Impact of the First Wave of the COVID-19 Pandemic on Trade between Canada and the United States

Abstract

We examine how the coronavirus disease 2019 (COVID-19) pandemic has affected trade between Canada and the United States, using a novel dataset on monthly bilateral trade flows between Canadian provinces and US states merged with COVID-19 health data. Our results show that a one-standard-deviation increase in COVID-19 severity (case levels, hospitalizations, deaths) in a Canadian province leads to a 3.1 percent to 4.9 percent fall in exports and a 6.7 percent to 9.1 percent fall in imports. Decomposing our analysis by industry, we determine that trade in the manufacturing industry was most negatively affected by the pandemic, and the agriculture industry had the least disruption to trade flows. Our descriptive evidence suggests that lockdowns may also have reduced Canadian exports and imports. However, although our regression coefficients are consistent with that finding, they are not statistically significant, perhaps because of the lack of variation as a result of similar timing in the imposition of restrictions across provinces.

Introduction

The coronavirus disease 2019 (COVID-19) global pandemic has resulted in unprecedented disruptions to the Canadian economy at the local, regional, and national levels. In response, policy-makers have urgently explored options to minimize its economic and public health repercussions and support Canadian consumers and businesses throughout the crisis.¹ The economic uncertainty created by the crisis can be particularly harmful for Canadian businesses that engage in international trade and depend on the global value chain as part of their business model.² Understanding the ramifications of the pandemic and the associated policy responses is crucial in evaluating the long-term prospects for Canadian trade and the Canadian economy in general.

Although we are still in the early stages of a potential multi-year economic slowdown, examining the early impacts of the first wave of the COVID-19 pandemic can help policy-makers understand how initial policy responses such as lockdowns have affected various sectors of the Canadian economy and direct resources more effectively to aid the recovery process. Although most government policies relating to the COVID-19 pandemic are directed at protecting public health, a robust evaluation of these policies also considers the secondary effects on various other sectors of the Canadian economy, such as international trade. If these policies lead to significant reductions in trade, quantifying these effects can provide valuable information to weigh these costs against the public health benefits of pandemic restrictions. Conversely, if trade is not significantly affected by these restrictions, this result can provide equally valuable insight to policy-makers deliberating the appropriate levels of policy response. With this motivation in mind, we use new data on monthly bilateral trade flows between Canadian provinces and their corresponding trading partners at the US state level to analyze the impact of the COVID-19 pandemic’s first wave on Canada’s international trade patterns.

This article encompasses four main research objectives contributing to the current sparse but growing literature on the relationship between the COVID-19 pandemic and trade. First, we assemble a novel dataset on monthly bilateral trade flows from the Government of Canada’s Trade Data Online (TDO) database and combine it with data on COVID-19 statistics from Canadian provincial and US state health reporting agencies to document the changes in trade patterns between January 2019 and September 2020 amid the emergence of the COVID-19 pandemic. Second, our analysis provides insight into the differences in origin-specific and destination-specific effects. For example, if trade is highly affected by conditions in the destination location, we would expect provincial exports to US states hit hardest by COVID-19 to be most adversely affected, irrespective of origin province. Conversely, if trade is highly affected by conditions in the origin location, exports from provinces such as Ontario and Quebec, where COVID-19 metrics are consistently highest among provinces, might be expected to suffer more than those coming from the Maritime provinces, where the severity of the pandemic is consistently lowest, irrespective of destination state. Third, we decompose our analysis of the pandemic’s effect on trade into two distinct channels: exposure to infection (such as the number of cases, deaths, or hospitalizations attributed to COVID-19) versus the severity of policy responses (such as restrictions on work, travel, or gatherings). This allows us to examine whether variation in COVID-19 policy measures instituted across provinces and states has resulted in differences in outcomes in trade flows. Finally, we segment our analysis using the more disaggregated two-digit North American Industry Classification System (NAICS) goods classification system to determine whether different types of industry (i.e., agriculture vs. manufacturing) experience differential effects of the pandemic’s severity on the resultant changes in trade in each industry.

From our descriptive trends analysis, we find large and persistent falls in bilateral trade between Canada and the United States, at both the national and the provincial or state levels, after the classification of COVID-19 as a global pandemic in March 2020. At the national level, year-over-year decreases in both exports and imports were close to 50 percent in the months after the initial outbreak of the pandemic. The manufacturing and mining, oil, and gas sectors were the hardest hit; decreases in trade flows in these sectors surpassed 60 percent. At the provincial level, we find a similar trend, with all provinces experiencing a minimum 25 percent decrease in exports, and some provinces experiencing as much as a 75 percent decrease in exports in the wake of the pandemic. We find similar magnitudes for the changes in provincial import flows.

We also find strong evidence that the decline of exports and imports is correlated with the severity of the COVID-19 pandemic. We find a negative relationship between monthly COVID-19 case levels and changes in trade flows across both provinces and time. Provinces experiencing higher population-adjusted case counts exhibit larger decreases in trade flows, with the largest decreases typically occurring in the months with the most severe COVID-19 outbreaks. We also find evidence of a positive relationship between monthly COVID-19 case levels and the severity of lockdown restrictions imposed by provinces over time.

With our descriptive analysis having revealed the broad negative relationship between trade and COVID-19, we next turn to a regression model for a more rigorous assessment of this relationship. Using a variety of COVID-19 health metrics, such as cases, deaths, and hospitalizations (all population controlled, per million inhabitants), we find that most of these measures of pandemic severity predict significant decreases in year-over-year trade flows for both Canadian exports and imports. Because our COVID-19 health metrics exhibit substantial differences in their mean levels, we interpret our coefficient estimates on the basis of a one-standard-deviation increase in each control variable for ease of comparison.³ Using this method, our regression results predict that a one-standard-deviation rise in the severity of the pandemic in Canada leads to a reduction of 3.1 percent to 4.9 percent in Canadian provincial exports and 6.7 percent to 9.1 percent in Canadian provincial imports.⁴ Similarly, we find that a one-standard-deviation increase in the severity of the pandemic in the United States results in a 2.3 percent to 7.1 percent decrease in Canadian exports and a 1.5 percent to 8.5 percent decrease in Canadian imports.⁵ If we exclude our estimates for US cases because they are not statistically significant at the 90 percent confidence level for Canadian exports or imports, the impact narrows to 6.3 percent to 7.1 percent for exports and 5.1 percent to 8.5 percent for imports.

Across Canadian provinces and US states, we find that the degree of lockdown stringency had a comparable (although typically smaller) economically negative impact on changes in Canadian export and import flows. However, none of our estimates are statistically significant at the 90 percent confidence level. This may in large part be due to a lack of variation across Canadian provinces and US states in the timing and severity of lockdown restrictions imposed during the first wave of the pandemic.

Building on the work of Baldwin and Freeman (2020) and Blit (2020), who find evidence of significant differences in the impacts of COVID-19 across industries, we further decompose our analysis at the NAICS two-digit industry level across the agriculture; mining, oil, and gas; and manufacturing industries. We find significant differences in the impact of pandemic severity on trade across industries. Although trade in the agriculture industry was largely unaffected by our COVID-19 health metrics, we find that higher COVID-19 case counts in both Canadian provinces and US states negatively affect Canadian exports in the mining, oil, and gas industry. We find the largest impact in the relatively more labour-intensive manufacturing industry, where increases in any of our Canadian health metrics (cases, deaths, and hospitalization rates) lead to decreases in both Canadian exports and imports and increases in US deaths and hospitalizations also predict decreases in Canadian exports and imports.

Similar to our pooled results, we find that lockdown policies were not statistically significant in explaining changes in Canadian export or import flows across any of these three industries. However, they are generally economically significant and consistent with our descriptive results. Taken together, these regression results suggest that measures imposed to protect public health during the first wave of the pandemic may have had additional negative effects on year-over-year trade growth above and beyond any direct effects of the pandemic’s severity for a given province–state pair. However, because of the lack of statistical significance, we cannot conclusively assert that the data validate this claim.

This article contributes to the growing literature on understanding the economic and public health effects of the COVID-19 pandemic in Canada. Baggs, Fung, and Lapham (forthcoming) examine the impact of decreased international travel due to COVID-19 on Canadian retailers and find the US–Canada border closure helped offset some of the losses retailers experienced during economic lockdowns. Important analyses of the pandemic’s effects in other areas of the Canadian economy by Lemieux et al. (2020), Gallacher and Hossian (2020), and Beland et al. (2020) have documented large drops in employment and worsening mental health stemming from COVID-19, especially for Canadians employed in industries in which the possibility of remote work is less likely. More recently, Frenette, Messacar, and Handler (2021) predict the long-term negative ramifications for youths’ earnings prospects from entering the labour market during a COVID-19–fueled slowdown of the economy.

Work specific to international trade has been more limited; however, reports by intergovernmental organizations such as the World Trade Organization and the Organization for Economic Co-operation and Development (OECD) have highlighted the potential risk to global trade from COVID-19. For example, OECD (2020) finds that the threat of re-nationalizing supply chains to ease the immediate burden of satisfying medical supply needs may make countries worse off in the long run if countries retaliate by severing productive cross-border supply chains for intermediate goods in other industries. Alternatively, trade may expedite economic recovery in countries struggling to overcome COVID-19 outbreaks. Using a general equilibrium model of world production and trade, Bonadio et al. (2020) find that increased import flows during times of stringent national or regional lockdowns can mitigate the economic losses caused by disruptions to domestic production capabilities. However, these same supply linkages may also prolong the negative impact of COVID-19 on trade, even after COVID-19 is under control domestically. Baldwin and Freeman (2020) warn that differences in the timing and severity of COVID-19 across countries create multiple waves of supply-chain contagion in which supply shocks in one country lead to reductions in foreign intermediary goods in another. Studies by Meier and Pinto (2020) and Friedt and Zhang (2020) have highlighted the role that global value chains (GVCs) play in magnifying the negative trade impact from COVID-19, focusing on China and the United States, respectively. A similar study conducted by Hayakawa and Mukunoki (2021b) expands to a larger set of countries but narrows the focus to the role of the COVID-19 pandemic in disrupting the GVC in the machinery products industry.

Perhaps most closely related to our work is that of Hayakawa and Mukunoki (2021a), who demonstrate the negative impact of COVID-19 on bilateral trade across a wide variety of countries in the first quarter of 2020. However, although they also examine year-over-year changes in trade flows, their pandemic severity metrics are limited to COVID-19 cases and deaths at the national level, do not address the impact that government polices (e.g., lockdown measures) have on trade flows, and focus solely on the first quarter of 2020, after which time much larger and wide-ranging changes in trade flows were observed.⁶

Our work builds on this burgeoning literature with multiple contributions. First, to our knowledge, this is the first article to describe the changes to monthly province–state trade flows between Canada and the United States in the wake of the COVID-19 pandemic. The Canada–US relationship provides an ideal case study for the early effects of the pandemic on bilateral trade, because it is one of the most well-established and widely researched trade relationships in the world. The stability of Canada–US trade before the pandemic provides an opportunity to more accurately isolate the impact of the COVID-19 pandemic on trade flows.⁷ Second, our work is unique in examining these changes at a more disaggregated level, to contrast the impacts on trade felt across various industries. Finally, this article is among the first to jointly produce a quantitative analysis relating COVID-19 cases and policy data, such as lockdown severity, to trade data throughout the course of the first wave of the pandemic. This approach provides valuable insight for researchers and policy-makers in designing and understanding the role of policy in response to the changing dynamics of the COVID-19 pandemic.

The remainder of the article is structured as follows: next, we present the data and descriptive analysis, and then we detail our regression specifications and empirical results. The final section concludes.

Data

In this section, we describe the data sources and document the data patterns of the relationship between COVID-19 and bilateral trade between Canada and the United States.

Data Sources

Our trade data, covering the 21-month period from January 2019 to September 2020, are compiled from the Government of Canada’s TDO database, which provides data on monthly bilateral trade flows between Canada and the United States, at the province–state level, for select industries at the NAICS two-digit classification level. Four industries are recorded in the TDO: agriculture, forestry, fishing, and hunting (NAICS Code 11); mining and oil and gas extraction (NAICS Code 21); utilities (NAICS Code 22); and manufacturing (NAICS Codes 31–33). Collectively, these industries accounted for an average of 93 percent of total exports between Canada and the United States and 69 percent of the total Canadian worldwide exports during this time period. We have similarly high levels of coverage for imports during this time period, an average of 94 percent of total Canadian imports from the United States and 47 percent of total worldwide imports. The underlying data used in the TDO are sourced from Statistics Canada and the US Census Bureau and updated monthly with only a three- to four-month lag, which enables us to study the contemporaneous effects of COVID-19 on bilateral trade (see Canada 2021b for details).

Our COVID-19 data are assembled from a variety of sources. The Canadian data come from the COVID-19 Canada website hosted by ESRI Canada, which collects province-level COVID-19 metrics using ArcGIS software to compile data directly from each province’s official health reporting services (see COVID-19 Canada 2020 for details). The US data are obtained from the COVID Tracking Project, which similarly amasses data directly from each state’s public health authority (see COVID Tracking Project 2020 for details). This provides us daily counts of positive cases and deaths, as well as supplementary metrics, such as the number of tests administered and the number of hospitalizations.⁸ For each province or state, we then aggregate daily counts to monthly totals to match to our monthly trade flow data, according to the source or destination location of each trade flow.

One particular challenge facing the Canadian and US economies as a result of the COVID-19 pandemic has been the introduction of lockdowns—restrictions on economic and social activity imposed by federal and provincial–state governments. To quantitatively assess the impact of these policy responses on trade, we use data from the Oxford COVID-19 Government Response Tracker, a unique index created by a panel of researchers and experts affiliated with the University of Oxford and the Blavatnik School of Government, to measure the degree of severity of these restrictions across a wide range of countries and regions over the course of the pandemic (see Blavatnik School of Government n.d. for details). We use the generalized Lockdown Stringency Index, a weighted index of indicators such as school and workplace closures; restrictions on travel, movement, or gatherings; or stay-at-home orders ranging in value from 0 to 100. This index value is reported for each province and state on a daily basis, which we similarly aggregate into a monthly average to match to each monthly trade flow.

Trade Data Patterns

To visualize the impact of the COVID-19 pandemic on trade flows between 2019 and 2020, in Figure 1 we plot the year-over-year percentage change in total exports and total imports between Canada and the United States.⁹ Beginning in March 2020, coinciding with the declaration of COVID-19 as a global pandemic causing countries to begin taking policy actions such as lockdowns, we observe a substantial drop in both exports and imports compared with the same month in the previous year. These percentage decreases in trade are substantially larger than those observed for Canadian gross domestic product (GDP), often by a factor of two to three (refer to Table A.1 in the online Appendix).

Figure 1:

Year-over-Year Percentage Change in Canadian Trade to the United States by Direction, National Level, 2019–2020

Source: Statistics Canada, US Census Bureau, Canada (2021b).

We further investigate whether these trade patterns occur across the three two-digit NAICS goods categories in our data: (a) agriculture; (b) mining, oil, and gas; and (c) manufacturing.¹⁰ For example, as a result of the vital role that temporary foreign workers play in the agriculture sector in Canada, the imposition of travel restrictions and quarantine measures to combat COVID-19 may have a more pronounced effect on production capacity in the agriculture sector than in the goods-producing sectors, where temporary foreign worker employment is small (Lu 2020). Alternatively, whereas demand for agricultural goods has remained high throughout the pandemic, demand in the mining, oil, and gas sector may have fallen dramatically as a result of work-from-home orders reducing the need to travel. In Figure 2, we observe significant heterogeneity in the response and persistence of the shock to trade flows across industry categories, which presumably reflects the differences in the impact of COVID-19 on the demand or supply for the goods in these industries. There was a sharp and persistent decline in the mining, oil, and gas and manufacturing trade flows, but only a modest decline in the agricultural sector, followed by a quick recovery.

Figure 2:

Canadian Bilateral Trade to the United States, by Industry: (a) Year-over-Year Percentage Change in Canadian Exports to the United States by Industry, 2019–2020, and (b) Year-over-Year Percentage Change in Canadian Imports from the United States by Industry, 2019–2020

Sources: Statistics Canada, US Census Bureau, Canada (2021b).

The previous trade figures clearly show a notable decline in trade beginning in March 2020, which coincides with the arrival of significant numbers of COVID-19 cases, and the resultant policy responses, in North America. To first explore the relationship between trade flows and COVID-19 cases at the national level, in Figure 3 we plot changes in total exports and total imports against the national COVID-19 case totals over time. We find compelling evidence that the decline of exports and imports is strongly correlated with the COVID-19 case severity. For example, during our sample period, the highest number of COVID-19 cases are reported in April and May, as are the largest decreases in exports and imports relative to the same month in the previous year.

Figure 3:

Canadian Trade versus Canadian Monthly COVID-19 Cases: (a) Year-over-Year Percentage Change in Canadian Exports to the United States and COVID-19, National Level, 2019–2020, and (b) Year-over-Year Percentage Change in Canadian Imports to the United States and COVID-19, National Level, 2019–2020

Notes: The rightmost observations in both graphs represent Quebec for April and May 2020. Slope coefficient = −0.0011, SE = 0.0001. COVID-19 = coronavirus disease 2019.

Sources: Trade data from Statistics Canada, US Census Bureau, and Canada (2021b); COVID-19 case data from COVID-19 Canada (2020) and COVID Tracking Project (2021).

Given differences across provinces in exposure to COVID-19, proximity to key US border crossings, and industry composition, a natural concern is that the trade patterns shown in Figure 1 and Figure 3 are driven by a selected sample of provinces. However, as is seen in Figure 4, across almost all provinces, the severity of year-over-year decreases in exports and imports is correlated with higher COVID-19 case numbers.

Figure 4:

Year-over-Year Percentage Changes in Canadian Trade with the United States and COVID-19 Cases by Province and Direction, 2019–2020

Note: COVID-19 = coronavirus disease 2019.

Sources: Trade data from Statistics Canada, US Census Bureau, and Canada (2021b); COVID-19 case data from COVID-19 Canada (2020) and COVID Tracking Project (2021).

Figure 5 further decomposes year-over-year changes in trade against COVID-19 cases per million at the provincial level for each of our two-digit NAICS industries. We find similarly negative relationships across all industries; however, this negative correlation is stronger in some industries (manufacturing and mining, oil, and gas) than others (agriculture). These differences can manifest for a variety of reasons. Safety measures aimed at slowing the spread of COVID-19 affecting the supply side of the economy may arguably have a more pronounced impact on labour-intensive industries that require significant in-person presence for production.¹¹ Moreover, the well-documented fall in earnings for consumers from COVID-19 has also resulted in decreased aggregate demand; however, these shocks are not symmetric across all industries. Eaton et al. (2016) study the effect of the 2008–2009 recession on trade and show that the large drop in trade that followed was largely caused by a shift away from durable product spending (e.g., manufactured goods products) toward non-traded sectors (e.g., housing). The heterogeneity in the transmission of these shocks across industries in the economy contributes to the differences observed in these figures.

Figure 5:

Year-over-Year Percentage Changes in Canadian Trade with the United States and Monthly COVID-19 Cases by Industry and Direction, 2019–2020

Notes: The rightmost observations in the graphs represent Quebec for April and May 2020. Each point represents a different province in a given month, between January 2020 and September 2020. COVID-19 = coronavirus disease 2019.

Sources: Trade data from Statistics Canada, US Census Bureau, Canada (2021b); COVID-19 case data from COVID-19 Canada (2020) and COVID Tracking Project (2021).

Finding evidence of the relationship between changes in trade flows and COVID-19 cases, we next look at the link between COVID-19 cases and the resultant policy decisions regarding lockdowns. Figure 6 plots the lockdown stringency index for each Canadian province over the course of the pandemic, with all provinces exhibiting similarly strong (yet not identical) escalations of lockdown measures in the early months of the pandemic, followed by varying degrees of relaxation over the summer and early fall of 2020.¹²

Figure 6:

Canadian Lockdown Stringency

Source: Blavatnik School of Government (n.d.).

Figure 7 plots the relationship between the number of monthly COVID-19 cases (per million inhabitants) in each province and the severity of the lockdown measures implemented by each province during that month. As expected, we find a positive relationship between cases and lockdown policy response in 2020.¹³ This finding lends credence to potential supply-side effects, with provinces hardest hit by COVID-19 cases reacting with the most stringent lockdown policies and therefore potentially experiencing larger effects on their trade flows.

Figure 7:

Canadian Lockdown Stringency versus COVID-19 Cases

Notes: The uppermost outliers represent Quebec in April and May 2020. Slope coefficient = 5.5551, SE = 0.0342. COVID-19 = coronavirus disease 2019.

Sources: COVID-19 lockdown data from Blavatnik School of Government (n.d.); COVID-19 case data from COVID-19 Canada (2020).

Taken together, these findings suggest that there is a strong negative relationship between changes in trade flows between Canadian provinces and US states and the severity of the COVID-19 pandemic, both in case statistics and in policy responses such as lockdowns in those regions. This negative relationship holds across multiple industry categories, across provinces, and for both exports and imports. In the next section, we outline a framework to empirically investigate this relationship.

Empirical Approach

To quantify the impact of COVID-19 on trade, we construct a set of regression specifications to examine year-over-year changes in trade flows across province–state pairs, controlling for COVID-19 health metrics and government policy response measures. Our health metrics include COVID-19 cases, average hospitalizations, and deaths, and the policy response is measured by our lockdown stringency index.

Across all regressions, our dependent variable is the year-over-year difference in monthly directional trade flows for each province–state pair in a given NAICS two-digit industry. Because of potential heteroskedasticity issues arising from the large variation in trade values between larger and smaller province–state pairs, we log-transform our dependent variable. Moreover, because of the relative frequency of zero values in our trade data, we use the following transformation:

$$log\Delta X_{psit}\equiv log\left(1+X_{psit}\right)-log\left(1+X_{psi(t-12)}\right),$$ (1)

where log ΔX_psit represents the change in logged trade flows (exports or imports) between province p and state s in month t between years 2019 and 2020.

Our regression specification in Equation (2) takes the form

$$log\Delta X_{psit}={\beta }_1\Delta COVID-{19}_{pt}+{\beta }_2\Delta Lockdow{n}_{pt},+{\beta }_3\Delta COVID-{19}_{st}+{\beta }_4\Delta Lockdow{n}_{st}+\Delta {\delta }_{psi}+\Delta {\epsilon }_{psit}$$ (2)

where COVID-19_𝑝𝑡 and COVID-19_𝑠𝑡 represents one of our COVID-19 health metrics,¹⁴ such as cases, deaths, and hospitalizations in each province p and state s, respectively, and Lockdown_𝑝𝑡 and Lockdown_𝑠𝑡 represent our lockdown stringency index values.¹⁵ We also include a province–state–industry fixed effect, 𝛿_𝑝𝑠𝑖. These fixed effects control for any unobserved heterogeneity across province–state–industry observations that may be correlated with both trade flows and COVID-19 severity and might otherwise bias our estimates.

We run separate regressions using each of our COVID-19 health metrics because the incredibly high level of correlation between these three measures, as shown in Table 2, presents issues with multicollinearity.¹⁶

Table 2:

Correlations: COVID-19 Statistics (per Million)

Variable	1	2	3	4
1. Lockdown	—
2. Cases	0.49	—
3. Average hospitalizations	0.37	0.86	—
4. Deaths	0.31	0.84	0.98	—

Note: COVID-19 = coronavirus disease 2019.

Sources: COVID-19 case data from COVID-19 Canada (2020) and COVID Tracking Project (2021); lockdown data from Blavatnik School of Government (n.d.).

Regression Results

The estimation results for Equation (2) using our pooled sample for Canadian exports and Canadian imports are presented in Tables 3 and 4, respectively. Columns (1) through (3) show the anticipated negative coefficient estimates when using cases, deaths, and hospitalizations as our various metrics for measuring the severity of the pandemic across provinces and states. However, our estimates for cases in US states are not statistically significant in explaining export growth, and cases and hospitalizations in US states are not statistically significant in explaining import growth.

Table 3:

OLS Estimates of Year-over-Year Export Growth Regressions, All Observations

Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Cases
Province	−0.000103* (5.64e-05)				−8.11e-05 (6.21e-05)
State	−8.85e-06 (1.23e-05)				−4.98e-06 (1.33e-05)
Deaths
Province		−0.000712* (0.000414)				−0.000664 (0.000439)
State		−0.000725*** (0.000268)				−0.000688** (0.000299)
Hospitalizations
Province			−0.00136* (0.000720)				−0.00129* (0.000779)
State			−0.000625** (0.000301)				−0.000601* (0.000335)
Lockdown
Province				−0.00155 (0.00533)	−0.00101 (0.00540)	−0.000367 (0.00537)	−0.000496 (0.00538)
State				−0.000251 (0.00484)	2.45e-06 (0.00485)	2.27e-05 (0.00487)	0.000261 (0.00485)
Constant	0.0219 (0.0295)	0.0251* (0.0148)	0.0337 (0.0210)	0.0656 (0.0621)	0.0586 (0.0629)	0.0398 (0.0645)	0.0434 (0.0643)
No. of observations	13,770	13,770	13,770	13,770	13,770	13,770	13,770
Adjusted R2	0.0719	0.0723	0.0722	0.0719	0.0718	0.0721	0.0721

Notes: Values are estimates of versions of Equation (2) for export trade flows between Canadian provinces and US states between January 2020 and September 2020. Cases, deaths, and hospitalizations are per-million population. Includes province–state–industry fixed effects. Robust standard errors clustered by province–state are in parentheses. OLS = ordinary least squares.

^*p = 0.10;

^**p = 0.05;

^***p = 0.01.

Source: Authors’ calculations.

Table 4:

OLS Estimates of Year-over-Year Import Growth Regressions, All Observations

Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Cases
Province	−0.000154** (7.53e-05)				1.91e-05 (8.43e-05)
State	−5.93e-06 (1.30e-05)				2.56e-05* (1.37e-05)
Deaths
Province		−0.00155* (0.000842)				−0.000751 (0.000862)
State		−0.000866** (0.000362)				−0.000197 (0.000400)
Hospitalizations
Province			−0.00326** (0.00134)				−0.00134 (0.00142)
State			−0.000513 (0.000346)				0.000296 (0.000381)
Lockdown
Province				−0.00218 (0.00535)	−0.00264 (0.00530)	−0.00140 (0.00536)	−0.00185 (0.00533)
State				−0.00414 (0.00490)	−0.00497 (0.00494)	−0.00423 (0.00489)	−0.00452 (0.00492)
Constant	−0.118*** (0.0330)	−0.103*** (0.0206)	0.0337 (0.0210)	0.144** (0.0637)	0.147** (0.0635)	0.129** (0.0648)	0.139** (0.0639)
No. of observations	13,770	13,770	13,770	13,770	13,770	13,770	13,770
Adjusted R2	0.0997	0.101	0.100	0.102	0.102	0.102	0.102

Notes: Values are estimates of versions of Equation (2) for import trade flows between Canadian provinces and US states between January 2020 and September 2020. Cases, deaths, and hospitalizations are per-million population. Includes province–state–industry fixed effects. Robust standard errors clustered by province–state are in parentheses. OLS = ordinary least squares.

^*p = 0.10;

^**p = 0.05;

^***p = 0.01.

Source: Authors’ calculations.

Using the coefficient estimates in Table 3, our results predict that each additional COVID-19 case per million in a Canadian province will reduce the log-plus-one of Canadian exports by 0.000103 compared with the previous year’s value in the same month. This is equivalent to an approximately 0.0103 percent decrease. Although this effect, when interpreted as causal, may appear small, it is perhaps better understood in the context that, as shown in Table 1, mean values of provincial cases are 263.95 per million over this time span, with a standard deviation of 476.36 per million. Therefore, an increase in provincial cases by one standard deviation of 476.36 is estimated to reduce Canadian exports to the United States by 0.0103 ´ 476.36, or approximately 4.9 percent. Thus, if we considered the effect of an identical one-standard-deviation increase happening simultaneously across all Canadian provinces, we would expect a 4.9 percent reduction in Canadian exports. A similar calculation estimates that a one-standard-deviation increase in Canadian monthly cases of 476.36 would be expected to reduce Canadian imports from the United States by 7.3 percent.¹⁷

Table 1:

COVID-19 Summary Statistics (per Million)

Variable	Mean (SD)	Range, Min–Max	No. of Observations
Province
Cases	263.95 (476.36)	0–2,789.06	27,540
Deaths	12.31 (43.39)	0–329.88	27,540
Hospitalizations	10.15 (27.9)	0–194.50	27,540
Lockdown	50.45 (24.09)	7.71–88.15	27,540
State
Cases	2,223.53 (2,620.43)	0–14,755.39	27,540
Deaths	56.96 (98.39)	0–934.05	27,540
Hospitalizations	71.92 (100.27)	0–812.77	27,540
Lockdown	49.66 (26.81)	0–92.93	27,540

Notes: This table contains the summary statistics for the COVID-19 health and lockdown measures across the 10 Canadian provinces and 50 US states (plus the District of Columbia) from January through September 2020. COVID-19 = coronavirus disease 2019.

Source: COVID-19 case data from COVID-19 Canada (2020) and COVID Tracking Project (2021). COVID-19 lockdown data from Blavatnik School of Government (n.d.).

Similarly, a one-standard-deviation increase in provincial deaths and hospitalizations results in a 3.1 percent and 3.8 percent reduction in Canadian exports, respectively, and a 6.7 percent and 9.1 percent decrease in Canadian imports. Alternatively, a one-standard-deviation increase in US deaths results in a 7.3 percent decrease in Canadian exports and a 8.5 percent decrease in Canadian imports.¹⁸ In column (4) of Tables 3 and 4, using our lockdown indices, we find the expected negative signs, but our coefficient estimates are not statistically significantly different from zero. However, unlike our health metrics, we note that the lack of statistical power for our lockdown measures may be a reflection of the relative lack of variation in lockdown policies across provinces in each month (regardless of variation in COVID-19 incidence), leading to larger standard errors in our regression analysis when using lockdown stringency as our explanatory variable.¹⁹ Using the point estimates, we would predict a one-standard-deviation increase in provincial lockdown stringency would decrease Canadian exports by 3.73 percent and Canadian imports by 5.25 percent, whereas a one-standard deviation increase in state lockdown stringency would decrease Canadian exports by 0.67 percent and Canadian imports by 11.1 percent. These estimates would be consistent with the idea that Canadian lockdown policies adversely affect both exports and imports to an economically significant degree, whereas US lockdown policy is more likely to affect Canadian imports, but not Canadian exports. However, although these point estimates suggest that there may potentially be an economically significant effect of lockdowns on trade flows, because of the large standard errors in this case, we are unable to confidently assert that these values are statistically significantly different from zero.

In columns (5) through (7) of Tables 3 and 4, we combine each of our COVID-19 health metrics with the lockdown policy measure. We find similar coefficient estimates; however, likely because of the relatively high correlation between the pandemic severity and lockdown measures, some of our estimates such as provincial cases and deaths lose their statistical significance.

One issue to address with our pooled results is the influence of potential outliers, specifically, the disproportionately high population-adjusted COVID-19 case counts in Quebec, particularly in April and May 2020, relative to all other Canadian provinces.²⁰ To test the robustness of our pooled results, we omit Quebec from our regression analysis and find similar results to those with our full sample. In most instances, our coefficient estimates are similar, but larger in magnitude, when omitting Quebec; for example, a one-standard-deviation increase in provincial deaths (outside of Quebec) now results in an 18 percent decrease in Canadian export flows and a 17.4 percent decrease in Canadian import flows, compared with 3.1 percent and 6.7 percent, respectively, with our full-sample estimates. Similarly, a one-standard-deviation increase in provincial hospitalizations predicts a 14.1 percent decrease in Canadian exports and a 20.1 percent decrease in Canadian imports when excluding Quebec, compared with 3.8 percent and 9.1 percent, respectively, with our full sample (refer to Tables A.4 and A.5). These results are not unexpected; because Quebec experienced such disproportionately large case and death counts compared with other Canadian provinces in many months but continued to produce and trade at less disproportionately suppressed levels, including Quebec in our pooled sample serves to slightly dampen the relationship between adverse COVID-19 outcomes and trade disruptions.

Another issue potentially complicating our estimation is the prevalence of zero values in our trade data. A substantial number of province–state–industry observations have zero trade flows in each corresponding month of both 2019 and 2020. However, we cannot be sure whether these zero-value observations arise from lack of domestic production in a given industry or from barriers to trade or prohibitive transportation costs. As a result, these zero-value cases do not add explanatory power to the relationship between pandemic severity and changes in trade flows in our analysis. To check the robustness of our results, we repeat our analysis on the subset of observations with non-zero values in corresponding months of both 2019 and 2020. This reduces our sample size from 13,770 to 9,550 observations of year-over-year changes in monthly trade flows at the province–state–industry level. We find this approach produces similar results for our COVID-19 and lockdown measures from the full sample.²¹

Table 5:

Summary Statistics by Month

Month	Lockdown Stringency		Cases (per Million)		Deaths (per Million)
	Mean (SD)	CoV	Mean (SD)	CoV	Mean (SD)	CoV
January	8.02 (0.75)	0.09	0.04 (0.08)	2.00	0 (0)	N/A
February	12.11 (1.55)	0.13	0.21 (0.43)	2.07	0 (0)	N/A
March	43.20 (5.32)	0.12	190.44 (115.70)	0.61	1.72 (1.53)	0.89
April	77.66 (7.33)	0.09	646.60 (801.14)	1.24	36.50 (63.61)	1.74
May	73.44 (4.32)	0.06	444.11 (817.11)	1.84	47.30 (97.25)	2.06
June	65.25 (5.71)	0.09	153.16 (192.78)	1.26	14.23 (30.41)	2.14
July	59.43 (5.50)	0.09	210.07 (215.54)	1.03	4.73 (6.08)	1.28
August	59.53 (5.55)	0.09	269.03 (239.70)	0.89	3.58 (3.66)	1.02
September	55.38 (5.39)	0.10	461.91 (448.21)	0.97	2.74 (3.11)	1.13

Notes: Means, standard deviations, and CoVs for the lockdown stringency index, cases and deaths are per million inhabitants across the 10 Canadian provinces for each month of 2020. CoV = coefficient of variation; N/A = not applicable; COVID-19 = coronavirus disease 2019.

Sources: COVID-19 data from COVID-19 Canada (2020) and COVID Tracking Project (2021); COVID-19 lockdown data from Blavatnik School of Government (n.d.).

Collectively, these results generally support an economically significant effect of COVID-19 on Canadian trade with the United States; however, some of our estimates, including those for provincial case numbers, are only statistically significant at a 90 percent confidence level. Given the heterogeneity in trade flow changes across industries observed in Figure 2 and the correlation between the severity of COVID-19 and trade flow changes across industries observed in Figure 5, these pooled results may be obscuring heterogeneous industry-specific impacts. We therefore run versions of Equation (2) separately for the agriculture (Table 6); mining, oil, and gas (Table 7); and manufacturing (Table 8) industries.

Table 6:

OLS Estimates of Year-over-Year Export and Import Growth: NAICS Code 11, Agriculture

Variable	Year-over-Year Expert Growth				Year-over-Year Import Growth
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Cases
Province	5.18e-05 (4.63e-05)				−0.000123 (0.000103)
State	4.23e-06 (9.07e-06)				6.50e-06 (1.98e-05)
Deaths
Province		0.000843* (0.000500)				−0.000912 (0.00104)
State		−0.000143 (0.000201)				−0.000860* (0.000510)
Hospitalizations
Province			0.00124 (0.000768)				−0.00203 (0.00168)
State			−0.000126 (0.000217)				−0.000503 (0.000540)
Lockdown
Province				−0.00187 (0.00370)				−0.00649 (0.00874)
State				0.00176 (0.00324)				0.00101 (0.00727)
Constant	0.00544 (0.0230)	0.0275** (0.0124)	0.0249 (0.0160)	0.0367 (0.0493)	−0.0645 (0.0532)	−0.0277 (0.0282)	−0.0273 (0.0388)	0.180 (0.114)
No. of observations	3,019	3,019	3,019	3,019	3,019	3,019	3,019	3,019
Adjusted R2	0.183	0.184	0.184	0.183	0.256	0.257	0.257	0.259

Notes: Values are estimates of versions of Equation (2) for export and import flows between Canadian provinces and US states between January 2020 and September 2020 for all non-zero observations. Cases, deaths, and hospitalizations are per-million population. Includes province–state–industry fixed effects. Robust standard errors clustered by province–state are in parentheses. NAICS = North American Industry Classification System; OLS = ordinary least squares.

^*p = 0.10;

^**p = 0.05;

^***p = 0.01.

Source: Authors’ calculations.

Table 7:

OLS Estimates of Year-over-Year Export and Import Growth: NAICS Code 21, Mining, Oil, and Gas

Variable	Year-over-Year Expert Growth				Year-over-Year Import Growth
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Cases
Province	−9.14e-05* (4.79e-05)				−2.13e-05 (0.000139)
State	−3.18e-05** (1.37e-05)				−2.60e-05 (2.35e-05)
Deaths
Province		−0.000570 (0.000548)				−0.000608 (0.00131)
State		−0.000886*** (0.000341)				−0.000597 (0.000839)
Hospitalizations
Province			−0.00112 (0.000967)				−0.00163 (0.00210)
State			−0.00101*** (0.000388)				−0.000419 (0.000733)
Lockdown
Province				−0.00751 (0.00548)				−0.0136 (0.0118)
State				0.00228 (0.00480)				0.0102 (0.0106)
Constant	0.0999*** (0.0373)	0.0578*** (0.0207)	0.0870*** (0.0310)	0.252*** (0.0756)	0.0635 (0.0662)	0.0412 (0.0455)	0.0515 (0.0530)	0.167 (0.149)
No. of observations	2,278	2,278	2,278	2,278	1,918	1,918	1,918	1,918
Adjusted R2	0.186	0.185	0.187	0.188	0.168	0.168	0.168	0.168

Notes: Values are estimates of versions of Equation (2) for export and import flows between Canadian provinces and US states between January 2020 and September 2020 for all non-zero observations. Cases, deaths, and hospitalizations are per-million population. Includes province–state–industry fixed effects. Robust standard errors clustered by province–state are in parentheses. NAICS = North American Industry Classification System.

^*p = 0.10;

^**p = 0.05;

^***p = 0.01.

Source: Authors’ calculations.

Table 8:

OLS Estimates of Year-over-Year Export and Import Growth: NAICS Code 31, Manufacturing

Variables	Year-over-Year Expert Growth				Year-over-Year Import Growth
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Cases
Province	−0.000109*** (2.70e-05)				−0.000258*** (4.52e-05)
State	1.54e-08 (5.84e-06)				−2.87e-06 (7.86e-06)
Deaths
Province		−0.00110*** (0.000295)				−0.00234*** (0.000651)
State		−0.000406** (0.000163)				−0.000899*** (0.000256)
Hospitalizations
Province			−0.00225*** (0.000436)				−0.00439*** (0.000969)
State			−0.000245* (0.000145)				−0.000634** (0.000253)
Lockdown
Province				−0.00269 (0.00343)				−0.00648 (0.00427)
State				−0.000995 (0.00308)				0.000377 (0.00372)
Constant	−0.0747*** (0.0127)	−0.0670*** (0.00898)	−0.0630*** (0.0101)	0.0796** (0.0380)	−0.113*** (0.0173)	−0.112*** (0.0154)	−0.0985** (0.0187)	0.104* (0.0535)
No. of observations	4,253	4,253	4,253	4,253	3,598	3,598	3,598	3,598
Adjusted R2	0.176	0.178	0.178	0.183	0.122	0.127	0.126	0.127

Notes: Values are estimates of versions of Equation (2) for export and import flows between Canadian provinces and US states between January 2020 and September 2020 for all non-zero observations. Cases, deaths, and hospitalizations are per-million population. Includes province–state–industry fixed effects. Robust standard errors clustered by province–state are in parentheses. NAICS = North American Industry Classification System; OLS = ordinary least squares.

^*p = 0.10;

^**p = 0.05;

^***p = 0.01.

Source: Authors’ calculations.

Consistent with the results found by Hayakawa and Mukunoki (2021a), we find no strong evidence that the severity of COVID-19, as measured by our health metrics, had any statistically significant effect on Canadian agriculture trade flows. Unlike the more volatile demand for durable goods, agricultural goods are composed of many necessities that are required for daily living, which are more likely to maintain stable levels over time. Similarly, stricter lockdown measures had no significant negative impact on agriculture trade flows, which in part reflects the nature of agricultural production in Canada and the United States being relatively capital-intensive (by a factor of three) compared with manufacturing (Baldwin et al. 2008), where these lockdown measures would presumably affect labour-intensive industries more acutely.

In our results for the mining, oil, and gas industry, we find some evidence that US deaths and hospitalizations each negatively affect Canadian export flows, with a one-standard-deviation increase in deaths or hospitalizations resulting in a 5.6 percent and 10.1 percent reduction in mining and oil and gas exports, respectively. Both Canadian and US cases decrease Canadian exports, although the former is only significant at the 90 percent confidence level. These results would be in keeping with theories of decreased demand for fuel products stemming from work-from-home orders and the introduction of mandatory quarantine requirements for international travel. Similar to our results for the agriculture industry, we also find that the severity of lockdown measures across Canadian provinces and US states did not have a significant impact on trade flows in the mining, oil, and gas industry.

The results for the manufacturing industry show the largest and most consistent negative impact of COVID-19 on trade flows. In particular, we find that the severity of cases, deaths, and hospitalizations in Canadian provinces has significant negative impacts on both Canadian exports and Canadian imports. Additionally, deaths in US states have a significant negative impact on Canadian imports. Alternatively, we find no support for lockdown measures significantly affecting Canadian exports or imports in the manufacturing industry. The in-person presence requirement and relatively labour-intensive nature of the manufacturing industry corresponds to larger negative impacts of the pandemic’s severity on trade flows compared with other industries. Interpreting these coefficient estimates, we predict that a one-standard-deviation increase in provincial cases results in a 5.2 percent decrease in Canadian exports and a 12.3 percent decrease in Canadian imports in the manufacturing industry. A similar one-standard-deviation increase in provincial deaths results in 4.8 percent less Canadian exports and 10.2 percent less Canadian imports in this category. The comparable values for US state deaths are a 4.0 percent reduction in Canadian exports and an 8.8 percent reduction in Canadian imports. Finally, we estimate that a one-standard-deviation increase in provincial hospitalizations decreases Canadian exports by 6.3 percent and decreases Canadian imports by 12.2 percent, whereas a comparable increase in US state hospitalizations results in a 2.5 percent fall in Canadian exports and a 6.4 percent drop in Canadian imports of manufactured goods.

Taken together, our industry-specific results provide support for a demand-driven mechanism as a central explanation of the results depicted in Tables 6, 7, and 8. During highly volatile times, consumers typically consume fewer durable goods but continue to consume necessities, consistent with our findings of a drop in manufacturing trade and a stable agricultural and oil and gas trade. Moreover, we find that Canadian case counts affect Canadian imports more severely than Canadian exports, which further supports this explanation. However, without stronger data to identify a causal relationship, this theory is still speculative.

In each industry, we find no statistically significant impact of provincial or state lockdown stringency on Canadian export or import flows, although the point estimates do line up with the expected lockdown effects suggested by our descriptive statistics. As with our pooled sample, this result may in large part be due to the fact that, despite significant differences in levels of COVID-19 severity (as measured by cases, deaths, and hospitalizations) across provinces and states, there is much less variation in the severity of the lockdown measures imposed (shown in Figure 6), particularly at the outset of the pandemic in March and April 2020. This lack of variation in our lockdown index may hamper our ability to estimate the true impacts of the lockdown restrictions during the first wave of the COVID-19 pandemic.

Conclusion

In this article, we contribute to the small but growing literature on the economic fallout from the COVID-19 pandemic, using timely data on bilateral trade flows between Canadian provinces and US states. We find that all Canadian provinces experienced a minimum 25 percent year-over-year decrease in monthly exports, with some provinces experiencing as much as a 75 percent decrease in exports in the months after the classification of COVID-19 as a global pandemic. Our descriptive results identify a strong negative relationship between the severity of COVID-19 in a province or state and export and import flows. We also demonstrate a positive relationship between the severity of our COVID-19 health metrics, represented by cases, hospitalizations, and deaths, and our quantitative measure of lockdown stringency across provinces.

We follow our descriptive results with a series of regressions to further evaluate the negative relationship between COVID-19 health metrics and government policy responses to trade. Across a variety of specifications, we find that our COVID-19 health metrics, such as cases, deaths, and hospitalizations, produce the expected negative effect on Canadian exports and imports observed during our sample period. Repeating our quantitative analysis at the more disaggregated two-digit NAICS level, we find a large degree of heterogeneity that is obscured in the aggregated trade data. We find the severity of our COVID-19 health metrics across provinces to be a much stronger and statistically significant predictor of decreasing trade flows in the manufacturing industry (and to a lesser extent, the mining, oil, and gas industry) than in the agriculture industry, where trade flows experienced the least disruption during the course of the pandemic.

Although our regression results on the impact of lockdown measures on Canadian trade are not statistically significant, the coefficient estimates suggest they are most likely economically significant and consistent with our descriptive results. One possible factor contributing to the lack of statistical significance of our lockdown estimates is the fact that, particularly in the first few months of the pandemic’s emergence in Canada, most provinces followed very similar public health responses and imposed similar restrictions, regardless of their provincial COVID-19 caseloads at the time. As a result, our lockdown stringency metric exhibits a low coefficient of variation across provinces over the time frame of our data, remaining between 0.06 and 0.13 in every month. This lack of variation in policy response may contribute to the lack of statistical power of our lockdown measures to explain the variation in trade flows observed in our data. Although this article focuses on the impacts of the first wave of the COVID-19 pandemic, future studies may be better equipped to analyze this channel, particularly because early indications suggest increasing variation in lockdown policies across provinces as the pandemic progresses into its second and third waves.

As the course of the pandemic evolves and more data become available, these relationships can be re-evaluated, and further insight into the effects of COVID-19 health metrics and policy measures on trade can be ascertained. Unfortunately, traditional control variables such as GDP that are typically used in gravity models of international trade will never be available on a monthly basis at the provincial and state levels; however, with enough time to collect more data and amass sufficient observations on annual or quarterly GDP figures, these estimates may be improved. Additionally, whereas data are typically not available at the more disaggregated regional level, similar analyses can be performed at the national level for other Canadian trading partners.