William Bernstein is the most well-known proponent of Liability Matching Portfolios (LMP), which is a specific kind of “safety first” retirement portfolio.
Safer assets (individual bonds or income annuities) are matched to specific retirement spending needs, and then the remaining discretionary wealth can be invested more aggressively in a “risk portfolio.”
LMP are often offered as an alternative to “probabilistic” portfolios. That is, a portfolio where based on past performance and market characteristics you will have a 99% chance of having enough funds at a certain point in time.
The problem with LMP is that it underestimates the costs substantially. It does this by constructing unrealistic scenarios and not actually matching liabilities.
This Wall Street Journal article shows how LMPs are often described as working in practice. On the surface, everything looks good. But when you dig into the details, you start to notice the shortcuts taken to make LMP look more viable.
- The investor has a $750,000 LMP portfolio. $250,000 will be spent in the next five years, until Social Security kicks in. The remaining $500,000 needs to match all liabilities until death.
- The investors needs $21,700 a year on top of Social Security. The LMP actually only matches 21 years of liabilities.
How many liabilities do you need to match?
If your going to match liabilities, first you need to decide what your liabilities are.
Do you think it is realistic to forecast your liabilities 30- to 40-years out? If you are currently 50, do you think your 20-year old self could have successfully forecast the dollar figure of your liabilities?
Inability to accurately guess expenses
The first problem that a retiree has is even figuring out what they will be spending money on. Most likely they are performing a straight line projection. Whatever expenses you have today is what you will have until the end of time.
Except we know this isn’t true. On the one hand, we know that retirees face increased medical expenses that generate liabilities that are difficult or impossible to predict.
Fidelity Benefits Consulting (2012) estimates that a representative couple retiring in 2012 will face almost $230,000 in health care expenses that are not reimbursed by Medicare. These costs do not include nursing home care, dental, vision or over-the-counter medications.
Does that mean your LMP need to account for an extra $75,000 in Year 16 of retirement, since that’s when you will break your hip? That medical spending will be lumpy. It will be $0 a year until you are 72 and then your health deteriorates and you find yourself with an additional $10,000 a year in out-of-pocket medical liabilities. (The Centers for Medicare and Medicaid say that an average retiree in poor health should expect to pay $10,000 a year in health care costs.)
But we also know that other retiree expenses don’t stay flat; liabilities go down. Aguiar and Hurst (2005) find that food expenditures decline 17% at retirement. Did you account for that? Oh, but Haider and Stephens (2007) found that people reduce spending on food when they retire by about 5–10%. So do your food liabilities decrease by 5% of by 17%? Which one should your LMP plan for?
There is a large body of research clearly establishing that consumption changes over time for retirees. In David Blanchett’s “Estimating the True Cost of Retirement” he digs into the data and shows us:
Not only does the overall consumption change but what we spend on changes as well:
So it seems like a fool’s errand to try to predict, 30-years in advance, what you will be spending your money on.
But it gets even worse when you take inflation into account.
CPI-E and Personal inflation rates
Do you think your local government will never raise property taxes? In New Jersey property taxes went up by 2.5%, which was much faster than the overall inflation-rate.
You were expecting costs to go up, that’s what inflation is all about. It is what TIPS and I-Bonds protect against. The problem is that there is a difference between the inflation rate you experience and the overall inflation rate experienced by every single person in America.
The BLS constructed an experimental Consumer Price Index for Americans 62 years of age and older (CPI-E). They found that over a five-year period (1990 to 1995) this CPI-E rose faster than other CPIs: The CPI-E rose 12.7% more than the CPI-W.
So even if you were able to perfectly predict what you were spending money on, you have difficulty predicting how much it will cost 30-years from now. And while you can TIPS that track the overall average American inflation, you will have difficulty buying financial instruments that track CPI-E, much less your own personal inflation rate.
So far things look dire but they’re going to get even worse. Let’s pretend you can actually calculate what your liabilities are going to be next year. And the year after. And the year after that, too. How many years out do you need to match liabilities for?
LMP explanations use unrealistic life expectancies to make the approach look more successful. In the Bernstein article, the example retiree only had an LMP big enough for 21 years of expenses. But you have far more than 21-years of liabilities. There are two factors that LMPs rarely take into account:
- The vast majority of retirees are not single. So you need to look at “couple life expectancy”. Your husband won’t be too happy if all the money runs out when you die and he’s still alive and has another decade of liabilities.
- Only 13% of couples are the same age. For a couple that gets married before age 40, the wife is (on average) 2–3 years younger than the husband.
You put these things together and it means that when the husband retires at age 65, the wife is 63, and the portfolio needs to provide 42-years of income for the couple in order to reduce the failure rate to 1%.
If you only have 21-years of liabilities matched in your LMP than 20% of the time the couple will outlive the LMP. That is, the LMP will fail 20% of the time.
(Numbers taken from https://www.aacalc.com/calculators/le.)
So in Bernstein’s example we have 42-years of liabilities but an LMP that only covers 21-years of liabilities.
That hardly looks like you’ve matched your liabilities, does it?
No floor: Some of the liabilities for some of the time
How do LMPs get around these issues? By not matching liabilities and instead relying on probabilistic success—the very thing the LMP is supposed to avoid.
In Bernstein’s article he says the LMP actually provides 28-years of liability matching. He arrives at that by assuming, based on historical probabilities, that the LMP will not only match liabilities but grow at 1% real per year. The LMP is composed of “TIPS and high quality corporate bonds”. Currently a 5-year AAA corporate has a 1.62 yield. Inflation is currently 1.1, so we’re looking at a 0.5 yield. That’s 50% of the yield Bernstein assumes to make his LMP stretch out to 28-years.
What does a retiree do if their LMP fails to match liabilities because growth rates don’t match assumptions?
And even if Bernstein’s growth assumptions for the LMP turn out to be accurate, what does a retiree do after Year 28 for the next decade of retirement?
The LMP answer: rely on your non-LMP portfolio to meet your liabilities.
Liability Matching Portfolios fail for individuals
The concept of matching liabilities works well for insurance companies, where the law of large numbers smooths out many of these issues. An annuity company doesn’t need to predict your spending pattern because they have thousands of other annuitants who will balance it all out.
But as an individual you don’t have that luxury. If any of your assumptions are wrong, then you won’t match your liabilities. And even being wrong by a very, very small amount can compound to large numbers for a 42-year retirement horizon.
Liability Matching is too expensive
Why does Bernstein’s LMP example only cover 21-years of retirement and then rely on probabilistic success thereafter? Because liability matching is simply too expensive for individuals.
To really construct a LMP you need to invest approximately 40-years of expenses into TIPS. (Of course, TIPS only go out to 30-years so to build a 40-year LMP you’ll need to engage in some complicated financial engineering.) That’s an extra 19 years of savings required beyond Bernstein’s example. That results in decades of extra employment to save up that money. The example retiree in Bernstein’s article needs $2,000,000 to construct a real LMP. But he only has $1,000,000 at age 65. Being a million short is not a trivial amount of money.
So LMPs fall back on probabilistic success, though they hide it under the rug. But if probabilistic success is good enough for a 94-year old retiree then why not for a 69-year old retiree? If you’re willing to live with a 20% chance of outliving your LMP, then why not take that same 20% failure rate and apply it to other areas of your retirement planning? The Trinity Study on withdrawal rates says that, if you’re willing to live with 20% failure rates, you can take out 7% a year (instead of the 4% a year usually bandied about). But then why does the “non-risky” part of your portfolio have a 20% failure rate?
