5 Determinants of the industrial structure of production across regions

View the most recent version.

This section asks whether the differences in the export intensity of different areas are related to differences in several structural characteristics related to productivity—plant size, product diversity and production-run length. It investigates this issue using multivariate analysis. In doing so, it also takes into account other variables that are believed to influence industrial organization or structure. Included in these are export intensity and regional binary variables for Atlantic Canada, Quebec and Western Canada. Ontario is the excluded region.

Structural differences may also reflect differences in the sizes of local markets, either in terms of intermediate demand from other manufacturing firms or final demand. This possibility is investigated by including a measure of the size of the manufacturing sector within each census division as measured by employment and whether the census division is urban. We also include in the model an interaction term between employment and whether the census division is rural or urban.

In order to investigate whether differences in the response to trade liberalization have had different effects across the manufacturing sector, we divide the manufacturing sector into five groups—natural resource-based, labour-intensive, scale-based, product-differentiated and science-based industries. Labour-intensive industries are those with low capital–labour ratios, low wages and considerable import competition. Natural resource-based industries are those that have relatively low value added relative to their inputs—industries like food processing that depend upon agricultural materials. Scale-based industries have large plants, high capital–labour ratios and economies of scale. Product-differentiated industries have high advertising-to-sales ratios and produce a large number of products per plant. Science-based industries have a large white-collar workforce and focus on research and development.⁶ In the regressions, each sector enters as its share of overall manufacturing employment in the census division, with natural resource-based industries as the omitted category.

Finally, the export intensity variable is also included as an interaction term with certain of the geographic binary variables—regional or urban—to test whether its effect is greater or less across regions. This serves as a more formal test of whether location relative to markets influences the industrial organization of production within geographic units.

5.1 Plant size

We first examine the relationship between higher levels of export intensity and average plant size using a series of econometric models estimated using a panel of census divisions (see Table 6). Models 1 through 3 are based on the between estimator and, as such, only take into account the effect of cross-sectional variation. Model 4 is estimated using a fixed-effects estimator, which only takes into account the effect of changes over time.⁷

Model 1 includes regional binary variables and controls for industrial structure. Examining differences across regions, we see that over the period the average plant size in census divisions found within Atlantic Canada, Quebec and Western Canada are significantly smaller than those in Ontario, which accords with the basic tabulations (see Table 3).

We then include a measure of the size of the manufacturing sector within each census division and whether the census division is rural or urban. We also include in the model an interaction term between employment and whether the census division is rural or urban. When this is done (Model 2), it is apparent that higher levels of manufacturing employment are associated with larger plant sizes in rural census divisions, reflecting the joint effect of larger downstream intermediate demand and possibly larger local final demand markets, which are likely correlated with the size of the manufacturing sector. In urban census divisions, plant size is also larger since Urban has a positive, albeit weakly significant, coefficient; however, here the relationship between employment and plant size is absent. Merely being part of an urban economy has a stronger effect on plant size than the size of the manufacturing sector within the urban census division.

Also included in Model 2 is the export intensity of census divisions, thereby permitting us to examine the impact of a region's being integrated into the world and North American economies. The coefficient on the export intensity variable is positive and highly significant. Export intensity and the average size of plants in census divisions are positively related. An increase in export intensity from 0.25 to 0.50 is associated with close to a $9 million increase in output, which is a non-trivial amount when compared with the average size of plants.

In Model 3, interaction terms between export intensity and regions are added. They indicate that the relationship between higher levels of export intensity and plant size is much weaker outside of Ontario, which is consistent with our theoretical expectations. It should be noted that inclusion of variables that allow the effect of export intensity to vary across regions eliminated any significant effect of the regional binary variables on plant size. That is, the consistent differences in the average plant size of census divisions across regions we observed in Table 3, and that are reflected in our parameter estimates in Models 1 and 2, are accounted for by the much weaker effect of trade on plant size outside of Ontario.

Also included in Model 3 is an interaction term between the urban binary variable and export intensity. Its coefficient is insignificant, indicating firms in urban areas, whose plant sizes tend to be larger, adjusted their plant sizes no more or no less than those in rural areas.

There remains the possibility that these results are due to unobserved fixed effects that are correlated with both average plant size and export intensity. Model 4, which corresponds to Model 3, is estimated using a fixed-effects model that controls for both cross-sectional and time- series effects. Time binary variables are also included to allow for long-term time trends in average plant sizes. Their inclusion allows for the possibility that long-term common time trends underlie both changes in export intensity and changes in plant size.

Table 6
Plant size models

The results presented in Model 4 are similar to those of Model 3. The fixed-effects model purges the results of cross-sectional variation and leaves the estimates mainly to capture the effect of changing export intensity over time. It is therefore noteworthy that export intensity is still significant in this formulation, thereby demonstrating that the changes in export intensity tended to increase plant size.

Perhaps the most significant conclusion to be drawn from Model 4 is that the effect of exports on plant sizes is insignificant outside of Ontario and Quebec. The coefficients on export intensity for Atlantic Canada and Western Canada are not significantly different from zero. For Quebec, the effect of export intensity on plant size is more muted than for Ontario, but remains statistically significant. The impact of growing exports on the average plant sizes of census divisions was limited to census divisions in Central Canada, with by far the largest effect being in Ontario. These results therefore show that adaptation to trade liberalization occurred primarily in Central Canada

5.2 Commodities per plant

It has been argued that trade barriers may not only lead to plants that do not fully exploit economies of scale, but that they may also lead to plants producing too many varieties of products.⁸ By doing so, plants can increase overall production to exploit economies of scale more fully at the plant level, all the while losing some of the scale economies that result from long production runs. Essentially, a trade-off occurs between suboptimal plant size and suboptimal production-run length that results from product packing at the plant level—with the outcome depending upon on the relative size of cost penalties in each area.

Baldwin and Gu (2008) demonstrate that declining tariff barriers will result in fewer products being produced within plants. We investigate here how this tendency varies across Canadian regions.

We regress the same set of explanatory variables used in the plant size model against the number of commodities per plant—the numbers equivalent averaged across plants within each census division (see Table 7). Model 1 is the simplest and includes only controls for industry and a set of regional binary variables to account for level differences across regions. Across regions, the average number of commodities per plant in Atlantic Canada, Quebec and Western Canada is statistically not different from that in Ontario.

Model 2 adds the level of export intensity across census divisions, as well as controls for the size of regions. Theoretically, market size is positively associated with the number of commodities (products) produced per plant (see Baldwin and Gu 2008). Therefore, our expectation is that census divisions with larger markets, as proxied by these variables, would have higher average counts of commodities per plant. However, controls for the size of the manufacturing sector in each census division and whether census divisions were rural or urban were insignificant.

Consistent with expectations, the coefficient on export intensity is negative and significant. Census divisions that are more export-intensive have fewer commodities per plant on average. The differences across regions in terms of product diversity on a cross-sectional basis are strongly related to the degree to which they are integrated into the North American market.

Table 7
Numbers-equivalent¹ commodities per plant model

To allow for a differential effect of export intensity on the average number of commodities per plant across regions and rural and urban areas, a set of regional interaction terms is included in Model 3. With the possible exception of Atlantic Canada, none of the regional, or the urban, interaction terms were significant, suggesting the effect of export intensity on the average number of commodities per plant did not vary systematically across geographic regions or urban- rural classes.

Model 4 is a fixed-effects model that corresponds in functional form to Model 3 that captures only changes over time (i.e., that has the cross-sectional effects removed). The effect of export intensity in Ontario was not significantly different from zero. The same was true of the other regions. As this indicated no significant difference in the effect of export intensity across regions, we estimated a variant of Model 4 with the interaction terms removed. The coefficient on export intensity in this model was also not significantly different from zero. Although there was a general decline in the commodity diversity of plants as indicated by the year binary variables, there was no discernable independent effect of changes in the export intensity of census divisions over time. This may, of course, simply mean that it is difficult to separate the impact of export intensity over time and the time trend reduction in commodity diversification—that they were both being caused by the same underlying causes or that export intensity was increasing with a trend and directly driving specialization.⁹

5.3 Length of production run

Previous analytical and empirical work suggests falling barriers to trade will result in increases in the length of production runs. In the models presented below, we test this further by looking at how variations in the export intensity across regions affect the average length of production runs within each region.

Production-run length is derived as the ratio of plant size divided by number of products produced (derived from our numbers-equivalent measure). Thus, we expect that the same variables that were related in a systematic way to plant size and number of commodities will also be related to production-run length. We therefore include binary variables for four broad regions (Atlantic Canada, Quebec, Ontario and Western Canada) and interact these with export intensity to test the differential effect of export intensity across regions. In light of our previous findings, our expectation is that higher levels of export intensity will have the strongest effect on Ontario and Quebec and the weakest effect on Western Canada and Atlantic Canada, which are much more isolated from North American markets.

The results are presented in Table 8. In Model 1, we include only controls for industrial composition and region. After controlling for the industrial composition of regions, the average length of production runs for census divisions in Atlantic Canada, Quebec and Western Canada was less than in Ontario.

In Model 2, we include variables that control for local market size, both in terms of the manufacturing sector and (indirectly) final demand. They show that production-run length increases with the size of manufacturing employment in rural areas and is typically higher in urban areas.

To test the effect of export intensity on the average length of production runs, Model 2 also includes the export intensity of census divisions. As with the analogous plant size model, census divisions with higher levels of export intensity also tend to be characterised by plants with longer production runs.

Table 8
Length of production run

Model 3 adds interaction terms between export intensity and regional binary variables. Regions that are more export intensive have average production runs that are longer than regions that are less export intensive. When regional interaction terms are included, substantial differences are seen to exist across regions. In Ontario, which is the excluded region in the regression, the effect of export intensity on average production-run length is significantly higher than in Atlantic Canada, Quebec or Western Canada. As with the analogous plant-size model, the interaction term between export intensity and urban was insignificant.

Shifting from the between to a fixed-effects estimator (Model 4) to purge the relationship of its cross-section dimension and examine just time-series dimension provides similar results. The main differences are that the effect of export intensity on the length of production runs is about halved. Nevertheless, it is still statistically significant. The effect of export intensity outside of Ontario, which was positive in Model 3, is not significantly different from zero in Model 4. It is also noteworthy that export intensity for Quebec has a positive effect on plant sizes, but the effect for the length of production runs is not significantly different from zero.

5.4 Exporters compared with non-exporters

The differences in regional performance described previously are derived from averages in each region that are derived from heterogeneous groups of firms. Baldwin and Gu (2008) predict that exporters and non-exporters may differ in terms of their reaction to increasing trade liberalization. In particular, a decline in tariffs is expected to lead to a decline in average plant size in non-exporters but has an ambiguous effect on existing and new exporters. The plant numbers equivalent should fall for non-exporters but the effect on exporters is ambiguous. A decline in trade costs increases the length of production runs of individual products at existing and new exporters. It reduces the length of the production run of individual products at non- exporters.

In order to test whether these predictions are confirmed across regions and whether there are differential responses of each group across regions, the model was estimated for plant size, number of products and the length of production run for exporters as opposed to non-exporters. The results are reported in Table 9. Only the fixed-effects Model 4 is reported, since we are primarily interested in how changes over time had a differential impact on the structural characteristics. The same set of independent variables that were used in Tables 6, 7 and 8 are employed but only the results on export intensity are reported in Table 9. Export intensity refers to the average export intensity of the census division; the structural attributes refer to the characteristics of exporters and non-exporters within the census division.

For non-exporters, changes in the trade intensity of the region are associated with either no change in plant size for Ontario or a reduction in plant size for the other regions, as was predicted by Baldwin and Gu (2008). The number of commodities falls for Ontario and most of the other regions—though these other regions do not experience a significantly lower reduction than does Ontario. The length of production run decreases as predicted, except in Ontario. In summary, the expected reactions for plant size and commodities are found across most of the outlying regions, but the changes are smaller in Ontario.

For exporters, the model had ambiguous predictions on the effect of trade liberalization on plant size; though at a national level, the empirical results indicated that exporters increased their plant size (Baldwin and Gu 2008). The regional estimates show that reactions to increased trade integration were generally positive but much larger in Ontario compared with both Atlantic Canada and Western Canada, though about the same as in Quebec. For length of production run where the model predicted increases for exporters, the impact of export intensity is once again positive across all regions, but greatest in Ontario. Contrary to the model's predictions of increasing specialization for non-exporters, the predictions for exporters is ambiguous, and the coefficients that relate changes in the number of commodities and trade integration are not significantly different from zero, nor are there significant differences across regions.¹⁰

Table 9
Exporting and non-exporting plant fixed-effects models

Not only do these results provide broad confirmation for the predictions of the model, they show that the reactions generally differed between Canada's industrial heartland and the other regions. Whether they involve the reactions of exporters or non-exporters, cross-regional differences reinforced the averages that were previously reported. Outside the heartland, non-exporters were likely to see their plant size reduce more and exporters were more likely to see their plant size increase less in response to an increase in trade intensity. Non-exporters saw the length of their production runs fall by larger amounts and exporters saw their length of production runs increase by smaller amounts.

 

⁶ For a description of the sectors, see Baldwin and Rafiquzzaman (1995).

⁷ We also estimated a first difference model to capture changes over time and obtained results that were qualitatively similar to those reported here.

⁸ Earlier work by Safarian (1966) for the 1960s noted that multinationals operating in Canada had short production runs. Scherer et al. (1975) compared several Western countries in the 1970s and found that Canadian production runs were generally lower than those in the United States.

⁹ This interpretation receives some support when the time binary effects are removed from Model 4. When this is done, export intensity becomes highly significant and the value for Ontario is significantly higher than for other provinces.

¹⁰ These results accord more closely with the predictions if the time series variables are removed.