Labor Mobility Across Canadian Provinces University of Groningen Faculty of Economics and Business DD MSc in International Economics and Business (IE&B) with Corvinus University of Budapest Master Thesis

(1)

University of Groningen

Faculty of Economics and Business

DD MSc in International Economics and Business (IE&B)

with Corvinus University of Budapest

Master Thesis

Labor Mobility Across Canadian Provinces

Author:

Laszlo Farkas

Student number:

S3233561

E-mail address:

l.farkas@student.rug.nl

Supervisor:

Tarek M. Harchaoui, PhD

Co-assessor:

Dr. László Lőrincz

(2)

Abstract

Interprovincial labor mobility constitutes an important vehicle for an economy to effectively react to shocks. The wide range of disparities in terms of economic structures, policy-making and labor institutional arrangements between its provinces make Canada a clean case to test whether interprovincial labor mobility constitutes an effective adjustment margin that contributes to a reduction in the labor market performance disparities. We apply a panel vector autoregressive model to the dispersion between national and provincial employment, participation and unemployment rates over the 1976-2016 period. In addition, we assess the robustness of our results by adding weekly earnings to the estimation procedures. While our results support the notion that labor mobility plays an important role in the reduction of labor market differences between provinces, the adjustment mechanism takes a long time and permanent asymmetries persist after the adjustment process. Despite many reforms implemented in recent decades in the Canadian labor market, our findings suggest the persistence of important frictions that prevent Canada to benefit from the sort of labor market flexibility that the U.S. possesses.

JEL Classification Numbers: E24, J64

(3)

Figures and tables

Figure 1: Population of Canadian states (2015) ... 5

Figure 2: Provincial and national unemployment, by provinces (1976-2016) ... 6

Figure 3: Employment as a proportion of total employment by provinces ... 7

Figure 4: Participation rate by provinces ... 8

Figure 5: Employment by provinces after recession from July of 1981 ... 10

Figure 6: Employment by provinces after recession from January of 1990 ... 11

Figure 7: Employment by provinces after recession from September of 2008 ... 12

Figure 8: Net interprovincial migration 1976-2015 ... 17

Figure 9: Unemployment Dispersion ... 18

Figure 10: Impulse response to a labor market demand shock ... 25

Figure 11: Impulse response to an employment-growth shock ... 26

Figure 12: Impulse Response to a Negative Employment Shock, including wage ... 29

Table 1: Interstate Migration in Canada, 2015-2016 ... 16

Table 2: Provincial unemployment heterogeneityand migration ... 19

Table 3: National unemployment and migration ... 20

Table 4: Output of pVAR(12) Δem, ur and pr ... 24

(5)

1. Introduction

Labor as a factor of production is perhaps the most important component in economic growth and in all economic activities. In this context, the issue of unemployment is not negligible, and has a special emphasis thanks to the ongoing automation of different economic activities, which is likely to remain on future agendas. In general, and especially in international

comparisons, the average national unemployment rate is the main indicator of interest in the field of labor economics. Even though, there are serious discrepancies in the unemployment rates across subnational units within the majority of countries worldwide. (Debelle and Vickery, 1998) In addition to other alternatives, the movement of workers from one province to another is a particularly important adjustment mechanism between and within countries. This research aims to reveal this mechanism in connection with the Canadian provinces.

The degree of internal labor mobility is largely determined by the domestic legal, linguistic, cultural geographic and demographic condition of a country. Considering that Canada is not only geographically the second largest country in the world, which makes internal mobility more difficult but also both linguistically and economically divided across the space, it is of importance to keep these factors of labor mobility in mind.

This paper investigates the role of domestic mobility of the labor force as an

adjustment mechanism for reducing differentials in labor market conditions across Canadian provinces. In order to investigate this relationship, we applied a panel vector autoregressive model, so as to quantify the pace at which the adjustment occurs. By using this model, we get insights about one important aspect of labor market flexibility shock sensitivity of the

Canadian provinces.

The evidence in this paper suggests that Canadian provinces respond flexibly to labor market changes, but the adjustment after the shocks is relatively long, suggesting the

presence of important impediments that hold back the Canadian labor market indicating the long-term impact of shocks. Interprovincial labor mobility plays a significant role in

alleviating asymmetric shocks in Canada, although the process can only be estimated at a 4-5-year horizon. Due to the different industrial structures, the provinces of Canada are heterogeneous based on labor market considerations, therefore, they react differently depending on the type of shock.

(6)

2. The Canadian Labor Market: Theory, Background

and Facts

2.1. Theory and Background

Provinces under the same jurisdiction and financial system can be seen as a series of small open economies with a fixed exchange rate. The provinces engage in trade with each other as well as internationally, some of them have less favorable balance sheets, composed they make up the countries aggregate economic situation. Provinces are both subject to state-specific (asymmetric), as well as aggregate (symmetric) shocks (Debelle and Vickery, 1998). Besides many benefits, the main disadvantages of common currency areas are that the sub-national units have no control over their monetary policy, and that they cannot manipulate their exchange rate in favor to the present economic situation (Eric Bond 2007). This issue is of relevance if there is an asymmetric (state-specific) shock within the country or a monetary union. As in the case of a symmetric shock affecting the whole country, monetary and exchange rate policies are reliable and effective tools that can benefit both the states and the country as a whole. By contrast, in case of an idiosyncratic state-or region-specific shock, these tools are not available. Though, other forms of adjustment can be facilitated by federal (or state) fiscal policies or by changes in factor flows and factor prices. In terms of shock, we assume that there are only labor demand shocks, related to changes in the demand for the state’s product. Since the provincial demographic situation does not change in the short term, any changes in labor supply are induced by interstate migration. (Debelle and Vickery, 1998).

Within this framework, our focus is on the factor flows, specifically the flow of labor. However, before we go there, let's take a look at the factors that determine the country's exposure to asymmetric shocks and the significance of stabilizing mechanisms, to make sure that Canada fulfills a framework of assumptions in which we can properly examine it. Moreover, this also enables us to better judge how important the labor migration mechanism is in the country's life.

First, we need to examine the magnitude of the risks of adverse asymmetric shocks threatening the country. The level of interprovincial trade integration is one determinant. Therefore, we need to observe how connected the Canadian provinces are in terms of their bi or multilateral trade flows. A highly integrated economic area by trade is also characterized by being substantially less vulnerable to asymmetric shocks, however as Paul de Grauwe has stated, the likelihood of developing them is also substantially lower (De Grauwe, 2016). One can measure the indicator trade integration by the portion of the total merchandise traded represented by interprovincial trade. In Canada this ranges from 19% to 55%, which is much less than the corresponding portion of intra-EU trade for the euro-zone countries (55% to 85%). The portion of international trade of Canadian provinces is high (45% to 81%), but this can be easily explained by the exceptionally high degree of integration to the US market (Eric Bond 2007).

(7)

real GDP growth1_{of nearly 10%, while its geographical neighbor Saskatchewan was in}

recession and its GDP contracted by 4% (Statistics Canada, Table 384-0038). The reason of the heterogeneity shows the natural resource based orientation of the Canadian economy (Eric Bond 2007). Figure 1 in the Appendix shows the comparison of provinces with

(Alberta, Saskatchewan and Newfoundland) and without (rest of the provinces) fossil energy resources. The chart strongly depicts the dependency of energy-provinces on world

commodity market developments (Kirby, 2016).

Now, we are turning to the adjustment mechanisms. First, the above-mentioned fiscal policy can provide an important tool in the treatment of asymmetric shocks. We would like to know the extent of this adjustment mechanism to objectively evaluate the role of labor

mobility as an adjustment mechanism. There are four sources of direct transfers in

redistribution of public funds, the Canada Health Transfer (CHT), the Canada Social Transfer (CST), Equalization and Territorial Formula Financing (TFF). The first to support specific policy areas such as health care, education, while Equalization and TFF programs provide unconditional transfers to less prosperous provinces and territories (Department of Finance Canada, 2016).

Based on the equalization program, provinces that are displaying above average GDP/capita need to transfer money into a fund which is then redistributed appropriately to the regions that are lagging behind. Receiving provinces are free to spend the funds according to their own priorities (Department of Finance Canada, 2011). However, since natural

resources are administered at the provincial level, serious disparities remain between provinces despite the Equalization Program (Eric Bond 2007). The persistent financial differences between the states and the inadequacy of the program prove that the distribution of payments and subsidies has not changed much since 2008, moreover the system itself is very controversial (Department of Finance Canada, 2016).

The second adjustment mechanism is the change in price and wage. If wages are flexible the wage claims of workers in a depressed region will fall, this subsequently causes prices to fall in that region. Similarly, wages rise in the region that is experiencing an economic upturn, Moreover, because of the excess demand for labor, prices will rise. The cut in prices stimulates demand for the depressed area’s products, making the region’s production more competitive. The opposite occurs in the thriving region (De Grauwe, 2016). In theory, this is a well-functioning mechanism, however in practice, the movement of payments has an ambiguous effect on labor mobility. Lower wages relative to the rest of the country increase labor demand in that province, in other words, in-migration. However, in the same time the effect increases as the incentive for labor out-migration, due to the lower return on work (Debelle and Vickery, 1998). In Canada, wages are generally flexible, last year the inflation-adjusted wage growth was -1.7 percent in Alberta, while Prince Edward Island experienced a more than 3 percent increase (Tencer, 2016). However, the low interprovincial trade limits the efficiency of the balancing mechanism involving prices and wages (Eric Bond 2007).

Third, firms (and capital) can relocate to regions with a relatively larger pool of unemployed workers in order to take advantage of the cheaper labor force. Although it is important to recognize that this is a highly questionable mechanism since the direction of the force cannot be determined beforehand. A region that has been hit by a negative idiosyncratic demand shock might end up being a less attractive location of production, especially if it’s a

(8)

company operating in the service sector. In addition, this mechanism is presupposed by the presence of the previous adjustment channel (Debelle and Vickery, 1998).

Finally, we are going to have a quick look on the traits of Canadian internal labor mobility. In general, internal migration is not hampered by serious legal obstacles, although the provincial/territorial licensing of regulated occupations remains a systemic barrier to inter-provincial labor mobility, it affects about 15-20% of employed Canadians.

(Employment and S.D. Canada, 2017). On average about 1.2 million Canadians move permanently each year, but much of the internal migration takes place within the provinces. The total internal migration is 4,2% whilst the inter provincial migration is 0,7 percent, which is less than half of the corresponding migration between US states2. (Department of Finance Canada, 2014) These facts indicate a general pattern that the percentage of the Canadian population moving between provinces has been in steady decline since 1977, when 1.5 per cent of the population was mobile to less than one per cent in 2012. Canadian demographics, characterized by an aging society plays a major role in this trend. The median age of

interprovincial migrants increased from 23.6 years in 1976/1977 to 29.5 years in 2010/2011 (Bohner, 2013). It is also important to note that employment is usually not the main reason why people move as only 11% of workers move for job-related reasons (Employment and Social Development Canada, 2017). From a geographic point of view, western provinces are considerably more mobile than average, with central provinces having a larger proportion of internal migration due to their size, while Atlantic provinces are less mobile than the average (Bayoumi, 2006).

2.2. Documentation of Fact

We now outline the broad trends and features of the Canadian labor market over the last forty years that began in 1976. We also highlight what has changed in the last 10-15 years that warrant an extension to the earlier studies by Pradas and Bayoumi.

First, it’s important to highlight that the population of Canada is distributed unevenly throughout the thirteen provinces/territories. Most of the population is concentrated in the southern part of the country, near the U.S. border, therefore the northern areas are largely uninhabited. Another important factor is that the territorial distribution of the provinces is also very heterogeneous. Primarily, because of historical reasons, the first provinces on the east coast are considerably smaller territories than the provinces located in inner Canada (Historical boundaries of Canada, 2015). Figure 1 shows the population of the provinces in 2015. One can observe that Ontario is an outlier, with the population which is more than one third of Canada's total population. Together with Quebec, the two provinces represent 63% of the whole population. At the same time the three northern territories (Yukon Territory,

Northwest Territories, Nunavut) is barely inhabited. They represent approximately 3 % of the Canadian population, while they account for 40 % of Canada's land mass (Government of Canada, 2010). Consequently, these territories were ignored from my study.

(9)

Figure 1: Population of Canadian states (2015)

Source: Statistics Canada, Table 051-0005

Figure 2 shows relative unemployment of the provinces over the past forty years. The blue line represents the provincial unemployment rate, while the orange line serves the national unemployment. There are permanent differences between provinces even if we take serious changes in time into account. In the Atlantic provinces (New Brunswick, Prince Edward Island, and Nova Scotia and Newfoundland and Labrador), the provincial unemployment rate is typically above the national unemployment rate. In contrast, in the Prairie provinces it is below the national level. In general, the development of the provincial unemployment rate is not highly dependent on the national unemployment rate. The overall coefficient of

determination (R2) between the provincial and national unemployment rates is generally low, 0.2, while within the panels it is 0.64 on average with yearly data. At the same time, huge differences are present between provinces, as evidenced by Quebec (0.85) and Prince Edward Island (0.45).

(10)

Figure 2: Provincial and national unemployment, by provinces (1976-2016)

Source: Statistics source: Canada. Table 282-0086

Figure 3 shows the development of the employment share of each province as a

(11)

Figure 3: Employment as a proportion of total employment by provinces

Source: Statistics Canada. Table 282-0088

(12)

participation rates of the Atlantic Provinces compared to other provinces may also indicate the discouraged workers and increasing the hidden unemployment.

(13)

Figures 5-7 display the last three large recessions in Canada and the employment response of the provinces. Apart from these shocks, there has been other slight contractions or weakness in the economic activity during the sample period, namely in 1986, 1995, and 2001, but they did not display the characteristics of deteriorating events that feed on each other to lead to cumulative economic decline, which is a defining requirement of recessions (Cross and Bergevin, 2012).

We are interested in three dynamics of post-recession employment rates. The short term (within one-two year) impact of shock to the province which shows the shock absorption capacity of the province. Second, the length of the shock effect and finally, the pace of the recovery of the province.

The 1981/1982 recession which had the intensity of four out of five - just as the next two -, according to Crossand Bergevin (2012). The shock was driven by desire to slow

inflationary pressures. At that time interest rates increased above 20 percent in both Canada and the US. The rates remained high until the spring of 1982 as the US Federal Reserve experimented with targeting money supply growth instead of interest rates (Cross and Bergevin, 2012).

(14)

Figure 5: Employment by provinces after recession from July of 1981

A decade later the economy started slowing in 1989. The cycle peaked in March 1990 before the monthly GDP dropped abruptly in April, which was followed shortly after by employment. The contraction exacerbated significantly in the beginning of 1991, when the combination of the introduction of the goods and services tax (GST) and the Gulf War sharply declined consumer spending. In the second quarter of 1991, improving figures

(15)

Figure 6: Employment by provinces after recession from January of 1990

The most recent recession is the Great Recession, which began at the end of 2007.

(16)

Figure 7: Employment by provinces after recession from September of 2008

(17)

3. The Literature Review

Having established the main documentation of facts, we now move to the review of the relevant literature with a focus on contributions that exploited the same methodology as ours.

Examining the nature of asymmetric shock adjustment, Blanchard and Katz (1992) built a model which has been widely accepted and used ever since to study the labor market conditions of different geographical units. Blanchard and Katz analyzed the behavior of the American state labor markets in relation to state-specific shocks. Their fundamental focus was that variations in relative employment levels across US states persist over time, while relative unemployment and activity rates are stationary variables, meaning unemployment and activity rates tend to return (Arpaia et al, 2014). These findings were supported by the fact that many national economies have a structure where, thanks to factor mobility asymmetric, shocks have only temporary effects on relative wages, unemployment and participation rates, while the effects on relative employment are permanent.

Following their observation, they made the following reasoning, “If asymmetric shocks have a permanent effect on employment but not on the unemployment and activity rates, the change in long run employment levels should be absorbed by changes in the working-age population” (Arpaia et al, 2014). Considering the fact that, labor demand shocks have limited influence on demographic trends, the response of relative population must reflect the response of labor mobility. Another important assumption of the model, that unexpected movements in employment within the year primarily reflect movements in labor demand. Blanchard and Katz found that these unexpected labor demand shocks generate a 4-year effect on unemployment and activity rate, followed by a 3-year adjustment, which can be considered a rapid process. They found that labor mobility has a significant adjustment effect among US states.

(18)

The work of Juan Jimeno and Samuel Bentolila (1998), which investigated the Spanish labor mobility across provinces in the period of 1976 to 1994, is also a significant piece of literature, in the field of our research. They showed that regional wages, relative unemployment and participation rates, are very persistent in Spain, while employment growth rates are not. They also found, the responses of the migration and participation rates to labor demand shocks seem to be significantly slower than in US states and EU regions, while the long-run employment level response is significantly lower. In fact, the mobility of the Spanish labor force is extremely different from that of the United States.

Debelle and Vickery (1998) applied the Blanchard-Katz (1992) approach same model to the Australian states over the 1979-1997 period using ordinary least squares, time series VAR. Among other things, they found that internal migration is influenced by the level of total unemployment and out-migration from a state occurs slowly and steadily. According to their results, Australian real wages are not significantly determined by labor market conditions, therefore the real wage channel has no significant role in Australia. There are permanent differences between the Australian states in terms of labor mobility, the most spectacular example of this, is the persistently high unemployment in Tasmania.

Perhaps the most relevant paper with this methodology for Canada is the comparative study by Prasad and Thomas from 1997. They researched why the Canadian unemployment is permanently higher than that of the US. On the one hand, they explained the difference with lower internal mobility and on the other hand with the low real wage flexibility typical for Canada. In addition, they revealed employment growth shocks were found to have a larger and more persistent effect on both employment levels and participation rates in Canada than in the United States. They also measured the role of unemployment insurance (UI) systems of the two countries and found, overall UI generosity has increased aggregate unemployment persistence in Canada.

Finally, Bayoumi et al. (2006) did a much similar research in 2006. The main findings of their research proved that in common with the US internal migration has a significant role in the adjustment to labor demand shocks across Canadian provinces. They also compared Canada's labor mobility rates with European countries and found, the adjustment mechanism is more important in Canada than in Spain, France, or Germany. Looking for internal relations, they revealed that the mobility of western provinces was significantly higher than that of the rest of the country and especially that of the Atlantic provinces.

(19)

4. Tests of Labor Mobility

4.1. Hypotheses

As explained before, the primary purpose of the paper is to determine whether interprovincial labor mobility plays an important role in alleviating asymmetric shocks in Canada. Based on the previous descriptive statistical picture and the existing literature, we formulate the following hypotheses:

The essence of asymmetric shocks is that the economic conditions of one or more provinces produce a radical change over the average of the country as a whole, in a short time. Given that research primarily focuses on labor market shocks, it is assumed that higher relative unemployment leads to higher out-migration, while in the opposite direction, higher relative employment results in immigration.

H1: The relative unemployment rate of a province positively correlates with the

out-migration from that province.

Secondly, while previous American, European and Canadian papers differed considerably in the effects of shocks and the effectiveness of adjustments, the duration of the shocks were roughly the same 3-4 years, in each geographic area. Based on this, it is expected that the generated shocks take most of their effects within a short time, followed by an adjustment period.

H2: The shocks take effect for three years after their occurrence.

Thirdly, on the basis of the theory described, the change in real wages also serves as an adjustment channel, so it is expected that by including relative wages to the original equation, labor mobility is declining, as the real wage channel absorbs some of the imbalance generated by the shock.

H3: The change in real wages serves as an effective adjustment channel for the Canadian

market

And finally, the documented descriptive statistics indicated - through both the long-term unemployment rate and shock sensitivity of the provinces - permanent heterogeneity across provinces. Based on this, we can conclude that there are persistent differences in the degree of mobility of labor regardless of the shocks.

H4: There are persistent differences in labor mobility between the provinces, regardless of

(20)

4.2. Preliminary Analysis

In this section, we provide a closer look at labor mobility, with the help of some well-established tests. First, we will have a look at Table 1, which is a labor mobility snapshot from 2015-2016. The size and the direction of migration are the main points of interest in this test. Newfoundland and Labrador, Ontario and British Columbia had positive net flows. British Columbia experienced the greatest immigration and it also had strong positive positions in the previous 3 years. Quebec had the most negative balance, and generally the negative net migration is a regularity in this province. In 1996, the population proportional migration was between -1,45 and 0,95 the same range at the same time was between -0,58 and 1,12 in case of Australia which has similar population geography. Snapshots from the previous decades can be found in Appendix at point 8 (Clemens, Labrie, Emes, 2016).

Table 1: Interstate Migration in Canada, 2015-2016

Source: Statistics Canada. Table 051 -0018

(21)

A survey by Ipsos Reid, for the Canadian Employee Relocation Council (O'Neill, 2014) found that more than half of the Canadians are not interested in moving for employment opportunities. Moreover, the decision of the 34 percent of Canadians who would consider a relocation to another province, is conditional on a 20% increase in pay. However, Quebec loses only relatively few residents annually, so the extremely low level of in-migration – probably due to the linguistic difference - explains the comparatively high level of net out-migration. It is worth paying attention to these trends, but since these processes are extremely slow, they do not pose a serious threat to the reliability of our subsequent tests (Clemes, Labrie, Emes 2016). Figure 6 shows the net migration over the whole sample period. On the left-hand panel, the total net migration is plotted. Over the last forty years, only two

provinces have positive net position in interprovincial migration, these are Alberta and British Columbia. They received more than half a million domestic migrants each, while Quebec approximately lost the equivalent amount.

Figure 8: Net interprovincial migration 1976-2015

The right-hand side panel displays the population proportionate change.The previous distribution seems to be stable, but Quebec's loss does not seem to be any more critical. Manitoba and Saskatchewan experienced 18 and 16 percent negative net flow respectively, despite the fact that most of their loss took place in the 1990s, they have recently shown a negative trend. The relative loss of Newfoundland is even bigger. Even though it is a much smaller province there has been a positive tendency since the late 2000s.

The second test is the test of unemployment dispersion. This is an essential part of our analysis, as according to the theory, provincial unemployment levels should converge to the national level on the long run. Persistent differences prove asymmetric economic

circumstances and ineffective labor flows. Figure 9 plots standard deviations of the provincial unemployment rate. Higher level of dispersion means higher heterogeneity in unemployment rates. If the labor force was perfectly mobile the standard deviation would be close to zero. The first observation is that, during the recessions - which are indicated by the grey areas - the dispersion is declining, as a result of the general and universal unemployment rate growth. After the shock, the level of dispersion increases due to the geographically different labor market response to it. The earlier mentioned crises of 1981, 1990 and 2008 are well-observed in the graph. The diverging effect may take up to 7-8 years after the crisis, which can be explained by slow and low labor flows. In fact, since the 2008 crisis, the

(22)

Figure 9: Unemployment Dispersion

Standard deviation of provincial unemployment rates from national rate

The third and final test is a pair of panel regression. First, we are interested in the correlation between the province proportionate net migration and the deviation of the

provincial unemployment from the national level. We regress the ratio of net migration into a province to the province’s population on the province’s relative unemployment rate

differential. According to the theory, the higher the unemployment in a province, the more encouraged the labor is to move-out from that province, consequently the coefficient on the unemployment rate differential should be negatively signed.

Second, we want to seeto what extent internal migration increases as a result of higher aggregate unemployment rate. We run a regression of the absolute value of the net migration ratio on the absolute value of the unemployment differential and the national unemployment rate. We used annual data between 1976-2016 with 10 panels according to the 10 provinces. The following fixed-effect panel regression has been utilized:

𝒏𝒆𝒕 𝒎𝒊𝒈𝒓𝒂𝒕𝒊𝒐𝒏_𝒊𝒕

𝒑𝒐𝒑𝒖𝒍𝒂𝒕𝒊𝒐𝒏 = 𝜶𝒊+ 𝜷(𝒑𝒓𝒐𝒗𝒊𝒏𝒄𝒊𝒂𝒍 𝒖𝒏𝒆𝒎𝒑𝒍𝒐𝒚𝒎𝒆𝒏𝒕 𝒓𝒂𝒕𝒆𝒊𝒕− 𝒏𝒂𝒕𝒊𝒐𝒏𝒂𝒍 𝒖𝒏𝒆𝒎𝒑𝒍𝒐𝒚𝒎𝒆𝒏𝒕 𝒓𝒂𝒕𝒆𝒕)+𝜺𝒕

(23)

cross-sectional dependence tests is also significant, so we used a fixed effect panel regression adjusted for Driscoll-Kraay standard errors. The test results are disclosed in the appendix at point 2-5 (Hoechle, 2007) while Table 2 reports the results of the above equation with an adjustment of the Driscoll-Kraay standard errors. The top panel shows the regression output, the lower the provincial constant terms. From the name of the variable/province to the right, one can see the corresponding coefficient in the first column and to the right the standard error. The error adjustment has only minimally changed the coefficient compared to the adjustment-free model.

Table 2: Provincial unemployment heterogeneityand migration

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Source: Author's calculation based on data from Statistics Canada

We find that almost all the provincial constant terms are significantly different from zero at the 1% level, and only Quebec is not significant at the 10% level, the R-squared is 50%. Since the provincial intercepts vary noticeably, we ascertained that the assumption about fixed effect was appropriate. The test of 9 pairs of equalities is also consistent with our priors on the presence of between-provincial heterogeneity. In case of this test, the null hypothesis is that all intercepts are equal, what we can reject in all cases, even the joint F-test is

significant, so we are sure that the intercepts are not all equal. These tests demonstrate, that some net interstate migration occurs even in the absence of the gap between the provincial and national unemployment.

(24)

We now turn to our second panel regression, and follow the same steps as we followed in the case of the first panel regression. We are curious about the degree of correlation between the total unemployment rate and the rate of relative migration, that is, to what extent higher internal mobility can be explained by the higher national unemployment rate. Thereby, we also gather information about the role played by aggregate shocks in the development of internal asymmetries. The second regression is written by the following formula:

|𝒏𝒆𝒕 𝒎𝒊𝒈𝒓𝒂𝒕𝒊𝒐𝒏_𝒊𝒕|

𝒑𝒐𝒑𝒖𝒍𝒂𝒕𝒊𝒐𝒏 =𝜶𝒊+ 𝜸|𝒑𝒓𝒐𝒗𝒊𝒏𝒄𝒊𝒂𝒍 𝒖𝒏𝒆𝒎𝒑𝒊𝒕− 𝒏𝒂𝒕. 𝒖𝒏𝒆𝒎𝒑𝒕| + 𝜹𝒏𝒂𝒕. 𝒖𝒏𝒆𝒎𝒑𝒕+ 𝜺𝒕

The provincial constant terms show a much weaker sign of interprovincial heterogeneity, as instead of one, four provinces are not significant (Nova Scotia New Brunswick, Quebec, Ontario), meaning the constants of almost half of the provinces do not differ much from zero, therefore their net migration can be explained purely by the volatility of the unemployment rate differences. The estimation results are reported in Table 3.

Table 3: National unemployment and migration

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

(25)

Based on theory, we can suspect that high aggregate unemployment deters people from emigrating, especially if they are liquidity constrained. However, none of the variables are significant of the second regression. We cannot state that higher national unemployment rate reduces the level of Canadian internal migration as our second regression is not significant. This also means that the national unemployment rate does not have a significant explanatory power for internal asymmetries.

Altogether, the three checks prove, that interprovincial labor mobility is significant. Crises have long-term impacts on unequal labor market developments among provinces. In addition, we found that provincial unemployment plays a significant role in interprovincial mobility.

4.3. A Panel Var Method

Backround

So far, we have examined the cross provincial heterogeneity of labor market conditions and the impact of labor flows using a static framework. Now, we are starting to explore the dynamics of labor flows. We accomplish this by applying a panel VAR model of the state labor markets that combines information on provincial employment, unemployment and participation rates. The panel VAR model is a framework, which defines the dynamic

interrelationship between stationary variables (Hill et al., 2011). For a more accurate picture, we use quarterly data (apart from yearly) for the period Q1 1976-Q4 2016, using Cansim database, the socioeconomic database of Canada’s national statistical agency, Statistics Canada as we did before3. Some essential information about the used panel database can be found in the Appendix at point 5 (mean, standard deviation skewness, kurtosis, median, max, min). In addition, a detailed summary data is presented in the Appendix at point 19 with the source of all the data used in the paper. We apply the panel VAR (pVAR) model outlined by Abrigo and Love (2015).

Time-series vector autoregression (VAR) models originated in the macroeconometrics literature as an alternative to multivariate simultaneous equation models (Abrigo and Love. 2015). Panel VARs have the same structure as VAR models, in the sense that all variables are assumed to be endogenous and interdependent, but a cross sectional dimension is added to the representation. The modern panel vector autoregression (pVAR) models have been increasingly used in the applied research, especially in the field of macroeconomics and finance. Panel VARs seem to be a particularly well-suited model, as they are able to capture both static and dynamic interdependencies, treat the links across units in an unrestricted fashion, easily assimilate time variations in the coefficients and in the variance of the shocks and account for cross sectional dynamic heterogeneities. However, the weakness of the model is that in any form it imposes restrictions. The Bayesian panel VARs leave the model

unrestricted but employ a shrinkage prior to effectively reducing the dimensionality of the coefficient vector (Canova and Ciccarelly, 2013).

(26)

We have three endogenous labor market variables, em, ur and pr. The first variable is the natural logarithm of employment in the province as a proportion of total employment. The second variable represents the difference between unemployment rate in the province to the national unemployment rate. In fact, it is defined as the log ratio of one minus provincial and one minus national unemployment rate. This formula is consistent with Debelle and Vickery and with the original creators of the model, Blanchard and Katz (1992). The third variable is a log ratio of the state participation rate to the national participation rate. All the variables are given in as deviation from the respective national means and they are multiplied by 100 to possess more manageable numbers. Here we present the actual formation of our variables: 𝑒𝑚 = ln (𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 𝑖𝑛 𝑡ℎ𝑒 𝑝𝑟𝑜𝑣𝑖𝑛𝑐𝑒 𝑡𝑜𝑡𝑎𝑙 𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 ) 𝑢𝑟 = ln ( 1 − 𝑝𝑟𝑜𝑣𝑖𝑛𝑐𝑖𝑎𝑙 𝑢𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 𝑟𝑎𝑡𝑒 1 − 𝑛𝑎𝑡𝑖𝑜𝑛𝑎𝑙 𝑢𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 𝑟𝑎𝑡𝑒 ) =ln(𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 𝑖𝑛 𝑡ℎ𝑒 𝑝𝑟𝑜𝑣𝑖𝑛𝑐𝑒/𝑙𝑎𝑏𝑜𝑟 𝑓𝑜𝑟𝑐𝑒 𝑖𝑛 𝑡ℎ𝑒 𝑝𝑟𝑜𝑣𝑖𝑛𝑐𝑒 𝑡𝑜𝑡𝑎𝑙 𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 /𝑡𝑜𝑡𝑎𝑙 𝑙𝑎𝑏𝑜𝑟 𝑓𝑜𝑟𝑐𝑒 ) 𝑝𝑟 = ln (𝑝𝑟𝑜𝑣𝑖𝑛𝑐𝑖𝑎𝑙 𝑝𝑎𝑟𝑡𝑖𝑐𝑖𝑝𝑎𝑡𝑖𝑜𝑛 𝑟𝑎𝑡𝑒 𝑛𝑎𝑡𝑖𝑜𝑛𝑎𝑙 𝑝𝑎𝑟𝑡𝑖𝑐𝑖𝑝𝑎𝑡𝑖𝑜𝑛 𝑟𝑎𝑡𝑒 ) = ln (𝑙𝑎𝑏𝑜𝑟 𝑓𝑜𝑟𝑐𝑒 𝑖𝑛 𝑡ℎ𝑒 𝑝𝑟𝑜𝑣𝑖𝑛𝑐𝑒 /𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑖𝑛 𝑡ℎ𝑒 𝑝𝑟𝑜𝑣𝑖𝑛𝑐𝑒 𝑡𝑜𝑡𝑎𝑙 𝑙𝑎𝑏𝑜𝑟 𝑓𝑜𝑟𝑐𝑒 /𝑡𝑜𝑡𝑎𝑙 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 )

The specific model and moment selection is an essential part of any pVAR model. If our variables are stationary, the model can be estimated using least squares applied to each equation if, however, y and x are nonstationary, then, the first difference of the variables should be used. A series stationary if the joint probability distribution (and consequently the mean) does not change over time, this characteristic is crucial, because we must obtain mean-reverting variables (Croarkin and Tobias, 2003). If the variables are nonstationary but

integrated to order one, meaning their first differenced form are stationary, and they are cointegrated at the same time, we can also construct a VAR model with them (Hill et al., 2011). For this investigation, we tested two characteristics of our variables. First, the stationarity by running four different tests (Im-Pesaran-Shin, Levin-Lin-Chu, Fisher, Breitung’s) on each of the variables, and second, the cointegration by a joint application of the Westerlund test. In the case of the Im-Pesaran-Shin, and the Levin-Lin-Chu tests we used the Schwarz criterion for lag selection based on the model article.

(27)

and the variation in relative employment levels is I(1) (integrated of order 1), meaning non-stationary series whose first difference is non-stationary (Blanchard and Katz (1992).

The Westerlund test (by Persyn, D. and J. Westerlund. 2008.) for cointegration, tests for the presence of long-run relationships among integrated variables with both a time-series dimension, and a cross-sectional dimension. The original three variables - with the lags selection by the Akaike information criterion – indicates no cointegration among the variables. Since the first variable did not fulfill the requirement of stationarity, we convert this variable to first differential form (Δem), which is also a useful step because the three variables are thus cointegrated. Our finding about the nature of the variables are consistent with the assumptions of Blanchard and Katz that relative employment growth and relative changes in the activity and unemployment rates tend to revert to their initial levels (their mean) after deviations (Arpaia, 2014). After studying the characteristics of our database, the final forms of the model specification we use is the following:

∆𝑒𝑚𝑖𝑡 = 𝛼𝑖,1+ ∑ 𝛽1,𝑠 𝑛 𝑠=1 ∆𝑒𝑚𝑡−𝑠+ ∑ 𝛾1,𝑠𝑢𝑟𝑡−𝑠 𝑛 𝑠=1 + ∑ 𝛿1,𝑠𝑝𝑟𝑡−𝑠 + 𝜀1𝑖𝑡 𝑛 𝑠=1 𝑢𝑟_𝑖𝑡 = 𝛼_𝑖,2+ ∑ 𝛽_2,𝑠 𝑛 𝑠=0 ∆𝑒𝑚_𝑡−𝑠 + ∑ 𝛾_2,𝑠𝑢𝑟_𝑡−𝑠 𝑛 𝑠=1 + ∑ 𝛿_2,𝑠𝑝𝑟_𝑡−𝑠+ 𝜀_2𝑖𝑡 𝑛 𝑠=1 𝑝𝑟𝑖𝑡 = 𝛼𝑖,3+ ∑ 𝛽3,𝑠 𝑛 𝑠=0 ∆𝑒𝑚𝑡−𝑠+ ∑ 𝛾3,𝑠𝑢𝑟𝑡−𝑠 𝑛 𝑠=1 + ∑ 𝛿3,𝑠𝑝𝑟𝑡−𝑠+ 𝜀3𝑖𝑡 𝑛 𝑠=1

Where i denotes the province and t is the time. In this system, each variable is a function of its own lag and the lag of the other variables. (Hill et al., 2011)

Following the optimal model selection, we need to accomplish the optimal moment selection as well, which means the right choice of lags (Belingher, 2015). Based on Andrews and Lu (2001), “consistent moment and model selection criteria (MMSC) for general method of moments (GMM) models are based on Hansen’s (1982) J statistic of over-identifying restrictions” (Časni et al. 2016). The test based on the three model selection criteria (MBIC, MAIC, MQIC), and in our case first-order panel VAR (pVAR(1)) is the preferred model, supported by the result of MBIC and MQIC. The result of the test is available in the appendix at point 9. However, we set our model to 12 lags and 24 instrumental lags, because the one-lag length does not provide a rich enough dynamic structure for the model. With the increased lag order, some of the coefficients became insignificant which weakens the

reliability of the model, although these changes have no effect on the response graphs, given that only the employment impulse is researched (Časni et al., 2016).

Finally, just like in case of the previous panel model, we want to obtain robust standard error. The tests found that our quarterly data is heteroskedastic and autocorrelated, so we have to use the cluster robust option for the pVAR(12) model. It is important to note that the response rates in the next chapter are based on the data for the entire country with ten

(28)

of a Canadian province. The estimation includes fixed effects between provinces by allowing the constant term α to vary across states.

Estimation Results

Building on the previous model selection tests, we fit a twelfth-order panel VAR model for the ordinary least squares of Δem, ur and pr with a twenty-four instrumental-lag specification using GMM estimation and cluster robust error term (Abrigo and Love, 2015).

The estimation has a total of 1389 observation with 10 panels. We have significant results for almost all the variables, except the response of Δem to the effect of a participation rate shock and the future response of unemployment to a shock on itself. The other results’ significance is at least five percent, which seems to be a convincing proof of the applicability of the model.4 The output of the pVAR (12) can be viewed in Table 4.

Table 4: Output of pVAR(12) Δem, ur and pr

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Source: Author's calculation based on data from from Statistics Canada

To truly make sure that we have reliable results, we perform the post-estimation tests. First, we run a Granger causality Wald test. Apparently, all our variables Granger-causal for the rest of the variables, at the one percent confidence level. This means any of our variable’s ( x) future values can be predicted by the past values of the variable ( x) and the rest of the variables ( y). The subsequent test is the Eigenvalue stability condition check. The test calculating the modulus of each eigenvalue of the estimated model. The pVAR model is stable if all moduli of the associated matrix are strictly less than one. Stability implies that the panel VAR is invertible and has an infinite-order vector moving-average representation. The stability test reports that the model satisfies stability condition, due to the fact, that all the eigenvalues lie inside the unit circle of the test (Abrigo and Love, 2015). The results of the tests can be found in the appendix at point 11-14.

4_{We also considered to use "GMM-style" instrument as proposed by Holtz-Eakin, Newey and Rosen (1988),}

(29)

Benchmark model

Before we evaluate the impulse response graphs based on quarterly data, we run the same model using yearly data. This model has already been implemented by Prasad and Thomas (1997) and Bayoumi et al (2006), but we also carry out it for comparison. This is a simplified version of the main model that only shows the main trends of the variables, helping us

understand their general movements, thus it is an excellent subject of comparison as a benchmark model. This model has the same model specification, except the regression was run in first order with 7 instrumental lags. This does not mean a serious change since the choice of a lag length is not crucial to the inferences generated from the VAR model (Debelle and Vickery, 1998).

All of the three response variables have 1% significance. Figure 10 depicts the response of Δem5, ur and pr to a shock resulted by one standard deviation increase in employment

growth. The relative employment shows a quick and dynamic increase after the shock, the effect of the impulse passes after 4 years. After the peak it shows a slight decline and then stabilizes 1 percent higher than it was at the time of the shock, indicating a 2% increase in the long-term employment level. Reiterating our earlier statement, while short-term employment growth can be partially explained by the fewer unemployed and inactive, the long-term sustainability of employment growth can only refer to population growth, which can only be a migration due to time constraints. The participation rate rises smoothly as a result of the shock, and reaches its peak at 0.4 unit, around 5-years after the impulse, then very slowly converting it to the baseline. The extremely slow convergence of pr suggests that - while the tests contradict with this – the variable has a kind of ‘quasi-nonstationary’ nature, which is in line with the findings by Bayoumi (2006). The provincial unemployment radically declines for four years after the stimulus and then shows a relatively rapid convergence over the next five years, although convergence is not completed even after 20 years. The movement of ur responsible for the increase of employment in the first years, however it cannot be explained by that in the long-run.

Figure 10: Impulse response to a labor market demand shock

(30)

From this figure, the following can be deduced: first, labor market indicators are flexible to shocks. Second, the shocks affect the province for about 3-4 years followed by a relatively slow adjustment, indicating the long-term impact of shocks. Third, the impact of the shock on employment produces a significant level of long term employment growth, indicating labor immigration. In this light, we are going to examine the output of our quarterly estimates.

Provincial impulse response

Figure 11 shows the response of relative employment, unemployment and participation rate to a one standard deviation increase in the employment growth (Δem), based on quarterly data. The provincial employment6 shows a gradual and significant increasing period, and peaks at 2% higher than the level right after the shock. However, much of the growth takes place in the first three years. Furthermore, the entire effect seems longer then on the previous graph, as it lasts for around 6-7 years instead of 4-5. The growth in employment caused by the shock proves to be permanent, indicating the presence of labor mobility. The relative participation rate shows a slight but very long-lasting movement, confirming our earlier observation that participation rate adjustment is relatively rigid. Undoubtedly, the trend in unemployment has the most radical change. As we have seen on the basis of annual data, unemployment shows a downward trend in response to a positive employment shock which is in line with our expectations. But the previous regression is based on relatively small samples and sparse observations, so there is reason to believe that the graph reflects rather the general, long term response of the provinces to a shock. Here, we see that right after the impulse, the relative unemployment rate increases for 1 year, followed by a downward trend alternately with employment growth, which hardly increased in the first year. As employment very slowly increases in the first year, and the relative unemployment grows parallel in the short term, we can conclude that a short but intensive labor flow is taking place in the first year. It reaches the original level at around 7.5 years, after this point it falls below the baseline and remains there for a considerable time.

Figure 11: Impulse response to an employment-growth shock

(31)

We can conclude that Canadian labor market indicators are sensitive to shocks, and they have long-term impacts on the provincial labor market situation, given that the total

adjustment can be measured in decades. In previous studies, stationary variables (pr and ur) take 7 (US) and 15 (EU) years to adjust based on Blanchard and Katz (1992) and Arpaia et al. (2015), respectively. Employment response to shocks is very dynamic between the first and the third year, which slows down, but keeps up to 6 years, followed by a slow and long-lasting adjustment, which results to a significant influx of migrants to the province. In addition to their dynamics, the direction of the variables is also notable. In principle, employment and participation rate are negatively correlated with unemployment rate,

however, the Canadian data have often shown the opposite in the short run (Hornstein, 2013). The only reasonable explanation for the short-term co-movement of participation rate and unemployment rate is the emergence of early migratory waves, as a result of significant job creation, due to provincial economic improvement. However, taking into account the uncertain movement of ur and pr variables with higher lag values, we cannot draw far-reaching conclusions without the closer examination of the co-movements of these variables. As a robustness check, we present the same impulse-graph based on the monthly data and the proportionate number of lags in the appendix, at point 7. The graph supports our main model as to the behavior of the variables are identical. However, as one can see, with the richer sample all variables take slightly lower maximum values.

4.4. Incorporating Wages

The previous model examines the movement of labor market conditions solely in the context of labor demand and supply. Thus, it is ignoring the potential influence of provincial wage levels on labor mobility. As discussed above, the change in earnings level is one of the adjustment mechanisms, although the effect of real wages on migration is ambiguous. Because the lower real wages do not just attract more firms and increase labor demand but also increase the out-migration stimulus due to lower returns. If there is a clear link between the labor market conditions and the wage level, which can have significant repercussions on the market, there is good reason to assume that this mechanism will replace / complement the labor mobility channel (Wiczer, 2014). In any manner, we are curious if wages have any impact on internal migration at all, in consequence we include, as in Debelle and Vickery (1998), real wages to our original pVAR model using quarterly data. The new variable is the following:

𝑤𝑎𝑔𝑒 = ln (𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑤𝑒𝑒𝑘𝑙𝑦 𝑒𝑎𝑟𝑛𝑖𝑛𝑔𝑠 𝑓𝑜𝑟 𝑡ℎ𝑒 𝑝𝑟𝑜𝑣𝑖𝑛𝑐𝑒 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑤𝑒𝑒𝑘𝑙𝑦 𝑒𝑎𝑟𝑛𝑖𝑛𝑔𝑠 𝑓𝑜𝑟 𝐶𝑎𝑛𝑎𝑑𝑎 )

The source of the data is the Cansim database again, and we used the same sample period.

(32)

moment selection criteria, the first-order model (with one lag) is determined to be optimal, but in the same way as before, we use higher number of lags7. The test result is disclosed in the appendix as Figure 9.

With the inclusion of Δwage, one coefficient from the original model has become insignificant: the pr’s effect on itself. In addition, the relationship between Δwage and the original three variables in both directions is highly questionable, given that only the change in the employment level has significant effect on the wage variable. This means - if we can give credit to our wage data - the labor market conditions played little role in determining the level of real wages, and the development of real wages have little impact on the labor market situation (Debelle and Vickery, 1998). Therefore, it implies that wage adjustments provide only a weak mechanism for the transmission of labor demand shocks, which is consistent with the findings of Bayoumi about Canada from 10 years ago (Bayoumi et al, 2006).

Table 5: Output of pVAR(12) Δem, ur, pr and Δwage

Even if its significance is limited in the perspective of the whole model, the inclusion of the wage variable appears on the response graph. The temporary increase in unemployment is smaller and its return to the starting position is quicker, followed by a greater decrease. The movement of the participation rate is unchanged, but its degree of growth has decreased slightly. The most important change for us is that long-term employment growth, which has been reduced by one unit, shows a downward pressure on interprovincial immigration.

(33)

Figure 12: Impulse Response to a Negative Employment Shock, including wage

These effects are consistent with those found by Blanchard and Katz on the US market. Although the Δwage variable has been multiplied by one hundred so that its movement can be observed, it shows minimal dynamics. It slightly increases with employment level, which is consistent with European and Australian experiences (Arpaia, 2014). This result is consistent with the theory that the presence of wages mitigates labor market movements, and absorbs the part of the asymmetries.

4.5. Provincial Fixed Effects

In the previous models, we allow the presence of permeant differences between provinces by letting the panel specific constant terms vary between our panels. We assumed, there are steady differences between provincial employment, unemployment and participation rate developments, for this reason, the fixed effect model was used.

(34)

provincial differences in terms of unemployment, participation rates and relative real wages. The result of the Hausman tests are reported in the appendix at point 15-18.

4.6. Hypotheses Evaluation

Using the results found in the previous sections, one can now evaluate the hypotheses laid out in section 4.1 of the paper. Based on the first panel regression of the preliminary analysis, the first hypothesis (H1) is accepted, as the coefficient of the unemployment difference variable is significant and negatively signed.

H1: The relative unemployment rate of a province positively correlates with the

out-migration from that province.

Regarding hypothesis 2 (H2), we can only partially accept the claim, given that the provinces' labor market have been significantly affected by the shocks, in the first three years, but on the basis of the quarterly data we have found that the shock affects the market only slightly but until the sixth year.

H2: The shocks take effect for three years after their occurrence.

The third hypothesis (H3) is clearly rejected, as there is a weak correlation between real wages and changes in labor market conditions.

H3: The change in real wages serves as an effective adjustment channel for the Canadian

market.

Finally, we can confidently accept the hypothesis about interprovincial heterogeneity, as it has been repeatedly documented in the paper, that the provincial constants differ substantially from one another.

H4: There are persistent differences in labor mobility between the provinces, regardless

of the shocks.

5. Conclusions

This paper has analyzed the interprovincial labor migration of Canada using a variant of econometric models utilized previously by a well-established literature discussed in our literature review section. Although, the research fits into a series of prior studies that apply some version of the methodology both by Blanchard and Katz from 1992.

(35)

Additionally, the labor mobility adjustment mechanism is debilitated by the migration defining preferences of Canadians which are not primarily job related.

The results of our empirical tests suggest that provincial labor market conditions relative to the national have a role in migration decisions. Specifically, the higher the provincial unemployment relative to the national in a province, the higher the probability of the out-migration of individual from it. This aspect of Canadian labor market is similar to that of Australian, though its degree of impact lags behind the later. We also tested the explanatory power of the national unemployment rate on the development of domestic migration and found, that it has no significant impact on it, as there is no significant correlation between the national unemployment rate and the development of internal migration.

We applied a panel VAR model, with the main goal to examine the provincial labor market flexibility, the internal dynamics of migration and the its role in asymmetric shock adjustment. We found that the Canadian labor market is relatively flexible, but labor market impulses have a prolonged impact on the provinces, which do not help resolve internal asymmetries. The extent of migration is a relevant and decisive mechanism to overcome Canadian cross-provincial asymmetric shocks. The rate of migration is comparable to that of the US and similarly to previous studies, migration is the most articulated in the first 3-4 years after the shock, the total adjustment takes about 10 years, which is more typical of European internal migration, than to the US or Australia (Dao et al., 2013). We also indicated that movements in relative wages across provinces have not been considerably determined by provincial labor market conditions and does not play an important part of the adjustment process.

(36)

6. Bibliography

Abrigo, M. R., & Love, I. 2015. Estimation of Panel Vector Autoregression in Stata: A Package of Programs.

http://paneldataconference2015.ceu.hu/Program/Michael-Abrigo.pdf

Arpaia, A., Kiss, A., Palvolgi, B., & Turrini, A. 2014. Labour mobility and labour market adjustment in the EU. European Economy. ISSN 1725-3187

Bayoumi, T., Sutton, B., & Swiston, A. 2006. Shocking Aspects of Canadian Labor Markets. IMF working paper, 06/83, IMF.

https://www.imf.org/external/pubs/ft/wp/2006/wp0683.pdf

Belingher, D. 2015. A pVAR model built on the ricardian approach to deficits in Central and Eastern Europe.Ecoforum.

http://www.ecoforumjournal.ro/index.php/eco/article/view/299

Beyer, R. C., & Smets, F. 2015. Labour market adjustments in Europe and the US: How different? ECB Working Paper Series, NO 1767 ed., Vol. March 2015, European Central Bank

https://www.ecb.europa.eu/pub/pdf/scpwps/ecbwp1767.en.pdf?719e5b79eff9f29c069 b4897958afbac

Blanchard, O. J., & Katz, L. F. 1992.Regional Evolutions (1st ed.) Brookings Papers on Economic Activity. Pp. 1–75.,

https://scholar.harvard.edu/lkatz/files/regional_evolutions.pdf

Bohnert, N. 2013. Migration: Interprovincial, 2009/2010 and 2010/2011.Report on the Demographic Situation in Canada.

http://www.statcan.gc.ca/pub/91-209-x/2013001/article/11786-eng.htm

Bond, E. 2007. The euro-zone, Canada, and the optimum currency area theory. Atlantic International Studies Organisation, Spring. ISSN 1710-4947

https://atlismta.org/online-journals/0607-journal-gender-environment-and-human-rights/the-euro-zone-canada-and-the-optimum-currency-area-theory/

Canova, F., & Ciccarelli, M. 2013. Panel vector autoregressive models, A survey. Working paper series, NO 1507, Vol. January 2013. European Central Bank. ISSN 1725-2806

http://edz.bib.uni-mannheim.de/daten/edz-ki/ezb/13/w-paper/ecbwp1507.pdf Casni, A. C., Dumicic, K., & Tica, J. 2016. The Panel VAR Approach to Modelling the

Housing Wealth Effect: Evidence from selected European post-transition economies, Nase Gospodarstvo our economy, NO. 4 ed., Vol. 62.

https://www.degruyter.com/downloadpdf/j/ngoe.2016.62.issue-4/ngoe-2016-0021/ngoe-2016-0021.pdf

(37)

https://www.fraserinstitute.org/sites/default/files/interprovincial-migration-in-canada.pdf

Croarkin, C., & Tobias, P. 2003. Handbook of Statistical Methods. NIST/SEMATECH. http://www.itl.nist.gov/div898/handbook/

Cross, P., & Bergevin, P. 2012. Turning Points: Business Cycles in Canada since 1926. C.D. HOWE Institute.

https://www.cdhowe.org/sites/default/files/attachments/research_papers/mixed/Comm entary_366_0.pdf

Dao, M., Furceri, D., & Loungani, P. 2013. Moving closer? Changing patterns of labour mobility in Europe and the US. VOX, CEPR's Policy Portal

http://voxeu.org/article/labour-mobility-europe-and-us

Debelle, G., & Vickery, J. 1998. Labour market adjustment: evidence on interstate labour mobility. Reserve Bank of Australia, Research Discussion Paper 9801. Economic Research Department.

https://www.rba.gov.au/publications/rdp/1998/pdf/rdp9801.pdf Department of Finance Canada. 2011. Equalization Program.

https://www.fin.gc.ca/fedprov/eqp-eng.asp

Department of Finance Canada. 2014.Jobs report: The State of the Canadian labour market. ISBN: 978-1-100-23184-6

Department of Finance, Canada. 2016. Federal Support to Provinces and Territories. Retrieved May 2, 2017, from https://www.fin.gc.ca/access/fedprov-eng.asp Department of Finance, Canada.2017. Federal Support to Provinces and Territories.

https://www.fin.gc.ca/fedprov/mtp-eng.asp

Employment and Social Development Canada. 2017. Labour Mobility and Occupational Licensing in Canada.

https://www.cfmws.com/en/AboutUs/MFS/ResourcesMFRCs/Documents/2017%20D ocuments/METSpouse/Employment%20Presentations/Employment%20and%20Socia l%20Development%20Canada%20-%20Labour%20Mobility.pdf

Fatás, A., & Decressin, J. 1995. Regional labor market dynamics in Europe. https://flora.insead.edu/fichiersti_wp/inseadwp1994/94-61.pdf Grauwe, P. D. 2016. Economics of Monetary Union (Eleventh edition).

Oxford University Press. ISBN 978-0-19-873987-6. pp. 70-97

Hill, R. C., Griffith, W. E., & Lim, G. C. 2011. Principles of Econometrics (4th ed.). John Wiley and Sons.

Chapter 13-15

Historical boundaries of Canada. 2015.

(38)

Hoechle, D. 2007. Robust standard errors for panel regressions with cross-sectional dependence. The Stata Journal, Number 3.

http://www.stata-journal.com/sjpdf.html?articlenum=st0128 pp. 281–312

Hornstein, A. 2013. Why Labor Force Participation (Usually) Increases when Unemployment Declines. Economic Quarterly, Volume 99, Number 1.

https://www.richmondfed.org/~/media/richmondfedorg/publications/research/economi c_quarterly/2013/q1/pdf/hornstein.pdf

Intergovernmental Affairs. Government of Canada. 2010. Difference between Canadian Provinces and Territories.

http://www.pco-bcp.gc.ca/aia/index.asp?lang=eng&page=provterr&doc=difference-eng.htm

Jimeno, J. F., & Bentolila, S. 1998. Regional unemployment persistence (Spain, 1976-1994) Labour Economics, Issue 1, Vol. 5.

Pp 25-51

Kirby, J. 2016. Canada’s non-energy provinces need to step it up. Macleans.

http://www.macleans.ca/economy/economicanalysis/canadas-non-energy-provinces-need-to-step-it-up/

Liebig, T., Puhani, P. A., Jauer, J., & Martin, J. P. 2014. Migration as an Adjustment Mechanism in the Crisis? A Comparison of Europe and the United States. IZA, Discussion Paper, No. 7921.

http://ftp.iza.org/dp7921.pdf

O'Neill, C. 2014. Staying Put: More than half of Canadians not interested in moving for employment opportunities.

https://c.ymcdn.com/sites/www.cerc.ca/resource/resmgr/News_Releases/April_30_Re vised_survey_FINA.pdf

Persyn, D., & Westerlund, J. 2008. Error-correction–based cointegration tests for panel data. http://www.uh.edu/~bsorense/st0146.pdf

pp. 232-241

Prasad, E. S., & Thomas, A. 1997. Labor Market Adjustment in Canada and the United States Working paper, WP/97/2, International Monetary Fund,

http://www.elibrary.imf.org.proxy-ub.rug.nl/doc/IMF001/04193- 9781451841725/04193-9781451841725/Other_formats/Source_PDF/04193-9781451890167.pdf

Statistics Canada. Table 281-0049. 2017.

http://www5.statcan.gc.ca/cansim/a26?lang=eng&retrLang=eng&id=2810049&&patt ern=&stByVal=1&p1=1&p2=-1&tabMode=dataTable&csid=

(39)

Tencer, D. 2016. Canada's Inflation Is Eating All Your Wage Gains, And Then Some. Huffingtonpost.

http://www.huffingtonpost.ca/2016/11/25/wages-canada-inflation_n_13208758.html Wiczer, D. 2014. Are Wages and the Unemployment Rate Correlated?

(40)

7. Appendix

1. Annual GDP growth of energy and non-energy provinces

source: www.macleans.ca

2. Modified Wald test for groupwise heteroskedasticity in fixed effect regression model (netmigpop, undiff)

3. Wooldridge test for autocorrelation in panel data (netmigpop undiff)

4. Breusch-Pagan LM test of independence (netmigpop undiff)

Prob>chi2 = 0.0000 chi2 (10) = 1718.60

H0: sigma(i)^2 = sigma^2 for all i in fixed effect regression model

Modified Wald test for groupwise heteroskedasticity

Prob > F = 0.0000 F( 1, 9) = 54.225 H0: no first-order autocorrelation

Wooldridge test for autocorrelation in panel data

Based on 40 complete observations

Breusch-Pagan LM test of independence: chi2(45) = 236.727, Pr = 0.0000

__e10 0.0506 0.0740 0.2918 0.2895 0.0436 -0.6966 -0.4527 -0.5966 -0.0864 1.0000 __e9 -0.4642 -0.3324 -0.6055 -0.3436 -0.4189 -0.4680 -0.6575 0.0109 1.0000 __e8 0.0418 -0.3856 -0.2688 -0.2610 -0.3959 0.5595 0.3529 1.0000 __e7 0.2249 0.2114 0.4009 0.3331 0.2742 0.6202 1.0000 __e6 0.1836 -0.0043 0.0834 -0.0786 -0.1834 1.0000 __e5 0.0310 0.4095 0.3033 0.1543 1.0000 __e4 0.1680 0.2835 0.7964 1.0000 __e3 0.1208 0.4625 1.0000 __e2 -0.0010 1.0000 __e1 1.0000

(41)

5. Pesaran's test of cross sectional independence

6. Description of data

7. Impulse response to 1 standard deviation increase in employment growth, monthly data (36 lag, 72 instlag)

Average absolute value of the off-diagonal elements = 0.299

(42)

8. Interstate migration snapshot, 1976-77, 1986-87, 1996-97