We would like to thank Herb Emery, Kris Inwood, David Jacks, Byron J. Lew, Irving Rosales, Scott Taylor, John Wilson and seminar participants at the University of Calgary, the 2010 Business History Conference, the 2011 Canadian Economic Association Meetings, the 2012 Asia-Pacific Economic and Business History Conference, and the “Globalization and the Making of Canada” Conference in Waterloo, Ontario, for helpful comments and suggestions. This research has benefited from funding from the University of Calgary Research Grants (Beaulieu) and Social Science and Humanities Research Council of Canada (Cherniwchan). We would also like to thank Michael Ata, Matt Krzepkowski, and Fatih Yilmaz for excellent research assistance. Email: email@example.com
Canada's trade policy at the end of the 19th century is commonly viewed as protectionist and extremely costly. In this paper, we employ the Anderson-Neary Trade Restrictiveness Index to re-examine this view. Based on product-level customs data, we show that Canadian trade policy between 1870 and 1910 was more restrictive than previously understood, but created smaller welfare losses than previously believed. These results are primarily driven by high tariffs on inelastic, non-competing import goods. Although Canada's tariff structure becomes more restrictive over the period, our findings indicate it was not as protectionist or as costly as once thought.
Structure tarifaire, expansion du commerce et protectionnisme canadien entre 1870 et 1910. La politique commerciale canadienne à la fin du 19e siècle est habituellement considérée comme protectionniste et extrêmement coûteuse. Dans ce mémoire, on emploie l'indice de restriction du commerce de Anderson-Neary pour réexaminer ce point de vue. A partir de données du service des douanes au niveau du produit, on montre que la politique commerciale canadienne entre 1870 et 1910 a été plus restrictive que ce qu'on avait pensé autrefois, mais qu'elle a créé des pertes de bien-être moindres que ce qu'on avait cru antérieurement. Ces résultats sont d'abord engendrés par des droits de douanes élevés sur des biens d'importation qui ne concurrencent pas les biens domestiques et dont la demande est inélastique. Même si la structure tarifaire du Canada devient plus restrictive au cours de cette période, les résultats montrent qu'elle n'est ni aussi protectionniste ni aussi coûteuse qu'on l'a déjà pensé.
The conventional view among economic historians is that Canada's trade policy became protectionist and had large efficiency costs following the National Policy tariff of 1879. Pomfret (1993) suggests that the welfare losses associated with the Canadian tariff were greater than 4%–8% of GDP during this period. There are two important problems with the evidence supporting this view. First, it relies almost exclusively on the import-weighted average tariff (AWT) as a measure of protection.1 This is problematic because the AWT does not accurately reflect the true level of protection or capture the structure of the tariff, meaning the case for protectionism may be overstated.2 Second, the welfare implications of the tariff are not directly measured from data, but are instead inferred from estimates from later periods based on changes in the AWT over time.3 As a result, we do not know how restrictive, or how costly, Canadian trade policy was at the end of the 19th century.
The purpose of this paper is to analyze the level and costs of Canadian trade policy at the turn of the 19th century using theoretically consistent measures of protection and welfare loss. To do so, we employ the Anderson-Neary Trade Restrictiveness Index (TRI) to measure protection.4 The TRI is an index measure equal to the uniform tariff, which if applied to all goods, would yield the same welfare level as the existing tariff structure. Moreover, it provides us with a simple means for calculating the static welfare losses associated with trade policy. This makes it possible compare the level, and costs of trade policy across time using an approach that is consistent with theory.
The conventional view of history, based on the AWT, also posits that the National Policy tariff established the protectionist structure of the tariff for the next 50 years. Although this interpretation may be correct, it is difficult to reconcile this view with the continued importance of the tariff as the main source of government revenue. Moreover, it is impossible to infer how protectionist a given trade policy is based on the AWT, because a high AWT is consistent both with high levels of protection for certain commodities and with high tariffs on non-competing import goods meant to maximize government revenue. In order to make inferences about how protectionist the tariff policy is, we compute the TRI for broad commodity classes and show how the structure of protection has changed over time.
The shift to protectionism at this time was not a uniquely Canadian phenomenon. Indeed, the end of the 19th century witnessed a shift to higher tariffs around the globe as many countries moved from revenue tariffs to protectionist trade policies in an effort to protect domestic producers from foreign competition.5 At the same time, the world was in the midst of one of the largest trade booms in history, with tremendous increases in both the volume and the variety of goods traded internationally.6 Together, these changes led to significant revisions in tariff schedules around the industrializing world as governments changed policy in an effort to protect domestic producers and account for new varieties of products. Unfortunately, little is known about how these changes affected the level, structure, and welfare consequences of protection offered by trade policy, because existing analyses are based on average tariff measures.
This paper provides the first evidence of how the shift to protectionism and expansion of trade during the first wave of globalization affected the level, structure, and welfare costs of trade policy. By employing the TRI methodology to examine changes in Canadian trade policy between 1870 and 1910, we have a simple means of identifying the channels through which changes in the tariff structure and the expansion of trade led to changes in the level of protection. We follow Kee, Nicita, and Olarreaga (2009) and Irwin (2010) and employ Feenstra's (1995) simplification of the TRI to measure protection. This approach has the advantage that it allows us to construct the index using observable data on imports, ad valorem tariff rates, and elasticities of import demand. Employing disaggregate data allows us to document changes in the structure of protection in response to the shift to protectionist trade policies and increases in the number of products traded. We compute TRIs for agricultural, manufacturing, and exotic goods industries as defined by Lehmann and O'Rourke (2011). They show that the tariff structure across these goods is key to understanding the impact of tariff policy on economic growth.
We find that the AWT understates the restrictiveness of trade policy measured by the TRI by as much as 13 percentage points, so in this sense, the tariff was more restrictive than previously understood. We compute the first estimates of the welfare costs associated with protection in Canada during our period of study and find that the static welfare losses arising from protectionism amounted to 0.7%–1.5% of GDP. Although they have not been measured previously, these estimates are drastically lower than the conventional view of welfare losses found in the literature. From 1875 to 1880, the higher level of trade restrictiveness was primarily driven by an increase in the average tariff across goods. After 1880, the changing structure of the tariff across goods resulted in a higher variance of the tariff and this became the main factor that increased the TRI. Although the TRI increased after the National Policy, the structure of the tariff continued to be restrictive with low deadweight loss because the highest tariffs tended to be applied to goods with inelastic demand. Tariffs on agricultural and industrial goods increased, but most of the tariff increases were on exotic goods – goods that were not produced in Canada and were primarily taxed for revenue purposes.
These results suggest a re-evaluation of the typical history of Canadian trade policy at the end of the 19th century. Based on the TRI, Canadian trade policy was more restrictive than previously understood and the measured static welfare losses are much lower than previously inferred. Moreover, contrary to the conventional view of history, we find that Canadian tariff revisions after the National Policy had significant effects on trade restrictiveness. Previous research has focused on the National Policy tariff as the key driver of protectionism in Canada at this time because it corresponds to the largest increase in the AWT prior to the Great Depression. Later revisions to the tariff schedule are ignored because they do not lead to similar changes in the AWT. We show that these tariff revisions did indeed play an important role in determining the level of protection because they affected the variance of the tariff structure. In particular, the Tariff Amendment Act of 1887 and Fielding's Tariff of 1897 altered the level of protection similar in magnitude to the National Policy tariff. Hence, while the National Policy was the beginning of higher Canadian tariffs, subsequent revisions to the tariff schedule also had a large impact on the level of protection offered by trade policy.
Our results imply that Canadian trade policy at this time is not as protectionist as is generally believed. Although agricultural and industrial goods received higher levels of protection during our period of study, the level of protection we measure with the TRI is driven by high tariffs on exotic goods. This implies that the tariff structure was not protectionist in the sense that the highest tariffs were not primarily protecting domestically produced goods as policy statements at the time suggest. Instead, high tariffs on inelastically demanded exotic goods are a means to maximize revenue, which is consistent with the fact that tariff revenue continued to be the largest component of government revenue during this period. This also suggests that welfare costs are lower than they would have been under a more protectionist trade policy.
This paper illustrates the importance of using a theoretically consistent measure of protection when examining how revisions to trade policy affect the economy. Our analysis also highlights the importance of employing disaggregated trade data to measure protection, because the structure of the tariff is the crucial factor. We show that, although the AWT understates the restrictiveness of trade policy, it is accurate when revisions to the tariff schedule are relatively uniform across all goods, such as the case of the increase in tariffs following the National Policy. In contrast, we find that the AWT significantly understates the level of protection when tariff changes are designed to increase protection for some products more than others, such as the change in protection resulting from the Tariff Amendment Act of 1887 and Fielding's Tariff of 1897. Hence, research relying on the AWT understates the effects of trade policy reforms that alter the structure of the tariff across goods.
Our paper contributes to an emerging literature documenting the extent of protection and the welfare costs of trade policy using the TRI. This method of measuring the restrictiveness of trade policy has been employed to examine trade policy across time and countries. The TRI has been used to examine protection in France and the UK in the 19th century (O'Rourke 1997), in the US between 1859 and 1961 (Irwin 2010), and various other countries around the world from the 1980s onward (Anderson and Neary 2005; Kee, Nicita, and Olarreaga 2008, 2009; Lloyd and MacLaren 2010). We add to this literature by constructing the TRI for a historic period and by demonstrating the importance of using disaggregate data to calculate the simplified TRI.
By constructing the TRI using historical data, our paper is closely related to the work of Irwin (2010). He constructs an annual TRI for the United States over the period 1859–1961 using data for up to 21 import categories. Our analysis differs from Irwin in three important respects. First, we examine how changes in the tariff schedule designed to alter trade barriers and to account for expansion in the number varieties traded affect the level and cost of protection. Although the United States underwent a shift to protectionism and a large expansion in trade during our period of study, these features are not examined by Irwin. Second, we employ customs data that are reported at the product level. This allows us to account for detailed changes in the tariff structure. As we show below, employing aggregated data will understate the effect of changes in tariff structure. Finally, we examine the robustness of our results to our choice of elasticities. Given that historical elasticity estimates are not available for our period of study, we follow Irwin's approach and calculate the TRI using modern import demand elasticity estimates in place of historical estimates. We employ the delta method and simulations to examine the robustness of our results and find that they are not being driven by our elasticities.
2. Canadian Trade Policy through the Lens of the AWT
Canadian trade policy at the end of the 19th century is commonly viewed by both economists and historians as protectionist. This view traces back to the “National Policy” tariff enacted by the Macdonald Conservatives in 1879 following a prolonged stagnation of the economy. The National Policy and many of the subsequent tariff policies implemented in the following 30 years had the stated intent of providing Canadian producers with protection from foreign competition. Public proclamations stated that Canada started actively protecting domestic industry in 1879 with the enactment of the National Policy. The Canadian Finance Minister at the time declared that trade policy was set “to select for higher rates of duty those [goods] that are manufactured or can be manufactured in the country” (McDiarmid 1946, 161; see also Pinchin 1970). These revisions to tariff policy correspond to increases in the AWT, leading many to conclude that protection was the both the intent and the result of tariff reform (see, e.g., Easterbrook and Aitken 1965; Norrie and Owram 1991). This can be seen in Figure 1, which presents the import-weighted average tariff in Canada from 1870 to 1910. As shown in the figure, the average tariff increased from approximately 14% in 1875 to over 20% by 1880, following the enactment of the National Policy.
This increase is often viewed as laying the foundation for the tariff schedule for the next 50 years. The AWT remained relatively stable after the National Policy, leading to the conclusion that the National Policy legislation was the main determinant of protection for years to come. For example, Taylor (1939) writes, “The general level of protection was essentially unchanged until 1930, with only minor changes to the tariff schedule.” The tariff schedule was modified in 1887 and 1894, with a major revision in 1897 when a preferential tariff for Great Britain was introduced. There was further revision to the tariff in 1904 and in 1907, by which time the government had established three levels of duties, the lowest being the British preferential, then scaling up to the intermediate and general tariffs. The intermediate tariff served as the basis for the negotiation of treaties with non-British countries. Given that these later revisions did not affect the AWT tariff by as much as the National Policy, they are viewed as secondary determinants of Canadian protectionism during this time.
Altogether, the evidence based on the AWT gives the typical view of Canadian history: the National Policy set the stage for Canadian protectionism over the next 50 years. But is this the right conclusion? There are two key reasons to dispute this view. First, the average tariff does a poor job of measuring the level of protection, particularly when tariffs are non-uniform (Anderson and Neary 2005). Given that many of the tariff revisions following the National Policy appear to have affected the structure of the tariffs rather than the average tariff, the AWT presented in Figure 1 may be giving a misleading view of Canadian tariff history. The raw data suggest that the National Policy and later revisions changed the tariff structure to a greater extent than is reflected by examining the AWT in Figure 1. For example, in their first session of Parliament in 1879, Macdonald's Conservative Party almost doubled the rates of protection for the textile sector and the iron and steel industry, but only increased the general rate from 17.5% to 20%. In the textile industry, cotton duties increased from 17.5% to rate of about 30%, while primary iron and steel went from a range of free to 5% to a range of 12.5% to 17.5%. Given the economic importance of textiles and iron and steel at the time, these revisions may have had a larger effect on the level of protection and the welfare costs of the tariff than implied by the AWT.
Second, protection is not the only possible motive for the observed changes in trade policy. Tariff revenue was the primary source of government revenue in the early post Confederation period. As Figure 1 shows, customs duties provided on average almost 70% of the total Canadian government revenues up until 1887 and declined but still represented about 60% of revenue as late as 1910. Because of the large transportation development debts assumed by the Dominion during this period, the revenue from import duties was essential to the government's nation-building objective, which suggests the Canadian government may have structured protection to maximize revenue rather than primarily protectng domestic industry. This means that the observed changes in trade policy are consistent with two competing explanations; differentiating between them is largely an empirical question, which we examine below.
The tariff regime observed in Canada following the National Policy is also considered to have led to large welfare losses. This conclusion is not based on direct evidence, but is inferred using estimates from later periods. Various studies have suggested that the welfare costs of protection in Canada ranged between $1 billion or 4% of GNP in the 1950s (Young 1957) to 10% of GNP in the 1960s (Wonnacott and Wonnacott 1967) to 4%–8% of GDP in the 1980s (Harris and Cox 1984). Pomfret (1993) comments that the welfare costs of protection in Canada were likely higher than this from 1859 to 1939 because levels of protection were higher. However, this conclusion relies on changes in the AWT that may not accurately reflect changes in protection over time. As described in the following section, using the TRI method, we will directly measure the static welfare costs of the tariff during this period.
3. Measuring Protection and Welfare Loss
To measure protection, we follow the approach of Kee, Nicita, and Olarreaga (2009) and Irwin (2010) and employ Feenstra's (1995)) simplification of the Anderson-Neary Trade Restrictiveness Index.7 With his simplification, the TRI can be written:
where n is the number of goods, is the price of good i, is the ad valorem tariff rate on good i, and is the change in import expenditures on good i resulting from a small change in the price of i.
Direct calculation of the TRI based on equation (1) is not possible, as is unobserved. Kee, Nicita, and Olarreaga (2009) show that equation (1) can be rewritten and expressed solely in terms of observables:
where denotes total imports of good i and is the own-price elasticity of import demand for good i. By re-expressing equation (1) in this manner, it is possible to calculate the TRI directly using data on imports, ad valorem tariff rates and elasticities of import demand.
This approach also provides a simple means for identifying the channels through which changes in tariff policy led to changes in the level of protection. As Kee, Nicita, and Olarreaga (2008) show, the TRI given by (2) can be rewritten:
where is the AWT, is the import-weighted variance in the tariff rate, and is the import-weighted covariance between tariffs and elasticities, where is the share of good i in total imports and is the import-weighted average elasticity. This decomposition makes it easy to identify whether changes in protection are being brought about primarily through changes in the average level of tariffs (), through changes in the set of goods high tariffs are applied to (σ2), or by placing tariffs on goods with relatively inelastic demand (ρ).
In addition, employing the simplified TRI provides a method for calculating the static welfare loss associated with trade policy. As Kee, Nicita, and Olarreaga (2008) indicate, the numerator in equation (2) is equal to twice the static deadweight loss (DWL) arising from protection. Hence, an approximate value of the welfare loss associated with protection can be calculated using
where is the ratio of imports of good i to GDP. Like the TRI given in (2), this equation can be calculated directly using observable data.
Equation (4) can also be manipulated so that the channels driving welfare losses can be identified. Again following Kee, Nicita, and Olarreaga (2008), DWL/GDP can be rewritten:
Like the decomposition presented in equation (3), this equation makes it possible to determine how changes in the level, dispersion, and placement of tariffs affect welfare.
The computation of the TRI and welfare loss measures given by (2) and (4) require data on GDP, imports, ad valorem tariff rates, and elasticities of import demand. Estimates of Canadian GDP used to calculate static welfare loss were obtained from Urquhart (1993, Table 1.1). The trade data are detailed customs data from the Tables of the Trade and Navigation of the Dominion of Canada. These tables report Canadian imports and duties at the product level throughout our period of study. Using product-level customs data is important because it allows us to observe changes to the tariff schedule that might be missed at higher levels of aggregation. Our trade data were taken from these tables at five-year intervals for the years 1870–1910.
We also employ the customs data to calculate the ad valorem tariffs that were applied to each product in each year. These tariff rates were calculated by dividing the value of duties collected by the value of imports for each product that was imported. We employ the calculated ad valorem tariff rates from customs data for two reasons. First, it is computationally necessary to calculate ad valorem rates from the customs data because the Canadian government employed a combination of specific and ad valorem tariffs and we do not have sufficient data to separately incorporate specific tariffs into our measurement of protection.8 Second, employing the calculated ad valorem tariff rate allows us to capture trade policy as it was actually applied on imported goods at the product level.
Table 1 provides an overview of the data used in our analysis. As can be seen in the table and in Figure 1, the import-weighted average tariff rate in Canada remained relatively constant until the implementation of the National Policy. The AWT increased from just over 13% in 1875 to just over 20% by 1880. Tariffs remained high throughout the 1880s, peaking again at just under 22% in 1890, before falling to 17.5% in 1895. The AWT fell to just over 16% by 1900 and remained relatively constant throughout the rest of the period studied.
Table 1. Data Overview
Imports, duty, and GDP are in nominal values. The price index was obtained from Urquhart (1993, Table 1.6).
Two features of Canadian trade flows that stand out in the data summarized in Table 1 are the dramatic increases in both the value of imports and the number of different products imported. These features are illustrated in Figure 2. As the figure shows, the number of goods imported increased rapidly, starting in 1875, but the value of imports did not increase dramatically for another 20 years.
Since the expansion of trade along the extensive margin is such a prominent feature of the data it is important that we are clear about how we measure the number of goods. We have been careful to ensure that we are documenting an actual increase in the number traded products rather than an expansion in the number of categories included in customs records. Rather than relying on the categories of goods used by customs officials of the period, we mapped the products into the Harmonized Commodity Description and Coding System (HS) at the 6-digit level on the basis of descriptor, and we treat each HS6 category as a unique product or good. This ensures we are using a consistent classification in all years that is not affected by the whims of customs agents of the period. Columns 8 and 9 in Table 1 compare the expansion in products and customs categories. During our period of study, there was a similar pattern of expansion in both customs categories and products. However, our approach ensures we are documenting an expansion in goods and not simply capturing the splitting of customs categories.9 This means the expansion at the extensive margin reflects an expansion across the static 6-digit HS categories.10
Although the growth in trade along the extensive margin is dramatic, it should not be considered surprising. As Jacks, Meissner and Novy (2010) point out, economic growth, and the reduction in trade costs contributed to a dramatic increase in trade between 1870 and 1913. This expansion of tradable goods is also reflected in the expansion of production along the extensive margin. Inwood (1995) indicates there was a rapid expansion of entirely new products in Canada during this period. For example, electrical goods did not exist, while rubber and paper products were a very small share of production in Canada in the 1870s. Products in these industries grew significantly over the next 40 years and were becoming widely produced by 1910. With respect to our import data, the types of industries with most of the expansion in the numbers of goods during this period were metals, textiles, chemicals, and vegetable products. Machinery and electrical products also saw a large increase in the number of goods imported. Given falling transport costs, strong economic growth, and the expansion in production along the extensive margin during this period, it is not surprising to find the dramatic growth of tradable goods along the extensive margin depicted in Figure 2. We examine the number of goods by industry in more detail in the results section and present the TRI and number of goods by broad industry grouping.
The TRI and welfare loss measures given by equations (2) and (4) also require estimates of the elasticity of demand for each product. Unfortunately, no such estimates exist for this period and so they cannot be calculated with available data. Hence, throughout most of the analysis we follow the approach of Irwin (2010) and calculate the TRI using modern import demand elasticity estimates in the place of historical estimates.
We obtain our estimates of import price elasticities from Kee, Nicita, and Olarreaga (2008), who estimate elasticities for a wide range of goods and countries (including Canada) for the period 1988–2001. They are classified according to the HS6 classification scheme, making it possible to match Canadian elasticity estimates to our reclassified import data. If a Canadian estimate is not available for a particular HS6 category or the existing estimate is an outlier,11 the average estimate for that category from the rest of the world was used. Products that could not be assigned to an HS6 category (owing to poor or non-existent correspondence between the data and the HS6 classification scheme) were treated as miscellaneous products and assigned the average elasticity value from the Kee, Nicita, and Olarreaga (2008) data.
Using the data from Kee, Nicita, and Olarreaga also provides us with a simple means to examine the robustness of our results to changes in elasticity values. In addition to reporting elasticity estimates, Kee et al. report estimated standard errors. The estimated standard errors allow us to construct a confidence interval for both the TRI and the DWL/GDP using the delta method. Let denote the vector of estimated elasticities that we use in our analysis. Applying the delta method to equations (2) and (4) yields
where is an diagonal matrix containing the variance of each estimated elasticity from Kee, Nicita, and Olarreaga, , as elements, is a vector of length n with typical element , and is a vector of length n with typical element . We employ equations (6) and (7) to construct 95% confidence intervals for our estimates. Because these confidence intervals capture how variation in our elasticities affect our estimates, they allow us to identify how changes in the elasticities influence our results.12
4.1. The Level and Cost of Protection
Our main results for the TRI are presented in Table 2. This table also includes the AWT and other commonly employed measures of protection for comparison. Our results indicate that the AWT understates the level of protection offered by trade policy in all years and that Canada's trade policy at the end of the 19th century was more protectionist than previously believed. Moreover, the correlation between the TRI and AWT is only 0.65 in our sample, meaning that changes in the TRI do not correspond perfectly to changes in the average tariff. Importantly, other commonly employed measures of protection, such as the average tariff on dutiable goods and the share of duty free imports, also fail to capture these changes.
Table 2. Average Tariffs, Trade Restrictiveness, and Welfare Losses
All values in percentages. Standard errors for both the TRI and DWL/GDP are reported in parentheses. These standard errors were calculated by applying the delta method to equation (2).
The difference in protection as measured by the AWT and the TRI can be seen clearly in Figure 3. This figure displays the AWT, the TRI, and the confidence interval (based on equation (6)) around the TRI (the dashed lines) in each year. Clearly, the AWT is less than the TRI in all years, meaning that previous studies have understated protection. Notice that the import-weighted average tariff is a relatively accurate measure of the restrictiveness of trade policy up to 1885, suggesting that the tariff was relatively uniform up to then. The AWT tracks the TRI quite closely until the Tariff Amendment Act of 1887, when there was a large increase in the TRI but the AWT levelled off. Moreover, the AWT lies within the TRI confidence interval in every year up to (and including 1885), but lies outside the confidence interval in the years that follow, meaning the two measures are distinct. This suggests that the structure of the tariff may be changing and that the variance of the tariff across industries my be getting larger. Figure 3 reveals that tariff revisions enacted after the National Policy had a much larger effect on the level of protection than previously thought. This is important because most previous studies of Canada have focused on the National Policy as the long-term determinant of the level of protection available in the Canadian economy. When the AWT is used to measure protection, the National Policy appears to have the largest effect on protection. Our results show that the Tariff Amendment Act of 1887 and Fielding's Tariff of 1897 altered the level of protection by a similar magnitude.
Unlike the National Policy, which largely increased protection through changes in the level of tariffs, later revisions primarily increased protection through changes to the dispersion of tariffs across goods. This can be seen by employing the decomposition of the TRI given in equation (3) and presented in Figure 4. From this figure it is readily apparent that the increase in protection resulting from the National Policy was driven primarily by changes in the level of tariffs. The changes in protection resulting from the Tariff Amendment Act of 1887 and Fielding's Tariff of 1897, on the other hand, were driven primarily by changes in tariff rates for a subset of goods. This can be seen from the increase in the variance of the tariff rate; such changes occur only as the set of goods subject to high tariffs changes. Such changes fit the historical record. As Pinchin (1970) indicates, the Tariff Amendment Act of 1887 and Fielding's Tariff of 1897 largely consisted of revisions to the tariff rates applied to some goods as opposed to schedule-wide changes in tariff rates. Moreover, the covariance between tariffs and elasticities are nearly constant for the entire period.
Table 2 also includes the first estimates of the welfare loss associated with Canadian tariff policy during this period. Our results indicate that Canadian protectionism led to substantial welfare losses. Losses were less than 0.6% of GDP prior to the National Policy, and ranged between of GDP thereafter. Like the TRI, the change in welfare loss was highest following the implementation of the National Policy, but later revisions to the tariff schedule, particularly the Tariff Amendment Act of 1887, also had significant effects on welfare. The ratio of static deadweight loss to GDP (the solid line) and the confidence interval (the dashed lines) are plotted in Figure 5.
Like the TRI, the sources of welfare loss vary by year. Figure 6 presents the results of a decomposition based on equation (5). This figure shows that changes in welfare are primarily driven by changes in the set of goods tariffs are applied to, rather than changes in the level of protection. This can be seen from the large changes in the variance of the tariffs and the covariance between import elasticities and tariffs after 1880; such changes occur only as the set of goods subject to tariffs changes.
4.2. The Structure of Protection
At first glance, the results presented above suggest that Canada was much more protectionist than previously thought. Moreover, this protectionism appears to be consistent with the idea, prevalent in the literature, that tariffs were extended to protect vulnerable Canadian producers (see Easterbrook and Aitken 1965 or Dales 1966 for classic references). Most of the changes in the level and costs of protection result from increases in the variance of the tariff rates across goods, which would arise from preferential tariffs being extended to certain industries to foster domestic production. However, the validity of this conclusion hinges on the exact structure of protection, which we turn to examine now.
First, consider how the structure of protection was affected by the expansion of trade along the extensive margin. Recall that the number of goods imported into Canada quadrupled during our period of study as a result of technological progress and falling trade costs. The first three rows of Table 3 show that the expansion of trade along the extensive margin was an important component of the overall value of imported goods. In fact, by 1890, goods that were not imported at all in 1870 (new goods), comprised nearly half of the value of imports. Although it is not shown in this table, new goods and old goods were approximately equal shares of imports until 1900, when new goods became a larger share of the value of imports than goods that were already imported in 1870. During this period there was very little attrition along the extensive margin; that is, most of the goods that were imported in 1870 continued to be imported in 1910.
Table 3. The Expansion of Trade: Old vs. New Goods
Nj is the number of goods traded in category j, Mj is category j's real imports in millions of 1900 dollars, and TRIj is the TRI for goods category j. Real imports were calculated using the deflators reported in Table 1. Column totals may not equal overall value, owing to rounding.
Table 3 also shows that old goods generally received systematically higher levels of protection than new goods. This structure of protection is what one might expect from a political-economy perspective as older, more established producers were able to obtain protection that was not available to those competing with imports of new varieties. However, when we examine the expansion of trade by commodity type using the classification scheme from Lehmann and O'Rourke (2011), it appears that this pattern is driven primarily by protection on old exotic commodities, as similar levels of protection were granted to old and new goods in both agriculture and industry. This suggests that the Canadian government was not enacting protection for old, established producers, but instead was setting tariffs primarily for revenue purposes.
To analyze how the structure of protection changed over time, we calculated the TRI for a number of broad commodity classes and grouped them into agricultural, industrial, and exotic commodities using the categorization of Lehmann and O'Rourke (2011).13 These results are shown in Table 4, which presents the TRI, number of goods traded, and value of imports by industry for 1870, 1890, and 1910.14 As we showed in the previous section, there was an increase in protection, as measured by the overall TRI from 1870 to 1890, and some decline by 1910. However, there are substantial differences in the level of protection across commodities. Agricultural and industrial commodities had much lower protection than exotic commodities. Given that exotics are goods, such as coffee and tobacco, with limited domestic production, which were primarily taxed for revenue purposes, this suggests that the observed changes in the level of protection were driven not by tariffs favouring specific domestic producers but by tariffs designed to increase government revenue.
Table 4. The Expansion of Trade
Nj is the number of goods traded in category j, Mj is category j's real imports in millions of 1900 dollars, and TRIj is the TRI for goods category j. Real imports were calculated using the deflators reported in Table 1. Column totals may not equal overall value because to rounding. Shares may not add to 100% because of rounding. Goods categories were created by aggregating data to HS2 as follows: Animal & Animal Products: 01-05; Vegetable Products: 06-08, 10-15; Foodstuffs: 16-17, 19-21, 23; Mineral Products: 25-27; Chemicals: 28-38; Plastics & Rubbers: 39-40; Hides, Skins, Leather & Furs: 41-43; Wood & Wood Products: 44-49; Textiles: 50-63; Footwear & Head gear: 64-67; Stone & Glass: 68-71; Metals: 72-83; Machinery & Electrical: 84-85; Transportation: 86-89; Miscellaneous: 90-99; Exotic goods: 09, 18, 22, 24.
Share of Overall Total
Plastics and Rubbers
Hides, Skins, Leather, Fur
Wood and Wood Products
Footwear and Headgear
Stone and Glass
Machinery and Electrical
Share of Overall Total
Share of Overall Total
This conclusion holds even for narrower commodity classifications. Most commodities had similar levels of protection throughout our period of study. Moreover, commodities such as textiles and metals, which are typically viewed as facing the highest level of foreign competition because of the high level of imports of these commodities, do not have systematically higher levels of protection. This provides further support for the idea that trade policy was primarily motivated by revenue considerations, not the protection of producers of favoured commodities.
4.3. The Magnitude of Canadian Protection: Some Context
While the results presented above outline the patterns of protection and welfare loss in Canada from 1870–1910, it is natural to ask whether the level of protection and welfare losses obtained from Canadian trade policy are high. To get a sense of the magnitudes of these estimates, we compare our results to estimates from other studies. Table 5 compares our estimates with those for Canada in later periods. These results show that Canada's recent trade policy was much less restrictive and much less costly than that at the end of the 19th century, when Canada's trade policy was more restrictive and tariff revenue was more important. Our estimates of welfare loss are much lower than estimates from mid-century, but this is likely attributable to differences in methodology.
Table 5. Canada's Protection over Time
All values reported in percentages.
*Reported estimate indicates welfare loss as a fraction of GNP. †Estimate for 1975 using data from Leacy (1983). Import-weighted average tariff values for 1950s and 1960s also constructed using data from Leacy (1983). Import-weighted average tariff values for 1980s constructed using data from CANSIM tables 227-0002 and 380-0034. DWL/GDP estimate for the 1950s from Young (1957). DWL/GDP estimate for the 1960s from Wonnacott and Wonnacott (1967). DWL/GDP estimate for 1976 taken from Cox and Harris (1985). DWL/GDP estimate for the 1980s from Harris and Cox (1984). Estimates for 1990 taken from Anderson and Neary (2005). Estimates for 1988–2001 taken from Kee, Nicita, and Olarreaga (2008).
Table 6 presents protection and welfare loss estimates for both Canada and the United States between 1870 and 1910. From the table it can be seen that Canada's TRI is much lower than that of the United States during this period, meaning that although Canada's tariff structure was more restrictive than previously understood, it was still less restrictive than that of other countries at this time. More important, even though Canada's trade policy was less restrictive, it had higher welfare losses than the United States during much of this period. This was likely driven by the size of the two economies and by the fact that imports were a smaller share of the US economy.
Table 6. Canada vs. the United States
All values are in percentages. US figures are taken from Irwin (2010).
Overall, our results show that Canada's trade policy at the end of the 19th century was more restrictive than previously thought. While the extent to which the average tariff understates protectionism varies with time, the TRI is higher than the average tariff in all cases. This suggests that previous studies that have relied on the average tariff for inference are understating the case for protection, particularly in years when tariff revisions primarily consisted of revisions for some goods, instead of changing the tariff level faced by all goods. Moreover, the results show that the average tariff does not move in perfect correspondence with the TRI; inference based on changes on the AWT alone will be invalid.
5. Robustness Checks
Throughout the analysis presented above, we employed modern elasticity estimates because disaggregated period elasticities are unavailable and cannot be calculated using existing data. While this allows us to match elasticities to each product in our data, it means that some of our elasticities lie outside the range of historical elasticity estimates for Canada: −2.8 to −0.2 (Marquez 1999). This can be seen in Figure 7, which plots the distribution of modern elasticities and range of elasticity estimates. Clearly, most of our elasticities fall within the historical range, but there are a number of outliers that may be driving our results. Moreover, most of our elasticities are at the upper end of the historical range of elasticities, meaning the distribution of elasticities across HS6 categories may be driving our results.
To address these issues, we proceeded in two ways. First, we examined the robustness of our estimates to outliers in the modern data by recalculating both the TRI and DWL/GDP using truncated elasticity data from Kee, Nicita, and Olarreaga (2008), where the data were truncated to ensure that all of the elasticities lie within the historical range. That is, we truncated the elasticities for each HS6 category so that the maximum value in the data is −0.2 and the minimum value in the data is −2.8. Second, we examined the robustness of our results to an alternative distribution of elasticities across HS6 categories using simulations based on the historical range of elasticities. The simulation procedure we employ is inspired by Irwin (2010), who suggests that estimated trade elasticities are relatively stable over time. Hence, a reasonable estimate of the TRI should be similar to a TRI calculated with a random draw of elasticities from the historical interval. Given that we do not know the shape of the historical distribution of elasticities, for our simulations we assumed that all historical elasticities are uniformly distributed over the interval . Each product was assigned a random elasticity drawn from the interval and the corresponding TRI was calculated. This procedure was repeated 10,000 times to create 10,000 distinct measures of TRI and welfare loss for each year. The results of these procedures are displayed in Figures 8 and 9.
Figure 8 displays the AWT, the original TRI and associated confidence interval as calculated above, the truncated TRI, and the mean simulated TRI. Like the original estimates, both the truncated and simulated TRIs are higher than the AWT in all years, and are imperfectly correlated with the AWT. Moreover, the Tariff Revision of 1897 still leads to a large increase in protection and Fielding's Tariff of 1897 leads to a decrease in protection in both cases, meaning our finding that later tariff revisions were important determinants of protection is robust to our choice of elasticities. However, both the truncated and simulated TRIs are much higher than the calculated TRI in all years because we have reduced the import weighted covariance between tariffs and elasticities by restricting the data to the historical range. Hence, while our choice of elasticities does not affect the pattern of protection over time, it does influence our estimate of the level of protection. As such, our calculated TRI may be considered a lower bound on the true level of protection.
The results of our welfare loss simulations paint a similar picture. Figure 9 displays the original welfare loss estimates and confidence interval presented above, the welfare estimates using the truncated Kee, Nicita, and Olarreaga (2008) data, and the mean simulated results. Our results using the truncated data display a trend similar to our original results, but are lower in nearly every year. This is unsurprising given that our welfare loss estimates depend explicitly on the average elasticity in the data (recall equation (5)); the truncated data has a lower average elasticity than the original data by construction. Our simulated results display a general trend similar to our calculated results, but are higher in every year because the average of the simulated elasticities is higher than the average elasticity in our original data. However, both the truncated and simulated estimates lie within the original confidence interval, suggesting that our original estimates reflect the true welfare losses arising from Canadian protectionism.
In order to ensure we use a consistent definition of products in all years, we were forced to aggregate the raw trade data from the Tables of the Trade and Navigation of the Dominion of Canada to the HS6 level. As we discussed previously, this involved matching our data to each HS6 category on the basis of descriptor. As can be seen in Figure 10, this leads to a substantial reduction in the number of goods reported in all years, which in turn leads to a natural concern: the reduction may drive our results by artificially manipulating the variance in the tariff rate in all years. To investigate whether aggregation is driving our results, we recalculated TRI and DWL/GDP using both disaggregated and aggregated data for all years. Throughout, we assume a constant elasticity of import demand for all goods; doing so allows us to focus on differences in protection created solely from differences in how the trade data are aggregated.
Figures 11 and 12 show how aggregation changes our measurement of protection and welfare loss. In Figures 11 and 12, the measures for disaggregated data and data aggregated to HS6 are nearly identical. This means our main analysis is not a product of aggregation.
One other aspect of Figures 11 and 12 is worth noting. While the TRI and DWL/GDP are nearly identical when calculated using disaggregated or HS6 data, higher levels of aggregation (HS3 and HS1) clearly understate the true level of protection and welfare loss. While this point has been made previously by Irwin (2010), our results show that the degree to which calculations using highly aggregated data understate protection or welfare loss is not constant. This means that studies using highly aggregated data in their analysis may fail to capture the full effect of the tariff structure on protection and welfare loss.
We examine the trade restrictiveness and welfare consequences of trade policy in a small open economy during the first wave of globalization (1870–1913) – a time of a historic transformation of trade policy from revenue- to protectionist-based policy and a time of unprecedented growth in trade. To examine the effects of trade policy on protection and welfare we construct a simplified version of the Anderson-Neary Trade Restrictiveness Index for Canada at the end of the 19th and beginning of the 20th century. We find that the historiography of Canada's trade policy is very different when considered through the lens of the TRI than when only the AWT is considered. The evidence based on the TRI reveals that Canada's trade policy at this time was more restrictive than previously thought. The AWT understates the level of protection by as much as 13 percentage points when compared with the TRI. However, we show that the AWT does a reasonable job of measuring the restrictiveness of trade policy when changes in the tariff are relatively uniform across products, as was the case in Canada up to 1885. Thereafter, as the variance of the tariff grew, the AWT became a poor measure of the restrictiveness of trade policy. Moreover, we compute the first estimates of the static welfare losses associated with tariff policy at this time and find the deadweight losses from protectionism amounted to be of GDP. These welfare costs based on the TRI are dramatically lower than is found in the literature, where the welfare costs are inferred from estimates based on later periods.
We present compelling evidence suggesting a re-evaluation of the typical history found in the literature on Canadian protectionism at the end of the 19th century. Most studies focus on the National Policy marked as the key driver of protectionism because it corresponds to the largest increase, in the AWT prior to the Great Depression; later revisions to the tariff schedule are ignored because they do not lead to similar changes in the AWT. The evidence presented here leads one to conclude that these revisions also played an important role in determining the level of protection. In particular, the Tariff Amendment Act of 1887 and Fielding's Tariff of 1897 altered the level of protection a similar amount. Hence, while the National Policy marked the beginning of Canadian protectionism, subsequent revisions to the tariff schedule also had a large impact on the level of protection offered by trade policy.
Moreover, we find that most of the increase in trade restrictiveness is from high tariffs on exotic goods with little to no domestic production. From 1875 to 1880, the higher level of trade restrictiveness was primarily driven by an increase in the average tariff across goods. After 1880, tariffs on agricultural and industrial goods increased, but most of the observed increase in the TRI is due to increased tariffs on exotic goods – goods that were not produced at home and had inelastic demand. This is not the protectionist trade policy that is presumed based on examining the average tariff. The tariff structure appears to have been designed to maximize revenue and is consistent with the observation that the tariff continued to be the most important revenue source throughout the period.
Our results are robust to the elasticities used in our analysis. While the elasticities we employed were estimated for a much later period, our robustness results show that the trends in protection and welfare loss are not sensitive to changes in the distribution of elasticities. However, the level of both the TRI and the DWL/GDP depend on the distribution; our calculated results are less than our simulated results in all years. This suggests that our calculated results represent plausible lower bounds for the magnitudes of protection and welfare loss.
Other measures that have been employed include the effective rate of protection, the coefficient of variation of tariffs and the non-tariff-barrier coverage ratio. Of these, the AWT is the most commonly employed because it is easily calculated from aggregate data by dividing the value of total duties collected by the value total imports.
As Irwin (2010) points out, the AWT suffers from four key limitations that make it a poor measure of the level and welfare costs of protection: (i) it is downward biased; (ii) it ignores dispersion in duties across goods, causing welfare costs to be understated; (iii) it lacks any economic interpretation; and (iv) it does not reflect the impact of non-tariff barriers on trade. For further discussion of the problems associated with using the AWT and other common measures of protection, see Anderson and Neary (2005).
Young (1957), Wonnacott and Wonnacott (1967), and Harris and Cox (1984) estimate the costs of trade policy in the 1950s, 1960s, and 1980s respectively. These estimates provide the basis for inference in earlier periods. The typical logic used in this inference is as follows: the AWT was higher between 1870 and 1910 than between 1950 and 1980, thus the welfare losses during this period must also be higher.
The TRI was pioneered by Anderson and Neary (1994). For a detailed overview of the theory underlying the TRI, see Anderson and Neary (2005).
For a discussion of the transition to protectionism around the world during this period, see O'Rourke (2000).
For an overview of the expansion of world trade during this period, see Estevadeordal, Frantz, and Taylor (2003).
It is important to note that by employing the simplified version of the TRI, we are focusing on the first-order effects of trade policy and ruling out cross-price effects and other general equilibrium interactions. In principle, this means our estimates may understate the true level of protection (Lloyd and MacLaren 2010). General equilibrium interactions could be captured using a computable general equilibrium (CGE) model, but Canadian data from the period of our study do not provide sufficient detail to allow for CGE modelling. More important, when calculated using CGE methods, the TRI is highly sensitive to the specification of the model (O'Rourke 1997), meaning any second-order effects will depend explicitly on our modelling choices. By using the simplified TRI and focusing on the first-order effects of trade policy, we are able to abstract from this issue.
Although this approach is common practice, it means that changes in tariff levels for some goods will reflect changes in import prices in addition to changes in trade policy. Price changes have been shown to have large effects when specific tariffs are widely used (Irwin 1998); however, in our case their effects should be small. Canada employed specific duties sparingly throughout our period of study.
For example, in 1870 the customs data contain one line for carriages, which we classified in HS category 871680. In 1910, the customs data contain nine lines for carriages, all of which which were all classified in the HS category 871680.
In Section 'Results' we discuss the robustness of our results to aggregating our data to the HS6 category.
We consider elasticity estimates to be outliers if they are greater than two standard deviations away from the average value in the Kee, Nicita, and Olarreaga (2008) data.
In Section 'Results' we present additional evidence documenting the robustness of our results to employing modern elasticity estimates.
We assigned the broad commodity classes to the three categories on the basis of descriptor, using the classification scheme from the appendix of Lehmann and O'Rourke (2011).
Table 4 includes only these three years because of space constraints. Results for all years can be calculated using the materials provided in the online appendix.