Articles

Tariffs and the Composition of Employment: Evidence from the Canada–US Free Trade Agreement

Department of Economics, Wilfrid Laurier University, Waterloo, Ontario

https://doi.org/10.3138/cpp.2018-080

Abstract

Section:

I analyze the effect of the Canada–US Free Trade Agreement (CUSFTA) on the composition of employment at the local labour market level in Canada. I construct local labour-market-level changes in tariffs for both exports and imports, exploiting regional differences in pre-CUSFTA industrial composition to obtain variation in the degree to which CUSFTA affected localities across Canada. I find that Census Divisions (CDs) that experienced larger Canadian tariff cuts against US imports experienced higher rates of self-employment. CDs with larger American tariff cuts against Canadian goods correspondingly experienced lower rates of self-employment. These self-employment effects are dampened in CDs with higher initial shares of educated people, immigrants, and female workers. I do not find any evidence that part-time employment is affected by CUSFTA tariff cuts. The findings in this article provide evidence that increasing trade openness between two developed countries can affect the prevalence of self-employment.

L’auteur analyse l’incidence de l’Accord de libre-échange entre le Canada et les États-Unis (ALECEU) sur la composition de l’emploi au niveau des marchés du travail locaux au Canada. Il établit des indicateurs de la variation des tarifs douaniers au niveau des marchés du travail locaux tant pour les exportations que pour les importations, en exploitant les différences régionales dans la composition industrielle antérieure à l’ALECEU afin de déterminer dans quelle mesure l’ALECEU a touché les localités du Canada. L’auteur constate que les divisions de recensement ayant connu des réductions plus importantes des tarifs douaniers canadiens sur les importations américaines affichent des taux plus élevés de travail indépendant. Les divisions de recensement ayant connu des réductions plus importantes des tarifs douaniers américains sur les produits canadiens affichent parallèlement des taux plus faibles de travail indépendant. Ces répercussions sur le travail indépendant s’atténuent dans les divisions de recensement présentant des proportions initiales plus élevées de personnes instruites, d’immigrants et de travailleuses. L’auteur ne relève aucune donnée qui démontrerait que les réductions de tarifs de l’ALECEU ont eu une incidence sur le travail à temps partiel. Les constatations de l’auteur témoignent du fait que l’ouverture accrue des échanges entre deux pays développés peut avoir une incidence sur la prévalence du travail indépendant.

Keywords: tariffs, local labour markets, free trade agreements

Mots clés : accords de libre-échange, marchés du travail locaux, tarifs douaniers

Introduction

Section:

In this article, I adopt a local labour market approach to understand how different local labour markets in Canada that were exposed to increased import competition or better export opportunities as a result of the implementation of the Canada–US Free Trade Agreement (CUSFTA) performed relative to less-exposed regions in terms of employment outcomes and composition. More specifically, I construct regional measures of exposure to changes in tariffs by exploiting the pre-CUSFTA industrial composition of each Census Division (CD), a subnational geographical unit that is equivalent in size to US counties. I then use these regional measures to estimate CUSFTA’s impact on local employment outcomes, such as the self-employment rate.¹

I find that Canadian tariff cuts increased the incidence of self-employment while decreasing the incidence of paid employment; correspondingly, American tariff cuts had the opposite effect, reducing self-employment and increasing paid employment. I further find that the responsiveness of self-employment to these trade shocks depends on a CD’s initial characteristics. Specifically, I show that CDs with higher initial female employment shares, educated population shares, and immigrant population shares do not respond as sharply to the tariff cuts with changes in self-employment.

Studying the effect of the CUSFTA on employment composition is an important, policy-relevant, and timely issue for several reasons. First, the recent resurgence in policy uncertainty relating to the renegotiation of the North American Free Trade Agreement (NAFTA), as well as Brexit, has made understanding the costs and benefits of trade agreements between developed countries a topic of key concern once again. Second, much existing literature on the benefits and costs of the Canada–US trade relationship, and indeed other trade agreements between developed countries, does not consider the impact of trade shocks using local labour markets. Finally, understanding how trade shocks affect the composition of employment, and self-employment in particular, can crucially change the understanding of how workers’ employment outcomes and a country’s welfare and productivity are affected by increased trade openness. I further elaborate on the importance of each of these motivations in the next several paragraphs.

Recent policy uncertainty resulting from the renegotiation of NAFTA, motivated by President Trump’s targeting of the trade agreement as unfavourable to the United States, has cast a spotlight on the trade relationship between Canada and the United States and has renewed interest in understanding how CUSFTA and NAFTA have benefited their member countries. In particular, given Canada’s tight economic connection to and reliance on the United States, understanding how Canada’s local labour markets have been affected by increased trade openness with the United States is of first-order importance. Although firm-level studies using CUSFTA as their setting continue to be conducted, analyses of how labour markets have been affected have not been done recently, to the best of my knowledge, and in particular have not taken a local labour market approach.

The local labour market approach taken in this article follows that of numerous recent studies, such as those of Topalova (2007), Mccaig (2011), and Kovak (2013), that recognize that trade exposure at the geographical level can allow economists to capture general equilibrium impacts of increased openness to trade that industry-based measures cannot. As Autor, Dorn, and Hanson (2013) point out, the local labour market approach of mapping regional exposure to trade to labour market outcomes at the regional level allows researchers to get at both the direct impacts of trade, through trade-affected industries, and the indirect, general equilibrium effects in the local labour market. If the productivity-driven gains from trade are being captured via indirect effects in the same local labour market, a local labour market approach focusing on regional exposure to tariff cuts may better capture these productivity gains and how they potentially transmit to workers and labour market outcomes. An important implication for this article is that shifting the variation in trade exposure from the industry to the regional level also allows for analyses of regional outcomes not directly related to industry-level outcomes, such as self-employment.

The analysis of how trade openness, particularly between two developed countries, affects the prevalence of self-employment has remained largely unstudied. Little is therefore known of any potential relationship between trade and self-employment. I argue that understanding such a relationship matters greatly in understanding the labour market consequences and potential welfare effects of trade shocks. If workers are entering self-employment in response to an import shock (such as a cut in Canadian tariffs against American goods) that reduces employment opportunities in wage employment, looking purely at unemployment as an outcome could conceal the extent to which labour markets and workers are affected by import competition. If workers are just as well off in self-employment, this may not matter for welfare. Unfortunately, recent research by Galindo da Fonseca (2018) shows that workers who become entrepreneurs when laid off have worse outcomes than those who self-select into entrepreneurship in other circumstances, in terms of number of employees employed as well as their business’s survivability. This finding corroborates work by Humphries (2018), who argues that pushing workers into self-employment using subsidies induces the entry of low-productivity self-employed workers, lowering earnings and welfare. These two articles strongly suggest that workers who are pushed into self-employment, whether through being laid off, getting subsidies, or being exposed to trade shocks, are not as productive in self-employment. Whether trade openness affects self-employment rates therefore has important productivity and welfare implications.

This article first contributes to work examining the causes of self-employment, particularly in Canada. Kuhn and Schuetze (2001) examine self-employment in Canada from 1982 to 1998 and argue that increasing instability in paid employment was a key factor driving increasing self-employment trends for men.² More recently, Bahar and Liu (2015) use a similar definition of self-employment as that used in this article and also argue that self-employment rates respond to business cycle fluctuations. Similarly, in this article, I find that changes in self-employment rates induced by CUSFTA tariffs are always accompanied by corresponding opposing changes in the prevalence of paid employment in affected CDs.

Also using Canadian data from a similar period, Lin, Picot, and Compton (2000), using mainly time-series analyses, find no evidence that macroeconomic conditions influence self-employment inflows and outflows. In contrast, by using a first-difference approach combined with cross-regional variation in exposure to tariff cuts, the results presented in this article show that one particular type of macroeconomic shock, the signing of a free trade agreement, does in fact have an effect on self-employment. This is also consistent with more recent evidence presented by Roed and Skogstrom (2014), who find that working in a firm that would be exogenously shut down increases the subsequent probability of a worker becoming an entrepreneur.

There is also a significant literature that has focused on examining the various effects of CUSFTA. Of particular relevance to this article are the studies conducted by Gaston and Trefler (1997) and Trefler (2004), who use industry-level variation in tariff cuts to examine how CUSFTA affected workers and firms.³ Almost all studies that focus on the outcomes of CUSFTA do so using industry-level variation in trade exposure and therefore do not look for local labour market effects. One exception is that of Baldwin, Brown, and Gu (2012), who construct CD-level manufacturing trade exposure measures to study CUSFTA’s effect on manufacturing plants. Their work does not extend to local labour market effects, however, focusing solely on manufacturing plant outcomes. In addition, this article extends their analysis by incorporating industries beyond manufacturing into trade exposure measures, capturing the full spectrum of tradeable goods.

The rest of the article proceeds as follows. The next section outlines the theoretical framework. The third section provides some institutional background on CUSFTA and describes the data used, the fourth section specifies the empirical methodology used, the fifth section presents and explains the empirical results, and the final section concludes.

Theoretical Framework

Section:

To outline some mechanisms and illustrate potential consequences of the effect of trade shocks on self-employment, it is helpful to have a theoretical framework, or model, in mind. To that end, I now briefly describe a variant of the model in Blau (1987). In his model, workers can allocate their inelastically supplied labour in two different sectors: the wage employment sector or the self-employment sector. Each sector produces their own good, the value of which is given by exogenously determined prices. The output in the wage employment sector is normalized to one, so that the price of the good in the self-employment sector is a price relative to the wage employment sector good. All workers are identically productive in the wage employment sector and therefore earn identical wages. In contrast, workers are heterogeneously productive in the self-employment sector, with their self-employment productivity assumed to have no bearing on their productivity in the wage sector. Wages in the wage employment sector are determined by setting aggregate labour demand equal to aggregate labour supply, and they depend in equilibrium on the relative value of the self-employment good and productivity in the two sectors.⁴

Two implications of the Blau (1987) model are relevant to this article. First, workers will allocate more labour to self-employment if they are correspondingly more productive or capable in self-employment.⁵ Next, as the relative price of the self-employment sector good goes up, this has the effect of lowering the average ability or productivity in the self-employment sector because there is now an additional incentive for all workers to allocate labour to self-employment.

To incorporate trade shocks into the Blau (1987) model, it is simplest to think of a tariff cut that leads to increased import competition (i.e., a Canadian tariff cut) as a decrease in the price of the wage employment sector good relative to that in the self-employment sector good. Similarly, an American tariff cut, which increases export demand for Canadian goods in the United States, can be thought of as an increase in the price of the wage employment sector good relative to that of the self-employment sector good.⁶ A Canadian tariff cut would, therefore, using the results in Blau, induce workers to allocate labour to self-employment by making it a relatively more attractive prospect than wage employment, whereas an American tariff cut would do the opposite by making wage employment relatively more attractive. It is important to note that Blau would also imply that Canadian tariff cuts would lower the average quality or productivity of self-employed workers because workers with relatively low productivity in self-employment would be disproportionately induced into increasing their self-employment labour allocation because workers with higher self-employment productivity would already have allocated much or all of their labour to self-employment. Similarly, American tariff cuts would serve to increase the average productivity of those who are self-employed as a result of selection out of self-employment and into wage employment by workers not as skilled in self-employment.

The model presented in Blau (1987) is highly stylized. I nonetheless use it as theoretical motivation, to show that even a simple model of labour allocation can generate predictions concerning trade shocks and selection into self-employment. In more recent work involving the trade-off between wage employment and starting one’s own business, Galindo da Fonseca (2018) incorporates search frictions, unemployment, and a range of other features into a model to analyze how unemployment shocks to workers affect those workers’ decisions to start a business. He shows that unemployed workers are more likely to start a business but are also less productive in those businesses. In both articles, the mechanism is similar, with selection into starting a business as a result of an import (or other negative demand) shock to wage employment output driven by reduced opportunities or benefits to remaining in wage employment and the end result being more low-productivity workers entering self-employment.

This section highlights one mechanism through which trade shocks can affect workers’ decisions to enter or exit self-employment. It also implies that this change in employment composition could also have significant welfare implications because workers who are induced into entering and exiting self-employment through tariff cuts are those who are least well suited to self-employment, in terms of their ability.

Background and Data

Section:

In this section, I first provide an overview of the circumstances and background of the CUSFTA. Readers interested in a more comprehensive description of CUSFTA should consult Brander (1991). I then describe the data used in this article.

Background on CUSFTA

The CUSFTA, at the time of its signing in 1988, was touted as the world’s largest free trade agreement. It bound together two countries that already enjoyed strong economic and cultural ties. Townsend (2007) calculates that, just before CUSFTA went into effect, Canada sent 72.8 percent of its total exports to and sourced 65.6 percent of its total imports from the United States.

CUSFTA built off of the existing Auto Pact from the 1960s, which significantly lowered tariffs in the automotive industry between Canada and the United States A desire for increased economic integration with the United States after the Auto Pact led to proposals for a more comprehensive free trade agreement. This desire was embraced by the Progressive Conservative government during the mid-1980s, led by Prime Minister Brian Mulroney.

Although the proposed free trade agreement was relatively uncontroversial in the United States, unlike its eventual successor NAFTA, public opinion concerning a potential trade deal with the United States was very divided in Canada. The corresponding political divide between the Progressive Conservatives, who supported CUSFTA, and the opposition Liberal Party and New Democratic Party, both of which were staunchly opposed to further Canada–US integration, led to a federal election in 1988 for which a potential CUSFTA was the principal issue. The 1988 election was therefore a de facto referendum on CUSFTA, with the survival and ultimate ratification of the free trade agreement heavily dependent on whether the pro-trade Progressive Conservatives retained power.

The incumbent Progressive Conservative Party did in fact retain power after the 1988 election, ensuring the passage of CUSFTA. The outcome of the election, however, as many articles, such as Breinlich (2008) and Lileeva and Trefler (2010), point out, was highly uncertain ex ante. This uncertainty greatly mitigates concerns that the tariff changes guaranteed by the Progressive Conservative victory in November 1988 were anticipated beforehand, reducing the possibility that CDs that would be affected in the post-CUSFTA period by tariff cuts would have pre-emptively adjusted their employment and other labour market outcomes. In addition, as Breinlich argues, CUSFTA was primarily designed to be an agreement concerning trade liberalization only, and it therefore did not come packaged with other economic policies that may have confounded estimates of tariff effects. CUSFTA is thus an ideal natural experiment in which to examine the impacts of a trade liberalization between two highly developed countries.

The increase in trade flows between the two countries was considerable, despite the relatively small change in the tariff levels. Romalis (2007), for example, calculated that the commodities that received the highest level of tariff preferences from CUSFTA experienced a 99 percent increase in the Canadian share of US imports. Similarly, Trefler (2004) shows that tariff concessions increased American imports flowing into Canada by 54 percent. Lileeva and Trefler (2010) find that, after a relatively constant level of Canadian exports to the United States extending into 1991, Canadian exports skyrocketed by a remarkable 75 percent from 1991 to 1996. When coupled with the fact that Canada has a small, open economy that already relied heavily on trade before CUSFTA,⁷ the impacts of even relatively modest tariff cuts with the United States can be interpreted as potentially quite economically meaningful and important.

Data Sources

The main measures of exposure to the CUSFTA in this article are CD-level changes in tariffs, both for American tariffs against Canadian goods and vice versa. I begin by using the Canadian and US tariffs for all Standard Industrial Classification (SIC) three-digit industries in the manufacturing sector from Trefler (2004).⁸ I augment these data with agriculture and resource industry tariffs for both Canada and the United States from Magun et al. (1988), which have previously been used in Beaulieu (2000). I use tariffs from 1987 for pre-CUSFTA tariff levels and 1996 tariff levels to represent trade barriers in the post-liberalization period.⁹ For agriculture and resources, I perform some additional aggregations to the two-digit level because of limitations in the Census data that I used to construct employment weights.¹⁰ Combining both data sources, I obtain tariff data for 1987 and 1996 for 119 industries, which I use to construct local labour-market-level tariff change measures.

I use CDs as the sub-national level of aggregation that represents local labour markets. As noted earlier, CDs are the Canadian equivalent of US counties, and they form the basis for the Commuting Zones used as local labour markets in Autor et al. (2013). CDs have also been used extensively as local labour markets in research, such as that by Marchand (2012). To create the CD-level tariff measures, I first take the SIC-level tariffs described previously and construct CD-level tariff changes using 1986 employment counts by SIC industry for each CD. These employment counts by CD and industry are obtained through a custom tabulation request from Statistics Canada, drawing from the 1986 Canadian Census. Using these custom data, I construct 1986 employment weights with which to create industry employment-weighted average 1987 and 1996 tariffs for each CD. Finally, the difference between the weighted 1996 and 1987 Canadian and US tariffs are calculated to form my main measures of trade exposure resulting from CUSFTA.

CD-level variables concerning labour market outcomes and demographics are obtained or calculated from the 1986 and 1996 Census profiles for each CD, available through Statistics Canada’s Data Liberation Initiative. The year 1986 is the nearest Census year to CUSFTA’s implementation in 1989. I use 1996 as the end date for this study because it minimizes any contaminating impact that the rise of China as a major exporter to the United States has in terms of diverting trade. I do not use 1991 as the post-CUSFTA period because of the recession that occurred from 1989 to 1992, which Gaston and Trefler (1997) argue was not caused by CUSFTA, but was an issue that confounded the earliest studies on the impact of CUSFTA. In this study, I therefore use 1986 and 1996 as the pre-CUSFTA and post-CUSFTA periods, respectively. I also focus on CDs within provinces, omitting those in the Yukon and Northwest Territories; this leaves 260 of 266 CDs, using 1986 CD definitions. Variables with monetary values (such as dwelling value, household incomes, and rent) are converted to 1996 Canadian dollars using the Canadian Consumer Price Index. I calculate the employment-to-population ratio of a CD as that CD’s total employment divided by the population aged 15 years and older. Finally, the self-employment rate used is the share of unincorporated self-employed individuals divided by all workers.¹¹

One issue that complicates the analysis is that CD definitions, although typically somewhat stable over time, changed substantially in certain provinces from 1986 to 1996. In particular, Statistics Canada undertook a significant overhauling of CDs in Quebec to more closely align them with CD definitions in other provinces. I therefore use custom Census profile data for 1996 from Statistics Canada that have been converted to 1986 CD definitions to maintain consistency in the definitions over time.

Trade Exposure Measures

My main measures of the change in trade exposure as a result of CUSFTA are of the form

Δ τ_{c} = \frac{{\sum_{i = 1}^{I^{t}} L}_{c i, 1986} Δ_{τ_{i}}}{{\sum_{i = 1}^{I^{t}} L}_{c i, 1986}} .

(1)

Δτ_c, the change in tariffs for CD c, consists of the change in tariffs for each SIC industry i, Δτ_i, in the set of traded industries, I^t.¹² These industry-level tariff changes are weighted by that industry’s employment share,

\frac{L_{c i, 1986}}{{\sum_{i = 1}^{I^{t}} L}_{c i, 1986}},

, on employment levels within each CD c in 1986. For expositional purposes, I denote

Δ τ_{c}^{C A N}

as the CD-level change in Canadian tariffs on US imports, and I refer to

Δ τ_{c}^{U S A}

as the CD-level change in US tariffs on Canadian imports.¹³ This measure of local labour-market-level tariffs is consistent with those used in recent research by Hakobyan and McLaren (2016) and Kovak (2013).

Summary Statistics

Table 1 presents some basic summary statistics, without weighting any CDs. Table A.2 of the Appendix also presents the same summary statistics but weighting by 1986 population levels as in the main regression results.

Panel A of Table 1 presents summary statistics for selected CD-level variables for 1986. The table illustrates the substantial degree of variation that exists in most of the outcome variables of interest. For example, the average employment-to-population ratio for CDs, weighting each CD equally, was 55.2 percent, but it ranged from a low of 31.9 percent (for Division 19 in Manitoba) to a high of 73.7 percent (for Peel Regional Municipality in Ontario). Other key variables display similarly high degrees of geographic variation in the pre-CUSFTA period of 1986. Panel B of Table 1, which presents descriptive statistics for CD-level variables in 1996, displays similar trends in terms of variability.¹⁴

Panel C of Table 1 provides the extent of CD-level average tariff changes, both Canadian and American, from 1987 to 1996. The average decline in Canadian tariffs against American goods was 3.2 percent, whereas the average decline in American tariffs was 1.8 percent. These values are consistent with those calculated by Magun et al. (1988), who found that Canadian tariffs would decline by 3.2 percent overall and that American tariffs would decline by 2 percent overall, from 1987–1996. They are, however, notably smaller than those found in studies that focus on manufacturing only, such as Trefler (2004). The main reason is that manufacturing industries underwent the largest tariff cuts as a result of CUSFTA, which I discuss later in greater depth.

These averages mask a considerable degree of variation across Canada. Canadian tariff cuts at the CD level, for example, varied from 0.3 percent to 8.5 percent. To further illustrate this point, I map CD-level tariff declines for Canadian and US tariffs from 1987 through 1996 in Figures 1 and 2, respectively. The maps clearly show that exposure to CUSFTA was not equally distributed spatially. Southern Ontario and Quebec, for example, were exposed to larger Canadian and US tariff cuts than Northern Alberta and Manitoba. Figures 1 and 2 also imply that a CD’s exposure to Canadian tariff cuts was correlated with that CD’s American tariff cuts. This is not surprising, given that an industry’s Canadian tariff cuts are positively correlated with that industry’s American tariff cuts. This can be clearly seen in Figure 3, where I plot each industry’s 1987–1996 Canadian tariff change with its corresponding 1987–1996 American tariff change. Figure 3 therefore implies that CD-level tariff exposure from Canadian and American tariffs will be correlated, explaining the similarities in Figures 1 and 2.¹⁵

Table 1: Summary Statistics

Variable	M (SD)	Range, Min–Max
Panel A: 1986 levels
Population	97,052.27 (213,689.4)	2,022–2,192,721
Proportion of immigrants in population	0.072 (0.064)	0.003–0.402
Female employment share	0.397 (0.026)	0.319–0.462
Labor force participation rate	0.631 (0.057)	0.470–0.786
Employment-to-population ratio	0.552 (0.079)	0.319–0.737
Proportion full-time employment	0.460 (0.077)	0.188–0.607
Proportion part-time employment	0.540 (0.077)	0.393–0.812
Proportion paid employment	0.900 (0.068)	0.613–0.978
Proportion self-employment	0.100 (0.068)	0.022–0.386
Proportion of working population with university degree	0.060 (0.027)	0.025–0.202
Proportion employed in manufacturing	0.178 (0.099)	0.008–0.443
Proportion employed in primary industry	0.165 (0.119)	0.004–0.528
Panel B: 1996 levels
Population	109,357.8 (240,032.4)	1,265–2,363,875
Proportion of immigrants in population	0.066 (0.068)	0.003–0.475
Female employment share	0.446 (0.019)	0.396–0.496
Labor force participation rate	0.638 (0.064)	0.480–0.810
Employment-to-population ratio	0.564 (0.086)	0.305–0.755
Proportion full-time employment	0.473 (0.073)	0.226–0.588
Proportion self-employment	0.106 (0.055)	0.026–0.333
Proportion of working population with university degree	0.083 (0.037)	0.033–0.258
Proportion employed in manufacturing	0.150 (0.085)	0.011–0.438
Proportion employed in primary industry	0.137 (0.105)	0.004–0.532
Panel C: 1987–1996 tariff changes
$Δ τ_{c}^{U S}$	−0.018 (0.009)	−0.052 to −0.003
$Δ τ_{c}^{C A N}$	−0.032 (0.019)	−0.052 to −0.003

Notes: All summary statistics are presented unweighted across all Census Divisions in the sample. Any statistics on levels or changes therefore do not necessarily reflect national levels or changes.

Source: Author’s calculations.

Differences in tariff exposure at the CD level stem, by construction, from pre-CUSFTA industrial composition by employment. An important determinant of differences in the extent of pre-CUSFTA tariff levels and subsequent CUSFTA tariff cuts at the CD level are whether that CD had more employment in agriculture and resources relative to manufacturing. Agriculture and resource industries did not experience large tariff cuts, largely as a function of their already low pre-CUSFTA tariff levels. Forestry imports from the United States coming into Canada, for example, were only assigned a tariff level of 0.1 percent in 1987. Similarly, Canadian agriculture tariffs declined from 2.2 percent in 1987 to 0.4 percent in 1996. In contrast, manufacturing industries experienced the largest tariff cuts on both sides of the border, owing to their much higher pre-CUSFTA tariff levels. On average, manufacturing had a Canadian tariff level of 6.7 percent against American goods in 1987, with a similar level of 3.2 percent for American tariffs against Canadian goods.¹⁶ In Figure A.1 in the Appendix, I provide additional graphical evidence that CD-level tariff cuts, both Canadian and American, were highly correlated with that CD’s 1987 pre-CUSFTA regional tariff level. The figure shows that tariff cuts from both sides of the border were almost perfectly predicted by that region’s initial tariff levels.

Figure 1: Change in CD-Level Canadian Tariffs against the United States, 1987–1996

Note: CD = Census Division.

Source: Author’s calculations.

Figure 2: Change in CD-Level US Tariffs against Canada, 1987–1996

Note: CD = Census Division.

Source: Author’s calculations.

Even within manufacturing, however, there exists considerable heterogeneity in terms of the size of tariff cuts faced between 1987 and 1996. Knitting mills experienced the largest tariff cuts on both sides of the border, with declines of 17.2 percent for Canadian tariffs and 10.1 percent for American tariffs.¹⁷ Other manufacturing industries, such as paper products, saw much smaller changes in tariffs. I further illustrate the variation in industry-level tariff cuts by listing in Table 2 the top and bottom five industries affected by both Canadian and US tariff cuts.

Figure 3: Canadian and US Industry-Level Tariff Changes, 1987–1996

Note: CDN = Canadian.

Source: Author’s calculations.

The data displayed in Table 2 further highlight the fact that much of the variation in the data stems from differences in tariff exposure in manufacturing; for both American and Canadian tariffs, the largest changes were often industries in manufacturing. For example, the industry that faced the largest Canadian tariff cuts from 1987 to 1996 was children’s clothing. Similarly, the industry that received the largest American tariff cuts was men’s and boy’s clothing. It is this variation in employment within manufacturing and that across the manufacturing, agriculture, and resources sectors that I exploit to construct local measures of exposure to tariff changes.

To corroborate whether self-employment is relatively insulated from the CUSFTA tariff cuts, I use the 1986 Census public use microdata file to examine in which sectors self-employed workers are employed. Approximately 24 percent of self-employed workers work in the agricultural and primary sectors, which only see very slight changes to tariffs. Only 5.4 percent of self-employed workers work in the manufacturing sector, which is the sector most affected by the tariff cuts. The remaining more than 70 percent of self-employed workers are in various service sectors, which are not directly affected by the tariff cuts.¹⁸ The data therefore strongly confirm that self-employment, at least from sectoral composition, is more insulated from trade shocks.

Table 2: Largest Industry-Level Tariff Changes, 1987–1996

Rank	Industry	Tariff Change
Canadian tariff cuts
1	Children’s clothing	−0.182
2	Women’s clothing	−0.177
3	Brewery products	−0.174
4	Men’s and boys’ clothing	−0.160
5	Other clothing and apparel	−0.152
US tariff cuts
1	Men’s and boys’ clothing	−0.138
2	Other clothing and apparel	−0.121
3	Women’s clothing	−0.113
4	Plastic and synthetic resin	−0.087
5	Broad-knitted fabric	−0.086

Notes: Industries are three-digit Canadian Standard Industrial Classification codes, except where aggregated as described in the “Data Sources” section.

Source: Author’s calculations.

Comparison of Self-Employment from 1986 to 1996

I examine differences in the composition of self-employment in both 1986 and 1996. For 1986, using the Census public use microdata file, I calculate that approximately 46 percent of self-employed individuals had paid help. Of self-employed individuals, 31 percent were incorporated; 77 percent were male, and 13 percent had at least four years of university education. The mean age of self-employed individuals was 44.37 years. The mean income of self-employed individuals, in 1996 Canadian dollars, was $33,950, and the mean income of other workers was $25,891.

Using the 1996 Census public use microdata file, I calculate that 39 percent of self-employed individuals had paid help, 32 percent were incorporated, 69 percent were male, and 18 percent had at least four years of university education. The mean age of self-employed individuals was 44.55. The mean income of self-employed individuals was $30,645, and the mean income of other workers was $27,807.

The differences in mean income between paid and self-employed workers suggests that the income benefit to self-employment decreased between 1986 and 1996. I confirm this in regressions run separately by Census year that control for industry and province fixed effects, age, sex, and university education. I find that although in 1986 self-employed individuals made $4,743 more in 1996 Canadian dollars, in 1996 the self-employed made $794 fewer; these differences in both years are statistically significant using robust standard errors.

Although the income gap between self-employed and paid workers appears to have changed in the pre- and post-CUSFTA period, caution should be exercised when interpreting this as evidence in favour of or against the push–pull mechanism described in the “Theoretical Framework” section. The data used in this analysis do not allow me to account for unobserved worker heterogeneity; for example, a decrease in self-employed incomes post-CUSFTA relative to paid workers’ income could be due to low-ability or low-productivity workers transitioning from paid employment to self-employment.¹⁹ In addition, even if incomes do not respond to CUSFTA tariff cuts, workers could still find it relatively more advantageous to enter self-employment if an increase in Canadian tariffs against US goods leads to a lower probability of finding a job in paid employment.²⁰ Unfortunately, pinning this mechanism down exactly requires data on job postings and workers searching for paid employment, which to my knowledge is not available for the time period and is beyond the scope of this article.

Empirical Methodology

Section:

Using the local employment-weighted tariff changes from Equation (1), I begin by estimating specifications of the form for employment-related outcomes:

Δ y_{c} = β_{0} + β_{1} Δ τ_{c}^{C A N} + β_{2} Δ τ_{c}^{U S} + {X^{'}}_{c} β_{3} + \sum γ_{p} p r o c_{c, p} + ε_{c} .

(2)

Δy_c is a first-difference outcome variable for CD c from 1986 to 1996. Estimating the specification in first differences removes any time-invariant heterogeneity across CDs. β₁ and β₂ capture the average effect of changes in Canadian and American tariffs, respectively, on changes to the outcome of interest. I control for a vector of control variables at the CD level in X_c in all regressions as well. These control variables include the female employment share, the proportion of the population that is foreign born, the share of employment in manufacturing, and the share of the working-age population that has a university degree, all in levels for 1986. This specification is consistent with that used recently by Autor et al. (2013) and Feler and Senses (2017) to examine the impacts of Chinese import competition on American local labour markets. Finally, I also include a set of province fixed effects,

\sum γ_{p} p r o v_{c, p}

that allow for unobservable differences in changes to the outcome variables from 1986 to 1996 across different provinces.

All regression results presented are weighted using each CD’s 1986 population. I use heteroskadasticity-robust standard errors in all specifications.

Results

Section:

Employment Effects

I first examine the impact of US and Canadian tariff cuts on regional employment outcomes. Estimates of specification (2) are presented in Table 3. These estimates suggest that there is no meaningful relationship between tariff cuts and changes to a CD’s employment-to-population ratio (column 1), labour force participation rate (column 2), or the proportion of workers who work full time (column 3). The effects presented in Table 3 therefore suggest that the impacts of CUSFTA on employment rates, labour force participation, and full-time employment status were negligible, consistent with prior industry-level studies such as that of Gaston and Trefler (1997).²¹

Table 3: Employment Effects

Dependent Variable	(1) Δ Employment-to Population Ratio Δ Employment-to-Population Ratio	(2) Δ Labour Force Participation Rate	(3) Δ Proportion Employed Full Time	(4) Δ Proportion Self-Employed
$Δ τ_{c}^{C A N}$	0.052	−0.168	0.283	−0.439***
$Δ τ_{c}^{C A N}$	(0.280)	(0.219)	(0.176)	(0.134)
$Δ τ_{c}^{U S}$	−0.095	−0.174	−0.280	0.831***
$Δ τ_{c}^{U S}$	(0.383)	(0.342)	(0.318)	(0.232)
Female employment share₁₉₈₆	−0.178*	−0.267***	0.173***	0.098**
Female employment share₁₉₈₆	(0.093)	(0.087)	(0.063)	(0.039)
Immigrant population share₁₉₈₆	−0.253***	−0.171***	−0.061***	−0.019*
Immigrant population share₁₉₈₆	(0.029)	(0.020)	(0.015)	(0.012)
Manufacturing employment share₁₉₈₆	0.055	0.011	0.018	0.023*
Manufacturing employment share₁₉₈₆	(0.034)	(0.023)	(0.017)	(0.013)
Proportion with university degree₁₉₈₆	0.083	−0.017	−0.174***	0.155***
Proportion with university degree₁₉₈₆	(0.066)	(0.049)	(0.045)	(0.027)
CD controls	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes
Observations	260	260	260	260
R²	0.834	0.722	0.627	0.634

Notes: Robust standard errors are reported in parentheses. CD controls include the 1986 female employment share, the 1986 immigrant population share, the 1986 manufacturing employment share, and the proportion of the working-age population with a university degree in 1986. $Δ τ_{c}^{C A N}$ is the CD-level weighted change in Canadian tariffs against the United States; $Δ τ_{c}^{U S}$ is the CD-level weighted change in US tariffs against Canada. CD = Census Division; FE = fixed effect.

*p < 0.10;

**p < 0.05;

***p < 0.01.

Source: Author’s calculations.

I find that Canadian tariff cuts induce a significant shift toward self-employment (column 4) and away from paid employment for workers. American tariff cuts, conversely, significantly shift workers into paid employment and away from self-employment. These magnitudes are also large and significant at the 1 percent level. The findings in Table 3 therefore indicate that workers were induced to enter and exit self-employment as a result of exposure to Canadian and American tariff cuts, respectively. The relative lack of effect on other employment outcomes such as the employment-to-population ratio would also be explained if workers are being forced out of paid employment and into self-employment as a result of fewer opportunities because of import competition from American goods and vice versa for export opportunities as a result of American tariff cuts.

To get a sense of the magnitude of the tariff effects, I calculate the self-employment rate effects of the average decrease in Canadian and American tariffs. The average Canadian tariff decrease of 3.2 percentage points, from Magun et al. (1988), would translate to a 1.4 percentage point increase in the self-employment rate for that CD. Similarly, the average American tariff decrease of 2 percentage points would imply a 1.7 percentage point decrease in the self-employment rate. For perspective and scale, using 1986 Census data and the same definition of self-employment, I calculate that 6.5 percent of employed workers were self-employed in 1986.

Although the two effects, evaluated at the simple mean tariff changes, seemingly more or less cancel out, it is also important to note that the larger coefficient on the American tariff variable implies that a CD that experienced equivalent Canadian and American tariff cuts would see an overall decrease in its self-employment rate. This implies that a trade liberalization episode with another developed country in which tariff cuts were equalized on both sides, such as CUSFTA, results in a net decrease in the proportion of employment made up of self-employed individuals.

For additional statistics and discussion concerning the employment effects from US and Canadian tariff cuts and which CDs benefited the most and the least from CUSFTA in terms of self-employment, see the “Geographic Variation in Tariff-Induced Self-Employment” section later in this article, where I discuss the geographic distribution of the impact of CUSFTA-induced tariff cuts on local self-employment rates.

Heterogeneity of Employment Effect Results

I now turn to a relatively brief discussion of potentially heterogeneous impacts of the CUSFTA-induced tariff cuts on the employment outcomes, as presented in Table 4. Specifically, I show that a CD’s initial 1986 population share of university-educated individuals, its 1986 female employment share, and its 1986 immigrant population share have an impact on how much self-employment is affected by CUSFTA. I do so by taking measures for these 1986 CD characteristics and interacting them with the tariff measures used in the main analysis.

First, in Panel A of Table 4, I find that CDs with a higher initial share of university-educated people in the population do not experience as large an impact on self-employment. The estimates show that higher education levels in the population actually serve to dampen both the positive effect of the Canadian tariff cuts on self-employment and the size of the shifts out of self-employment induced by the American tariff cuts. The heavier effects on less educated workers echo the patterns found in the United States with Chinese imports by Autor et al. (2013). This also lends some support to the notion that workers who are induced by business cycle fluctuations into and out of self-employment may not be as productive or able as those who enter self-employment for non-business-cycle reasons and is consistent with recent Canadian work by Galindo da Fonseca (2018).²²

It is possible that men and women react differently to trade shocks, and in particular that they sort into and out of self-employment differently in response to tariff cuts.²³ In Panel B of Table 4, I therefore analyze whether gender plays a role in the selection into self-employment by interacting each CD’s tariff cuts with its 1986 female employment share. I find that CDs with higher initial female employment shares also do not experience shifts into and out of self-employment that are as large as those with lower initial female employment shares. This is consistent with the Norwegian evidence of Roed and Skogstrom (2014), who found that workers who are displaced because of employer bankruptcy are 3.7 percentage points more likely to be entrepreneurs four years later if they are men, but are only 1.8 percentage points more likely to be entrepreneurs if they are women.

Finally, in Panel C of Table 4, I investigate whether CDs that are proportionately composed of more immigrants are more or less likely to have self-employment rates that respond to trade shocks. This line of inquiry is motivated by work by Cadena and Kovak (2016), who find that immigrants were able to help negate the negative impact of the Great Recession in US cities. I interact the tariff measures with each CD’s 1986 immigrant population share and find that, as with education and female employment, immigrant shares serve to dampen the impacts of CUSFTA on self-employment. Note that this does not imply that immigrants are not overall more likely to be self-employed, as Borjas (1986) and others have found, but rather that immigrants may not be as inclined as non-immigrants to being influenced into or out of self-employment as a result of trade shocks. The findings in Panel C of Table 4 stand in contrast to those of Cadena and Kovak (2016), however, and suggest that the role of immigrants as labour market arbitrageurs may be different in Canada or may not operate through the channel of self-employment.

Taken together, the results in Table 4 suggest that CUSFTA had stronger impact on men, less educated individuals, and non-immigrants.

Geographic Variation in Tariff-Induced Self-Employment

In this section I further discuss how the costs (and benefits) of the CUSFTA are distributed across Canada’s CDs.

To illustrate my point, I first take the tariff effect coefficients from column 4 of Table 3, which presents the estimates of the effect of Canadian and US tariff changes on CD-level changes in the local self-employment rate. I then calculate a predicted change in the local self-employment rate for each CD, based on that CD’s exposure to CUSFTA-induced Canadian and American tariff cuts. More specifically, for each CD, I calculate

Δ \hat{S} E_{c} = {\hat{β}}_{1} Δ τ_{c}^{C A N} + {\hat{β}}_{2} Δ τ_{c}^{U S} .

(3)

I interpret ∆ŜE_c as the overall impact of CUSFTA on each CD c’s self-employment rate, allowing for CUSFTA’s impact to vary through both import (Canadian) and export (American) tariff cuts at the CD level. I take these predicted changes in self-employment rates for each CD and visually illustrate the geographic variation in these predicted values in Figure 4; on the map, darker blue areas represent CDs that have the largest increases in the predicted self-employment rate. As expected, there is considerable geographic variation across Canada in terms of the net effect of CUSFTA.

Much of southern Ontario and Quebec, which were relatively heavily concentrated in manufacturing, experienced the largest net increases in self-employment as a result of tariff cuts from CUSFTA. Certain areas within the Maritimes also exhibit a particularly strong, CUSFTA-induced increase in the self-employment rate. Conversely, a band of CDs in Alberta, Saskatchewan, and Manitoba were predicted to have close to zero, or even slightly negative, changes in self-employment as a result of CUSFTA; these are displayed in Figure 4 as the lightest coloured regions. This is likely the result of these CDs’ heavier concentration in primary, non-manufacturing industries. Overall, the picture painted in Figure 4 is one in which certain regions display some trends in terms of its impact from CUSFTA, but also one in which considerable variation exists even within provinces. It is also relatively clear that the most and least affected CDs, in terms of self-employment, are not necessarily concentrated in one region or province, a fact that I further illustrate in the analysis in the next paragraph.

Table 4: Employment Effects, Heterogeneity

Dependent Variable	(1)Δ Employment-to-Population Ratio	(2)Δ Labour Force Participation Rate	(3)Δ Proportion Employed Full Time	(4)Δ Proportion Self-Employed
Panel A: by 1986 population share with bachelor’s degree
$Δ τ_{c}^{C A N}$	−0.896	−0.663	0.272	−1.654***
$Δ τ_{c}^{C A N}$	(0.510)	(0.411)	(0.357)	(0.253)
$Δ τ_{c}^{C A N} \times {educ}_{c}$	15.17**	7.951*	1.428	14.52***
$Δ τ_{c}^{C A N} \times {educ}_{c}$	(6.090)	(4.717)	(3.806)	(2.868)
$Δ τ_{c}^{U S}$	0.671	0.222	−0.684	3.442***
$Δ τ_{c}^{U S}$	(0.934)	(0.745)	(0.699)	(0.463)
$Δ τ_{c}^{U S} \times {educ}_{c}$	−14.93	−7.779	2.909	−31.33***
$Δ τ_{c}^{U S} \times {educ}_{c}$	(11.67)	(8.683)	(6.754)	(5.082)
CD controls	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes
Observations	260	260	260	260
R²	0.835	0.728	0.632	0.683
Panel B: by 1986 female employment share
$Δ τ_{c}^{C A N}$	−6.393**	−2.256	1.166	−8.372***
$Δ τ_{c}^{C A N}$	(2.634)	(2.334)	(2.093)	(1.523)
$Δ τ_{c}^{C A N} \times {female}_{c}$	15.74**	5.211	−1.891	18.81***
$Δ τ_{c}^{C A N} \times {female}_{c}$	(6.365)	(5.211)	(4.986)	(3.662)
$Δ τ_{c}^{U S}$	8.140	0.838	−5.418	19.43***
$Δ τ_{c}^{U S}$	(5.232)	(4.871)	(4.197)	(2.970)
$Δ τ_{c}^{U S} \times {female}_{c}$	−20.24	−2.749	12.00	−44.38***
$Δ τ_{c}^{U S} \times {female}_{c}$	(12.68)	(11.63)	(9.913)	(7.100)
CD controls	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes
Observations	260	260	260	260
R²	0.840	0.726	0.635	0.688
Panel C: by 1986 immigrant population share
$Δ τ_{c}^{C A N}$	−0.870**	−0.678***	−0.022	−0.972***
$Δ τ_{c}^{C A N}$	(0.317)	(0.243)	(0.244)	(0.167)
$Δ τ_{c}^{C A N} \times {imm}_{c}$	11.36***	6.546***	4.135**	4.653***
$Δ τ_{c}^{C A N} \times {imm}_{c}$	(2.340)	(1.836)	(1.854)	(1.134)
$Δ τ_{c}^{U S}$	1.074*	0.402	−0.015	2.056***
$Δ τ_{c}^{U S}$	(0.600)	(0.461)	(0.480)	(0.317)
$Δ τ_{c}^{U S} \times {imm}_{c}$	−14.58***	−7.771**	−4.355	−10.82***
$Δ τ_{c}^{U S} \times {imm}_{c}$	(4.079)	(3.383)	(3.639)	(1.978)
CD controls	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes
Observations	260	260	260	260
R²	0.857	0.739	0.642	0.664

Notes: Robust standard errors are reported in parentheses. CD controls include the 1986 female employment share, the 1986 immigrant population share, the 1986 manufacturing employment share, and the proportion of the working-age population with university degrees in 1986. $Δ τ_{c}^{C A N}$ is the CD-level weighted change in Canadian tariffs against the United States; $Δ τ_{c}^{U S}$ is the CD-level weighted change in US tariffs against Canada. CD = Census Division; FE = fixed effect.

*p < 0.10;

**p < 0.05;

***p < 0.01.

Source: Author’s calculations.

Figure 4: Predicted 1986–1996 Change in Self-Employment Share, from 1987–1996 Tariff Changes

Note: Darker blue areas represent Census Divisions that have the largest increases in the predicted self-employment rate.

Source: Author’s calculations.

I summarize the top and bottom five most affected CDs in Table 5. First focusing on the CDs with the largest predicted self-employment rate change, Panel A of the table shows that the five CDs predicted to see the largest increases in the self-employment rate were primarily located in Quebec or the Maritimes. The CD with the highest predicted CUSFTA-induced increase in the self-employment rate is Saint John County in New Brunswick, with a predicted increase of 0.011. Of the top five most positively affected CDs, two are in Quebec,²⁴ two are in the Maritimes,²⁵ and one is in British Columbia. The patterns exhibited by this list largely mirror the visual patterns displayed in Figure 4, as described previously.

Because CDs vary somewhat widely by population,²⁶ I restrict the sample to only the top 50 CDs by 1986 population levels and re-rank the most positively affected CDs by the change in predicted self-employment rate in Panel B of Table 5. Halifax County, in Nova Scotia, with a predicted self-employment rate increase of 0.007, is the CD predicted to have the largest CUSFTA-induced self-employment rate increase in the large CD ranking in Panel B. The regional patterns displayed in Panel A are still somewhat consistently present in Panel B as well; two of the five most affected large CDs are still in Quebec, with more in Nova Scotia (Halifax County). The smaller Maritimes and British Columbia CDs from the unrestricted ranking are now replaced by larger Ontario CDs.²⁷ Overall, the rankings in this panel reflect the pattern of Southern Ontario, Quebec, and Maritimes CDs having the highest predicted increases in self-employment, roughly consistent with the top panel and the map in Figure 4.

Table 5: CDs Least and Most Affected by CUSFTA by Self-Employment

Rank	CD	Province	Predicted Δ in Self-Employment Rate
Panel A: all CDs
1	Saint John County	NB	0.011
2	Division No. 6	NL	0.009
3	Montmagny	QC	0.008
4	Powell River Regional District	BC	0.007
5	Montmorency No. 1	QC	0.007
...
256	Division No. 8	MB	−0.008
257	Frontenac	QC	−0.008
258	Victoria County	NB	−0.009
259	Carleton County	NB	−0.009
260	Division No. 9	MB	−0.009
Panel B: top 50 CDs by 1986 population only
1	Halifax County	NS	0.007
2	Hull	QC	0.006
3	Thunder Bay District	ON	0.005
4	Levis	QC	0.005
5	Peel Regional Municipality	ON	0.005
...
46	Division No. 2	AB	−0.002
47	Division No. 8	AB	−0.002
48	Algoma District	ON	−0.003
49	Lambton County	ON	−0.005
50	Frontenac County	ON	−0.007

Note: CD = Census Division; CUSFTA = Canadian–US Free Trade Agreement.

Source: Author’s calculations.

The most negatively affected CDs, which in this case refer to CDs that have the largest predicted decrease in self-employment from CUSFTA, are also listed in Table 5. Panel A of Table 5 shows that three of the five CDs predicted to have the most negative predicted changes in self-employment are from Quebec or New Brunswick, with the other two coming from Manitoba. The distribution of least affected CDs across regions and provinces bears some similarity to the most affected CDs. This illustrates that, even within provinces and regions, the variation in CDs that experience different predicted outcomes in self-employment as a result of the CUSFTA tariff cuts is large.

Turning again to Panel B of Table 5, I examine the CDs with the most negative predicted changes in self-employment in the top 50 CDs by population. Although earlier the most positively affected CDs still largely maintained their regional composition between Panels A and B, this is not the case in the most negatively affected CDs. Restricting only to large CDs shows that the most negatively affected CD is Frontenac County in Ontario, with a predicted change in self-employment rate of −0.007. In fact, three of the five least affected large CDs are now in Ontario, with the remaining two being from Alberta (Divisions 2 and 8).

Robustness Checks

In this section, I discuss various robustness checks that I perform on the main set of employment results.

I present three sets of robustness checks in Table A.4. First, I verify that the weighting of observations does not drive my results by running unweighted versions of the regressions run for the results in Table 3. Panel A of Table A.4 presents these results and shows that the self-employment results are even stronger and larger in magnitude. Because CDs vary widely in their populations, though, I prefer the estimates shown in Table 3 and use them as my preferred estimates.

Next, in Panel B of Table A.4, I include a measure of how far a CD is from the nearest US border crossing. Specifically, the measure is the logged distance from that CD’s centroid to the nearest crossing. The results in Panel B also show that although the magnitudes are very slightly affected, the sign and significance of the coefficients for the self-employment column are maintained.

In Panel C of Table A.4, I control for sectoral composition shifts that may also have occurred as a result of CUSFTA. In some results showing sectoral composition effects from the tariff cuts that have been omitted for brevity, I find that CUSFTA tariff cuts may have induced changes in the employment shares of the primary, construction, finance, and other services industries. If some industries, in particular the primary sector, more heavily feature unincorporated self-employment regardless of CUSFTA, sectoral composition shifts could potentially explain all of the impact of the tariff cuts on self-employment found in Table 3. I therefore include, as additional controls, changes in the employment share of the primary, construction, finance, and other services sectors.²⁸ The results in Panel C show that the inclusion of the sectoral controls barely affects the magnitudes and has no effect on the sign or significance of the self-employment coefficients. I therefore conclude that the tariff cuts had an impact on self-employment above and beyond that induced by sectoral compositional shifts.²⁹

One concern could be that the tariff cuts induced migration, but particularly migration that changed the population’s composition, thereby potentially also affecting the population’s propensity to enter or exit self-employment. This has been found to be important in work by Greenland, Lopresi, and McHenry (2019), within the context of the US experience with Chinese imports. I therefore verify that no compositional changes in the population occurred as a result of CUSFTA. In Table A.5, I regress the tariff cut measures against each CD’s change in female employment share, immigrant population share, and university-educated population share. I find that CUSFTA had no statistically significant effect on any of these measures, implying that the population in CDs hit by tariff cuts was not noticeably changed in composition.

As another robustness exercise, in Table A.6 of the Appendix I leave out provinces one at a time to verify that the results in column 6 of Table 3 still hold and are not driven by trends in CDs in one particular province. For example, if the deepest tariff cuts occurred in Ontario and Ontario also experienced trends in self-employment for reasons not related to CUSFTA, this could explain the regression results in Table 3. Although the maps displaying tariff cut intensities show that tariff cuts at the CD level were reasonably dispersed throughout Canada and indeed within provinces as well, I nonetheless perform this check. Table A.6 shows that this is not the case and that, even leaving out provinces one at a time, Canadian tariff cuts increase self-employment rates and American tariff cuts decrease self-employment rates.

In Table A.7 in the Appendix, I re-estimate the main employment results using Tobit. Because the employment variables are all shares, their changes are constrained to have values between –1 and 1. I therefore re-estimate the Table 3 specifications using Tobit, where I set the upper limit to be 1 and the lower limit to be -1. I find that the results are qualitatively the same in terms of sign and crucially that the self-employment coefficients in column 4 remain highly statistically significant.

Table A.8 of the Appendix revisits the main employment results but using a logged version of the tariff measures. Specifically, I take the log of main tariff measures plus one, and then take the difference. This is very similar to the main measure of regional tariff changes as used in Kovak (2013). The results exhibit the same patterns as the main results from Table 3 in terms of sign of coefficients and statistical significance. I again find that cuts to Canadian tariffs increase self-employment rates, whereas American tariff cuts do the opposite.

Instead of separately examining the impacts of the Canadian and US tariff cuts separately, in Table A.9 I introduce a measure that examines the Canadian and US tariff cuts together, which is the Canadian tariff cut minus the American tariff cut, or $Δ τ_{c}^{C A N}$ . The results using this measure are reported in Table A.9. These results confirm the main findings in Table 3, which argue that Canadian tariff cuts increase self-employment, and US tariff cuts decrease self-employment.

Discussion

Section:

The employment results in Table 3 provide evidence that tariff reductions affect the proportion of workers in self-employment. This echoes other work in this area, such as that by Roed and Skogstrom (2014), who found that workers in Norway respond to negative employment shocks by entering entrepreneurship. Similarly, Kuhn and Schuetze (2001) find that Canadian men were more likely to be induced into self-employment when faced with more instability in paid employment. My work echoes these mechanisms and previous article by also demonstrating that improved economic conditions, in the form of lower American tariffs, induce workers to exit self-employment and enter paid employment. Conversely, worsened conditions in the form of higher import competition resulting from lower Canadian tariffs push workers to enter self-employment, much like in the Norwegian case.

It is also worthwhile to note that, even though the results from the previous section imply that CUSFTA led to a slight increase in local labour markets’ self-employment rates, the estimates actually imply that Canada should see an overall increase in paid employment and a decrease in self-employment from an in-kind mutual tariff reduction with the United States. In every specification and outcome examined in this article, the American tariff cut coefficients were always larger in absolute magnitude than those for the Canadian tariff cuts; this suggests that a 1-percentage-point reduction in Canadian tariffs coupled with a 1-percentage-point reduction in American tariffs faced by a CD would actually reduce the prevalence of self-employment and increase that of paid employment.

If this is the case, why was the average predicted self-employment rate change from CUSFTA slightly positive? This largely stems from the fact that, on average, Canadian tariff cuts were deeper than American tariff decreases. Magun et al. (1988), in an Economic Council of Canada study, calculates that Canadian tariffs would be cut by 3.2 percentage points, whereas American tariffs would decline by 2 percentage points; Canadian tariffs, therefore, declined by 1.2 more percentage points than their American equivalents. This discrepancy was principally the result of higher pre–free trade agreement tariff levels in Canada; Beaulieu (2000) calculates, for example, that Canadian manufacturing tariffs against the United States were 6.7 percent, whereas American manufacturing tariffs against Canada were 3.9 percent. CUSFTA’s mandated tariff reductions thus, mechanically, favour increasing American import competition more than increased export opportunities for Canadians.

I now turn to a brief discussion of the strengths and weaknesses of the main empirical specification. My specification is very similar to those commonly used in the trade and local labour markets literature. For example, Autor et al. (2013) also use a first-difference specification to estimate the impact of Chinese import competition on a variety of employment outcomes. The benefits to this approach are that one is better able to pick up changes that might require longer term adjustment, such as, potentially, self-employment.

One potential weakness is that it is possible that there is some selection in that relatively good workers induced into self-employment as a result of Canadian tariff cuts (and the reverse for American tariff cuts) are more likely to survive for longer. However, if the main mechanism described is valid, that relatively low-productivity workers are selecting into self-employment because of the tariff cuts, then the survival bias would likely remove the workers with the lowest productivity (in self-employment) from self-employment. This implies that such selection issues must lead to an underestimate of the main estimates of the impacts of CUSFTA on self-employment, as a result of long-differencing not taking into account these employment dynamics. In addition, if these lowest productivity workers do not survive in self-employment and leave, they must be exiting the labour force, be unemployed, or be re-entering paid employment. Because I also do not observe any robust effects on unemployment or labour force participation, the estimates in this article suggest that this selection issue cannot be driving my results, and at most understates them.

Conclusion

Section:

In this article, I investigate the costs and benefits of the CUSFTA using a local labour market approach. I find that the composition of employment was affected by the tariff cuts, in terms of paid versus self-employment as well as sectoral employment composition.

The usefulness of using regional tariff exposure, as opposed to its industry-level counterpart, is highlighted in the relative lack of impact on traditional labour market outcomes, such as wages and employment and the importance of CUSFTA in driving more indirect, general-equilibrium regional outcomes such as the prevalence of self-employment, an outcome that would not easily map onto industries alone. This emphasis mirrors the approach taken by Autor et al. (2013), who use the same features of regional trade exposure, in their case Chinese import competition, to analyze non-traditional regional outcomes such as take-up of social benefits. This article therefore follows in their footsteps but also contributes to the literature by applying this approach to a trade liberalization episode between two developed countries, which is a type of trade shock and setting that remains relatively understudied using the local labour market approach.

My work also helps add to the existing body of knowledge relating to self-employment, finding that trade openness can also play an important role, inducing workers to enter or leave this non-standard form of employment. In particular, this study does this in a Canadian, and more broadly a developed country, setting. My findings also corroborate a literature that concludes that individuals are more likely to become self-employed or entrepreneurs during downturns in the business cycle. This finding may be particularly important in light of recent evidence by Galindo da Fonseca (2018) that suggests that entrepreneurs who enter entrepreneurship as a result of involuntary unemployment have worse outcomes, such as lower employment and higher exit rates. Understanding whether trade liberalization, particularly between developed countries, affects entry into self-employment could also help shed light on how trade affects aggregate productivity, and it can also inform policy-makers seeking to understand the ramifications of policies designed to promote self-employment or entrepreneurship.³⁰

Finally, as Canada looks to be poised to ratify or engage in additional free trade agreements, some of which also involve other developed countries, these results will continue to be of use to relevant stakeholders seeking to understand the implications of these agreements for the composition of employment.

Acknowledgements

Data analysts at Statistics Canada provided expert assistance in obtaining custom data tabulations. I am also very appreciative of the generous assistance from the Wilfrid Laurier University library staff, particularly Michael Steeleworthy. I thank Joseph Marchand, Brian McCaig, Tammy Schirle, Justin Smith, and participants at the 2018 Canadian Economics Association Conference for helpful comments. I gratefully acknowledge funding from the Laurier Centre for Economic Research and Policy Analysis. All errors are my own.

Notes

1 In this article, I use self-employment rate to refer to the proportion of workers who are self-employed but also unincorporated. This is consistent with the definition of self-employment used in related Canadian work by Bahar and Liu (2015).

2 In related work, Schuetze (2015) analyzes Canadian data, documents longer labour market continuation among self-employed men, and delves into some explanations.

3 Other studies that have investigated how industry-level exposure from CUSFTA has affected various outcomes include Beaulieu (2000), Head and Ries (2001), and Townsend (2007).

4 Specifically, equilibrium wages increase in the relative value of the self-employment good and technology in the self-employment sector, because this raises the opportunity cost of wage employment, and wages therefore have to rise to combat the resulting drop in labour supplied to wage employment.

5 Blau (1987) also allows for the possibility that workers choose to allocate all of their labour to wage employment or self-employment.

6 This is an interpretation of trade shocks’ theoretical effects that is in line with that taken by Davidson, Matusz, and Shevchenko (2008), who examine the role of trade shocks in determining how firms and workers match in an environment that has search frictions and worker and firm types.

7 Data from World Bank (2019) shows Canada’s international trade as a share of GDP was 53.1 percent in 1986, increasing to 70.3 percent in 1996.

8 Together, this consists of 105 manufacturing industries at the three-digit level.

9 This is largely for practicality, because the Magun et al. (1988) data for non-manufacturing tariffs begin in 1987.

10 For additional information on the aggregations conducted, see Table A.1 in the Appendix for details on which three-digit SIC codes map onto the aggregated industries I use for agriculture and resources in this article.

11 The 1986 and 1996 Census profiles both group incorporated self-employed workers as paid workers. Focusing on unincorporated self-employment is also consistent with the definition of self-employment used in the United States and a recent Canadian study by Bahar and Liu (2015).

12 In this article, these industries include all SIC three-digit industries in agriculture, resources, and manufacturing.

13 Throughout this article, Canadian tariffs refer to Canada’s tariffs against American goods. Similarly, US or American tariffs refer to US tariffs against Canadian goods.

14 I should note that averages displayed here may not reflect national, aggregate trends from other sources for several reasons. First, publicly released Census profile data (including custom tabulations) are rounded to the nearest 5 because of confidentiality requirements in Canada. This implies that there is some intentional measurement error introduced by Statistics Canada. In addition, I present unweighted summary statistics, and CDs can vary by population, as can be seen from the standard deviations presented in Table 1 for population. Finally, self-employment, income, and other variables from the Census are self-reported and may therefore diverge from figures calculated from tax records and administrative sources. Finally, I also exclude the territories in my analysis and summary statistics.

15 This may introduce some multicollinearity between the two tariff measures. I would first note that there are CDs for which the Canadian and American tariff cuts differ substantially. For example, Drummond, Quebec, has a decrease in Canadian tariffs of 3.4 percentage points and a decrease in American tariffs of 7.8 percentage points. Also, if this is the case, this will reduce the precision of my estimates. As I later show, however, because the main results are statistically significant at conventional levels already, the potential presence of this issue will serve to strengthen my results. Finally, as Table A.3 in the Appendix shows, including only one tariff measure (US or Canadian) at a time qualitatively changes the findings, in terms of sign; this is likely due to the correlation between the two tariff measures. This highlights the importance of including both tariff variables in my specifications.

16 See Beaulieu (2000) for additional details.

17 The tariff measures here, and in the main tariff measures used in most of the results in this article, are percentage point changes, not percentage changes. The 17.2 percent decline in Canadian tariffs for knitting mills is therefore a 17.2 percentage point decline, not a 17.2 percent decline from initial tariff levels.

18 These proportions are similar to those found in 1996, with 15.4 percent working in agriculture and primary sectors, 5.55 percent in manufacturing, and the remainder working in service.

19 This assumes that worker ability or productivity is at least positively correlated across employment types.

20 Conversely, workers could find it less beneficial to enter self-employment if American tariff cuts lead to paid jobs being relatively easier to find.

21 This stands in stark contrast to studies looking at developing countries as the origin of trade shocks, such as Autor et al. (2013).

22 The specifications used in this article are all 10-year long differences, implying that business cycles should be differenced out. Nonetheless, the CUSFTA-induced positive and negative trade shocks brought on by US and Canadian tariff cuts, to the extent that they resemble booms and busts in business cycles, cause changes to self-employment that are also consistent with the business-cycle-based literature. It is also therefore reasonable to conjecture that the self-selection mechanisms operating in those settings are also at play in this article.

23 Chan (2018), for example, finds that women (in particular, low-wage-earning women) are more likely than men to exit an industry affected by import competition.

24 Montmagny and Montmorency No. 1 CDs are the third and fifth most positively affected CDs, respectively.

25 In addition to the aforementioned Saint John County, the second most affected CD is Division No. 6 in Newfoundland and Labrador.

26 From Table 1, the standard deviation of CD-level 1986 population is 213,689.

27 Brant County is in third place with a predicted increase of 0.006, and Peel Regional Municipality is in fourth with a predicted increase of 0.006.

28 In estimates not presented in this article, I find that these sectors’ employment shares are affected by tariff cuts in a statistically significant manner.

29 I should note here that in Panels A and C, some of the coefficients for labour force participation (in Panel A only) and the full-time employment share (in Panels A and C) register significant coefficients for the Canadian tariff variable. These are typically less significant than the self-employment variables and are not consistently so across specifications, as in the main results in Table 4. I therefore do not particularly discuss or emphasize them out of caution.

30 Galindo da Fonseca (2018) uses a model to show that a policy that encourages entry into entrepreneurship from unemployment would also reduce productivity because of less productive entrepreneurs.

References

Section:

	Autor, D., D. Dorn, and G. Hanson. (2013). “The China Syndrome: Local Labor Market Effects of Import Competition in the United States.” American Economic Review 103(6):2121–68. https://doi.org/10.1257/aer.103.6.2121. Google Scholar
	Bahar, S., and H. Liu. 2015. “Unincorporated Self-Employment in Canada, 1989–2010.” Canadian Economy in Transition Series, Statistics Canada Catalogue No. 11-622-M-No. 31. Ottawa: Statistics Canada. Google Scholar
	Baldwin, J., W. Brown, and W. Gu. (2012). “Geographic Market Access and the Effects of Trade on Length of Production Run, Product Diversity and Plant Scale of Canadian Manufacturing Plants, 1974–1999.” Journal of Economic Geography 12(2):455–84. https://doi.org/10.1093/jeg/lbq059. Google Scholar
	Beaulieu, E. (2000). “The Canada-U.S. Free Trade Agreement and Labour Market Adjustment in Canada.” Canadian Journal of Economics/Revue canadienne d’économique 33(2):540–63. https://doi.org/10.1111/0008-4085.00028. Google Scholar
	Blau, D. (1987). “A Time-Series Analysis of Self-Employment in the United States.” Journal of Political Economy 95(3):445–67. https://doi.org/10.1086/261466. Google Scholar
	Borjas, G. (1986). “The Self-Employment Experience of Immigrants.” Journal of Human Resources 21(4):485–506. https://doi.org/10.2307/145764. Google Scholar
	Brander, J. (1991). “Election Polls, Free Trade, and the Stock Market: Evidence From the 1988 Canadian General Election.” Canadian Journal of Economics/Revue canadienne d’économique 24(4):827–43. https://doi.org/10.2307/135695. Google Scholar
	Breinlich, H. (2008). “Trade Liberalization and Industrial Restructuring through Mergers and Acquisitions.” Journal of International Economics 76(2):254–66. https://doi.org/10.1016/j.jinteco.2008.07.007. Google Scholar
	Cadena, B., and B. Kovak. (2016). “Immigrants Equilibrate Local Labor Markets: Evidence From the Great Recession.” American Economic Journal: Applied Economics 8(1):257–90. https://doi.org/10.3386/w19272. Google Scholar
	Chan, J. (2018). “Does Import Competition Worsen the Gender Gap? Evidence from Matched Employer-Employee Data.” Economics Letters 165:13–16. https://doi.org/10.1016/j.econlet.2018.01.024. Google Scholar
	Davidson, C., S. Matusz, and A. Shevchenko. (2008). “Globalization and Firm Level Adjustment with Imperfect Labor Markets.” Journal of International Economics 75(2):295–309. https://doi.org/10.1016/j.jinteco.2008.02.004. Google Scholar
	Feler, L., and M. Senses. (2017). “Trade Shocks and the Provision of Local Public Goods.” American Economic Journal: Economic Policy 9(4):101–43. https://doi.org/10.1257/pol.20150578. Google Scholar
	Galindo da Fonseca, J. 2018. “Unemployment, Entrepreneurship and Firm Outcomes.” Cahiers de recherche 2019–04, Département de sciences économiques, Université de Montréal. Google Scholar
	Gaston, N., and D. Trefler. (1997). “The Labour Market Consequences of the Canada-U.S. Free Trade Agreement.” Canadian Journal of Economics/Revue canadienne d’économique 30(1):18–41. https://doi.org/10.2307/136358. Google Scholar
	Greenland, A., J. Lopresi, and P. McHenry. (2019). “Import Competition and Internal Migration.” Review of Economics and Statistics 101(1):44–59. https://doi.org/10.1162/rest_a_00751. Google Scholar
	Hakobyan, S., and J. McLaren. (2016). “Looking for Local Labor Market Effects of NAFTA.” Review of Economics and Statistics 98(4):728–41. https://doi.org/10.1162/rest_a_00587. Google Scholar
	Head, K., and J. Ries. (2001). “Increasing Returns Versus National Product Differentiation as an Explanation for the Pattern of U.S.-Canada Trade.” American Economic Review 91(4):858–76. https://doi.org/10.1257/aer.91.4.858. Google Scholar
	Humphries, J. 2018. “The Causes and Consequences of Self-Employment over the Life Cycle.” PhD dissertation, University of Chicago. Google Scholar
	Kovak, B. 2013. “Regional Effects of Trade Reform: What Is the Correct Measure of Liberalization?” American Economic Review 103(5):1960–76. https://doi.org/10.1257/aer.103.5.1960. Google Scholar
	Kuhn, P., and H. Schuetze. (2001). “Self-Employment Dynamics and Self-Employment Trends: A Study of Canadian Men and Women, 1982–1998.” Canadian Journal of Economics/Revue canadienne d’économique 34(3):760–84. https://doi.org/10.1111/0008-4085.00098. Google Scholar
	Lileeva, A., and D. Trefler. (2010). “Improved Access to Foreign Markets Raises Plant-Level Productivity … for Some Plants.” Quarterly Journal of Economics 125(3):1051–99. https://doi.org/10.1162/qjec.2010.125.3.1051. Google Scholar
	Lin, Z., G. Picot, and J. Compton. (2000). “The Entry and Exit Dynamics of Self-Employment in Canada.” Small Business Economics 15(2):105–125. https://doi.org/10.1023/A:1008150516764. Google Scholar
	Magun, S., S. Rao, B. Lodh, L. Lavallee, and J. Pierce. 1988. “Open Borders: An Assessment of the Canada-U.S. Free Trade Agreement.” Discussion Paper No. 344, Economic Council of Canada, Ottawa. Google Scholar
	Marchand, J. (2012). “Local Labor Market Impacts of Energy Boom-Bust-Boom in Western Canada.” Journal of Urban Economics 71(1):165–74. https://doi.org/10.1016/j.jue.2011.06.001. Google Scholar
	Mccaig, B. (2011). “Exporting Out of Poverty: Provincial Poverty in Vietnam and U.S. Market Access.” Journal of International Economics 85(1):102–13. https://doi.org/10.1016/j.jinteco.2011.05.007. Google Scholar
	Roed, K., and J. Skogstrom. (2014). “Job Loss and Entrepreneurship.” Oxford Bulletin of Economics and Statistics 76(5):727–44. https://doi.org/10.1111/obes.12042. Google Scholar
	Romalis, J. (2007). “NAFTA’s and CUSFTA’s Impact on International Trade.” Review of Economics and Statistics 89(3):416–35. https://doi.org/10.1162/rest.89.3.416. Google Scholar
	Schuetze, H. (2015). “Self-Employment and Retirement in Canada: The Labour Force Dynamics of Older Workers.” Canadian Public Policy/Analyse de politiques 41(1):65–85. https://doi.org/10.3138/cpp.2013-070. Link, Google Scholar
	Topalova, P. 2007. “Trade Liberalization, Poverty and Inequality: Evidence from Indian Districts.” In Globalization and Poverty, ed. A. Harrison, 291–336. Chicago: University of Chicago Press. Google Scholar
	Townsend, J. (2007). “Do Tariff Reductions Affect the Wages of Workers in Protected Industries? Evidence from the Canada-U.S. Free Trade Agreement.” Canadian Journal of Economics/Revue canadienne d’économique 40(1):69–92. https://doi.org/10.1111/j.1365-2966.2007.00400.x. Google Scholar
	Trefler, D. (2004). “The Long and Short of the Canada-U.S. Free Trade Agreement.” American Economic Review 94(4):870–95. https://doi.org/10.1257/0002828042002633. Google Scholar
	World Bank. 2019. “Trade (% of GDP).” World Bank national accounts data. At https://data.worldbank.org/indicator/NE.TRD.GNFS.ZS?locations=CA. Google Scholar

Appendix

Additional Figure

Figure A.1: Local Tariff Changes, 1987–1996, and Local Tariff Levels, 1987

Notes: Circles represent population sizes of each CD, with larger circles for larger CDs. CD = Census Division.

Source: Author’s calculations.

Additional Tables

Table A.1: Aggregated Agriculture and Resource Industries

Aggregated Industry and 3-Digit SIC Industry	3-Digit SIC Code
Agriculture
Livestock farms	011
Other animal specialty farms	012
Field crop farms	013
Field crop combination farms	014
Fruit and other vegetable farms	015
Horticultural specialties	016
Livestock, field crop, and horticultural combination farms	017
Services incidental to agriculture
Services incidental to livestock and animal specialties	021
Services incidental to agricultural crops	022
Other services incidental to agriculture	023
Fishing and trapping
Fishing industries	031
Services incidental to fishing	032
Trapping	033
Logging and forestry
Logging	041
Forestry services	051
Mining
Metal mines	061
Non-metal mines	062
Coal mines	063
Oil and gas
Crude petroleum and natural gas industries	071
Quarry and sand pits
Stone quarries	081
Sand and gravel pits	082
Mining services
Service industries incidental to crude petroleum and natural gas	091
Service industries incidental to mining	092

Note: SIC = Standard Industrial Classification.

Table A.2: Summary Statistics, Weighted by Population

Variable	M (SD)	Range, Min–Max.
Panel A: 1986 levels
Population	565,743.4 (693,160.1)	2,022–2,192,721
Proportion of immigrants in population	0.155 (0.115)	0.003–0.402
Female employment share	0.423 (0.029)	0.319–0.462
Labor force participation rate	0.664 (0.053)	0.47–0.786
Employment-to-population ratio	0.596 (0.072)	0.319–0.737
Proportion full-time employment	0.507 (0.062)	0.188–0.607
Proportion part-time employment	0.493 (0.063)	0.393–0.812
Proportion paid employment	0.933 (0.042)	0.613–0.978
Proportion self-employment	0.067 (0.042)	0.022–0.386
Proportion of working population with university degree	0.095 (0.040)	0.025–0.202
Proportion employed in manufacturing	0.188 (0.085)	0.008–0.443
Proportion employed in primary industry	0.079 (0.079)	0.004–0.528
Panel B: 1996 levels
Population	631,889.9 (748,060.3)	1,265–2,363,875
Proportion of immigrants in population	0.169 (0.142)	0.003–0.475
Female employment share	0.459 (0.016)	0.396–0.496
Labor force participation rate	0.652 (0.051)	0.48–0.81
Employment-to-population ratio	0.585 (0.066)	0.305–0.755
Proportion full-time employment	0.512 (0.055)	0.226–0.588
Proportion self-employment	0.085 (0.034)	0.026–0.333
Proportion of working population with university degree	0.130 (0.055)	0.033–0.258
Proportion employed in manufacturing	0.155 (0.066)	0.011–0.438
Proportion employed in primary industry	0.065 (0.078)	0.004–0.532
Panel C: 1987–1996 tariff changes
$Δ τ_{c}^{U S}$	−0.021 (0.010)	−0.052 to −0.003
$Δ τ_{c}^{C A N}$	−0.045 (0.020)	−0.085 to −0.003

Notes: All summary statistics are calculated using 1986 population of each Census Division as weights, as in the regressions.

Source: Author’s calculations.

Table A.3: Employment Effects, with Tariffs Separately

Dependent Variable	(1) Δ Employment-to-Population Ratio	(2) Δ Employment-to-Population Ratio	(3) Δ Labour Force Participation Rate	(4) Δ Labour Force Participation Rate	(5) Δ Proportion Employed Full Time	(6) Δ Proportion Employed Full Time	(7) Δ Proportion Self-Employed	(8) Δ Proportion Self-Employed
$Δ τ_{c}^{C A N}$	0.00433		−0.254*		0.144*		−0.0264
$Δ τ_{c}^{C A N}$	(0.280)		(0.131)		(0.0843)		(0.0693)
$Δ τ_{c}^{U S}$		−0.0179		−0.425**		0.143		0.174
$Δ τ_{c}^{U S}$		(0.254)		(0.207)		(0.156)		(0.111)
CD controls	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	260	260	260	260	260	260	260	260
R²	0.834	0.834	0.721	0.721	0.626	0.623	0.609	0.613

Notes: Robust standard errors are reported in parentheses. CD controls include the 1986 female employment share, the 1986 immigrant population share, the 1986 manufacturing employment share, and the proportion of the working-age population with university degrees in 1986. $Δ τ_{c}^{C A N}$ is the CD-level weighted change in Canadian tariffs against the United States; $Δ τ_{c}^{U S}$ is the CD-level weighted change in US tariffs against Canada. CD = Census Division; FE = fixed effect.

*p < 0.10;

**p < 0.05.

Source: Author’s calculations.

Table A.4: Employment Effects: Robustness Checks

Dependent Variable	(1) Δ Employment-to-Population Ratio	(2) Δ Labour Force Participation Rate	(3) Δ Proportion Employed Full Time	(4) Δ Proportion Self-Emp.
Panel A: with no population weights
$Δ τ_{c}^{C A N}$	−0.522	−0.514**	0.422**	−0.697***
$Δ τ_{c}^{C A N}$	(0.326)	(0.225)	(0.205)	(0.167)
$Δ τ_{c}^{U S}$	0.214	0.0166	−0.603	1.495***
$Δ τ_{c}^{U S}$	(0.524)	(0.353)	(0.376)	(0.290)
CD controls	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes
Observations	260	260	260	260
R²	0.551	0.450	0.375	0.575
Panel B: with distance from border control
$Δ τ_{c}^{C A N}$	−0.0362	−0.248	0.257	−0.404***
$Δ τ_{c}^{C A N}$	(0.267)	(0.223)	(0.178)	(0.128)
$Δ τ_{c}^{U S}$	0.248	0.138	−0.178	0.692***
$Δ τ_{c}^{U S}$	(0.422)	(0.382)	(0.319)	(0.234)
CD controls	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes
Observations	260	260	260	260
R²	0.841	0.734	0.629	0.642
Panel C: with sectoral composition controls
$Δ τ_{c}^{C A N}$	0.210	−0.0498	0.304*	−0.391***
$Δ τ_{c}^{C A N}$	(0.248)	(0.194)	(0.180)	(0.126)
$Δ τ_{c}^{U S}$	−0.372	−0.385	−0.369	0.819***
$Δ τ_{c}^{U S}$	(0.390)	(0.328)	(0.309)	(0.218)
CD controls	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes
Observations	260	260	260	260
R²	0.878	0.746	0.657	0.685

*p < 0.10;

**p < 0.05;

***p < 0.01.

Source: Author’s calculations.

Table A.5: Population Composition Effects

Dependent Variable	(1) Δ Female Employment Share	(2) Δ Immigrant Population Share	(3) Δ Share of Population with Bachelor’s Degree
$Δ τ_{c}^{C A N}$	−0.0686	−0.430	−0.0758
$Δ τ_{c}^{C A N}$	(0.104)	(0.276)	(0.0949)
$Δ τ_{c}^{U S}$	0.107	0.364	0.0913
$Δ τ_{c}^{U S}$	(0.195)	(0.490)	(0.186)
CD controls	Yes	Yes	Yes
Province FE	Yes	Yes	Yes
Observations	260	260	260
R²	0.831	0.801	0.882

Source: Author’s calculations.

Table A.6: Self-Employment Effects, Leaving Out Provinces One at a Time

	Dependent Variable
Province Left Out	(1) NL	(2) PE	(3) NS	(4) NB	(5) QC	(6) ON	(7) MB	(8) SK	(9) AB	(10) BC
$Δ τ_{c}^{C A N}$	−0.393***	−0.441***	−0.455***	−0.435***	−0.676***	−0.353*	−0.360***	−0.429***	−0.412***	−0.537***
$Δ τ_{c}^{C A N}$	(0.134)	(0.134)	(0.144)	(0.151)	(0.153)	(0.179)	(0.129)	(0.131)	(0.130)	(0.129)
$Δ τ_{c}^{U S}$	0.766***	0.834***	0.888***	0.839***	1.140***	1.089***	0.678***	0.768***	0.698***	0.941***
$Δ τ_{c}^{U S}$	(0.229)	(0.233)	(0.251)	(0.257)	(0.253)	(0.323)	(0.231)	(0.233)	(0.228)	(0.224)
CD controls	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	250	257	242	245	184	211	237	242	241	231
R²	0.834	0.834	0.721	0.721	0.626	0.623	0.609	0.613	0.638	0.655

Notes: The dependent variable is change in proportion self-employed Robust standard errors are reported in parentheses. CD controls include the 1986 female employment share, the 1986 immigrant population share, the 1986 manufacturing employment share, and the proportion of the working-age population with university degrees in 1986. $Δ τ_{c}^{C A N}$ is the CD-level weighted change in Canadian tariffs against the United States; $Δ τ_{c}^{U S}$ is the CD-level weighted change in US tariffs against Canada. CD = Census Division; FE = fixed effect.

*p < 0.10;

**p < 0.05;

***p < 0.01.

Source: Author’s calculations.

Table A.7: Employment Results, Tobit

Dependent Variable	(1) Δ Employment-to-Population Ratio	(2) Δ Labour Force Participation Rate	(3) Δ Proportion Employed Full Time	(4) Δ Proportion Self-Employed
$Δ τ_{c}^{C A N}$	0.0515	−0.168	0.283*	−0.439***
$Δ τ_{c}^{C A N}$	(0.271)	(0.213)	(0.171)	(0.130)
$Δ τ_{c}^{U S}$	−0.0950	−0.174	−0.280	0.831***
$Δ τ_{c}^{U S}$	(0.371)	(0.332)	(0.308)	(0.225)
CD controls	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes
Observations	260	260	260	260
Log-pseudolikelihood	614.19	639.48	674.92	760.62

Notes: Robust standard errors are reported in parentheses. All results estimated using Tobit, with upper limit set to 1 and lower limit set to –1. CD controls include the 1986 female employment share, the 1986 immigrant population share, the 1986 manufacturing employment share, and the proportion of the working-age population with university degrees in 1986. $Δ τ_{c}^{C A N}$ is the CD-level weighted change in Canadian tariffs against the United States; $Δ τ_{c}^{U S}$ is the CD-level weighted change in US tariffs against Canada. CD = Census Division; FE = fixed effect.

*p < 0.10;

***p < 0.01.

Source: Author’s calculations.

Table A.8: Employment Effects by Logged Tariffs

Dependent Variable	(1) Δ Employment-to-Population Ratio	(2) Δ Labour Force Participation Rate	(3) Δ Proportion Employed Full Time	(4) Δ Proportion Self-Employed
$Δ l n (τ_{c}^{C A N} + 1)$	0.0439	−0.174	0.295	−0.455***
$Δ l n (τ_{c}^{C A N} + 1)$	(0.296)	(0.232)	(0.184)	(0.143)
$Δ l n (τ_{c}^{U S} + 1)$	−0.0789	−0.186	−0.280	0.846***
$Δ l n (τ_{c}^{U S} + 1)$	(0.393)	(0.354)	(0.326)	(0.241)
CD controls	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes
Observations	260	260	260	260
R²	0.834	0.722	0.627	0.634

Notes: Robust standard errors are reported in parentheses. CD controls include the 1986 female employment share, the 1986 immigrant population share, the 1986 manufacturing employment share, and the proportion of the working-age population with university degrees in 1986. $Δ l n (τ_{c}^{C A N} + 1)$ is the CD-level weighted change in logged Canadian tariffs against the United States plus one; $Δ l n (τ_{c}^{U S} + 1)$ is the CD-level weighted change in logged US tariffs against Canada plus one. CD = Census Division; FE = fixed effect.

***p < 0.01.

Source: Author’s calculations.

Table A.9: Employment Effects, Combined US–Canada Tariff Measure

Dependent Variable	(1) Δ Employment-to-Population Ratio	(2) Δ Labour Force Participation Rate	(3) Δ Proportion Employed Full Time	(4) Δ Proportion Self-Employed
$Δ τ_{c}^{C A N} - Δ τ_{c}^{U S}$	0.0303	−0.334	0.284*	−0.249**
$Δ τ_{c}^{C A N} - Δ τ_{c}^{U S}$	(0.287)	(0.213)	(0.145)	(0.115)
CD controls	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes
Observations	260	260	260	260
R²	0.834	0.718	0.627	0.618

Notes: Robust standard errors are reported in parentheses. CD controls include the 1986 female employment share, the 1986 immigrant population share, the 1986 manufacturing employment share, and the proportion of the working-age population with university degrees in 1986. $Δ τ_{c}^{C A N}$ is the CD-level weighted change in Canadian tariffs against the United States; $Δ τ_{c}^{U S A}$ is the CD-level weighted change in US tariffs against Canada. CD = Census Division; FE = fixed effect.

*p < 0.10;

**p < 0.05.

Source: Author’s calculations.

Journal Menu ☰

Join the Conversation

Articles

Tariffs and the Composition of Employment: Evidence from the Canada–US Free Trade Agreement

Abstract

Acknowledgements

CEA Member Access

Article History

Article Tools

Top Articles

Journal Menu ☰

Join the Conversation

Articles

Tariffs and the Composition of Employment: Evidence from the Canada–US Free Trade Agreement

Abstract

Acknowledgements

CEA Member Access

Article History

Article Tools

Related Content Search

Find related content

Top Articles

Cookies & Privacy