My Bloomberg Opinion column “Fight the Backlash Against Retirement Saving Nudges: Everyone Benefits When People Save More for Old Age” takes issue with a piece by Andrew Biggs of the American Enterprise Institute. A few days after my column came out, Michael Newman from Bloomberg Opinion invited Andrew Biggs to respond. That led to this dialogue by email (lightly edited). I am grateful to Andrew Biggs for permission to publish this discussion.
Andrew: The disagreement may be a bit narrow, in that I’m not really against auto-enrollment as a general policy. For middle-income people who do need to save but might forget to enroll in their 401(k), it’s a good idea.
My issue is that applying auto-enrollment to low earners, as with the states’ proposals for auto-IRA plans, could have real downsides in that you’re pushing people to save when they may not really need to and when the costs to reduced take-home pay could be more substantial. Kind of wonky, I admit! But given the momentum behind the state plans, I think it’s worth looking at.
Miles: There may not be a lot of disagreement. I agree with Andrew that pushing low-income earners to save for retirement isn't needed since for them Social Security looks pretty good compared to their usual income. However in my Bloomberg piece, though I am not fully explicit on this, I take the view that if a retirement saving plan allows people to borrow from it to make down payments on cars and houses, the automatic enrollment may serve a useful role in helping them put together that down payment. What this means, though, is that the lower income the earner, the more it should be structured with the idea that the ability to take loans from one's own "retirement saving account" is the main benefit, not the retirement saving itself.
There is no question that all of this only makes sense from a behavioral economics point of view. If everyone were a rational optimizer, none of this would needed and none of it would even work, since people would just opt out whenever that was optimal. There are two departures from being a rational optimizer in play: information processing costs (of which "forgetting" is one, but only one) and internal conflict. In an internal conflict between parts of oneself, automatic enrollment can give the edge to the part of oneself that wants to save more.
Andrew: I realized my first email didn't really get at where I disagreed with your piece. For the average employee I think it's arguable that auto-enrollment worked well, depending on your take on mortgage debt. But at the low end the additional consumer/auto debt was substantially more than employee contributions and almost equal to combined employee/employer contributions. I don't find an easy benign explanation for this. That's the main issue I have.
On top of that, I'm not sure whether additional mortgage debt is due to the ability to borrow against TSP assets, to lower down-payments (maybe due to lower liquidity) or maybe just to excess optimism ("I'm rich!"). I don't know if the data exist to really answer that question, though in theory some admin data on borrowing against TSP balances could help.
Miles: The way I read the data is based on the idea that those who are financially strapped usually pay the minimum down payment. If that is the case, owing more money on a car or house would have to represent either
a higher total purchase price,
becoming able to borrow for the car or house rather than having to pay in cash, or
less plausibly, the required down payment becoming less.
The only plausible reasons for the value of houses and cars they purchase to be higher are either that they realize they are richer because of the match or that the ability to borrow from the thrift savings plan makes it easier to put together a bigger down payment. Every one of these possibilities reflects the set of choices these individuals have for buying houses and cars expanding.
The story where forced retirement saving makes it harder to put together a down payment so they make a smaller down payment depends on people who are not in the retirement savings plan making more than a minimum down payment. That seems unlikely to me.
Expanding the range of choices low-income individuals have for buying houses and cars could be either a good thing or a bad thing. It might have been a bad thing that the Thrift Savings Plan expanded the range of options for low-income individuals to buy houses or cars. I wouldn't jump to that conclusion. After all, owning a house can be cheaper than renting and for many people a car that won't break down can be crucial for reliably getting to work from areas with low house prices or rents. But if expanding choices for buying houses or cars is a bad thing for these low-income folks, it can be remedied by fixing the design of the Thrift Savings Plan. For example, if research suggested borrowing for a house was a good thing, but borrowing for a car was a bad thing, modify the rules so they can borrow for a house but not for a car.
By the way, Congress is working on a 401(k) reform package that looks mostly sensible to me. I wonder about your thoughts on that.
Andrew: I have a piece in today’s Washington Post looking at the state plans. My three main points are that there’s not a ton of evidence that low-income Americans are undersaving for retirement; this doesn’t mean auto-IRAs are a bad idea, but it does mean that the plans’ downsides should probably be given greater weight than if there were a huge undersaving problem to solve. Second, I cite the Beshears et al paper on the debt issue, in particular for less-educated workers. And third, the states’ haven’t given much (or any) thought to how the auto-IRAs will affect eligibility for means-tested transfer programs. It doesn’t take much additional assets or income to be disqualified from TANF, SNAP, Section 8 or Medicaid. Beneficiaries may have to spend down their accounts until they regain eligibility, which means state budgets may benefit more than the poor will.
In terms of the savings/debt issues you raise, it would great to have more data. I wish some of the states setting up auto-IRA plans would do as Seattle did with the minimum wage and design a study from the get-go. I’m not confident on that. It also would be great if Beshears et al continued to track these federal employees over time; I don’t know whether they’re able to do that.
But sticking to what we know so far, I see the results for the less-educated auto-enrollees as a problem. Consider that that increase in consumer debt alone offsets almost 2/3rds of employee contributions. It’s hard to find a benign explanation for that. Maybe if it’s a wealth effect then the results will be smaller for auto-IRAs where there’s no employer match, but still I find that troubling. Add auto loans and you’re offsetting 82% of total contributions and 175% of employee contributions. Maybe they’re buying bigger cars, maybe making lower down payments (or both!) An auto is more consumption than asset, as I see it. In both these cases, I don’t think the ability to borrow against TSP assets plays into things.
Maybe, as you say, the higher auto/home loan balances are due to a wealth effect. But the employee is only truly richer by the amount of the employer match, which is under 4% of pay, whereas additional auto/home loan balances are equal to about 20% of pay. I’d like to see how these evolve over time, but if the principal result is workers borrowing against their new retirement savings to buy bigger/better homes and cars, that’s kind of underwhelming.
Miles: When you say “increase in consumer debt alone offsets almost 2/3rds of employee contributions,” I notice I have a different reading. I read the paper as saying that by the time the data were cut by income level there wasn't much precision to know what was happening with consumer debt. Certainly there could be a problem, but there is at least one benign possible explanation: statistical chance in the data.
I want to be clear that my story about more expensive houses and cars isn't primarily about a wealth effect; it is primarily about being able to get together a bigger down payment and therefore being able to get a car or house loan on a more expensive car or house. Even without being any richer, they would have liked to borrow more for a house or car but weren't allowed to without the bigger down payment. Having a little more money for a down payment allows one to borrow quite a bit more. It isn't clear that this is a bad thing, though it might be.
Andrew: If you saw the Post piece, I make three points: first, there’s solid evidence (see Bee and Mitchell and Brady et al) that most retirees, including lower income ones, have replacement rates above what financial advisors typically recommend. So there’s a hurdle of justifying the need to save more that’s usually ignored. Second, there’s the issue of how the account balances interact with means-tested benefits; that could be a real issue, which neither the states nor the federal government have looked at. And third, there’s the issue of whether auto-IRA accounts lead to higher net saving.
On the last, obviously we’d all like more and better data – in particular knowledge of the actual assets the households hold and how savings and debts evolve over time. The Beshears et al paper only had data through a bit over four years, which may not be enough time to really tell, but still I think it’s important that they released what they had.
I don’t really buy the argument in the paper that larger homes should be considered a form of saving that offsets the higher debt. Obviously a home is an asset, but it’s also a consumption good that carries a lot of tax/maintenance costs along the way. Moreover, many/most retirees don’t move out of their homes, so mostly what they’ve got is simply a larger home. And if you look at work by Mitchell and Lasardi and others, they consider households carrying more debt into retirement a major (maybe the major) issue with retirement income adequacy.
You think (and probably to a degree you’re right) that the higher savings are making it easier to employees to make a home down payment, so maybe they’re able to buy rather than rent or to buy a bigger home. Beshears analyzed federal workers, most of whom own homes (even among those with less than a high school education) so I’m not sure on that. What I’d like to see is not simply changes to TSP contributions over time, but some measure of account balances to see if people are borrowing against those funds. I know I did in making my first home purchase, but I just don’t know for sure.
In terms of less educated workers, the degree of debt increase just seems too large to be helpful. Debt rises by three times more than total TSP contributions and seven times more than employee-only contributions (perhaps relevant, given that auto-IRAs can’t have an employer match). Moreover, much of that seems to be auto and consumer debt, where the housing-as-asset argument doesn’t apply.
I’m not exactly against auto-IRAs; on one had, we know that saving through work is more effective than setting up IRAs on our own, but we also know there’s leakage as employees shift between employers. Auto-IRAs get around that. At the same time, too many of the states seem in rah-rah mode rather than really weighing the issues and working out practical problems like means-tested benefits. So I think there’s still work to be done.
Miles: Your point about interaction with safety-net programs is a good one. And I agree that one should question whether or not retirement savings plans should be designed to also help people get into bigger houses. That points to the broader issue that the precise design of retirement savings programs matters a lot. And, as you say, if low-income folks have adequate retirement savings (in part because social security looks pretty good to them compared to what they are living on before retirement), the side effects from design details can be a big deal compared to any benefit from encouraging extra retirement saving. If those side effects are good, that case should be made. If the side effects are bad from other design features, that could weigh heavily.