NYU Tax Policy Colloquium, week 1: paper by Brigitte Madrian

Yesterday we had the first session of the 2015 NYU Tax Policy Colloquium.  This is the 20^th time (in 20 straight spring semesters) that I have done the colloquium.  This year, I am happy to be co-leading it with Alan Viard.  (The full roster of people I’ve co-led it with in the past, ranked by how many weeks or years they’ve done it, except that the last two are tied, is David Bradford, Alan Auerbach, Mihir Desai, Bill Gale, Rosanne Altshuler, and Kevin Hassett.)

One sign of institutional health, I suppose, is that we can’t accommodate all of the students who want to enroll.  Luckily from an overall institutional standpoint, the law school ties my hands so that I can’t act in accordance with what would otherwise be my revealed preference of just letting everyone in.  (The problem is that the feel of the class changes for the worse if the class size grows too great.)

Although lots of other tax policy colloquia around the country have followed in our wake – maybe 20 or more, although they may not all be operating at present or in a given year – we still have a unique format, in which the author doesn’t present the paper, but rather a lead commentator focuses the conversation on a couple of main topics, one at a time.  We consult with each other first, then with the author, and then we run the show at 4 pm.

Not having the author present is admittedly a tradeoff.  I personally find the conventional twenty minute presentation upfront to be boring – including, and indeed especially, when I am the author and presenter at someone else’s session – if I have already read the paper carefully (or wrote it).  And not having the presentation induces the audience either to (1) invest more in reading it in advance or (2) not to come.  The key, for it to be a positive strategy, is that there must be enough people out there who choose option (1) rather than option (2).

For me, the tiebreaker against the author presentation is that, if you do that and then follow it up with your own commentary, the audience is just going to be sitting around forever before it gets to participate.  I don’t find that to be either fun or interesting, whether I’m in the audience or at the front.  So in my view, not shared by everyone who runs a tax policy colloquium, skipping the author presentation verges on being a necessary cost of leading and initially directing the discussion, rather than just asking for audience questions.

Anyway, on to yesterday’s session.  The author was Brigitte Madrian, presenting “Does Front-Loading Taxation Increase Savings?  Evidence from Roth 401(k) Introductions.”

While I thought it was a really good session, I don’t comment here about the content of a given session, because that would be inconsistent with our policy that they are off the record.  So here are some thoughts about the paper and the issues it raises, mainly based on expanding my notes for the session (I was in the lead this week).

The paper reports on the findings of a study of 11 firms that initially had just “pre-tax” retirement saving plans under Internal Revenue Code section 401(k), but then added Roth options.

To illustrate how these two types of section 401(k) plans work, suppose that my marginal tax rate is 35 percent at all times.  Suppose I contribute $100 to a pre-tax plan, and withdraw it after retirement once the money has doubled to $200.  I get to deduct the $100 contribution, generating a $35 reduction in my current year tax liability.  Thus, the cost to me is only $65.  Then, when I withdraw the money, it is taxable at 35%, so I end up with $130.

Under Roth, my contributions are not deductible, but the withdrawals are not taxable.  Thus, it turns out to be exactly identical – again, assuming my marginal tax rate is the same at all times, and ignoring various statutory differences between the two types of plan – so long as my nominal contribution – matching my “true” after-tax cost under the pre-tax option - is $65.  Once again, I have $65 less in my pocket up front, but end up with $130 in retirement once the money has doubled.

Suppose that, under a switch to Roth, I still nominally contributed $100, and thus ended up with $200.  I would then have changed the amount of consumption that I am actually from the present to the future.  Absent more to the story, it would violate consistent rational choice for me to contribute the same nominal $100 under pre-tax as under Roth, given that, in the rational choice model, I am aiming for a particular consumption flow, and wouldn’t change this merely due to form if my true opportunity set was the same in both cases.

Anyway, the paper finds evidence that people didn’t reduce their nominal contributions by reason of the firms’ adding Roth to pre-tax employer retirement plans.  The apparent violation of consistent rational choice (unless otherwise explainable) might come as a surprise if not for twenty years of excellent empirical research into retirement savings behavior, much of it by Madrian herself, that finds even greater violations of consistent rational choice regarding even simpler decisions, such as whether or not to save for retirement.  (I discuss this evidence and its broader implications in a recent paper that you can find here.)

I proffered two discussion topics: the paper’s empirical analysis, and the broader policy implications.

1) Empirics.

The paper is based on studies of 11 firms that added a Roth option between 2006 and 2010.  As one would expect given the skill of the researchers, it is normalized for age, salary level, and gender.  It compares people who were hired just before the Roth option was added, to those who were hired 12 months later (and thus at the same time of year) when the Roth option was available.

Suppose marginal tax rates are entirely fixed, even for the future, will be the same for each individual all times, and that there are no relevant differences between the actually available pre-tax and Roth style 401(k) plans.  Then any individual who contributes $X under pre-tax should contribute $X(1 – t) under Roth, where t is that individual’s marginal tax rate.

To know how much employee contributions should drop when a Roth option is made available, we have to know how much they utilize it.  E.g., suppose everyone’s marginal tax rate is 35%, and that given everyone’s hypothesized complete indifference they are simply told which type of plan they were in.  Then contributions should drop by 35% if everyone is put in Roth, 17.5% if it’s half-Roth and half-pretax, and zero if no one uses Roth.

The paper shows that only about 15% of the money that employees were contributing ended up in Roth plans after the option was introduced.  Hence, if we posit that everyone has a marginal tax rate at all times of 33.3%, the expected drop in pretax contributions would only be 5%.

This alone might be small enough for one to worry about losing it amid the statistical noise.  (The authors had only limited data, and published to get the ball rolling in terms of academic study of the issue – they not only acknowledge the need for further research, but very likely will be doing some of that further research themselves.)

Now let’s add the fact that pretax and Roth plans are not entirely identical in practice.  For example, Roth is better if you expect your tax rate at retirement to be higher than it is at the time of contribution.  Optionality – pretax versus Roth – inherently makes retirement saving through the employer plans more attractive (although admittedly this might lead not just to increased true saving via substitution effects, but also reduced true saving via income effects).

Anyway, the bottom line is that one can’t be sure the “rational adjustment” effect would actually show up here in the data.  So while I believe the result, this reflects my priors, as much or more than what we learn from the study.

One bit in support of independently crediting the paper’s finding was that pretax contribution rates did not discernibly differ as between firms which had different levels of Roth utilization.  I would guess the authors’ reason for not highlighting this potentially very helpful point more (although they do mention it) was that the amount of data involved was admittedly limited.

They did apparently find, although not reported in this paper, that peculiar behavior around 2008, when the financial crisis hit and (obviously) may have strongly affected people’s thinking, does not appear to have a measurable effect.

2) Policy implications

(a) Zero long-term budgetary cost – or zero “unintended” long-term budgetary cost? – The paper concludes: “These results raise the possibility that governments may be able to increase after-tax private savings while holding the present value of taxes collected roughly constant by making savings non-deductible up front but tax exempt in retirement, rather than vice versa.”

Just as a quibble, if people make (in real terms) more use of Roth than pre-tax 401(k) plans, because their fixed nominal contributions are equivalent to making what are actually greater tax-adjusted contributions, implies that there might be a positive long-term budgetary cost.  To be sure, this is an “intended” budgetary effect if we think of policymakers as wanting to encourage greater utilization of the retirement savings provisions.

(b) Pre-tax vs. Roth if Congress uses time-limited budget windows. – Because Congress tends to use short-term, rather than infinite horizon, budget windows, Roth saving “looks” cheaper than pre-tax in budgetary terms even if the true  long-term cost is the same.  Both Congress and particular legislators, making supposedly budget-neutral proposals, have not been shy about exploiting this point to engage in bogus “deficit reduction” or “budget neutrality.”  So it is plausible that, even if using Roth in lieu of pretax increased private saving, it might induce increased government dissaving.

(c) Other pre-tax versus Roth issues – The tax benefits for retirement saving under Roth are less truly pre-committed than those under pre-tax.  We can be confident that in, say, 2030, Congress won’t increase particular individuals’ taxes because they got deductions in 2015 from using pre-tax retirement savings plans.  But Roth exemption would effectively remain on the table in 2015 even if Congress didn’t explicitly and directly renege.  An example would be partly or fully replacing the income tax with a VAT or an X-tax (i.e., an individual-level progressive consumption tax that effectively combines a VAT with low-wage subsidies).

This difference could be either good or bad from the standpoint of evaluating Roth.  It makes the tax benefit less credible, but leaves Congress with more discretion.  (Those are basically two different ways, with seemingly opposite normative implications, of saying the same thing.)

One thing I don’t like about Roth style exemption is that it can result in effectively exempting extraordinary returns that people have a limited opportunity to generate.  E.g., Mark Zuckerberg would have done a whole lot better with Roth than pre-tax treatment for his initial investment in Facebook.  This is a real issue when you recall the evidence suggesting that, say, Romney (along with lots of other financial sector insiders) appears to have played games with the IRA contribution limits by grossly undervaluing initially contributed assets.  The answer, when using Roth-style savings rules, is to require that only true arm’s length asset purchases and contributions be permitted.

.

(d) Construing the choice set more broadly – It makes perfect sense for this paper to research pre-tax versus Roth, and then in its conclusion to spell out the main direct implication of its empirical finding.  But obviously policymakers should be thinking in broader terms than just pre-tax versus Roth.  The big items on the agenda today are Social Security, the use of “nudges” such as automatic enrollment in employer plans, and whether to continue using income tax “incentives” to encourage / increase retirement savings.

But note that income tax “incentives,” such as pretax and Roth, actually result in tax-neutral treatment of retirement saving – not in tax-favoring it – if one’s marginal tax rate is the same at all relevant times.  Thus, despite my general admiration for Raj Chetty’s work (with coauthors such as John Friedman) in this area, it makes my teeth ache a bit they describe the income tax rules (for example, here)  as providing “subsidies” for retirement saving.  This is only so (a) relative to an income tax baseline that discourages saving, or (b) when one uses a pre-tax plan and one’s marginal tax rate is lower at retirement than in the year of the contribution (which is, admittedly, likely to be a common scenario).

(e) Income tax vs. consumption tax – As discussed in my article on behavioral economics and retirement saving  a lot of evidence suggests that people have a great deal difficulty with intertemporal choice.  This has important implications for the income tax versus consumption tax policy debate that have not as yet been widely recognized.