NIRS Misinformation: DB Twice as Efficient as DC

Analysis is badly flawed

The National Institute on Retirement Security (NIRS) recently published a study,  “A Better Bang for the Buck 3.0”, which claims to prove that typical defined benefit (DB) pension plans have twice the “economic efficiency” of typical defined contribution (DC) plans[1]:   

“[I]t would be 96 percent … more expensive for a typical DC plan … to deliver the same level of retirement income as a typical DB plan.”

“Hence, DB plans should remain a centerpiece of retirement income policy and practice, given the persistent advantages in economic efficiency.”

They’re saying, for example, that $10,000 deposited into a typical DB plan trust will generate nearly twice the retirement income as $10,000 deposited into a typical DC plan trust. Same people, same markets over the same period. 

Their metric for efficiency is the percentage of annual income that must be contributed to generate a given amount of retirement income.  They do their calculations for a homogeneous population of female teachers who retire at age 62 with 30 years of service in the same year.

The trade press has covered the news extensively and uncritically[2].  The two authors of the study have impressive credentials and are prominent in the government (public) pension plan community.

Three swings-and-misses

The authors list three sources for the superior economic efficiency of DB over DC:

1. DB plans have a “longer investment horizon” than individuals in a DC plan

2. DB plans benefit from “professional management and lower fees from economies of scale”

3. DB plan participants benefit from longevity risk pooling in retirement

The “longer investment horizon” and “professional management and lower fees” are claimed to result in superior investment returns.  Together, they account for 85% of the asserted additional efficiency of the typical DB plan over the typical DC plan. 

Strike 1: DB plans’ longer investment horizon leads to efficiency

Suppose that, for your DB plan, you believe, based on your sophisticated predictive models, that your professionally managed low-expense portfolio is going to earn 6.8%[3] per year on average.  You calculate that, if it does earn 6.8% every year, a contribution of $10,000 this year on behalf of your employee Beatrice will be sufficient to fund the pension benefits she’s going to earn this year[4] that she will start collecting at retirement.   

Beatrice’s identical twin, Caroline, is in your typical defined contribution plan with participant-directed investments.  You contribute $10,000 to her DC plan account.

The authors assume that, as Caroline gets closer to retirement and her tolerance for risk presumably declines, she shifts her participant-directed investments to lower risk assets in accordance with the glidepath of a target date fund. In contrast, because your DB plan is “ageless,” you never have to downshift its investment risk, allowing it to earn, forever and always, the higher returns associated with the riskier assets. 

The authors claim that with the longer time horizon, the riskier assets in DB lead to higher returns[5], which lead to higher income per dollar contributed, which is their definition of greater efficiency.  

But it’s not that simple.

Risk premium is not a free lunch 

Your contribution of $10,000 into the DB plan on behalf of Beatrice was not the only cost for her corresponding DB plan benefits.   You have also issued the DB plan a guarantee, or in finance-speak, a put: If the invested $10,000 doesn’t earn enough to pay for the associated benefits you promised, you must make up the difference.

This guarantee is real, and very expensive, especially with today’s low market interest rates.  The cost of Beatrice’s benefit is much higher than the $10,000 contributed.  If Beatrice is 45, I estimate that the real cost of Beatrice’s promised benefit for the year would be over three times higher[6] than what the authors would calculate based on their 6.8% return.

The DB plan looks much less efficient with the cost of the guarantee properly included.

Fairly priced investments with less risk and a lower expected return are not less “economically efficient” than investments with more risk and a higher expected return. It only looks that way in the NIRS analysis because investment risk is ignored, or implicitly assumed to be zero. 

---

Aside: Not properly reflecting DB guarantee leads to intergenerational inequities

In practice, the cost of the DB plan guarantee described above is borne by future generations.  In the example above, Beatrice’s pension accrual for the year is part of her pay package for the services (e.g., teaching) she rendered during the year.  But only $10,000 is paid by the users of her services.  The significant remainder of the cost, the put, is the responsibility of future generations who did not directly benefit from her services during the year.

---

Strike 2: Typical DC plans have higher fees and participants make bad investment decisions

The authors assert that DB plans realize “higher net investment returns due to professional management and lower fees from economies of scale.” This accounts for about 60% of the purported efficiency advantage of DB. 

The argument:

• DC plan account investments are assumed to have higher investment management fees than DB trusts; and

• Participants are assumed to be bad investors who underperform DB investment professionals. For example, participants sell when they should buy and vice versa. The authors call participants’ poor investment choices “behavioral drag.”

Even if you accept these as fact and believe them a problem, you could simply not allow participant investment discretion in your DC plan and invest the assets as you would in a DB plan.  Or you could only offer low-fee investment options, like passively managed index funds (that might even outperform DB plan active management), and educate your participants so they have the best chance of investing wisely.  And/or you could default participants into an investment option that is similar to a DB plan and allow inertia to prevent bad decisions. 

The authors acknowledge that high fees and bad investment decisions in DC plans are less of a problem than they used to be, and even define an “ideal” DC plan for which these issues are not present. By their calculation, this ideal DC plan is 37% more costly than the DB plan to generate a given amount of retirement income, much lower than the 96% additional cost of the “typical” DC plan that is the main focus of the study. But the 37% figure is still 37 percentage points too high.

To sum up, this problem is either easily solved or nonexistent. No reason, here at least, to adopt a DB plan.

Strike 3: Longevity risk pooling makes DB more efficient 

Maybe this one’s a foul tip into the catcher’s mitt. It’s fair to say that longevity risk pooling adds efficiency.  You can get higher levels of lifetime income with a given pot of money if you pool longevity.  No additional investment risk required.

But it’s not true that longevity pooling can’t be accomplished in DC plans.

Longevity pooling means that the people who die early leave money in the pool (instead of a legacy for their heirs) that is used to pay for lifetime income to those who live longer.  This can be viewed as income insurance, and the longevity pool can be viewed as analogous to a housefire insurance pool – the people whose houses don’t catch fire pay for the people whose houses do catch fire, but everyone is protected from the financial loss of a fire. 

Traditional DB plans inherently pool longevity.  It is also possible to pool longevity within, or associated with, a DC plan.  One way would be to buy individual income annuities from an insurance company with DC balances at retirement, or even starting during one’s working years.  Like DB plans, income annuities inherently pool longevity. 

The authors, however, dismiss income annuities as inefficient, because of fees and expenses (a possibly fair argument in some cases) and lower implicit returns than in aggressively invested DB plans (a bad argument – see above). 

Modern tontines[7] would be another mechanism for pooling longevity risk during DC plan decumulation. These have already been implemented in many contexts in the U.S. and abroad under varying labels[8]. For regulatory reasons, tontines might not be currently possible in a U.S. corporate 401(k) plan, but they probably could be implemented in a public DC plan context, which is the universe in which the authors practice and the primary focus of the NIRS, and in IRC Section 403(b)(9) plans sponsored by church organizations. 

Economic efficiency does not depend on plan type

Arguing that either type of plan is more economically efficient than the other seems wrong on its face.  Providing income in retirement generally involves saving and investing during one’s working years and taking income during retirement.  Whether that process takes place inside a DB trust or a DC trust, on its own, can’t mean much.  There is nothing inherent in either DB or DC that significantly limits investments or precludes longevity pooling during retirement.  For corporate plan sponsors, however, DB may actually be less economically efficient than DC because of overpriced Pension Benefit Guaranty Corporation (PBGC)[9] premiums.

Nothing in this essay should be read as judgement that DB is an inherently bad plan structure.  Rather, economic efficiency doesn’t seem a valid criterion to evaluate differences between DB and DC, especially for public plans.  

DB advantages over DC might include facilitating workforce management, e.g., with periodic early retirement incentives, or skewing a retirement plan fixed available spend in favor of older workers, if that were a goal for some reason. Or DB may be preferred by a paternalistic employer that doesn’t mind effectively running an insurance company subsidiary and footing the bill for expensive guarantees. There are probably advantages I’m missing.

Investment risk allocation between sponsors and participants is key consideration

The allocation of investment risk and how it’s to be managed may be the most important consideration in determining an appropriate plan structure.  Good decision making thus requires accurate measurements of costs and risks. 

The NIRS study, and general practice within the public DB universe, assumes away the very real cost of DB guarantees. The result is a massive understating of DB plan cost. In the broad context of U.S. public pension plans, inadequate accounting for risk leads to consequential expensive policy mistakes. 

---

[1] A high level and essentially correct description of, and comparison between, DB and DC plans can be found here

[2] Pensions & Investments, BenefitsPRO, PLANSPONSOR, Chief Investment Officer, SmartAsset

[3] This is the authors’ modeling assumption

[4] The annual contribution in the study is calculated as a constant percentage of increasing pay over time, but the same principles apply.

[5] An excellent paper by Zvi Bodie regarding this type of claim:  Wishful Thinking About the Risk of Stocks in the Long Run

[6] For the actuaries: I calculate the present value factor of a deferred-to-62 annuity for a 45-year-old female teacher under PubT 2010 / MP2021 at 2022 to be 4.0 using 6.8% discounting and 13.1 using 2.5% discounting (to approximate a risk-free return rate, which is high as of this writing). Both factors would be about 8% higher with a 2.3% post-retirement COLA.  Feel free to check me and let me know if I got it wrong.

[7] See Fullmer: Tontines: A Practitioner's Guide to Mortality-Pooled Investments

[8] For example, see The Ambachtsheer Letter, June 2021: Transforming Capital Accumulation Plans Into Lifetime Income Plans: From Hope to Reality At Last

[9] Required under ERISA, sponsors of private sector DB plans buy insurance that provides promised income to participants, up to specified limits, if the sponsor of an underfunded plan becomes insolvent.