A different approach to portfolio construction.

By Joel Bruckenstein

    Most financial advisors have a decent grasp of portfolio construction and asset allocation. They know that Harry Markowitz started the ball rolling with his 1952 academic paper entitled Portfolio Selection, and that 38 years later he was awarded a Nobel Prize, along with Merton Miller and William Sharpe for, their work on what is now known as Modern Portfolio Theory.
    Essentially, Markowitz argued that risk/return characteristics should be evaluated at the portfolio level, not the individual security level. Using volatility as a proxy for risk, and expected returns as a return component, Markowitz illustrated how an efficient frontier can be graphed if one knows the correlations between the asset classes.
    The efficient frontier is a set of optimal portfolios. Theoretically, all are equally efficient. So how do we, as advisors, decide which one to use with clients? Well, suppose a couple engages you to determine how much money they need to accumulate for retirement. Given their assets and their ability to save, you determine that they must achieve a 9% nominal rate of return on their assets and their savings to reach their goal.
    Under this scenario, the clients might choose to select a portfolio with the required anticipated return along the efficient frontier. This will offer them the portfolio(s) that can deliver the return they need with the least amount of risk.
    Another common use of the efficient frontier is for advisors to first evaluate a client's tolerance for risk, and then select a portfolio on the efficient frontier that coincides with the client's risk tolerance. In this case, the advisor is looking for the greatest achievable level of return available at a given risk level.
    The two examples above are flip sides of the same coin. In the first instance, we are supplying a required rate of return, and the model is telling us the "optimal" or lowest level of risk we need to take in order to achieve that return In the second instance, we are supplying the level of risk we are willing to take, and the model supplies us with the highest possible return we can achieve for that level of risk.
    Gravity Investments, LLC, a firm that specializes in something they call "Diversification Intelligence," has built a software application called G-sphere that attempts to extend the utility of the efficient frontier for advisors and their clients. Rather than presenting the whole efficient frontier and leaving it to advisors to select the "right" portfolio along the frontier, G-sphere is designed to pinpoint the "best" efficient portfolio residing on the frontier.
    The G-Sphere approach is built upon the same foundations as the traditional approach to portfolio optimization, but it adds some unique ingredients to the mix. First, it starts by optimizing for diversification, and then it adjusts the asset weightings to account for the differences in attractiveness of the various asset classes. A utility factor, in this case the Sharpe ratio, is used to measure an asset's relative attractiveness. This differs from the traditional approach in that it seeks to explicitly quantify diversification.
    Under the G-Sphere methodology, rather than arriving at a continuum of efficient portfolios (the efficient frontier), the software will solve for a single efficient portfolio on the frontier, one that maximizes the benefits of diversification while providing acceptable returns (as illustrated in Figure 1).
    Another unique aspect of the software is the three dimensional modeling approach it takes. All assets are plotted from a starting point. The starting point is the risk-free rate of return, represented here as Cash (see Figure 2). Advisors set their own risk-free rate of return assumptions (the expected return on cash).
    The various assets are depicted in Figure 2 as lines extending out from the cash starting point. The length of the line graphically represents the Sharpe ratio. The longer the line, the higher the Sharpe ratio; so in Figure 2 MID has a higher Sharpe ratio than GI.

The angle between the assets indicates their correlations. An acute angle represents a high correlation. The greater the angle, the lower the correlation. A 90-degree angle represents no correlation. A 180-degree angle represents a negative one, or perfectly uncorrelated asset.
    In this illustration, for example, MID shows a very high correlation with MB, but it appears uncorrelated with SPX.

All of this sounds complicated, and the math behind it is, but essentially, what G-sphere provides that differentiates it from other software is a three-dimensional graphical depiction of the portfolio that equates portfolio diversification with symmetry. The size and shape of the portfolio indicate the anticipated risk-adjusted return of the portfolio. While the traditional efficient frontier graph only depicts risk and return, the G-sphere 3D model also displays correlations among the assets clearly. Under this methodology a large, symmetrical shape represents a "good" portfolio.

Now that we've summarized the underpinnings of the model (full details are available at www.gravityinvestments.com) let's turn our attention to the software itself. The list price of G-sphere is about $4,500 per year. This includes the software as well as the data that supports it. When you subscribe to G-Sphere, is comes with a full set of data on U.S. and Canadian equities, U.S. and Canadian mutual funds (open end, closed end and ETF's) and futures data. The numbers are updated regularly, and advisors have access to that data through an Internet connection to the Gravity servers.
    Once the software is installed, you can create individual client portfolios or model portfolios. The first step is to name a portfolio and create a type (model or client, for example), select a home currency and add comments as necessary. Next, you type in the portfolio assets or import them from a spreadsheet. You can enter ticker symbol and a number of shares, or you can enter a value, and the software will calculate the number of shares based upon the current price. If all of the holdings are equally weighted, you can just type them in and then tell the software to weight them equally. If you wish to place any constraints on the optimization process, you can enter the constraints here.
    The next step is to enter the time series you want to use to examine the data. You can choose a single series, or multiple series, and you can weight them to your liking; you can even use different intervals for different time periods. For example, if you wanted to use a time series for 2004, 2005 and 2006, you could use monthly data for 2004, weekly data for 2005 and daily data for 2006. In a case where an asset has a limited data set, the software can extrapolate the data over a longer time period if you want it to.

Next, you can condition the data. If you want you use a historical benchmark for comparison, you might first run an unconditioned portfolio. Then, you could run the same one using tools to shrink the range of returns or range of risk. You also can run the model using more than one measure of risk. Standard deviation is the traditional measure of risk, but advisors can also choose semivariance and maximum drawdown as a measure of risk. Semivariance represents standard deviation with respect only to adverse price changes, as opposed to standard deviation, which makes no distinction between good (price appreciation) and bad (price depreciation) volatility.

Maximum drawdown is designed to look at the worst-case scenario. It answers the question: "What if an investor bought the portfolio at a market peak and held it for a period of time through a bad decline, selling at the bottom; how much of the portfolio would be left?" The software even allows you to alter the cross-correlations for a time series, overriding the historical data for that period.

Next, you would run the optimizer and analyze the results. If the results are satisfactory, your work is almost done; if not, you might want to make some changes to the assumptions, or run multiple scenarios, using different measures of risk, different time series, etc. Analytical tools allow the advisor to view all of the portfolio statistics commonly taken into account (portfolio rate of return, Sharpe ratio, etc.) as well as some less familiar ones like Inter-Portfolio Correlation (IPC). This number, a measure of diversification usually presented as a range between 1 and -1, is converted into a percentage for easier viewing here. According to Michael Kerins, regional sales director at Gravity: "The typical advisor-constructed portfolio we see generally starts off with an IPC of 15% to 25%. Through the use of G-Sphere, advisors can boost that number to anywhere between 35% to 60%."