i i 1. years of service. the less risky the optimal portfolio. in this case the optimal portfolio offers an additional feature.Chapter 3 Asset Allocation: The Mean–Variance Framework 83 approach for incorporating this preference is to add the shortfall constraint introduced in Section 3. For example.n 1. ( 1 W 2 2 W ) T. The optimal portfolio is the same for all these horizons. the shorter the time horizon. . • The asset weights are the same for the investment horizons where the shortfall constraint is not binding.n i j i j i.j )½ Just as in the previous case. The constraint to limit this probability to be at most K would be given by N (H .

n 2 j 1. L i j 2 W T) K =T [T ( i = 1.n i L i L i. • We need to be careful in interpreting risk and return. The weight on the liability is not a choice variable. For large. the plan’s liability can be viewed as a known stream of future payments.L 2 L 2 L )] 1 2 The key implications of this problem are: • When the liability is added into the problem. In essence. The return here is actually the percentage change in the ratio of the plan’s surplus (Assets – Liability) to its assets. the beneficiaries have already earned the right to receive them. it is dictated by the value of the liability relative to the value of the plan’s assets.84 Chapter 3 Asset Allocation: The Mean–Variance Framework and other aspects of their employment. a large portion of the future payments can be projected with a high degree of confidence because they are fully vested—that is.n i i + L i j i.n i 1 designated i (liabilities/assets) 1 W 0 L N (H . and the main driver of its value will be the level of interest rates. The program is: Maximize by choosing subject to the investor’s utility given their level of risk aversion the weights of the n risky assets the budget constraint the long-only constraint the probability of a return less than or equal to H being at most K where the portfolio return is the weighted return over horizon T the portfolio variance is the weighted covariance over horizon T. Projecting and valuing the plan’s liability are difficult tasks because the benefit formulas are complex and the final payments are contingent on future events pertaining to the plan sponsor and its employees. This translates into the equations: Maximize by choosing subject to W 2 W i i i 1. Bonds act as hedging assets. the pension plan is short a long-term bond. we now have the constraint that the liability weight is fixed. The portfolio now includes this liability. Therefore. ( where W W 2 2 W ) T. the optimal solution reflects lower stock weights and higher bond weights. its movements are highly correlated with the bond asset classes. The liability is added to the portfolio with a negative weight because it is effectively a short position. well-established plans. j j 1. In addition to the budget constraint that requires the asset weights to sum to 1. and the optimal portfolio includes a heavier allocation to these asset classes than was the case in the absence of the liability. while risk is now the variance of these changes. therefore. at least to a first approximation.n i 1. . The value of the liability is assumed to be impacted primarily by interest rates. however.

T 1.645 T) . According to the normal distribution. This is discussed in later chapters.075 0. the 90 percent confidence intervals for wealth at various horizons are shown in the second and third columns of the following table.375 0. the range from X to Y is a Z percent confidence interval.288 Wealth Lower Bound 0. Note that while the lower confidence bound declines initially and then rises gradually. the logarithm of wealth at the end of T periods has a normal distribution with mean ( T) and variance ( 2T).1 11. the distribution does not spread out evenly above and below the mean—it is highly skewed to the upside.504 Probability of Loss 35. we simply convert from log wealth to wealth using the exponential function.645 standard deviations of the mean. the path along the way may be volatile.748 1. Horizon (T) 1 5 10 30 Log Wealth Mean 0. Moreover.8 34.e( T 1. In the model developed here.4 Note that interim drawdowns can be difficult for investors to endure emotionally and can be a serious financial problem if they coincide with periods of negative cash flow. So the 90 percent confidence interval for wealth is [e ( T 1.095. Applying the same parameters of mean returns and standard deviation.8 2.8% 27.645 T) ] To illustrate. this reflects the strong upward skew in the lognormal distribution.036 5. Note that introducing positive serial correlation in the return distribution would lead to higher long-term variance estimates and correspondingly wider confidence levels. bad results are not impossible. letting = . As a result. While the probability of loss over long periods may seem small.4% 20. 90 percent of the outcomes lie within 1.645 T. The Probability of Being Down 25% or 50% Horizon 1 5 10 20 30 25% Draw down 8. implying that = .20.991 57.2 7.565 Upper Bound 1. the risk of experiencing a large drawdown before the horizon is quite large. the upper confidence bound rises rapidly and at an accelerating rate.8 32.697 0. The confidence interval is not symmetric around mean wealth—the lower confidence bound is much closer to the mean than is the upper confidence bound. assuming we start with wealth of $1.2 33.0 50% Draw down 0.” In statistical parlance.250 Mean 1.0% 3. Again. there is a 34 percent probability of being down at least 25 percent and a 7 percent chance of being down by 50 percent at some point over a 30-year horizon.0 The table also shows that the probability of loss declines rapidly as the horizon increases. . In fact.100 1. To put it differently.608 2.Theory in Practice Drawdowns and Confidence Intervals for Wealth and Log of Wealth Chapter 3 Asset Allocation: The Mean–Variance Framework 85 It is often useful to summarize uncertainty with a statement such as “wealth will be between X and Y with Z percent probability.075 and = .645 T] To find the corresponding confidence interval for wealth. investment risk should not be taken lightly.750 2.586 17.498 3.776 0.9 7. Thus the 90 percent confidence interval for the logarithm of wealth is [ T 1.2 5.

With a little modification. investor preferences about return and risk. varying time horizons. n i 1 W 2 2 W ) T. it is a direct derivation of the normal return distribution. but it is important to derive the mechanics in order to fully understand the mean–variance tool—both its features and its shortcomings. and asset–liability relationships. . it may be used to manage downside risk. The model is powerful.n j 1.86 Chapter 3 Asset Allocation: The Mean–Variance Framework For short investment horizons it may not be possible to satisfy a shortfall constraint. constant mean and variance. The end-of-chapter problems will test the reader’s skill in applying the spreadsheet to answer further questions. The Excel Outboxes showed how to apply the basic equations in a spreadsheet format.n i j i j i. In this situation the program may be modified to minimize the probability of a shortfall in place of maximizing utility: Minimize by choosing subject to where the probability of a portfolio return below the threshold the weights of the risky assets the budget constraint the long-only constraint the portfolio return is the weighted return over horizon T the portfolio variance is the weighted covariance over horizon T. Summary This chapter introduced the motivation. shortfall constraints. In its simplest form.n where W i i + L L W [T ( i 1. In its simplest form. In the basic model. there are issues with the model.L 2 L 2 L )] 1 2 In this version of the asset allocation problem we are no longer trading off risk with the expected return of our portfolio. Practice Summary This section has illustrated the application of the M–V framework in solving several important investment problems. We discussed a lot of math. it incorporates varying returns. mathematics. Finally. As we saw earlier. j 2 j= 1. and many asset classes and may be used to compute optimal mixes of assets. We can expand the application to include multiple time horizons. we can use the framework to calculate the most efficient mixes of assets and solve for the optimal mix of assets in terms of risk and return. 2 W T) 1 (liabilities/assets) i 0 T i designated i 1. the resulting allocations are the same through time. and practice of mean–variance portfolio optimization. instead we focus only on the probability of eroding the asset–liability ratio. This translates into the equations: Minimize by choosing subject to N(H. and liability management. It assumes that (log) returns are normally distributed with known. However.( i i L 1. different horizons will result in different allocations. And although shortfall risk may be managed.n i L i L i. results are tied to the specified return and risk estimates.

125.2 to examine what happens to the level of certainty equivalent wealth (CEW) as the parameter . The payoff is P(n) = 2n where n is the number of tails. Go to Yahoo and download the price series for 10 randomly selected stocks (mix the sectors you choose and so on). What happens? 7. use the spreadsheet created in Section 3.B)). and 10 stocks. Problems 1.T. the payoff W3. Visit the Vanguard Web site and take the investor questionnaire (www. the complexity of the problem grows dramatically when we do so. if heads is the result of the first flip. The formula for the (S) (B) (S.xls. Using the <OP> worksheet in Chapter 03 Excel Outboxes. Create an optimization problem spreadsheet with the shortfall constraint. 5.. some of these assumptions can be relaxed.xls) to save the optimization results for five risky assets with expected returns from 0. How much would you be willing to pay to play this game? Flip Sequence H T. Techniques for developing inputs to the model—more accurate than simply relying on historical averages—are discussed in Chapter 4. b. Use the N risky asset spreadsheet template (the <5AEF> worksheet in Excel Outboxes. The template is provided in the <SF-Template> worksheet in the Chapter 03 Chapter Questions. Calculate and plot the portfolio standard deviation for portfolios of 1 stock and 2. N 0 1 2 3 P(n) $1 $2 $4 $8 3. Methods for solving more complex asset allocation problems are introduced in Chapter 5. The table shows the results for the first four potential flip sequences. Use the two-asset spreadsheet to review the impact of correlation on the shape of the feasible and efficient frontiers and the optimal portfolio mix. For example.T.B))/( (S)2 + (B)2 2 (S) (B) (S.05 through 0.. You have been invited to play a coin-flipping game. 6.H . Unfortunately. Why would investors hold only portfolios where Target ? MVP 8. and reoptimize the portfolio.H T. Many of these are straightforward. n = 0 and P(n) = $1. To illustrate how utility changes with changes in the inputs. set the risk aversion coefficient equal to 0 (zero).Chapter 3 Asset Allocation: The Mean–Variance Framework 87 Fortunately. remove the long-only constraint. a. The coin is flipped until heads appears for the first time. . and the probability of heads change. What is the difference between the two scenarios? c.xls file.com and follow the “Go to the site” link for personal investors.H T. What key characteristics about you as an investor is the questionnaire trying to determine? 4.T. 2.Vanguard . Solve the program for each target level of return and copy the weights into the results table for two scenarios: no short sale and short sales allowed. then → Planning and education → Create your investment plan then complete the investor questionnaire). Chart your results. MVP is W(S) = ( (B)2 5.

88 Chapter 3 Asset Allocation: The Mean–Variance Framework E F G H I J A 1 2 3 4 5 6 7 Inputs 8 9 10 Program 11 12 13 14 15 16 17 18 19 20 21 22 23 B C D Mean-Variance Optimal Portfolio Five risky assets No riskless asset Short sales Shortfall constraint Horizons Horizon T Objective MAX E [μ] .0500 μω σω = = 0.5∗ J23∧ 2. J22 0. J23. TRUE ) The “TRUE” indicates the cumulative distribution.2000 Portfolio 1.05 In Solver.λσ2 MAX 10 Program Objective 1 λ 11 12 Variables 13 ω 14 15 Constraints 16 ∑ω = 17 Prob(<0) = 18 19 20 Where 21 22 23 E F G H I J 1 = 0.2000 = <= 1. That is. is 5 percent: 0.0000 0.2000 Tbond 0.2000 SP500 0. The target value. To calculate the shortfall probability. entered into cell J18. enter the following in cell H18: NORMDIST (0.0882 Crop 0.2000 SP600 0. limit the probability of a negative return to be less than or equal to 5 percent.0000 0.λσ2 = λ Variables ω Constraints ∑ω Prob(<0) Portfolio Crop SP500 SP600 Tbill Tbond = = = <= Where μω σω = = Constraint Shortfall constraint: Assume the threshold is zero and the target is 5 percent.18831 Tbill 0.0995 0.1062 . add the constraint H18 J18. The completed worksheet should appear as follows: A B C D Mean-Variance Optimal Portfolio 1 Five risky assets 2 No riskless asset 3 Short sales 4 Shortfall constraint 5 Horizons 6 7 Horizon T 8 Inputs 9 E [μ] .

The liability is added to the portfolio with a negative weight because it is effectively a short position. entered in cells F14:K14. a. and 30-year investment horizons using a 1.Chapter 3 Asset Allocation: The Mean–Variance Framework 89 For charting purposes the formulas for the portfolio return and standard deviation should be copied into cells to the right of the program range. Data!H17 : M22). 9. F26 : K 26) where F14:K14 is the cell range containing the weights and F26:K26 references the cell range of the expected asset returns in the worksheet. TRANSPOSE(F14 : K14)))} where Data!H17:M22 references the cell range of the variance–covariance matrix on the “Data” worksheet. The portfolio return formula in cell J22 is now G8∗ SUMPRODUCT (F14:K14. Return statistics for the liability have been assumed. The template for the worksheet is provided in the <AL Template> worksheet in the Chapter 03 Chapter Questions. 25-. Therefore. Variables The decision variables.λσ2 1 = ω Constraints Crop SP500 SP600 Tbill Tbond Liability Portfolio ∑ω Prob(<0) = = = <= Where σω μω = = Where The portfolio now includes the liability. so there is no formula to enter. the portfolio return and standard deviation formulas need to be updated. are the weights for the assets and liability. Solve the program for 1-. b. Explore the optimal asset weights for differing horizons where the shortfall constraint is met. Create an optimization framework for an asset–liability problem. Using the shortfall constraint spreadsheet created in Problem 8.0 risk aversion parameter. set the shortfall constraint = 0 (K = 0) and T = 1. The portfolio standard deviation formula in cell J23 is now { SQRT (G8∗ MMULT (MMULT (F14 :K14. 5-. What happens? 10. 10-. . 15-. The asset weights are the solution.xls file and should appear as follows: A 1 2 3 4 5 6 7 8 Inputs 9 10 Program 11 12 13 14 15 16 17 18 19 20 21 22 23 B C D E Mean-Variance Asset Optimal Portfolio Five risky assets No riskless asset Short sales Shortfall constraint Horizons Horizon T Objective F G H I J K MAX λ Variables E [μ] . 20-. then copy the weights into a results table.

λσ2 1 = 0.1102 For collecting the data. 20-.0527 0. For this example.0000 ∑ω Prob(<0) = = = <= 0. The formula entered into cell H17 is the sum of the asset and liability weights: SUM(F14 : K14) The target value. Suppose the liability in Problem 10 was perfectly positively correlated with both the S&P 500 and S&P 600 and had a standard deviation of 0.06. to sum to 0 rather than summing to 1 as they did in the case of assets only. entered into cell J17. 10-. 25-. That is. assume that the assets just cover the liability.0406 λ Variables ω Constraints Crop 0. a.2000 SP500 0. and 30-year investment horizons.0000 0. inclusive of the liability.2000 Portfolio 0.90 Chapter 3 Asset Allocation: The Mean–Variance Framework The weight on the liability is also not a choice variable: It is dictated by the value of the liability relative to the value of the plan’s assets. enter 1 into cell K14: 1. the liability had the following correlations: . How do the weights on stocks in this example compare with earlier examples? Explain why they may be different. is 1 plus the weight of the liability: 1 K14 The worksheet is now complete and should appear as follows: A 1 2 3 4 5 6 7 8 Inputs 9 10 Program 11 12 13 14 15 16 17 18 19 20 21 22 23 B C D E Mean-Variance Asset Optimal Portfolio Five risky assets No riskless asset Short sales Shortfall constraint Horizons Horizon T Objective F G H I J K 1 MAX E [μ] . then copy the weights into the results table. 15-.33603 SP600 0.0000 0. 5-.2000 Tbond 0.2000 Tbill 0.0000 Constraint Budget constraint: The budget constraint now requires the portfolio weights. What is the definition of return in this example? 11.2000 Liability –1.0500 Where σω μω = = 0. the formulas for the portfolio return and standard deviation may be copied into the results template to the right of the program range. b. c. Solve the program for the 1-. Therefore.

895 Modified 0.0 SP600 0. s = 1.016 0.0 Appendix 1 Returns.0 0.0 0. ignoring the possibility that the distribution was received earlier in the period and reinvested.017 0. That is.0 0.0 0. For a single unit of the asset.0 0.0 0. including price appreciation and distributions.n.0 TBond 0.0 0.0 1.000 1.2 0. and Sample Statistics A.0 0. Compounding.960 1.0 0.895 Reoptimize the portfolio for each of the following initial portfolios: TBill Allocation 1 Allocation 2 Allocation 3 Allocation 4 Allocation 5 Allocation 6 0.0 0.0 0. The gross return over the multiperiod horizon is the product of the gross returns. Suppose we have a sequence of returns.000 0. (1 hi (1)) (1 hi ( 2 )) … (1 hi ( n)) The logarithmic return in period s is ri(s) = ln(1 + hi(s)).960 0.0 0.0 1.0 0. this can be expressed in terms of the asset price: hi (t ) {Pi (t ) Pi (t 1) Di (t )}/Pi (t 1) This definition assumes that the distributions occur at the end of the holding period.0 0. Returns The total holding period return (HPR) of an asset is defined as the dollars earned over the period.Chapter 3 Asset Allocation: The Mean–Variance Framework 91 Asset TBill Corp SP500 SP600 TBond Original 0.0 SP500 0. hi(s). .2 0.0 0.2 1.2 0. The gross return in period t is simply (1 + hi(t)).0 1.2 0.0 1.016 0.040 0.0 0.0 Corp 0. the return of a position in asset i over the period from t where Vi(t) is the value of asset i at time t and Di(t) is any distributions made over the holding period. divided by the dollars invested: HPR {Ending value hi (t ) Beginning value {Vi (t ) Vi (t 1) Distributions} / Beginning value 1 to t is Di (t )}/Vi (t 1) In general.

718. Mean Return The mean is calculated as the average return of the sample: i (1/T ) t ri (t ) Variance and Standard Deviation The variance of a random variable is measured as the average squared deviation from the mean: 2 2 (1/ (T 1)) t ( ri (t ) i i) In computing the sample variance it is standard to divide by (T 1) rather than T because one degree of freedom was used in computing the mean. For continuously compounded returns (that is. the notation here corresponds to the parameter definitions established in Section 3.2 and summarized in Appendix III. rearrange this equation: ri ln (Vi (T ) /Vi (0)) /T C. Continuous Compounding Logarithmic returns implicitly reflect continuous compounding—that is. ri(t). The standard deviation is simply the square root of the variance: i 2 i The variance of a sequence of returns may be calculated from the variance of the single-period return under the assumption of independent and identically distributed random variables. .92 Chapter 3 Asset Allocation: The Mean–Variance Framework B. defined as logarithmic returns. the formula for variance over the period 1 through T based on the one-period variance is 2 2 T 1 1. the expected log return is denoted by . The general compounding formula is Vi (t ) Vi (0) (1 ri /t )T /t where Vi(0) = the initial value of asset i. For example. t = 1. As the number of compounding periods goes to infinity. Sample Statistics14 Given a sample of log returns. the corresponding formula is S12.T 14 ( S1 2 M12 )T M1 2T With returns.T For a sample of gross returns. we can calculate sample statistics to estimate risk and return parameters. A common application is annualizing standard deviations calculated from monthly returns. the formula converges to V (t ) V (0)exp( rT ) i where exp = 2. This formula reflects earning a return of (ri/ t) in each of (T/t) subperiods. r. To calculate the continuously compounded rate of return on asset i per period over the holding period from 0 to T. T = the holding period. subdivision of the return period into finer and finer increments. and t = the number of compounding periods within the holding period (T). ri = the rate of return on asset i. Of course sample statistics can be computed for any data series.T. logarithmic returns).

1):Cell(i.j (ri.T)) =COVAR(Cell(i. j /( i j ) By construction the correlation coefficient lies between 1 and 1.t i )2 ) j = (1/T) i. the Chapter 03 Excel Outboxes spreadsheet contains historical data.1): Cell(i. 15 .T).1):Cell(j.15 This formula is exact if gross returns are lognormal.t) represents the cell reference for the return of asset i at time t. where lnHPR ln (1 HPR ) We use M and S here to distinguish them from the corresponding measures for log returns. Correlation The pure measure of association is correlation.j /( ) Here Cell(i. The sample covariance subtracts the sample mean from each series and averages the product of corresponding deviations: i. for the following assets: Asset Treasury bills Corporate bonds Large stocks (S&P 500) Small stocks Treasury bonds Abbreviation TBill Corp SP500 SP600 TBond Range C25:C840 D25:D840 E25:E840 F25:F840 G25:G840 Two series are provided for each asset: the HPR and the logarithmic HPR (lnHPR). Note that in our standard notation.T).1):Cell(j.T)) =VAR(Cell(i.T)) =CORREL(Cell(i.1): Cell(i. = log of the mean gross return = ln(M).1):Cell(i. The “Data” worksheet provides monthly returns.Cell(j. It is pure in the sense that it removes the variation of return to express the association only in terms of the direction: r i . For illustration. j (1/T ) t ( ri . j = i . D. Covariance The basic measure of association is covariance.t t i j = i.t )(rj. from January 1926 through December 1993.T)) (ri .t t (1/(T (1/(T =AVERAGE(Cell(i. Application in Excel Sample Statistics and Excel Formulas Statistic Return Variance Standard deviation Covariance Correlation i Equation = (1/T) 2 i 2 i Excel Formula )2 i i t ri(t) 1)) 1)) (r t i .T)) =STDEV(Cell(i.1):Cell(i. Otherwise it is a useful approximation.Cell(j.Chapter 3 Asset Allocation: The Mean–Variance Framework 93 where M is the mean and S is the standard deviation.j i. (t ) i )( rj (t ) j ) This measure captures both the volatility of the variables and their directional relationship.

000 0.1112 1.0520 0.005 0. from all the possible sets of values satisfying the constraints.0220 0.0576 0.0363 0. standard deviations. Often you will have several cases or variations of the same problem to solve.0161 -0.0000 0. Parameters and Assumptions Parameters are the constant coefficients of the model.002 0. 2.000 0.005 0.0256 0.0857 -0.0458 0.004 0.0030 0.0000 0.0028 0.0679 SP500 0.0000 0.8364 0.0093 TBond 0.0761 Liability 0.000 0.0885 0.1003 0.1997 Arithmetic SP600 0.0857 0.0000 0.004 0.3067 TBill 0.000 0.053 0.0000 0.0099 0. are entered in the range A1:M22.0342 0.0000 0. correlation table.0761 Corp 0.0000 0.1071 0.0000 0. that optimize (maximize or minimize) the objective.075 0. Decision variables: what is changed to optimize the objective.075 0.3067 Tbill 0.000 0.003 0. and covariance matrix.1712 0.004 0.004 0.000 0.94 Chapter 3 Asset Allocation: The Mean–Variance Framework The mean returns.0093 Tbond 0.0038 0.053 0.000 0.1188 0.0000 -0.8585 -0. They enter into the calculation of the objective function and/or the constraints.0043 0.000 0.8380 0. Assumptions.0514 0.002 0.000 0. they are held constant in the optimization.0043 0.000 0.1071 0.1723 0.000 0.0436 1.002 0.1019 0. Constraints: limitations on the decision variables.0220 0. 4.0027 0.8461 1.1784 0.2262 1.990 0.003 0.0099 0.2225 0.060 CORRELATIONS 1 2 3 4 5 COVARIANCE 1.040 0.0576 0. The working version should look like this: Summary Statistics OBS Monthly Annual DATE α σ α σ ρ Corp SP500 SP600 TBill TBond Liability Ω Corp SP500 SP600 TBill TBond Liability Corp 0. this resource is typically money and .200 0.0027 0.003 0.000 0. or summary statistics.001 0.001 0.900 0.094 0.0141 0. 3.1997 Logarithmic SP600 0. Parameters: constants in the formulas.0196 0.003 0.075 0.0885 0.0256 1.0047 0.1784 0.0259 -0.0196 0.000 0.0964 -0.005 1 2 3 4 5 Appendix 2 Optimization Optimization models have five key elements: 1.0568 0.0000 1.8364 0.900 1.1003 1.0135 0.045 0.0679 SP500 0.0161 0.006 0.1712 0.000 0.003 0.1619 0. Decision Variables Decision variables usually measure the amounts of resources to be allocated to some purpose or the level of some activity.8585 1. The optimization process finds the set of decision variable values.2225 1.000 0.1688 0.004 0.200 0. 5. Although you may change these values. and the parameter values will change in each problem variation.002 0.1184 0. Objective: the measure to optimize.1851 0. In finance.

if the objective is to minimize the risk of the portfolio. In most models constraints play a key role in determining what values are allowed for the decision variables. 000 / $10. i 0. the weights for this simple two-asset portfolio are S B $ Stocks/ ($ Stocks $ Bonds/ ($ Stocks $ Bonds) $ Bonds) $6. A shortfall constraint can be used to construct a portfolio that limits the expected probability of earning a return below some threshold level. Constraints Constraints reflect limits on the decision variables. has two elements: how the objective is measured and what you are seeking to do. 000 60% 40% 100% Notice that because the money is fully invested in stocks and bonds. • The expected behavior of the portfolio. For example. For example. or for each asset i. indicating the percentage invested in asset i relative to the Investment in asset i /Total investment For example. you specify a formula involving the decision variables. For example. the measure would be the portfolio standard deviation formula and the goal would be to minimize. To define a constraint. the longonly constraint.000 in stocks (S) and $4. Then you impose a condition (< = . or set of decision variable values. constraint and is expressed as S + B = 1. what values the objective can attain. For these assets the weights have to be positive. 000 S B Objective The objective. expressed as a function that depends on the decision variables. the requirement that the asset weights total to 1 is called the budget. 000 $4. Variable Constraints Constraints can be used to limit the following: • The value of the decision variables in aggregate. = or > = ) and limit on the formula’s value. • The value of specific decision variables. Most optimization algorithms start with a feasible solution and then find another feasible solution by changing the decision variables. or full investment. a shortfall constraint can be included in the optimization program. Solution A solution. if you invest $6. the investor may be prohibited from taking a short position in some or all of the available assets. If the new feasible . if the investor dislikes losses (the likelihood of negative returns) rather than volatility (the total variation in return). 000 / $10. and.000 in bonds (B). that satisfies all the constraints is called a feasible solution.Chapter 3 Asset Allocation: The Mean–Variance Framework 95 is expressed as a weight. For example. ($ Stocks $ Bonds) / ( $ Stocks $ Bonds) $10. total amount invested: i i . 000 / $10. as a result.

In many cases. The specific assumptions include: 1. the new feasible solution is kept and the process repeated until changing the decision variables does not improve the objective. no set of decision weights satisfies all the constraints. the objective is unbounded. change the initial values of the decision variables before solving. In some cases the program is not properly specified: • If the program is underconstrained. • If the program is overconstrained. Problem 6 at the end of the chapter was an example of an underconstrained program. . Proportionality: Output = Constant x input. These variables do not represent an optimal solution. though. In practice. An optimal solution is a feasible solution where the objective function reaches a maximum (or minimum) value. N j 1. Certainty: The problem can be modeled. 2. Divisibility: The decision variables take real rather than integer values. 2 0. In other cases.96 Chapter 3 Asset Allocation: The Mean–Variance Framework solution increases (decreases) the value of the objective function when maximizing (minimizing). but this is not always possible. Linear Programming A linear program (LP) is an optimization in which both the objective function and the constraints are linear. N i j i . 5. the optimizer would continue to run forever because changing the decision weights would continue to improve the objective. The decision variables reported are the last feasible solution the optimizer reached before stopping. it will not be clear that the solution produced by the optimizer is global. A local optimal solution is a feasible solution with the best objective function “in the vicinity”—that is. The global optimal solution (if there is one) is the feasible solution with the best objective function value among all feasible solutions. A simple way to check is to start the optimizer from different locations in the solution space—that is. this means that the portfolio variance is positive—that is. See Problem 11 at the end of the chapter for an P i 1. j example. Problem 9 at the end of the chapter was an example of an over-constrained program. there are no other nearby feasible solutions with better objective function values. 4. Additivity: Total output = i Output(i). The objective and constraints are linear. 3. In theory. See Problem 11 at the end of the chapter for an example. the optimizer will stop running after a set time limit or number of iterations and report that the program did not converge. it may be enough to find a good solution—one that is better than the solution you are using now and offers insights into the problem at hand. ideally global. Optimization Failure Optimization programs like Solver are designed to find optimal solutions. For portfolio optimization. Or the solution space is not well defined: • If there are many local optimal solutions. the inputs are not well defined: • Quadratic optimization requires that the objective function be convex.

. . 16 If Solver. Select Help from the Solver Results menu for a description of the reports available and their content. does not appear as an option on the Tools menu. it needs to be copied from the Excel install disk. . Quadratic Programming A quadratic program (QP) involves optimization of a quadratic objective function subject to linear constraints. . Select Help from the Options menu for a description of the optimization parameters and their settings. Also imbedded in the solution is specialized information called shadow prices. The mean–variance problem is a quadratic problem. load or save problem definitions. If Solver Add-in does not appear as an option.16 The Solver menu appears as follows: The Solver menu parallels the optimization program: Optimization Objective Decision variables Constraints Solver Set Target Cell: Equal To: By Changing Cells: Subject to the Constraints: The Options button allows you to control certain aspects of the solution process. select Add-Ins. Reports provides a detailed summary of the solutions that are available on the Solver results screen. and define parameters for both linear and nonlinear problems. . Solver has default settings that are appropriate for most problems. from the toolbar select ools then Solver. . .Chapter 3 Asset Allocation: The Mean–Variance Framework 97 Issues involving the constraints include whether they are binding. and then check the box for Solver Add-in. To start Solver. Excel Solver Solver is a numerical optimization program included with Excel. .

these steps can be repeated to add the target return and no short position constraints. the menu should appear as follows: . let’s solve the two risky asset program. Enter the cells $F$12:$G$12 in the By Changing Cells: field. Click Min for Equal To:. and enter $G$15 in the Constraint: field. To enter constraints. The goal of the program was to identify a portfolio along the frontier by finding the portfolio weights that minimized the portfolio standard deviation for a given level of return. click the Add button. When complete. If desired. The Add Constraint menu should appear: To add the full investment constraint. follow these steps: Enter cell $G$8 in the Set Target Cell: field. select = as the relation. enter $E$15 in the Cell Reference: field.98 Chapter 3 Asset Allocation: The Mean–Variance Framework To illustrate the use of Solver. This translates into Solver as follows: Optimization Objective Variables Constraints Solver Set target cell: Equal to: By changing cells: Subject to the constraints: $G$8 MIN $F$12:$G$12 $E$15 $E$16 = = $G$15 $G$16 To fill in the menu.

0600 = = 1.16 percent in stocks and 93. The worksheet should appear as: A 1 2 3 4 5 6 7 8 Program 9 10 11 12 13 14 15 16 17 18 19 20 21 B C D Mean-Variance Efficient Frontier Two risky assets No riskless asset Short sales E F G Objective Variables MIN σω = Stock 0. in cells F12 and G12.9384 ω Constraints Σω μω = = Portfolio 1.0000 0.84 percent in bonds has an expected return of 6 percent. states that the portfolio allocated 6. Repeating this process for each target return produces a reasonably complete frontier. Select Keep Solver Solution and click OK.0600 Where μω σω = = 0.0600 0.0745 The solution. These weights can be copied to the results table and represent one point along the frontier. To solve for additional points at different target return levels.Chapter 3 Asset Allocation: The Mean–Variance Framework 99 Click Solve. change the value in cell G16 and resolve. You should now see a Solver Results menu: Solver has found a solution that satisfies the constraints.0000 0.0616 0. .0745 Bond 0.

Natural logarithm function 2.j/( i j)) .100 Chapter 3 Asset Allocation: The Mean–Variance Framework Appendix 3 Notation Investment Pi(t) W B S n T U t i E[ ] Ri ri(t) H K L Xi(t) ln exp Price of asset i at time t Wealth Bond Stock Number of assets Investment horizon Utility function or utility Time Weight of asset i Coefficient of risk aversion (= ½ (1 )) Risk aversion parameter Expected value Cumulative return for asset i (= t = 1. base for natural logarithm 2 i )) Statistical N i i 2 i i i.718.j Cumulative normal distribution Expected continuously compounded return for asset i 2 1 Log of expected gross return for asset i ( ( i 2 i )) Variance of (log) return for asset i Standard deviation of return for asset i Covariance of return between asset i and asset j Correlation of return between asset i and asset j (= i.j i.T ri(t)) Log return for asset i at time t Shortfall constraint threshold return level Shortfall constraint probability level Liability Random variable for asset i at time t ( N ( i .

Sign up to vote on this title

UsefulNot useful- Private Equity Final Report
- Strategic Asset Allocation
- Mean Variance Optimization
- Private Equity in South Africa
- Understanding Asset Allocation - Canto
- Police and Fire Pension Briefing
- 24
- 00G Rolf Wetzer Treatise on Tactical Asset
- A17_20120702_Protected Flexible 80 July
- 15. Pension Funds
- Syllabus Booth
- Project1-AssetAllocation-Assignment.docx
- 2016-02-01 Investment Strategy Outlook
- Robust and Adaptive Algorithms for Online Portfolio Selection
- CPFA Curiculum
- Opportunity Fund
- CalPERS Total Fund Monthly Update, Mar. 31, 2013
- ArtificialIntelligence-book2
- ArtificialIntelligence-Book2
- The Norway Model
- Manager
- 20151227a_006101004.pdf
- Exam1 Topics
- Utsp Seminar
- Paired-Switching for Tactical Portfolio Allocation
- Contract Theory
- NaN
- Solution Corporate Reporting Strategy Nov 2007
- Rocky-Mountain-Power--Karl-A-McDermott
- 64639_1975-1979
- Sample Chapter 03