in the risky asset for an investor with a risk-aversion degree of 10?
The solution is:
That is, this investor will invest 47.3 % in the risky asset and 52.7 % in the riskless asset. By varying the degree of risk aversion, we can obtain the full set of optimal portfolios.
1.8.3 Mean-Variance Optimization with Growing Liabilities
Equation 1.10 displayed the formulation of the problem when the asset owner is concerned with the tracking error between the value of the assets and the value of the liabilities. Similar to Equation 1.18, a general solution for that problem can be obtained as well. Here we present a simple version of it when there is only one risky asset. The covariance between the rate of growth in the liabilities and the growth in assets is denoted by δ, and L is the value of liabilities relative to the size of assets:
(1.20)
It can be seen that if the risky asset is positively correlated with the growth in liabilities (i.e., δ > 0), then the fund will hold more of that risky asset. The reason is that the risky asset will help reduce the risk associated with growth in liabilities. For instance, if the liabilities behaved like bonds, then the fund would invest more in fixed-income instruments, as they would reduce the risk of the fund.
Example: Continuing with the previous example, suppose the covariance between the risky asset and the growth rate in the fund's liabilities is 0.002, and the value of liabilities is 20 % higher than the value of assets. What will be the optimal weight of the equity allocation?
It can be seen that, compared to the previous example, the fund will hold about 14 % more in the risky asset because it can hedge some of the liability risk.
EXHIBIT 1.5 Optimal Weights of Risky Investment and Degree of Risk Aversion
By changing the degree of risk aversion in the first example, we can obtain a set of optimal portfolios, as shown in Exhibits 1.5 and 1.6.
It can be seen that at low degrees of risk aversion (e.g., 4), the investor will be investing more than 100 % in the MSCI World Index, which means a leveraged position will be used. In addition, we can see the full set of expected returns and volatility that the optimal portfolios will assume, which is referred to as the efficient frontier.
The points appearing in Exhibit 1.6 correspond to various degrees of risk aversion. For instance, the optimal risk-return trade-off for an investor with a degree of risk aversion of 4 is represented by a portfolio that is expected to earn 10.5 % with a volatility of about 15.4 %.
EXHIBIT 1.6 Expected Returns and Standard Deviations of Optimal Portfolios
1.8.4 Mean-Variance Optimization with Multiple Risky Assets
It turns out that a similar graph will be obtained even if the number of asset classes is greater than one. In that case, the graph will be referred to as the efficient frontier. The efficient frontier is the set of all feasible combinations of expected return and standard deviation that can serve as an optimal solution for one or more risk-averse investors. Put differently, no portfolio can be constructed with the same expected return as the portfolio on the frontier but with a lower standard deviation, or, conversely, no portfolio can be constructed with the same standard deviation as the portfolio on the frontier but with a higher expected return.
Example: In this example, the set of risky asset classes is expanded to three. The necessary information is provided in Exhibit 1.7. The figures are estimated using monthly data in terms of USD. The annual riskless rate is assumed to be 1 %. Note that these estimates are typically adjusted to reflect current market conditions. This example is meant to illustrate an application of the model.
Using the optimal solution that was displayed in Equation 1.18, the optimal weights of a portfolio consisting of the three risky assets and one riskless asset can be calculated for different degrees of risk aversion. The results are displayed in Exhibit 1.8.
A number of interesting observations can be drawn from these results. First, notice that the optimal weights are not very realistic. For example, for every degree of risk aversion, the optimal portfolio requires us to take a short position in the MSCI World Index. Second, the optimal investment in the HFRI index exceeds 100 % for some degrees of risk aversion considered here. Third, unless the degree of risk aversion is increased beyond 40, the optimal portfolio requires some degree of leverage (i.e., negative weight for the Treasury bills). Finally, the bottom two rows display annual mean and annual standard deviation of the optimal portfolios. These represent points on the efficient frontier.
EXHIBIT 1.7 Statistical Properties of Three Risky Asset Classes
Source: Bloomberg and authors' calculations.
EXHIBIT 1.8 Optimal Weights and Statistics for Different Degrees of Risk Aversion
Source: Authors' calculations. (Note that because of rounding errors, the weights do not add up to one.)
As we just saw, mean-variance optimization typically leads to unrealistic weights. A simple way to overcome this problem is to impose limits on the weights. For instance, in the example just provided, we can impose the constraint that the weights must be nonnegative. Unfortunately, when constraints are imposed on the weights, a closed-form solution of the type presented in Equation 1.18 can no longer be obtained, and we must use a numerical optimization package to solve the problem.6
If we repeat the example but impose the constraint that weights cannot be negative, the resulting optimal portfolios will reflect those displayed in Exhibit 1.9.
It can be seen that the weight of the MSCI World Index is constantly zero for all degrees of risk aversion. This means that portfolios that are on the efficient frontier in this case do not have any allocation to the MSCI World Index. Another important point to consider is that the optimal portfolios do not have the same attractive risk-return properties. By imposing a constraint, the resulting portfolios are not as optimal as they were when there were no constraints.
The mean-variance optimization approach discussed in this section can be presented in different forms. The Appendix at the end of this book provides two alternative methods that have appeared in the literature. The advantages of the approach presented in this section are twofold. First, as Equations 1.18 and 1.19 showed, simple closed-form solutions can be obtained when there are no constraints. Second, the approach can be easily expanded to take into account preferences for higher moments of the probability distributions of asset returns. This will turn out to be important for our purpose, as alternative investments