Does Migration Make Regional Inequality Worse?
No. But Sudden, Unexpected Shocks to Migration Might!
Today I’m going to review a paper from MIT that argues very interestingly that domestic migration, far from equalizing labor supply and demand around the nation, actually exacerbates regional inequality, specifically inequality in unemployment rates. They key finding of the paper is that a sudden spike in in-migration into a city from other, distant cities, is associated with a follow-on decline in the receiving city’s unemployment rate. This decline fades within a few years.
Adam Ozimek thinks this means:
I am less sure [that migration equalizes unemployment], because I see downsides to a falling population and upsides to a growing population. A new study from economist Greg Howard gives some empirical validity to my doubts.
That is, Adam thinks, and a reasonable reading of this 79-page tome could suggest, that this paper indicates that falling population (or, put another way, negative net domestic migration) is associated with worse economic outcomes.
I think he’s misreading the paper, and indeed the paper’s own author may be overstating what he’s actually proven though, to be clear, the paper is exceptionally thorough and well-researched. The author has produced a lengthy, exhaustive paper which is extremely honest about its own limitations and problems, is scrupulous about checking for possible errors, and that has left me extremely convinced of its core, empirical finding.
To repeat that core finding: when there is a sudden spike in in-migration into a city that is unassociated with previous migration patterns, it pushes the unemployment rate downwards relative to what we’d expect given past economic performance.
However, I worry there are some key shortfalls in this paper’s ability to settle the debate. The key problem the author is trying to solve is that the population-economic relationship is endogenous: they almost certainly both cause changes in each other. It’s a chicken-or-the-egg question. But see, when normal people hear “chicken-or-the-egg,” they hear “unsolvable problem.” When economists hear that, they think, “Instrumental variable time!”
Adam summarizes the author’s instrumental variable pretty well:
Migration from metro A to metro B in a given year is predicted by combining historical migration patterns from metro A to metro B, and annual migration from all other counties to county A in that year. Then migration to metro B is predicted for metro C, D, and so on. Finally, all the individual county migration rates to metro B are summed. This creates an estimate of annual migration to metro B that is not related to current economic conditions in metro B, but is driven by factors outside the metro. If this sounds a little confusing, that’s OK. Just know that this makes for an important new innovation for estimating the effect of local population growth.
The one caveat I’d make to Adam’s comments is that, as the author repeatedly acknowledges, this method isn’t actually strictly new, but variations have actually been used for many studies of many different migration-related topics.
Now, this instrumental variable, the author acknowledge, has some shortfalls. For example, if two geographically distant cities have similar industry composition, and some flagship industry experiences a major economic shock, their bilateral flows to each other will be biased. The math on this is boring, but the upshot is this will overstate the strength of the model’s results, giving a misleadingly positive estimate of what inmigration does to employment. That is, their measured effect is an “upper bound” on the actual effect size. They also acknowledge that their model can’t control for the complex network effects of hundreds of cities. Outflows from a given county could go to a variety of cities; if they chose San Francisco instead of LA, that will look like a loss for LA, even if LA was actually outcompeting every city other than San Francisco. This bias has the same result: it will overstate the strength of the model.
Then we need to think a bit about the migration estimate itself. The author used IRS Statistics of Income data. Using IRS tax data excludes tens of millions of individuals, including many migrants, and is especially likely to exclude people not in the labor force in either of the two years, or who have unusual circumstances of some kind. Furthermore, IRS data collection methodologies have changed several times over the time period specified. This estimate will track employed migrants more completely than non-employed migrants, with the result that you will be more likely to ignore cases where inflows may have actually boosted unemployment, because many of those unemployed/discouraged/NILF migrants don’t exist in the IRS dataset. This creates yet another bias, which may lead the results to be overstated.
The author also suggests that ACS migration data isn’t available for counties, which is weird, because it is, just not on an annual basis.
Also note I’ve been careful to talk about inmigration. The study is not about out-migration, or even net-migration. There is no evidence provided that an out-migration-led shock would be similar to an in-migration led shock. The author also finds out-migration is less sensitive to labor demand shocks than in-migration. Furthermore, for the headline example of Great Recession unemployment impact that forms a centerpiece of the study, the author creates a counterfactual of what unemployment would have been without a “migration accelerator,” but only briefly mentions that, by the way, this simulation radically altered estimated inmigration, but kept outmigration patterns identical, as if outmigration patterns were immune to economic shocks or unrelated to inmigration patterns. This one-side-of-the-coin approach is methodologically understandable, but, again, should reduce our mental estimate of the extremity of the results.
The Appendices go into great depth testing different specifications. No single specification change or robustness check overturns the identified result, a very good sign. However, it seems reasonable that the sheer number of un-controlled and identified endogeneity problems would, if all simultaneously controlled for, almost certainly amount to a great reduction in the size of estimates.
Next up, let’s think about the exact question being asked by this paper. A central claim of the paper is that free migration may worsen regional inequality in a currency union. Which is weird, since that’s not what was tested. What was tested was actually shocks to free migration. That is, the paper is not actually exploring the effect of long-run, persistent outmigration. Nor is it exploring the determinants of a given level of migration. Nor does it give us any guidance about whether high net inflows may lower unemployment. Rather, this paper is exclusively focused on the short-run effects of shocks to migration.
Let me give an example. Ohio has had negative net interstate migration for nearly every year since 1900. This paper gives zero evidence about whether that has raised or lowered Ohio’s unemployment rate both because (1) the paper isn’t about net rates and (2) it is only about short-term shocks, not long-run trends. However, if Ohio were annexed into Pennsylvania and the state government offices were all closed down with their functions shifted to Harrisburg, the ensuing negative shock to inflows would create even more negative spillovers; the declining inflows (added labor supply) would not fully offset lost employment, and so unemployment would rise by more than you would expect from just the layoffs.
It’s a very narrow focus. It’s only about shocks. So if you think this paper is about persistent outflows, you’re wrong.
Oddly enough, the author seems confused as well:
That word “persistent” is doing a lot of work. What the author means is that, by about 3–5 years after an inflow shock, the amount of unemployment change driven just by local labor demand is small, but spillovers remain meaningful. However, by year 6–7, all migration spillovers have vanished.
But when we talk about “persistent differences in regional outcomes,” no regional economist means “differences lasting 3–5 years.” We mean differences lasting 30–50 years. This paper has zero explanatory power for those, because it doesn’t even show that persistently low inflows drive high unemployment, it just shows that actively declining inflows drive high unemployment.
That said, I do think persistently low inflows are bad for employment; it just needs to be noted that this paper does not venture any answer to that question.
Furthermore, the paper always controls for gross migration rates. Think about that. It offers zero explanatory power about whether or not we would benefit from higher overall rates of migration, because every estimate is a specified effect for a given gross migration rate! So if we raised gross migration by 30,000 people moving to San Francisco, this paper predicts a dramatic change in regional inequality for the ensuing several years, and better economics for San Francisco. But if we raised gross migration by 100 more people moving to every metro area (roughly equivalent), there’d be virtually no change whatsoever in regional inequality. All that matters for this paper is a given metro area’s disproportionate capture of inflows.
This is very useful knowledge. It means, if we want to be super-invasive technocrats, we can preserve the many benefits of higher migration that would accrue from migration vouchers by only making those X number of those vouchers applicable to a given city. That is, only allow voucher-moves equivalent to X% of the prior-year’s migration. This would be a ridiculous and invasive program and does raise questions about the viability of a voucher program for migration as I’ve called for. I take that issue seriously. At the same time, I don’t think the real effect of higher mobility would be more concentrated inflows. I think we’d get less concentrated inflows, as the people receiving vouchers likely would still face other constraints that would prevent an Everyone-Goes-To-San-Francisco scenario.
The key mechanism the author identify for explaining why inflows boost employment is housing. Inflows drive up local housing prices, demand for houses, and hence employment in construction. This seems totally plausible to me. They also find that this effects vary based on “local housing price elasticity,” which basically means how easily you can add new housing. Areas with little available land and tight zoning have low elasticity, and vice versa. The result is that inflow shocks in low elasticity areas (think San Francisco) drive up prices and employment more than in high elasticity areas (think Houston). In other words, if cities had loose zoning rules and lots of available land, inflow shocks would not boost employment nearly as much. In other other words, the observed effect of migration shocks boosting regional inequality is a product of strict zoning rules and the scarcity of land as much as it is anything else. This, then, suggests that tight zoning in combination with strong local boom industries creates extremely high rents for landowners, who are likely to be non-migrants.
This matters, because the paper points out that migration adds volatility to non-migrant incomes under their model. This is a cost to migration. But if that volatility is itself driven by tight land use rules, and the result of those rules is super-normal returns for non-migrant landowners, then all we’re really seeing is a transfer among non-migrants from non-landowners to landowners.
Finally, my last note is almost a philosophical criticism. The author notes that his model assumes that migration is unanticipated. That is, everybody is surprised when these in-migrants show up. He also, like the traditional models he rightly critiques, assumes that local employment begins after local residence. Neither of these assumptions seem reasonable to me.
Local businesses, and especially real estate, put in lots of effort to predict the next season or year of economic activity. Developers in particular engage in market research, and often have multi-year production timelines. There are some genuinely unexpected inflows (for example, Hurricane Katrina migrants to Houston), but many inflows, even sudden shocks, are at least partially expected. This chart seems telling:
Huh, that’s weird, even before there was a labor demand shock, inmigration was already rising! This problem does not exist in all the various charts and specifications, but does crop up in several of them.
But beyond this expectations problem, which the author acknowledges several times at some length, there’s a second problem, maybe.
I say maybe because I’ve tried to game out this next problem in my head and I’m not sure exactly what effect it would have. I’ll call it the “Recruitment Problem.”
The Recruitment Problem is simple: both the traditional framework and this study’s new framework wrongly assume that inflows come seeking employment, rather than having already obtained employment.
The classic story presented by the author assumes that people arrive in a city unemployed and obtain employment after arrival. This ordering, residence and then employment, is assumed under both traditional and new models. It is the basic underpinning of the idea that inflows should, at least in the short term, increase unemployment.
This model is true for some people, but not most. If you trust the answers people give to the Census Bureau about why they moved (and, to be honest, I’m a bit skeptical, but it’s all we’ve got), then the correct ordering becomes clear. In 2016, 3.8 million people reported changed residences for a new job or for a job transfer; that is, the new job preceded the move. They firmly expected a specific new employment. Meanwhile, just 530,000 people moved to look for work or due to a lost job.
But even many decades ago, long-range recruitment was common, and in-migrants who did not obtain employment often wouldn’t claim residency, and wouldn’t file taxes in the new area. Many people receive a job offer prior to migration. That is, local labor demand has direct access to non-local labor markets.
This is true even in unskilled labor! In fact, the classic historical case of this in international migration is the Bracero Program, focused on Mexican unskilled labor, and the classic domestic case is 1920s-1970s recruitment of non-union unskilled or semi-skilled Appalachian workers by automakers in Detroit.
The reason the unemployment rate may fall, then, is partly that migration does not monotonically increase number of unemployed workers, but rather monotonically increases the number of employed workers! Local labor demand in fact has a very long reach.
Think of it this way. Kuat Drive Yards is located in the city of Kuat, and makes Super Star Destroyers. There is a demand shock for Super Star Destroyers due to imperial military buildup. So KDY has to recruit workers. The “labor demand” does not occur when an employed person walks in their door, it occurs when KDY begins its search! So KDY sends out recruitment bulletins over the Holonet that they need individuals who speak binary and work well with automated machinery at a semi-skilled level, and who aren’t politically compromised (i.e. can get clearance). Tatooine is far from the rebellion and has lots of people who have to work with complex machinery every day just to get water to survive, so makes good recruiting ground. A KDY recruiter finds a talented young moisture farmer and hires him, but our young moisture farmer complains he can’t move just yet because (1) he wants to go to Taashi station to pick up some power converters and (2) his uncle won’ let him leave until after the next harvest. So KDY maybe took 1 month to find applicants, found somebody they like, and now wait 6 months for him to actually walk in the door and start working.
The labor demand occurred prior to the migration, but the measured labor demand (i.e. change in destination employment characteristics) occurred 7 months later! This is a big endogeneity bias! I don’t know what the average turnaround time is between “we need to hire someone” and that person walks in the door and starts working, but I’d think it’s at least 1–2 months, possibly higher. Some positions go unfilled for an extremely long period of time.
This being the case, official employment measures will fundamentally mis-measure the timing of labor demand shocks. This error doesn’t matter much at all for most things; it’s just a few months. But when you’re talking about YoY changes over long distances and making strong claims about the exact timing of an effect where causality could plausibly run either direction, it matters quite a bit to know the order in which labor demand expression, actual employment, and residence occur.
Broadly speaking, this paper is an extremely valuable contribution to the study of population geography and economics. It shows fairly compellingly that sudden inflow shocks do not create unemployment: in fact, they may reduce unemployment, and they produce meaningful economic spillovers. The paper demonstrates this for both domestic and one case of international migrants. The main apparent mechanism of this effect is the housing market, where tight zoning rules drive up the price of housing and employment in housing-related sectors. There is comparatively little benefit of higher inflows to other sectors, and indeed traded goods employment may actually shrink! It may be that traded goods producers are sensitive to local land costs (farmers, large manufacturing plants come to mind).
On the broader question of how population change impacts economic growth, this paper offers no answers. Sorry, Adam! With no test for outmigration or net migration, and indeed no population specification at all, this paper simply offers no answers one way or another on whether population growth or decline has meaningful effects on a region. However, it does seem to suggest that low rates of inflow may be concerning, regardless of net change in population. This is in keeping with lots of research (and my own personal priors), that population churn matters a very great deal. This paper suggests that certainly inflow shocks, and possibly high levels of inflows generally, have positive impacts on many labor market outcomes. The trick, then, is not how to raise population, but rather how to induce positive inflow shocks to struggling areas.
PS — Also, if I read the study right, inflow shocks won’t have as positive employment effects for localities with tons of land and loose zoning? That doesn’t seem right to me, but maybe I’m misreading.
Check out my Podcast about the history of American migration.
If you like this post and want to see more research like it, I’d love for you to share it on Twitter or Facebook. Or, just as valuable for me, you can click the recommend button at the bottom of the page. Thanks!
Follow me on Twitter to keep up with what I’m writing and reading. Follow my Medium Collection at In a State of Migration if you want updates when I write new posts. And if you’re writing about migration too, feel free to submit a post to the collection!
I’m a native of Wilmore, Kentucky, a graduate of Transylvania University, and also the George Washington University’s Elliott School. My real job is as an economist at USDA’s Foreign Agricultural Service, where I analyze and forecast cotton market conditions. I’m married to a kickass Kentucky woman named Ruth.
My posts are not endorsed by and do not in any way represent the opinions of the United States government or any branch, department, agency, or division of it. My writing represents exclusively my own opinions. I did not receive any financial support or remuneration from any party for this research.
