Don’t Kill the Golden Goose! Saving Pension Plans

Abstract

Defined-benefit (DB) pension plans are an endangered species; they are perceived as too risky and costly. But the emerging substitute, the defined contribution plan, has many shortcomings. The risk of DB plans can be controlled, first, by modeling the liability in terms of its market-factor exposures through surplus (asset minus liability) optimization. Then, sponsors may hold the minimum-risk position (a liability-defeasing portfolio) or they may move up on the efficient frontier—taking equity and other risks. The economic cost of a DB plan also needs to be managed, but it is a matter of managing the size of the pension promise; it is not an asset allocation problem.

Defined-benefit (DB) plans are like a goose that lays golden eggs—monthly retirement income at a decent level, guaranteed for life. The gradual disappearance of these plans is a tragedy for employees and for society because they are the only practical way to provide an adequate retirement benefit. Defined-contribution (DC) plans are not a suitable replacement.

DB plans offer forced savings, the sharing of longevity risk through annuitization, institutional-quality investment strategies, and institutional fee levels. Although DC plans could theoretically replicate all these features, in today’s legal and regulatory environment in the United States, there is no way this replication will actually happen. In fact, most DC-plan participants retire with plan balances so small they provide only a small supplement to Social Security income; they aren’t really pension plans at all.

A back-of-the-envelope estimate shows that from sharing longevity risk alone, a given dollar amount of retirement benefit is 35 percent cheaper to provide through a DB plan than through a DC plan.

DB plans are also valuable to employers because the economics of labor negotiation requires that this 35 percent (or greater) savings be shared between employer and employee. Thus, the total cost to the employer of obtaining a unit of labor is lower with a DB plan.

DB plans are disappearing because employers perceive that the financial risk of being a DB-plan sponsor is unacceptably great. For illustration, employers point to the “perfect storm” of 2000–2002, during which equity prices fell by 50 percent from peak to trough while the present value of pension liabilities was rising dramatically because of the decline in long-term interest rates.

However, these market events resulted in a perfect storm only because DB plans were poorly hedged. If assets had been selected with market risk exposures closer to those of the liability, there would have been little damage to pension plans over this period.

To save DB plans, sponsors should use existing—although often poorly understood—technology to: (1) hedge the market-related risks in the liability that can be hedged and (2) use “surplus optimization” to rationally take additional risk in pursuit of higher returns.

To hedge the risks that can be hedged, one must understand that the liability is someone else’s asset and can be modeled as one would any asset—namely, as the sum of the riskless rate, exposures to various market risks (in this case, inflation risk and real interest rate risk), and a residual or alpha portion. A portfolio of nominal U.S. T-bonds and Treasury Inflation-Protected Securities can be designed that will hedge the liability quite satisfactorily.

Note that the market value of the liability is the value relevant to this analysis. This approach is required to build a hedge portfolio and eliminate most of the risk from DB-plan sponsorship. Market-value accounting is a good thing; it provides the transparency that allows risks to be managed by using available market tools.

The hedge portfolio is, however, an extreme position. Its yield is too low to be acceptable to most sponsors. Surplus optimization, which is like asset-only optimization but with the liability included as an asset held short, should be used to assess the decision to hold equities and other risky assets. The sponsor can try to add value by taking “surplus beta risk”—that is, equity or equitylike risk not needed to hedge the risks in the liability—and by taking active (alpha) risk.

A financially strong plan sponsor can afford to take more surplus beta risk than a weak one because the strong company can afford the higher plan contributions that will be required if the risk does not “work out” (that is, if the stock market performs poorly). In general, however, equity allocations in DB plans should be lower than they are in current practice.

Most DB portfolios, then, should have a longer duration and less in equities than they do. Although the expected return from such a strategy is slightly lower than in today’s equity-heavy plans, theworst-case scenarios are much less bad. By holding assets that are closer to the liability in terms of their risk exposures, DB plans can be saved because the risk of sponsoring them will have been managed to acceptable levels.

Notes

¹ Since this was first written, the Pension Protection Act was passed, moving us closer to a mark-to-market system for funding; the FASB (Financial Accounting Standards Board) is still reviewing its standards for GAAP accounting under the act.

² We’ll digest rather than repeat these technologies in this article. They deal with economic views of the liability and lessons from economic accounting with respect to both risk and cost control and with surplus optimization and dual-duration matching of that economic view of the liability with respect to risk control. For more complete but more technical discussions, see Waring (2004a, 2004b), Siegel and Waring (2004), and the unpublished manuscript cited as Waring (forthcoming 2007).

³ Dallas Salisbury, president and CEO of Employee Benefit Research Institute, in a speech entitled “The State of U.S. Retirement” (25 October 2005), New York City.

⁴ In theory, virtually everything a DB plan does could be done in a DC plan. In practice in the United States today, DC plans aren’t anywhere close. Other countries do better; Australia is a good example with its 9 percent mandatory contribution level.

⁵ The same analysis would apply if we assumed a midcareer age instead of age 65.

⁶ We are using the Retirement Protection Act of 2000 mortality table, which is also mandated by the Pension Protection Act. It has mortality improvement projection through 2014. The relevant interest rate is the full forward rate curve for U.S. T-bonds; as of this writing (October 2006), the Treasury yield curve is relatively flat and close to 4.8 percent for 10- to 30-year bonds, so we have used 4.8 percent as the assumed interest rate, getting an annuity factor of 11.8.

⁷ Although a retiree could replace this employer-provided annuity with a commercial annuity, it isn’t clear that it would be an advantage to the employee, even presuming that the employer shared in the gains from annuitization. Individual purchases of annuities are often subject to high sales loads. Often, the annuity tables are “tilted” against purchasers, to protect or overprotect the insurer from annuitant selection bias. Of course, insurance companies also want to profit, so they will charge a “spread” or risk charge. There is risk: Insurance companies offering annuities, although regulated, are usually regulated in a book-value rather than a market-value framework, which gives rise to substantial credit risk over the long life of an annuity. And as we can observe, in any event, few employees actually arrange annuitization on their own.

⁸ The tools supporting better investment policies, which include proper asset-class “building blocks” consisting of index funds and thoughtfully chosen active funds, as well as families of premixed asset allocation funds that are on the efficient frontier by design (today, usually called “life-style” or “life-cycle” funds), are described in Waring and Assaf (1992), Waring (1994), Waring, Siegel, and Kohn (2004), and Waring and Siegel (2006). To the best of our knowledge, Waring and Assaf (1992) is the first incidence of these tools and configurations in the literature, and it draws on work performed by Waring and presented to clients and at conferences as early as 1989, suggesting that life-style funds may owe their origin to this effort. If others were independently working on the concept before then, which wouldn’t surprise us, we’d like to know about it.

⁹ Funds offered in DC plans are typically retail mutual funds, with retail pricing. A commingled trust fund (CTF) structure, which may follow the same strategy as a given retail mutual fund but with lower fees and other expenses, can be used to achieve cost savings for DC-plan investors. Although this opportunity is well known, it is not as widely used as one would expect, probably because it generates objections from the record keeper—who is often also the mutual fund provider.

¹⁰ Reckoned in the simplest possible manner, this calculation assumes a zero investment return. Of course, the investment path in a pension plan is more complex, with continuing cash flows, but our estimate is both directionally correct and sized at least somewhat appropriately.

¹¹ Except that the employer-match component of an employee’s DC plan balance may, in some cases, be subject to vesting requirements.

¹² However, the authors imagine that there are really very few industries where portability truly serves the employers’ interest, which is to retain experienced employees.

¹³ Note, however, that these data refer to the balances in the plan in which the individual is a participant at time of retirement. It does not take into account DC balances that he or she might have from plans associated with previous employers. But the point remains: Even if actual totals are twice these single-plan figures, the amount is still way too little. Moreover, one’s last employer is typically also the employer one worked for the longest—and at the highest rate of pay (for EBRI reports, seehttp://www.ebri.org/publications/srs/).

¹⁴ We sometimes joke that, by construction, the employees are always fully funded in a DC plan—but they might not like the benefit level! Such gallows humor calls attention to the difference between the liability associated with a desired level of retirement spending and the assets that would be accumulated in most DC plans.

¹⁵ Strictly speaking, there are more than two interest rate–related risk factors or durations at issue, but real interest rates and inflation are the two that matter most, so we focus on them.

¹⁶ For plans that are in deficit, the best approach (for semantic consistency) is still to think of assets minus liabilities as the pension surplus; but the surplus has a negative value.

¹⁷ These unhedgeable, residual, or “alpha” risks involve estimating nonmarket factors that affect benefit levels and benefit timing in the participant population. With good estimation practices, the errors should be unbiased over time; that is, the mean should be zero and the volatility should be small if dealt with each year. The largest risk is usually considered to be mortality risk. When sponsors use ordinary (unmodified) life-expectancy tables, they routinely experience longer-living employees and thus surprise costs, which is not a zero-mean experience. Modified life tables that incorporate an estimate of increasing life expectancies should go a long way toward returning this error to a mean of zero and reducing the surprise. Life-expectancy forecasts, however, will always have some error. Residual risks also include the possibility that benefits will be negotiated upward, as well as other liability surprises that occur when prior years’ estimates of pay, population, and other variables require updating.

¹⁸ “Economically sensible measures of periodic pension cost” are an economically sensible version of what actuaries refer to as “normal cost” (see Note 31).

¹⁹ For an investor with a long-term liability, such as a pension fund, long-term U.S. T-bonds are safe and short-term bonds (usually referred to in finance as yielding the risk-free rate) are risky (see Modigliani and Sutch 1966). Thus, we frame this equation in terms of the “long-term risk-free rate,” which is the yield to maturity on long-term T-bonds or strips (principal-only bonds). In the notation of Ibbotson Associates (2006), the long-term risk-free rate consists of the short-term risk-free rate plus a “horizon premium,” the extra return required by investors for investing in long-term rather than short-term Treasuries.

²⁰ We have dropped the minus signs in the transition from Equation 4 to the current discussion to follow the convention used in the bond market.

²¹ Waring (2004a) originally estimated equities as being close to (17, 0). Based on research since that time, the (8, 0) values in the text are an improvement. Equity duration is a challenging issue for many analysts. Many argue that it is zero, yet the same people will estimate stock–bond correlations as 0.3 or 0.4. If equities have a nonzero correlation with nominal bonds, then, mathematically, they must have a nonzero nominal duration. And if equities have a nonzero correlation with TIPS, which is easily demonstrated mathematically, they must have a nonzero real-interest-rate duration. (Both facts taken together suggest a difference between real-rate duration and inflation duration.) So, although we cannot yet estimate or measure equity durations with the precision we are accustomed to when calculating bond durations, we can satisfy ourselves that the estimates we have used here are in the ballpark.

²² These estimates are for a prototypical pension plan (not any specific actual plan) and are from Goodman and Marshall (1988).

²³ Because one can vary the durations of the bond portfolios at will, one can also vary the weights of the holdings so that the total fixed-income (bonds plus TIPS) portfolio still has a dual duration of (17.5, 7.6). Thus, a whole family of solutions, not merely the solution shown here, is available.

²⁴ The accounting measure called “pension expense” today is anet value including normal cost, investment returns on the assets, interest cost on the liability, and actuarial gains and losses. The fact that these items are netted to a single value called “pension expense” is somewhat controversial and may be addressed by the FASB.

²⁵ The real cost, risk adjusted, is not changed by holding equities. Risk premiums from the risky-asset portfolio are not relevant to the discount rate used in valuing the plan; only the expected return of the liability-matching or hedging portfolio is relevant.

²⁶ These criteria are the “two conditions of active management” of Waring, Whitney, Pirone, and Castille (2000) and Waring and Siegel (2003) but with managers instead of securities as the unit of selection.

²⁷ That is, the quantities and durations of the nominal bond and TIPS subportfolios are engineered so that the inflation dollar duration of thetotal portfolio (including equities) equals that of the liability and the real-interest-rate dollar duration of the total portfolio also equals that of the liability.

²⁸ The probability, given an investment policy decision made today, of needing additional contributions in the future to support today’s funded ratio or level of surplus can be directly read, however, from Figure 2. Specifically, at any point in time, a vertical cross-section of the chart represents a probability density function (the difference between two lognormal distributions). Thus, the portion of that cross-section line under a line drawn at the level of today’s funded status represents the probability of additional contributions being required at that point in the future as a result of downside investment risk. The area above represents the probability ofreduced future contributions from good returns (relative to holding a fully hedged zero-surplus-beta portfolio). Generally, the probability of future contributions being required can be determined from this distribution for any funded ratio or level of surplus. This knowledge can dramatically improve the sponsor’s ability to make proper risk–return trade-offs among investment policies.

²⁹ This effect is not a reduction in the “present value of future contributions” but a reduction in the “expected future value of the then present value of future contributions.” Although a mouthful, the distinction is important.

³⁰ In our experience, sponsors intuitively use the correct, economic notion of cost in conversation even if they cannot precisely define it.

³¹ Actuaries refer to this periodic cost as “normal cost” or “service cost.” It is a function of what is sometimes referred to as the “full economic liability,” which can be broken down into smaller, component measures of the liability. One of these smaller measures is usually used for cost purposes and contribution purposes. Cost is one of several elements of “pension expense,” along with interest costs, investment returns, and supplemental costs. Interest cost and investment returns are not really costs in any strict sense of the word. Supplemental costs are simply the changes in the liability required to reflect updates and improvements in estimates of mortality, length of service, and so on. Although actuarial normal cost measures would not usually be the same as a market or economic cost measure, a functional, albeit complex, relationship exists between the two: Actuarial measures have to follow the economic measures—at least over long periods of time. The economic cost is whatever it is, regardless of accounting, and it will show up even if manipulated for deferral. Relating actuaries’ and accountants’ versions of all pension measures to economic measures is a full subject in itself and treated in Waring (forthcoming 2007).

³² Waring (forthcoming 2007) shows in detail how the economic and accounting versions of the balance sheet, income statement, and cash flow statement entries for the liability are related to each other.