IMF advocacy for immigration: some caveats

The other day I came across mention of a chapter in the IMF’s latest World Economic Outlook on the economics of immigration.  It turned out to be only half a chapter (from page 183) but it had some interesting discussion and material.  It probably won’t surprise anyone that although immigration is a long way from the IMF’s core responsibilities, the Fund is pretty gung-ho on the benefits.  My own stance is, of course, more skeptical: I doubt the economic benefits to recipient countries are typically anywhere near as large as the enthusiasts make out, and in New Zealand’s specific case I think the evidence increasingly suggests that high rates of immigration in the post-war decades (continuing to today, as strong as ever) have been detrimental to the economic fortunes of New Zealanders as a whole.

Whatever the truth is, the IMF might want to be a little more careful as to how they present the material in support of their claim.  The World Economic Outlook is one of the flagship documents of the Fund, widely circulated and discussed (including at the Executive Board level) before release, not just some obscure researcher’s working paper.

And so I was interested to find this

There is a positive association between long-term real GDP per capita growth and the change in the share of migrants (Figure 4.16, panel 1).

I noted the careful wording (“association” rather than “causal relationship”) and went looking for the chart.


Which left me a little puzzled. The text said there was a relationship, illustrated by this chart, between increases in migrant numbers and real GDP per capita growth.  But the chart itself showed only a relationship between total real GDP growth and immigration.  That sort of positive relationship was hardly surprising –  growing migrant numbers raises the population, which tends to raise total GDP –  but the interesting question was surely about per capita growth.  And it wasn’t just a labelling error –  it was easy enough to reproduce the Fund’s chart, showing what the labels said it showed.

And so I started digging around.  The IMF uses a group of 19 advanced countries, the choice presumably limited by data availability (so, for example, Belgium, Italy, Iceland, Japan and Korea –  among advanced OECD countries –  aren’t included).  And they focus on 1990 to 2010, presumably also to ensure that all the data (for some of the background modelling) was available.  And rather than total migrant numbers, they look only at migrants over 25.  In the charts that follow, I just follow their basic approach/time period.  In terms of country selection, I’d note just two caveats: first, a large part of Luxembourg’s GDP is produced by people who don’t live in Luxembourg but commute in each day (so GDP per capita can be a bit misleading) and in Ireland the corporate tax system has helped contribute to a huge gap between GDP and GNI –  and it is the latter that measures income accruing to the Irish. Finally, it is worth noting that more than half the countries in the sample are in the euro, and 1990 to 2010 covered the period of convergence to the euro, and then the full effects of imposing a common interest rate on quite different economies.

With those methodological notes out of the way, what did the chart of growth in GDP per capita over 1990 to 2010 look like when graphed against the change in the migrant share of the (over 25) population over that period?


The red dot (hard to see, but lower right) is New Zealand.  The outlier (top right) is Ireland.  Excluding Ireland, there is almost no relationship at all, and certainly not one that would pass any tests of statistical significance.

And in case you do want to include Ireland, bear in mind that the big surge in immigration to Ireland came after the most rapid growth in GDP, or GDP per capita.


Over 1990 to 2010, Ireland’s strong growth in real per capita GDP was pretty clearly not caused by immigration.

GDP per capita has its limitations, and the Fund –  and most immigration advocates –  typically argue that the most valuable gains from immigration arise from improvements in productivity. So using, the Fund’s period, the Fund’s sample of countries, and the Fund’s immigration variable, what did the (simple bivariate) relationships look like with, first, real GDP per hour worked growth, and then (often regarded as the best sort of productivity growth) multi-factor productivity (MFP) growth?

(In all these chart’s I’m drawing GDP and productivity data from the Conference Board’s Total Economy Database.)

Here is the relationship with real GDP per hour worked.

imf migration and gdp phw.png

Again, New Zealand is in red (lower right) and Ireland is the outlier.

This relationship might not be very statistically significant either but –  at least excluding the Irish outlier –  if anything it runs in the wrong direction for the IMF story.  Over this period, and on this particular measure/sample, there was a modest negative association between immigration and labour productivity growth.

And what about MFP growth?


That relationship, especially excluding Ireland, is even more negative than the one for labour productivity.  And just to confirm that even on MFP, the immigration surge in Ireland came well after the best of the MFP growth.


All in all, on these measures, for this sample of countries, over this period, there doesn’t seem to be much left of the IMF’s story.  Yes, immigration obviously tends to make economies bigger in total, but there is little sign in the informal analysis that it has made them more productive, and thus made the average individual citizen of the recipient country better off.

Of course, where it can the IMF likes to rest its claim on more sophisticated analysis than that.  Later in the chapter, they report that

Recent research suggests that migration improves GDP per capita in host countries by boosting investment and increasing labor productivity. Jaumotte, Koloskova, and Saxena (2016) estimate that a 1 percentage point increase in the share of migrants in the working-age population can raise GDP per capita over the long term by up to 2 percent.

The recent IMF “spillover note” that is drawn from is available here.  The authors use much the same countries, the same immigration variable, and the same sample period as in the WEO analysis above.   They also focus on the level of real GDP per capita, and the level of productivity, not just growth over a particular period.  Their approach has a number of advantages over earlier studies (including the focus just on advanced countries) and as the Fund notes, the estimated real GDP per capita gains are less than in some previous studies.

I’m not a technical modeler, so I’m not going to try and unpick the paper on those grounds.  My simple proposition is that the results do not, even remotely, ring true.

Here is a chart from the paper showing the stock of migrants in the sample countries.


Think about France and Britain for a moment.  Both of them in 2010 had migrant populations of just over 10 per cent of the (over 25) population.  If this model was truly well-specified and catching something structural it seems to be saying that if 20 per cent of France’s population moved to Britain and 20 per cent of Britain’s population moved to France (which would give both countries migrant population shares similar to Australia’s), real GDP per capita in both countries would rise by around 40 per cent in the long term.  Denmark and Finland could close most of the GDP per capita gap to oil-rich Norway simply by making the same sort of swap.    It simply doesn’t ring true –  and these for hypothetical migrations involving populations that are more educated, and more attuned to market economies and their institutions, than the typical migrant to advanced countries.

(The study produces similar results for real GDP per person employed, but they do not test the relationship with either GDP per hour worked, or with MFP.  The authors suggests that the gains from immigration come through an MFP channel, but this seems doubtful –  especially over this period, and this sample, given the bivariate MFP growth chart above).

There are other reasons to be skeptical of the results in this IMF paper.  Among them is  that there is a fairly strong relationship between the economic performance of countries today and the performance of those countries a long time ago.  GDP per capita in 1910 was a pretty good predictor of a country’s relative GDP per capita ranking in 2010, suggesting reason to doubt that the current migrant share of population can be a big part of explaining the current level of GDP per capita (and some of the bigger outliers over the last 100 years have been low immigration Korea and Japan and high immigration New Zealand).    In fact, I’ve pointed readers previously to robust papers suggesting that much about a country’s economic performance today can be explained by its relative performance 3000 year ago.  How plausible is it that so much of today’s differences in level of GDP per capita among advanced countries can be explained simply by the current migrant share of the population?

And then, reverting to bivariate charts (but one from a relatively recent IMF working paper)


Total hours growth is not just determined by immigration, but differences in immigration rates account for a large part of the differences in population growth, and growth in total hours worked, over long periods of time such as those in this chart.  There is just no sign, over that period and those countries, of the longed for link between productivity growth (here MFP/TFP) and growth in anything remotely linked to differences in the volume of immigration.

To revert just briefly to the IMF WEO chapter, one of the advantages of looking at just 18 or 19 countries is that the authors should be able to illustrate their point, at least impressionistically, by reference to individual pairs of countries.  If nothing else, it is a bit of a realism check, but it can also help make the overall story seem more persuasive. But there is nothing of that in the IMF’s discussion and advocacy.   And nor is there any effort to deal with what might reasonably look like problems for the story.  The two countries with the largest increases in migrant share over 1990 to 2010 were Ireland and New Zealand.  In Ireland, as I’ve shown, the immigration clearly came after the peak productivity gains –  perhaps a case of sharing the gains, but hardly one where immigration looks causative.  And New Zealand, well….readers know the New Zealand story.  In 1990 we were supposed to be well-positioned to catch up again with the other advanced countries –  the sort included in the IMF sample –  and we had a big migration surge (by international standards of pretty good quality migrants) and yet over the full 20 years we’ve had among the lowest productivity growth rates (labour and MFP) of any of these countries.

Perhaps there are some other countries, or country pairs, where the intuitive case is more compelling.  It is shame the IMF didn’t put in the effort to find them