Keenan Viney
Abstract
“Regional Economic Shocks and Migration” by Andre Bernard investigates the economic factors that induce migration across regions. Rather than studying migration across provinces, this paper uses the Longitudinal Administrative Database (LAD) which has individual data across census agglomerations (CAs) and census metropolitan areas (CMAs). Theory suggests that the decision to migrate is some function of the probability of finding a job in the new region and the expected income that this new job would bring. Bernard’s paper concentrates on the economic factors that induce migration, whether prevailing regional economic conditions or personal income change can explain the probability of an individual moving.
The data for this study covers 2000-2008 and several broad themes emerge from the results. First, regions with smaller populations tend to have higher migration rates. These migrants are attracted to large metropolitan areas but at the same rate as other Canadians.
Table 2 shows, what is effectively, a transition matrix of migration between regions of different sizes, it reveals that people overwhelmingly move from smaller to larger CAs. Secondly, the economic conditions in origin and destination region were similar on average; there is no consistent migration due to high regional unemployment (compared to the national rate) or a one dollar per hour wage decrease over the previous two years, neither factor make those living in the effected region more likely to migrate. However, although regional economic conditions do not predict migration, personal income changes do predict the migration of workers. If personal income declines by 30% or more, a person is 82% more likely to leave their region within one year compared to someone in the same region with a stable income. Even when the income decrease is between 20 and 30% across two years, this still increases the probability of migration within a year by 46%, which is less but still a substantial increase in the chance of making the decision to migrate. A related result, is the finding that an increase in personal income over two years of 30% tends to increase the probability of migration by 46%. From these statistics we can conclude that the region’s economic status does not significantly impact the decision to migrate while changes in personal income, both positive and negative, make people more likely to migrate.
The results presented so far hold when the date is disaggregated by sex but different patterns emerge when examining three age categories separately. The three groups; those aged 20-34, 35-44, and 45-54 still do not migrate based on regional economic disturbances but personal income shocks affect each group differently. In response to a personal income decrease of 30% over two years, the oldest cohort increases their probability of migration by 98% whereas the youngest cohort increases their probability of migration by only 64%. Surprisingly, those aged 35 to 44 have the largest change in migration due to the income shock at 106%. Bernard suggests that since middle aged people tend to have stable and upward moving wages, income decreases are coming from layoffs which are more likely to induce migration. These results should be taken with caution because even though the youngest cohort does not respond as readily to personal income change they are the most mobile group.
The final group that Bernard examines are the regional migration decisions of recent immigrants. Unlike other Canadians, those who arrived in the last 10 years do seem to respond to a 1% increase in the regional unemployment rate by increasing their probability of migration by 10%. Bernard suggests that this result may be driven by immigrants being more mobile which would make them sensitive to regional economic conditions. This explanation is puzzling given that the probability of an immigrant migrating between regions, due to a 30% decrease in income, is only 71% which is less than the general population. If it is true that immigrants are more mobile than other Canadians, we would expect immigrants to have a larger migration response due to personal income decreases.
The incidence and distribution of migration can have pronounced economic effects and thus migration requires careful consideration by policy makers. As a first pass, “migration serves as a market adjustment mechanism” by allowing labour to move to regions with the strongest demand (Bernard, 2011). Migration is simply a method of allocation which pushes the economy toward more efficient outcomes. Thinking about migration as solely a market clearing device is really only valid in a partial equilibrium framework. Sufficient emigration from a region depletes the tax base which can cause disruption of municipal services. Another negative effect is if all of those leaving the region are of a certain type, for example high-skilled or young workers, this could mean that a temporary economic shock could have long term structural consequences. Policy then, should not restrict regional migration and thereby decrease allocative efficiency, rather it should seek to minimize these second order effects which tend to induce long term regional economic malaise.
To counter the negative effects of migration from a region a policymaker must address the difficulties that arise that are highlighted by Bernard’s paper. First we see that migration due to personal income shocks would shift the age structure. Middle aged people migrate at a higher rate which Bernard suggests is driven by outright layoffs as opposed to hours or wage reductions. One possible solution would be to increase Employment Insurance (EI) duration in order to facilitate a longer search before the decision to migrate is made. The problem with changing the duration or benefits of EI is that it would not reliably induce someone to change their migration decision when EI payments cease. Giving someone longer to look for a job may simply result in a better job match in the region they will subsequently move to. A similar problem arises if we try to better prepare young people to work within their own regions. Policymakers could, in principle, orientate the education system towards the industry in each particular region. Once again we have a policy that would be highly inefficient and would exacerbate structural unemployment as the regional and national industrial mix changes over time. Policy that tries to influence the migration decision ultimately disrupts allocative efficiency in the labour market which decreases welfare.
To address the diminished tax base and resulting interruption of municipal services, a second order effect of migration, Federal policy has consistently used transfer payments to arrive at a more equal distribution of government services. The problem with this approach is that regions with an influx of regional migrants do not receive the requisite level of government spending which acts to decelerate growth. Rather than subsidizing declining regions while growing regions struggle with a lack of public investment there may be organizational structures which would mitigate changes in services due to tax base fluctuation. Governments at all levels can focus on ensuring that services are scalable and standardized so that, like workers, they are mobile and less vulnerable to changes in regional government revenue.
The difficulties in making specific policy recommendations are borne out of the econometric issues of the Bernard paper. The first problem with the results is that the personal income changes are not exogenous. As an example, even though young people are the most mobile cohort they are also more likely to voluntarily take pay cuts in order to work fewer hours. This means that the study is biased in its estimation of the young cohort’s response to income changes because whether the change was voluntary or involuntary is not observed and hence is not separated in the regression. Said another way, if the changes in personal income are voluntary then we cannot reasonably assume that they are random, and then the changes in personal income are related to individual characteristics. Ideally we would find a natural experiment which would be some exogenous force that would randomly vary people personal income. This would ensure that the estimate of the effect of personal income on migration is not biased. Note that a similar experiment is not necessary when we are looking at the effect of the regional unemployment rate or average regional wage because individuals effectively take these variables as exogenous.
Bernard also acknowledges that the logit model lacks sufficient control variables. Not having a individual’s education characteristics, for instance, is a problem because other studies have found that more educated people are more mobile (Dion and Coulombe, 2008). The lack of control variables is due to the limited nature of the Longitudinal Administrative Databank that was used as the data sources of this study. A consequence of insufficient controls is omitted variable bias which we cannot sign. One further complication with this data set, is that we cannot be sure that personal income changes are coming entirely from labour market outcomes. Personal income, as reported to the government, may be changing due to variation in all sorts of non-labour income that cannot be separated out of LAD data.
One extension of this work would be to eliminate the a priori assumption that migration is a push phenomenon. An analogous discussion has happened in the self employment literature but here Bernard is implicitly assuming that the conditions in the origin region induce an agent to migrate. Without some strong theory to the contrary, we cannot rule out the possibility that high wages in low unemployment regions signal people to migrate into the region. So rather than looking at the increase in the probability of migration due to a past change in income, the pull theory would look at those who have recently migrated into a region and see how much their income has increased or if average wages are higher in the destination region.
To further explore the migration decision of Canadians the research must develop the data, remove restrictions not based on theory and attempt to find truly exogenous variation. It is clear that without controlling for individual characteristics Bernard’s method is placing a large amount of systemic variation in the error term and thereby biasing the results. To avoid this, a new dataset needs to be created which links individuals in LAD to their observed characteristics which would be available in other data sources. A future study should also avoid restricting attention to migration as a push phenomenon, rather both cases should be explored and their relative importance measured. Finally, an ideal future study would find some natural experiment in which people’s income is changed randomly across region, only the can we claim that personal income is sufficiently exogenous and that causation is only moving in one direction.
References
Bernard, A. Statistics Canada, (2011). Regional economic
shocks and migration (75-001-X) Perspectives on
Labour and Income.
Dion, P. & Coulombe, S. (2008) Portrait of the mobility
of Canadians in 2006: Trajectories and characteristics
of migrants (91-209-X) Report on the Demographic
Situation in Canada: 2005 and 2006
DeepForest.io 