Home > Library > Papers  

Asset Allocation Models: A Stable Value Perspective

Introduction

If asset allocation decisions are challenging for investment professionals-and they are-they can be overwhelming for laymen who have no specialized investment training. For the tens of millions of ordinary Americans who save for retirement in defined contribution plans, that's disturbing news-and good reason to embrace the computerized investment advice models that make professional investment expertise widely and affordably available.

These models, generally accessible via the Internet, have the power to enhance the value of defined contribution plans to plan participants. They also benefit society at large by improving the financial security of the nation's current and future retirees. To perform with the thoroughness and precision that plan participants deserve, however, the models must be as complete and accurate as possible.

A challenge to meeting this goal is accurately modeling stable value investment products, one of the most popular investment options in 401(k)s and other defined contribution plans. Stable value funds have risk and return characteristics distinctively different from those of other asset classes. While correctly modeling those risks and return expectations may be difficult, advice modelers, as ERISA fiduciaries, owe plan participants a statutory duty of care and loyalty in constructing their models and deciding on their inputs. Fortunately, the benefits of doing so easily justify the effort.

Executive Summary

Computer-based investment advice models benefit participants in defined contribution plans by helping them make difficult asset allocation decisions for which most participants have no formal training. However, these models can perform as intended only when they correctly model risk and return characteristics for all investment choices in a defined contribution plan, including stable value funds.

Modeling the risk and return factors for stable value funds can be difficult, but it is imperative. Stable value funds differ distinctly from other financial assets, and advice providers, as fiduciaries under ERISA, owe plan participants a statutory duty of care and loyalty in constructing their models and deciding on their inputs in ways that reflect these differences.

The defining characteristics of stable value funds are (1) their book value benefit responsiveness, which allows participants to transfer or withdraw assets at book value even when market values are lower; (2) the serial correlation of the volatility of their expected returns with the length of the investment horizon; and (3) their ability to deliver investment returns comparable to intermediate-term bond funds combined with volatility risk comparable to that of money market funds.

The ideal advice model, in addition to accurately modeling stable value investment products, will incorporate the personal preferences of its users, including all appropriate investment horizons. These would include not only the time horizon for meeting retirement goals, but also for auxiliary goals such as the purchase of a home or funding higher education. The correct method for incorporating multiple investment horizons is multi-period stochastic programming, which is not currently the industry standard.

In addition to lacking multi-stochastic programming, most advice models today fail to incorporate semi-variance and downside risk calculations for projected returns, which, if deployed, would have important benefits to model users. Such analysis would allow modelers to more accurately depict not just the expected result of a particular asset allocation strategy, but also the volatility of returns en route to that result.

In summary, the Stable Value Investment Association supports the use of investment advice models. However, only those models which are complete and accurate will best serve investors. Models which accurately depict the risk and return characteristics of stable value assets, and incorporate multi-period stochastic programming, semi-variance analysis and downside risk analysis, will be most complete and accurate.

The Ideal Model

The purpose of an asset allocation model is straightforward: to suggest to the user a mix of stocks, bonds, cash and other available asset classes in proportions that will maximize expected investment returns within the constraints of that investor's investment horizon and tolerance for risk. The ideal model would project, for each potential allocation strategy developed, every possible return and volatility outcome, and the probability of success assigned to each. It also would allow the user to view and adjust all the relevant variables relating to his or her personal financial preferences.

This objective, so simple to describe, is extremely difficult to achieve. It requires data inputs in two entirely different and individually complex areas: investment expectations and personal preferences.

Personal Preferences

The ideal advice model, unattainable in practice, would incorporate all personal preferences relating to wealth. The most fundamental of these is the individual's investment horizon, meaning the length of time over which the investor wishes to achieve an investment goal. This is not always the time remaining until retirement. While defined contribution plans owe their preferential tax treatment to their function as retirement plans, investors use them for a variety of other purposes, including saving for the down payment on a house or funding education expenses. Both applications reflect rational financial decisions; matching contributions that many employers make to defined contribution plan accounts, as well as the tax-deferred status of the plans, allow individuals to save more in such plans than they could in taxable accounts. In addition, withdrawals for the non-retirement goals cited above are not subject to the federal government's 10% premature withdrawal excise tax¹. For these reasons, asset allocation models should allow users to choose any investment horizon they desire, including multiple investment horizons.

In the language of mathematics, the ideal product would be a multi-period stochastic model. The user would be able to input different preferences applicable to different investment horizons. To continue the example from the preceding paragraph, the model could factor in a conservative risk profile for that period of time when the investor is saving for a down payment on a house, and a more aggressive risk profile, or preference, thereafter.

The ideal model should also allow investors to specify a wealth target, the amount of money they intend to invest, and the level of risk they are willing to accept in their investment portfolio. The model would then calculate the investor's odds of attaining these objectives. A distribution showing expected wealth at the investment horizon, given the investor inputs, would be one useful way of displaying this probability.

Of course, some personal preferences will be difficult for any practical model to incorporate. While one investor may consider it important that the value of his investment portfolio never be lower on his birthday than it was a year earlier, for example, it would be unreasonable to expect a model to incorporate so specific a constraint.

That said, a more generalized version of what amounts to the same thing is not unreasonable. If the journey, not just the destination, matters-and it does to almost everyone, though to varying degrees-then the ideal model must allow the investor to specify path constraints. For example, the investor should be able to input as a personal preference that the value of his or her account should rise by at least, say, 3% each year, or decline by no more than 5%. Incorporating semi-variance modeling techniques might allow users to specify limits to their downside risk, which would significantly improve the model's ability to reflect user preferences.

To illustrate the importance of being able to specify path constraints, consider the investor saving for a down payment on a first home. Assume this investor is planning to buy a house five years from now, given her income expectations and outlook for real estate prices. However, if there should be a housing market downturn or a steep drop in mortgage rates-or if the investor identifies a particular property priced below market-she wants the flexibility to act sooner. In order to enjoy this flexibility, the investor should be able to specify that she wants her investment portfolio to earn a specific minimum return each and every year.

Taking into account the limitations imposed by personal preference, the ideal model will identify the asset allocation mix that optimizes the investor's expectation of meeting his or her objectives. Of course, the optimal allocation may still produce a zero percent probability that the investor will attain those objectives. For example, no asset allocation strategy is likely to produce more than a zero probability of accumulating $50,000 over a period of five years with an investment of $100 per month.

Investment Expectations

Experience shapes our expectations for the future, and we can do no better than to project the future based on assumptions about how it will resemble and differ from the past. The ideal model considers all relevant data, both about the economy as a whole and about the particular investment funds and asset classes available to the investor. (As it is unlikely that any two people would have exactly the same views on how the future will unfold, it is no criticism of the modeler if his or her views differ from those of the investor.)

The nature of the data required to model investment expectations depends on the model's fundamental approach. Virtually all models currently available use mean-variance optimization (MVO) analysis, which requires three inputs: the expected return itself, plus the variance or standard deviation of the expected return and its correlation with the returns of all other asset classes in the model. However these inputs are derived and used, the method should be consistent for all assets modeled. It should be obvious that the actual returns, variances and correlations of a specific investment option should not be ignored on the basis of preconceptions about the financial markets. Each actual investment fund is unique, and the realities of particular funds should not be ignored in favor of generalizations about assets classes.

Stable Value Characteristics Significant for Asset Allocation Models The defining characteristic of a stable value fund, a feature it shares only with money market funds, is that participant transactions take place at amortized cost and accrued interest. This characteristic is termed book value when applied to stable value funds. This unique characteristic of stable value and money market funds enables them to offer an assurance of preservation of principal and accumulated earnings to participants that no other asset class can offer. By contrast, all other investment options in defined contribution plans conduct participant transactions at market value. Any advice model that does not capture stable value's principal preser vation characteristic falls short of its mission.

Note that in exchange for offering a book-value investment product, most defined contribution plans place certain restrictions on the direct transfer of assets out of the stable value investment option and into a competing option, such as a money market fund. These restrictions are necessary to protect stable value investors fro m the destabilizing impact of those plan participants who might otherwise be tempte d to arbitrage the two types of funds during periods of rapidly rising interest rat es. As a practical matter, however, plan participants in most time periods have ful l freedom to transfer to any option available in any plan, or to make any withdrawa l permitted by the plan, at book value.

Expected Return

The most significant determinant of expected return for a stable value investment product is the portfolio of securities in which it is invested. These assets under lie the benefit-responsive wraps, or insurance contracts, that support the product 's book-value guarantee. Most stable value funds invest in fixed-income securities that have a duration of two to three-and-a-half years and investment-grade credit ratings averaging AA or higher. The funds invest both in publicly issued bonds a nd in traditional, privately placed guaranteed investment contracts, or GICs. Th ese securities produce returns that are generally comparable to those of an inte rmediate-term bond fund.

Despite the similarities of their portfolios, stable value managers generally charge far lower investment management fees than managers of intermediate-term bond funds. T his raises the expected returns for stable value funds relative to intermediate-term bond funds. However, that gain is partially offset by the cost of the stable value fu nd's benefit-responsive guarantee.

In contrast to stable value funds, money market funds invest in securities that have a lower average duration than those held by an intermediate-term bond fund. Accordingly, they also have lower expected returns. This is an important distinction. The high er returns produced by stable value funds in comparison with money market funds has a major, favorable impact on wealth accumulation over the course of a long investment horizon.

Volatility of Returns and Cross-Correlation of Returns

Estimates for the volatility of stable value returns are dependent upon the length of the investment horizon, as the fund's returns are serially correlated. This seri al correlation results from the use of crediting rate formulas-formulas that determ ine what portion of the underlying portfolio's performance will be credited to part icipant accounts-that amortize any difference between book and market values. Most crediting rate formulas in use today keep the crediting rate the same unless there are changes in the stable value fund's expected cash flows.² Positive cash flows w ill move crediting rates in the direction of current market rates, but with a lag d irectly related to the duration of the fund's underlying investment portfolio. The total return on a similar "unwrapped" portfolio would display much greater volatility.

As the length of the investment horizon increases, so does the expected volatility of stable value returns. For periods of five years or less, the standard deviation would be very low-1% or less. For a 10-year horizon, standard deviation would be ab out 2%, and for 20 years or more it would be about 3% to 3.25%. These estimates are based on historical analyses of the Bankers Trust GIC index and a surrogate stable value return series dating back to 1926 and constructed from intermediate-term Tr easury note yields. The analysis is that recommended by Richard Wendt as a means of correcting for serial correlation.³ These estimates are also consistent with the standard deviation of returns for short-term investments.

Estimates of the correlation between stable value returns and returns for other asset classes can be calculated using the same basic approach as that used for calculating v ariance. However, basic characteristics of stable value funds provide valuable insight s into the expected outcomes. We know, for example, that stable value returns will be positive in any period, thanks to the benefit-responsive book-value guarantee of stabl e value funds. We also know that stable value returns tend to move in the same directi on as interest rates for securities of comparable duration because as older investment s mature they are replaced with investments yielding current rates. Moreover, the mone y-market-like volatility of stable value returns suggests that those returns should ha ve a lower correlation with asset classes that exhibit more volatile returns. As a res ult, the expectation is that stable value returns will have a relatively high correlat ion with those for short-term investments, a somewhat lower correlation with bonds, an d very low correlation with stocks.

The empirical evidence supports these hypotheses. The correlations between the annual returns for stable value funds and stocks, bonds and money market investments from 198 3 through 2000 were as follows:

Correlation	Stocks	Bonds	Cash
SV with:	0.0	0.4	0.7

Similarly, the correlation coefficients between the surrogate stable value return series and stocks, bonds and cash from 1926 through 2000 were:

Correlation	Large Cap	Small Cap	LT Corp	Int Gov	T bills
SV with:	0.0	0.0	0.5	0.6	0.7

Despite the seeming clarity of this issue, different modelers have reached radically different conclusions about stable value variance and correlation. The Stable Value I nvestment Association would like to conduct a dialogue with the modelers to identify the reasons. If the differences are the result of assumptions about which reasonable people may disagree, they must be accepted. However, if the differences are the resul t of a methodological error by one party or another, they should be corrected. Obviou sly, the SVIA has no desire to falsely characterize stable value products or distort the role they should play in participant portfolios. Nor should modelers wish to dilu te the value of the advice they offer by mischaracterizing an important and popular investment option.

Conclusion

We have acknowledged the vital role investment advice models can and should play in maximizing the value of defined contribution plans to plan participants. We have discussed characteristics of the ideal advice model, and characteristics of st able value relevant to modeling. We conclude as we began: Models that are as comp lete and as accurate as possible will serve participants best. A complete model w ill recognize the value to many participants of an investment return that is always positive. An accurate model will recognize that stable value has higher return s than money market funds, yet offer the same principal protection. An accurate model also will recognize that stable value has lower volatility than intermediate term bond funds, while offering comparable returns.