Journal Issue
Share
Article

Canada: Selected Issues

Author(s):
International Monetary Fund. Western Hemisphere Dept.
Published Date:
July 2018
Share
  • ShareShare
Show Summary Details

A CLOSER LOOK AT LABOR PRODUCTIVITY IN CANADA1

A. Recent and Historical Trends

1. The strong cyclical upturn helped boost labor productivity growth in Canada last year.2 As oil prices stabilized and aggregate demand was boosted by accommodative monetary and fiscal policies, output growth rebounded strongly to 3 percent in 2017 (top among the G7). With output growth outpacing labor input growth, labor productivity also rebounded strongly to 1.7 percent.

Labor Productivty Growth

(2010=100; underlying data are constant local currencies)

Sources: Haver Analytics

2. Nonetheless, from a long-term perspective, Canada’s productivity remains much weaker than the leading G7 economies. OECD estimates of labor productivity (at constant purchasing power parity) suggest that Canada enjoyed the second highest productivity level after the U.S. in the 1970s. However, labor productivity has since grown more slowly in Canada than in other economies, and as a result, Canada has fallen behind France and Germany, and its productivity gap with the U.S. has widened from around 12 percent in the 1970s to 24 percent today.

G7 Economies Labor Productivity

(Constant 2010 PPPs in U.S. dollar)

Source: OECD.Stat

3. Looking forward, an aging population will put a drag on growth in Canada. Statistics Canada projects that the share of working age population (ages 15–64) would fall from around 67–68 percent today to 60 percent in 2040 (under its medium-growth scenario). This implies that growth of the working age population would decelerate from 1 percent to less than 0.5 percent a year. Were labor productivity growth to stay the same (an average of 1.2 percent in 2010–17), GDP growth would decelerate to 1.7 percent a year, significantly below the average growth rate of 2.2 percent over the past one and a half decades. To reverse the negative implications of an aging population, Canada will need to boost labor productivity.

Canada: Long-term Growth

(Contribution to GDP growth in percent, annual average)

Sources: Statistics Canada; Haver Analytics; and IMF staff calculations.

4. Using data from Canadian firms, as well as industry-level and provincial data, we analyze labor productivity and business investment trends. We ask the following three questions.3

  • Q1. What is restraining labor productivity growth? We look at labor productivity trends using a shift-share approach and growth accounting approach (Section B).

  • Q2. How has firms’ productivity been associated with investment? We estimate markup measures as proxy for firm-level productivity and examine how markups are associated with business investment (Section C).

  • Q3. How important is the technology and productivity diffusion channel from the U.S. to Canada? The U.S. is the most important trade partner for Canada. Many studies show that trade is an important channel for technology diffusion (IMF, 2017). We attempt to find the evidence of spillover effects from the U.S. to Canada (Section D).

B. What is Restraining Labor Productivity Growth?

5. The structure of Canadian industry has evolved markedly over the past decades. As in many other advanced economies, Canada’s industry structure has shifted towards the service sector, with the share of manufacturing industry halved from over 20 percent in the 1960s to about 10 percent today. Given that the growth of labor productivity in manufacturing industry has been much higher than most of other industries, its shrinking share could explain part of the long-term declining trend of labor productivity growth.

6. To analyze if aggregate productivity growth has been driven by specific sectors (“within effects”) or changes in the sector composition of employment (“shift effects”), we employ a shift-share decomposition analysis (see Box 1). We compare the pattern of labor productivity growth (excluding the public sector) in Canada with that in the U.S., and split the sample before and after 2000 when the stagnation of productivity growth became evident.

  • Overall, the shift-share analysis indicates that the productivity slowdown mostly results from “within effects,” rather than “shift effects.”

  • The “within effects” contribution of the mining, oil and gas sector fell from 0.3 percentage points (annual average) in 1984–2000 to –0.1 percentage point in 2000–16 (Table 1). Similarly, the “within effects” contribution of the manufacturing sector decreased sharply from 0.5 percentage points in the first period to 0.2 percentage points in the second period. In both sectors, “shift effects” were small or negative, reflecting a decline in employment shares.

  • In the U.S., the contribution of the mining, oil and gas sector and the manufacturing sector has also fallen, but less so than in Canada.

  • Encouragingly, “shifts effects” have turned positive in Canada in 2000–16 (from -0.1 percentage points in the first period to 0.2 percentage points in the second period). This implies that more labor was allocated to sectors with higher labor productivity growth, contributing to higher aggregate-level productivity growth.

Table 1.Productivity Growth Decomposition(Percentage points)
TotalAgricultureOil & MiningManufacturingConstructionUtilitiesOther sectors (incl. services)
Canada
1984–2000
Productivity growth within sector1.710.020.320.540.040.010.79
Employment share shifts–0.18–0.02–0.18–0.14–0.05–0.010.22
Cross-sectoral–0.140.00–0.10–0.080.000.000.04
Total change in productivity1.400.000.050.31–0.01–0.011.05
2000–2016
Productivity growth within sector0.510.04–0.140.16–0.050.000.50
Employment share shifts0.16–0.020.20–0.390.180.000.19
Cross-sectoral–0.13–0.01–0.06–0.06–0.020.000.02
Total change in productivity0.550.010.00–0.290.110.000.72
United States
1984–2000
Productivity growth within sector1.870.060.220.67–0.020.070.87
Employment share shifts0.14–0.02–0.09–0.210.06–0.050.44
Cross-sectoral–0.27–0.02–0.13–0.190.00–0.020.08
Total change in productivity1.730.020.010.270.030.001.40
2000–2015 1/
Productivity growth within sector1.070.020.000.46–0.08–0.030.71
Employment share shifts0.120.000.10–0.25–0.030.000.30
Cross-sectoral–0.090.000.00–0.130.010.000.04
Total change in productivity1.110.020.090.08–0.11–0.031.06

Comparable US 2016 data not available.

Sources: OECD Stat. and IMF staff calculations.

Comparable US 2016 data not available.

Sources: OECD Stat. and IMF staff calculations.

Decomposition of Labor Productivity Growth

Source: OECD and staff calculations.

Note: Comparable US 2016 data not available. Total growth divided by period length shown.

7. Using growth accounting, we also ask whether the weakened productivity growth reflected multifactor productivity or capital accumulation. Growth accounting estimates by Statistics Canada show that the deceleration in labor productivity growth in the 2000s mainly reflected a decrease in multifactor productivity (in part due to the impact of the 2008–09 global financial crisis), while capital intensity (defined as capital stock per total hours worked) increased its positive contribution to overall labor productivity growth. Since 2010, multifactor productivity growth has recovered, but instead, capital intensity growth has slowed markedly, with its contribution to overall productivity growth halved. The decline in capital intensity growth was broad based: all industries (except for few industries, such as mining, oil and gas, and transportation) had lower capital intensity growth, with some (such as manufacturing, construction, retail and finance) incurring negative growth.

Canada: Sources of Labor Productivity Growth 1/

(Percent, annual average)

1/ The business sector, which includes the whole economy less public administration, non-profit institutions and the rental value owner-occupied dwellings.

2/ Growth in capital stock per total hours worked.

3/ Growth in the educational attainment and work experience of the labor force.

Sources: Statistics Canada CANSIM 383–0021; and IMF staff calculations.

Capital Intensity Growth by Key Industries

(Period average, percent)

Sources: Statistics Canada CANSIM Table 383–0021

1/ The size of bubbles indicate the size of GDP.

C. How Does Firm Productivity Relate to Investment?

8. Several factors could explain the slowdown in business investment growth. Among them, aggregate demand, financial constraints, and policy uncertainty are generally viewed as the most important drivers of the dynamics of business investment (see for example, IMF, 2015). In the aftermath of the global financial crisis, a recovery in economic activity in Canada’s trading partners (most importantly, the U.S.) was weak. Moreover, although U.S. economy finally began to gain momentum around 2014, global oil prices dropped sharply, hitting the Canadian economy. More recently, heighted trade tensions in the North America region may have contributed to dampening business investment.

9. We look at another possible factor: Canadian firms’ productivity. Because measuring productivity using micro firm-level data has some challenges (for example, due to the lack of data for firms’ output prices), we instead estimated firms’ markups (defined as price over marginal costs), following the empirical framework developed by De Loecker and Warzynski (2012). As well argued in De Loecker and Warzynski (2012), firms’ productivity and markups are closely associated, as productive firms tend to set higher markups.4 Due to data constraints, our analysis below focuses on markups for large Canadian (non-energy) firms using Worldscope database (see Appendix I for more detail). Note that energy firms are excluded from our analysis because we are mainly interested in the productivity trend in the non-energy sector.

10. We found some evidence of close correlations between markups and business investment in our sample data. It appears that markups and business investment are positively associated, but less so when markups are too high. This could suggest that highly profitable firms with too high markups might not be motivated to invest. Accordingly, positive relations between markups and business investment are not necessarily monotonic, but nonlinear. We test this hypothesis in the empirical analysis below.

Business Investment and Markup, 2015

(In logarithm)

Source: Authors’ estimates.

11. Furthermore, consistent with firm-level data, some industry-level indicators suggest waning competitiveness of Canadian industries.

  • The measure of firm entry and exit rates in Canada (the number of firms that start business or go out of business as a percent of total firms) shows a downward trend. Shortly after the global financial crisis, the entry and exit rate recovered somewhat, but has resumed a downward trend in most sectors since 2012. There is no clear consensus on why entry and exit rates have fallen, but the decreased entry and exit may signal weakened firm dynamism.5

  • The import share of Canadian goods and services in the U.S. market has been on the steady decline. The share of Canadian non-energy goods and services in the US import market has been steadily on the decline, with the share of China steadily rising. Using a shift-share analysis, Barnett and Charbonneau (2015) analyzed that the loss of Canada’s market share was attributed more to weakened competitiveness than to a change in US consumers’ preference. 6

Business Sector: Entry and Exit Rate

(Percent, excludes energy and mining companies)

Source: Statistics Canada CANSIM Table 527–0001

Business Sector: Entry Rate by Key Sector

(Percent)

Source: Statistics Canada CANSIM Table 527–0001

U.S.: Imports from Canada, Mexico, and China

(Percent share in the US market, three-quarter moving average)

Sources: U.S. Census Bureau; and Haver Analytics.

12. We estimate Tobin Q’s type investment models to examine whether these markup and industry competitiveness measures could be considered as part of business investment determinants. We embed markups and industry level competitive indicators, such as entry and exit rates and China’s import penetration rates, in a standard Tobin Q’s type investment model (Appendix II). The sample period is 1997–2016. The estimation methodology is OLS with fixed effects, and regression results are reported in Table 2.

  • The coefficient on markups is positive and highly significant, indicating positive relations between markups and business investment (Model 1). The positive association is stronger for exporters than domestic firms (Model 2 and 3). For exporters, Model 2 suggests that a 10 percent increase in markup is associated with 0.6 percentage points increase in the investment to capital stock ratio for non-energy exporters.

  • Evidence of nonlinearity. Model 4 (all firms) and Model 5 (exporters) show that the coefficient on the quadratic term of the markup variable is negative, indicating that if markup is too high, business investment would decline. This said, this result should be treated as tentative because the estimated coefficients are not statistically significant.

  • Industry level competitiveness and business investment. The coefficient on the entry and exit rate is positive but not statistically significant (Model 6).7 We also tried FDI inflows (as a percent of GDP)—our hypothesis is that higher FDI inflows directly boost business investment but also indirectly by enhancing competition. But the estimated coefficient is nearly zero and not statistically significant (Model 7). We thus attempted to interact these two variables and found that the coefficient on the cross term is positive and highly significant. This would suggest that higher FDI combined with entry and exit rates are associated with higher business investment, possibly through positive competition effects (Model 8).

  • Impact of China’s import penetration. In Model 9 (all firms), the coefficient on China’s important penetration is negative and highly significant: a one percentage point increase in China’s share in total U.S. imports led to a 0.2 percentage points decline in Canada’s investment ratio. This competition effect is more evident for exporters (Model 10) than for domestic producers (Model 11).

Table 2.Results of Firm Level Panel Regressions 1/(Excluding energy firms)
Dependent variable: log investment to capital
BaselineNon linear effectsEntry and Exit Rate and FDI InflowsChinese Imports
All Model 1Exporters Model 2Domestic Model 3All Model 4Exporters Model 5All Model 6All Model 7All Model 8All Model 9Exporters Model 10Domestic Model 11
Log markup0.071***

(0.03)
0.060**

(0.03)
0.148

(0.11)
0.087***

(0.03)
0.082**

(0.03)
0.071

(0.07)
0.073

(0.06)
0.079

(0.08)
0.132***

(0.05)
0.073

(0.05)
0.346**

(0.17)
Log markup * markup–0.001

(0.00)
–0.001

(0.00)
Lagged log return on assets0.176***

(0.02)
0.177***

(0.02)
0.145***

(0.05)
0.176***

(0.02)
0.177***

(0.02)
0.168***

(0.02)
0.170***

(0.03)
0.167***

(0.02)
0.176***

(0.02)
0.173***

(0.02)
0.147**

(0.06)
Log effective interest rate–0.160***

(0.02)
–0.152***

(0.02)
–0.218***

(0.07)
–0.160***

(0.02)
–0.152***

(0.02)
–0.154***

(0.03)
–0.164***

(0.03)
–0.152***

(0.03)
–0.113***

(0.03)
–0.102***

(0.03)
–0.194**

(0.09)
Tobin’s q0.149***

(0.02)
0.189***

(0.03)
0.101**

(0.04)
0.149***

(0.02)
0.190***

(0.03)
0.128***

(0.04)
0.129***

(0.03)
0.117***

(0.04)
0.091***

(0.02)
0.154***

(0.03)
0.027

(0.05)
Log REER–0.258*

(0.14)
–0.240*

(0.13)
–0.333

(0.70)
–0.255*

(0.14)
–0.237*

(0.13)
0.324

(0.25)
0.079

(0.20)
0.368

(0.25)
0.748**

(0.29)
0.890***

(0.29)
–0.023

(1.12)
US output gap0.029***

(0.01)
0.037***

(0.01)
0.032***

(0.01)
0.042***

(0.01)
0.041***

(0.01)
0.048

(0.03)
Chinese imports share in the U.S.–0.168***

(0.06)
–0.155**

(0.06)
–0.378

(0.27)
Entry and Exit Rates0.005

(0.02)
–0.014

(0.02)
FDI Inflows–0.000

(0.00)
–0.009*

(0.00)
Entry&Exit Rates * FDI Inflows0.001**

(0.00)
Constant–1.638***

(0.61)
–1.761***

(0.61)
–1.235

(3.23)
–1.665***

(0.61)
–1.793***

(0.61)
–4.316***

(1.29)
–3.121***

(0.86)
–4.224***

(1.29)
–5.723***

(1.25)
–6.374***

(1.24)
–2.000

(4.89)
Memorandum items:
Observations402635005264026350030763774293321121760352
Fixed effectsYesYesYesYesYesYesYesYesYesYesYes

Robust standard errors in parentheses, with ***, **, * indicating significance level at 1 percent, 5 percent, and 10 percent level, respectively.

Robust standard errors in parentheses, with ***, **, * indicating significance level at 1 percent, 5 percent, and 10 percent level, respectively.

D. How Important is Technology and Productivity Diffusion from the U.S. to Canada?

13. R&D activity in Canada remains relatively weak, especially compared to that in the U.S. R&D stock in the U.S. was about 9 percent of GDP in the early 1970s, and has since increased to 17 percent of GDP in 2016, although the pace of growth has decelerated since the global financial crisis. In Canada, R&D stock was much lower, compared to the U.S., at about 3 percent of GDP in the early 1970s. R&D stock in Canada has since increased, but only to about 9½ percent of GDP by 2008, and has been flat over the past decade.

Canada: Labor Productivity and R&D Stock 1/

Sources: Statistics Canada; and Haver Analytics

1/ R$D stock is defined as the stock of non-residential intellectual property products in the private sector (CANSIM 0310005).

U.S.: Labor Productivity and R&D Stock 1/

Sources: Statistics Canada; and Haver Analytics

1/ R$D stock is defined as the stock of non-residential intellectual property products in the private sector.

14. What about spillovers from the U.S. to Canada? Even though R&D stock in Canada remains lower than that in the U.S., Canada should have been benefitted from technology and knowledge spillover effects across its southern border, and such spillover effects may have been magnified as the economic ties between the two economies have become stronger since the Canada-United States Free Trade Agreement (1987) and North American Free Trade Agreement (1994).

15. We test productivity spillover channels from the U.S. to Canada. Using data from 10 Canadian provinces, labor productivity growth models are estimated (Appendix III).8 The sample period is from 1990 to 2015, and the annual data are averaged for five 5-year periods (a balanced panel data set). We are interested in long-term relationships and thus want to exclude cyclical effects arising from short-run business fluctuations. Our hypothesis is that labor productivity growth or real R&D expenditure in the U.S. would affect Canada’s productivity through trade channels. We calculated Weighti,j as Canadian province i’s total trade (=exports + imports) with U.S. state j as a percent of province i’s GDP, which is used to weight U.S. labor productivity growth or real R&D expenditure (as percent of employment) in each U.S. state.

16. The regression results are presented in Table 3.

  • Models 1–4 present the results of OLS fixed-effects models. Consistent with our priors, U.S. labor productivity growth and U.S. R&D spending growth are positively and significantly correlated with Canada’s labor productivity growth. Note that coefficients are greater if U.S. labor productivity growth and U.S. R&D spending are weighted by the trade share, which is consistent with possible productivity and technology spillover channel through trade.9

  • Because our dataset has a relatively short sample period combined with large cross section data (“small T, and large N”), as robustness checks, we also employ the Generalized Method of Moments (GMM) estimator (Models 5–10). The results confirm robust relationships between U.S. productivity growth or U.S. R&D spending growth and Canada’s labor productivity growth as estimated coefficients are much larger and more highly significant compared to OLS estimates (Models 5–8).10

  • In Model 9, both simple average U.S. productivity growth and weighted U.S. productivity growth are included. The coefficient on the former turns to be insignificant, while the coefficient on the latter remains highly significant, which suggests evidence of the productivity spillover through trade channels.

  • In Model 10, we also found similar evidence for R&D spending, with the coefficient on the weighted U.S. R&D spending greater than that on the simple U.S. R&D spending. However, these coefficients are not significant, suggesting that productivity spillover channels are more complex and factors beyond R&D and trade could matter.

Table 3.Results of Provincial Level Spillover Regressions 1/
Dependent variable: log labor productivity growth (five year average)
OLS Fixed Effect EstimatesGMM Estimators
Model 1Model 2Model 3Model 4Model 5Model 6Model 7Model 8Model 9Model 10
Log. initial labor productivity (GDP/hour worked)–14.270**

(2.988)
–11.848**

(2.050)
–12.757**

(3.251)
–8.522**

(3.143)
–26.788***

(5.733)
–22.554***

(2.993)
–19.344***

(2.364)
–13.034***

(2.294)
–25.551***

(6.187)
–18.235***

(2.298)
Change in net-migration in provinces to population1.980**

(0.939)
1.773*

(0.962)
0.987

(1.143)
0.997

(1.230)
1.493**

(0.681)
1.331**

(0.557)
1.090

(1.720)
0.690

(0.837)
1.421**

(0.647)
1.012

(1.500)
Exports to other provinces (growth rate)0.054

(0.043)
0.056

(0.045)
0.096**

(0.036)
0.107**

(0.048)
–0.025

(0.027)
0.010

(0.018)
0.035*

(0.018)
0.108***

(0.023)
–0.010

(0.038)
0.054**

(0.028)
Rule of Law: Estimate4.296***

(1.112)
5.512***

(0.839)
0.976

(2.437)
1.365

(2.715)
3.734***

(1.202)
6.778***

(1.021)
0.226

(1.832)
0.961

(1.558)
5.452***

(1.848)
0.515

(1.987)
Unweighted US labor productivity growth0.055*

(0.028)
0.158***

(0.059)
0.072

(0.093)
Weighted (with X+M/GDP) US labor productivity growth0.080*

(0.045)
0.325***

(0.088)
0.208**

(0.099)
Unweighted US R&D per employee growth0.035***

(0.010)
0.071***

(0.012)
0.051

(0.032)
Weighted (with X+M/GDP) US R&D per employee growth0.093*

(0.050)
0.486***

(0.138)
0.181

(0.248)
Memorandum items:
Observations60605050505040405040
R-squared0.5830.5750.6300.552
Adjusted R-squared0.4540.4420.4820.373
AR(1) p-value0.1760.2170.08550.2720.2860.168
AR(2) p-value0.07260.8660.3100.1040.8270.228
Sargan p-value0.1410.3150.02680.02280.2090.0133
Hansen p-value0.9880.8770.9540.9870.9930.981

Robust standard errors in parentheses with ***, **, and * indicating significance level at 1 percent, 5 percent, and 10 percent, respectively.

Robust standard errors in parentheses with ***, **, and * indicating significance level at 1 percent, 5 percent, and 10 percent, respectively.

E. Conclusion

17. Key empirical findings in this chapter are as follows.

  • The recent slowdown in labor productivity growth can be explained mainly by a deceleration in capital intensity growth. Disaggregated analysis shows that the slowdown in capital intensity and labor productivity has been broad-based and not attributed to particular industries.

  • The firm-level panel data analysis suggests positive relations between markups and business investment, together with some evidence of nonlinearity.

  • We also found some evidence that lackluster business investment in recent years could also be associated with weakened firm dynamism (as reflected in lower entry and exit rate).

  • We found evidence that is consistent with spillover effects from U.S. productivity or U.S. technology investment to Canada productivity through trade channels.

18. The Canadian authorities’ renewed efforts to boost labor productivity are welcome. The authorities are rightly focusing on measures to enhance innovation activity, given that spending on R&D and the density of industrial robots in Canada are much lower than top performers in OECD economies.

OECD: Research and Development Spending

(Percent of GDP)

Sources: OECD Statistics.

OECD: Industrial Robot Density, 2015

(Number of multiple industrial robots per 10,000 employees in manufacturing industry)

Source: International Federation of Robotics

19. But more can be done to enhance investment, innovation, and the diffusion of new technologies.

  • On the external trade front, a heightened level of trade tensions in the North America region is a concern, given the importance of trade in supporting Canada’s productivity through the diffusion of technology. Earlier resolution of trade disputes will help put Canada’s growth path on a higher and sustainable path. In addition, a further reduction of the preferential treatment given to domestic firms and suppliers could also be considered to maximize potential benefits from trade.11

  • By analogy, although this chapter has not covered internal trade issues, our analysis suggests that reducing barriers to internal trade should also be important to enhance the diffusion of technology across Canadian provinces. The implementation of the Canadian Free Trade Agreement should be accelerated, aimed at reducing barriers to internal trade, investment, and labor mobility, and harmonizing regulations and standards.

  • Another area that requires concerted efforts by all levels of jurisdiction is product market reform. OECD Study (2016) shows that Canada’s regulatory framework is relatively restrictive compared to other OECD economies (Table 4). There is apparently room to reduce barriers in professional, retail, and network services, and regulatory protection of incumbents by streamlining licensing and registration requirements.

Table 4.OECD Product Market Regulation Indicators, 2013 /1(Index from zero to six, with higher index indicating less competition friendly regulatory stance)
Canada 2/U.S.Best G7OECD Average
Overall market–0.11.41.61.11.5
State control–0.31.92.71.62.2
Barriers to entrepreneurship–0.41.31.61.21.7
Of which:
Barriers in services sectors0.43.73.03.03.3
Regulatory protection of incumbents0.11.42.10.61.3
Barriers in network sectors0.23.03.01.02.7
Antitrust exemptions0.30.72.90.00.4
Other barriers to entrepreneurship–0.70.90.90.51.5
Barriers to trade and investment0.51.00.50.40.5
Of which:0.0
Explicit barriers to trade and Investment0.50.50.30.10.0
Barriers to FDI1.01.00.50.10.0
Tariff barriers0.00.00.00.00.0
Other barriers to trade and investment1.51.50.80.20.0
Differential treatment of foreign suppliers1.81.81.30.10.0
Barriers to trade facilitation1.21.20.20.20.0
Professional services1.13.11.12.02.0
Accounting1.23.51.32.42.3
Architect1.73.30.81.81.6
Engineer1.32.60.81.11.3
Legal0.23.21.52.63.0
Network services–0.41.71.41.72.1
Airlines0.01.01.00.31.0
Telecoms–0.40.50.30.80.9
Electricity0.93.41.51.82.4
Gas–1.01.50.91.62.5
Post0.12.72.32.52.5
Rail–1.32.33.02.43.5
Road–1.20.80.82.61.9
Retail services0.52.51.92.52.0
Registration and licensing requirements2.25.74.83.03.5
Specific regulation of large outlet–2.30.0..4.62.3
Protection of existing firms1.03.03.02.42.0
Regulation concerning shop opening hours1.12.30.00.51.2
Price controls2.44.01.71.71.6
Promotions/discounts–1.30.00.03.01.3
Source: OECD Product Market Regulation database,

The 2013 questionnaire contains around 1,400 questions on economy-wide or industry-specific regulatory provisions. A bit more than 700 of the questions are used to compute the economy-wide PMR indicator and the NMR indicators on sector regulation.5 All of these questions are closed questions that can either be answered with numerical values (e.g. the number of bodies that need to be contacted to start a business) or by selecting an answer from a pre-defined set of menu (e.g. the question whether a specific regulation exists can be answered with ‘yes’ or ‘no’). The qualitative information is transformed into quantitative information by assigning a numerical value to each possible response to a given question The coded information is normalized over a zero to six scale, where a lower value reflects a more competition-friendly regulatory stance.

“Green” indicates below the OECD average; and “pink” indicats above the OECD average.

Source: OECD Product Market Regulation database,

The 2013 questionnaire contains around 1,400 questions on economy-wide or industry-specific regulatory provisions. A bit more than 700 of the questions are used to compute the economy-wide PMR indicator and the NMR indicators on sector regulation.5 All of these questions are closed questions that can either be answered with numerical values (e.g. the number of bodies that need to be contacted to start a business) or by selecting an answer from a pre-defined set of menu (e.g. the question whether a specific regulation exists can be answered with ‘yes’ or ‘no’). The qualitative information is transformed into quantitative information by assigning a numerical value to each possible response to a given question The coded information is normalized over a zero to six scale, where a lower value reflects a more competition-friendly regulatory stance.

“Green” indicates below the OECD average; and “pink” indicats above the OECD average.

Box 1.Shift-Share Analysis

Aggregate productivity growth can be decomposed into growth of sectoral productivity, and gains from the reallocation of labor resource across sectors (Fagerberg, 2000 and Andersson, 2006).

ΔLPt/LP0 ≡ (LPtLP0)/LP0Aggregate labor productivity growth
=1LP0Σsj0ΔLPjtSum of sector productivity growth (“within effects”)
+1LP0ΣLPj0ΔsjtSectoral shifts in employment shares (“shift effects”)
+1LP0ΣΔLPjt*ΔsjtCross-sectoral component (“co-movement effects”)

where sjt is the employment share of sector j at time t; labor productivity LPt is defined as gross value added per total hours worked (excluding the public sector).

The first component measures the effect of labor productivity growth within each sector holding the employment share constant (“within effects”). The second component measures the impact on aggregate labor productivity growth resulting from the movement of labor across sectors holding the level of labor productivity in each sector constant (“shift effects”). Finally, the cross-sectoral component measures the effect of the change in both labor productivity and the employment share (“co-movement effects”). The positive sign of the cross-sectoral component indicates that the “within effects” and “shift effects” are complementary, thus labor resources are reallocated towards sectors with sectors with higher productivity growth.

For both Canada and the U.S., we decompose aggregate productivity growth into key 6 sectors: agriculture, forestry, fishing, and hunting; mining, quarrying, and oil and gas; manufacturing; construction; utilities; and other sectors including services.

Appendix I. Estimating Markups

1. Markups are estimated at the firm level following De Loecker and Warzynski (2012). Their method can accommodate a large class of price-setting models and does not need to rely on very restrictive assumptions. Consider the following cost-minimization problem for firm i in period t:

where Pitl and rit are the prices of labor and capital respectively, Yit (·) is production technology, and Yit¯ is a scalar. Note that the Lagrangian parameter associated with the first-order condition can be interpreted as a direct measure of marginal cost (MC). We define the markup as the price for the output good over the marginal cost, which are both not observable. However, by solving the first order condition, we get a simple expression for the markup as following.

where θitι the output elasticity of laborit and αitι is the share of expenditures on laborit in total net sales.

2. We can directly observe αitι, calculated as total wages divided by net sales, where total wages are calculated as the number of employees in each firm multiplied by the average industry- level wage. However, θitι is not directly observable and needs to be estimated. We estimated the following production function by industry (in logarithm).

where ωit is the firm’s productivity, and at is idiosyncratic shock. The estimated coefficient β^ι is the output elasticity of labor input, θitι, which we can use for calculating markups.

3. However, in estimating this production function, there could be potential correlation between labor input and unobserved firm-specific shocks, ωit, leading to biased estimates. To solve this endogeneity problem, Olley and Pakes (1996) assume that the firm can observe productivity term ωit and adjust inputs depending on ωit. We can rewrite the production function as following.

where ϕit(iit,kit) = β0 + βkcapitalstockit + ωit (investmentit, capitalstockit), and ηit is idiosyncratic shock. We approximate ϕit(.) with a second-order polynomial. The partially linear production function above can then be estimated by OLS. Since ϕit(.) controls for unobserved productivity, the error term is no longer correlated with the inputs.

4. In the Worldscope database, data for real output is not available, and thus we used “net sales” deflated by industry-level GDP deflators. All other nominal variables such as capital and investment are also deflated using industry-level deflators. Data for industry-level wages is from Statistics Canada. The sample period is 1997–2016 (annual data). Because we are interested in the productivity trend of non-energy firms, our sample dataset excludes energy firms.

Caveats

  • The output elasticity is estimated using industry-level panel data, and constant though the sample period. Ideally, time varying elasticities for each company could be estimated, but such data are not available. Thus, the up and down of this markup measures is purely driven by a change in labor shares (e.g., lower labor share means, higher θitX/αitX and thus higher markups).

  • It is assumed that cost minimizing firms take wage setting as given, which may not always hold if firms’ wage setting is determined by their productivity levels.

Data Description

Worldscope Data

VariableDescriptionWorld scope codes
OutputAnnual net sales or revenueWC01001
LaborThe number of employeesWC07011
InvestmentCapital expendituresWC04601
Capital stockNet property, plant and equipmentWC02501

Data sample

Number of firmsAvg. Net Sales (C$ mill.)Avg. Investment (C$ mill.)
Total (non-energy)945931472
Exporters6351,200608
Non-exporters3106628

Industry level data

  • GDP deflator. CANSIM Table 383–0021

  • Wages. CANSIM Table 382–0006

Appendix II. Firm Level Analysis: Regression Model Specification and Data

1. We run firm-level panel regressions, using data from the Worldscope database for publicly traded Canadian companies. We restrict our sample to include non-financial firms (excluding energy firms). The sample period is from 1997 to 2016 (annual data), and the panel dataset is unbalanced. The following standard Tobin Q’s investment model is estimated.

where i denotes a company, lit is fixed investment, Kit fixed capital stock, dt time fixed effect, ηi firm fixed effect, and vit idiosyncratic shock. MUit is a vector of firm level markups that we discussed in the previous section. Xit is a vector of determinants of investment, including return on assets (reflecting profitability), the effective interest rate on debt (reflecting cost of borrowing), U.S. output gap; and Tobin’s Q. Zit is a vector of market-level competitiveness variables, including firm entry and exit rates, foreign direct investment (FDI) inflows, and China’s import share in the United States (all at the industry level).

Data Description

Macro and industry data

  • Real effective exchange rate. In logarithm. Source, Haver Analytics.

  • United States output gap. In percent. Source, IMF World Economic Outlook database.

  • Entry and exit rates. Calculated as NumberofEntrant+ExitsNumberofActiveEmployerBusiness×100% by industry. Source, CANSIM Table 527–0001.

  • Foreign Direct Investment inflows. FDI inward inflows as a percent of GDP (constant terms) by industry. Source, CANSIM Table 376–0122.

  • Share of Chinese imports in the United States. Calculated as U.S.importsfromChinaU.S.domesticGDP+U.S.importsU.S.exports by industry. Trade data are from UN Comtrade; and U.S. industry GDP data are from the Bureau of Economic Analysis.

Industry classification. (1) agriculture, forestry, fishing, and hunting; (2) construction; (3) manufacturing; (4) wholesale trade; (5) retail trade; (6) transportation and warehousing; (7) utilities; (8) information and cultural industries; (9) professional and scientific services; (10) administrative and support, waste management, remediation services; (11) arts, entertainment, and recreation; (12) accommodation and food Services

Firm-level data from Worldscope

VariableDefinitionWorld scope codes
Investment to capital ratio (logarithm)Capital expenditures as the ratio of lagged net capital stock (property, plant, and equipment)WC04601/WC02501
Effective interest rate (logarithm)Interest expense as the ratio of total debtWC01251/WC03255
Return on assets (logarithm)Lagged return on assetsWC08326
Exporter dummyFirms with foreign Sales for at least three consecutive yearsWC08731
Tobin’s Q(Equity market value + liabilities market value) / (equity book value + liabilities book value)(WC08001 + WC03351) / (WC03501 + WC03351)
Appendix III. Provincial Level Analysis: Regression Model Specification and Data

1. Following the setup used in many growth studies (for example, Barro and Lee 2001), we estimate the following model:

where i, j, and t denote Canadian provinces, US states, and time, respectively; μi denotes an unobserved fixed effect capturing time-invariable heterogeneity across Canadian provinces; and εitIID(0,σg2) is a white-noise error term.

2. The dependent variable, ΔLPit, is labor productivity growth for each Canadian province. Explanatory variables are: ΔLPjtUS, labor productivity growth for each US state; ΔR&DjtUS real research and development expenditure per employee growth for each US state; and ΔXit, control variables including the initial level of labor productivity; the growth rate of interprovincial trade; the change in net migration inflows as a percent of population; and the rule of law index (a proxy for the quality of institutions).

3. We estimate the above equation using OLS with fixed-effects and the Generalized Method of Moments (GMM) estimator. Among various GMM estimators, we choose the difference GMM estimator from Arellano and Bond (1991), because it addresses problems related to inconsistent estimators due to variable endogeneity and a relatively short sample period combined with a fairly large cross section data (“small T, and large N”). We use as instrument the initial level of labor productivity.

Data Description

  • Canada productivity growth. Growth rate of real GDP divided by hours worked. Gross domestic product at market prices, chained 2007 comes from CANSIM Table 384–0038. Data for provincial GDP prior to 1980 were estimated using nominal GDP data (CANSIM Table 384–0015) divided by the GDP deflator for Canada. Actual hours worked (all Jobs, both sexes, 15 years and over) is from the Labor Force Survey.

  • Exports and Imports as a percent of GDP. Canada nominal exports and imports from the US by province are from https://open.canada.ca/data/en/dataset?sort=metadata_modified+desc&q=CIMT&organization=statcan. Nominal GDP is from CANSIM Table 384–0038.

  • Change in net-migration in provinces to population. Net migration is from CANSIM Table 051 -0018, and total population is from CANSIM Table 051–0001.

  • Growth rate exports to other provinces. CANSIM TABLE 384–0038.

  • Rule of Law. Worldwide Governance Indicators database, World Bank Group.

  • U.S. productivity growth. Growth rate of real GDP divided by number of employees. GDP (chained 2009) by state is from Gross State Product database (Bureau of Economic Analysis). Number employed is from U.S. Regional Household Employment database (total U.S., states and selected areas).

  • U.S. R & D per employee growth. Total R & D expenditure (constant PPP prices) is from OECD Statistics. Number employed is from U.S. Regional Household Employment database (total U.S., states and selected areas).

References

Prepared by Jorge Alvarez, Yurani Arias Granada, Kotaro Ishi (all WHD); and Sanjana Goswami (University of California, Irvine).

In this chapter, unless otherwise indicated, labor productivity is measured as output per total hours worked.

There are many other channels to explain labor productivity. For example, using Canadian provincial data, Petersson et al (2017) present empirical evidence that an increase in female labor force participation is positively associated with labor productivity growth. Ishi and Mariscal (forthcoming) present some evidence that public investment in infrastructure could boost economic growth and complement private investment.

Our focus on firms’ productivity and markups is in part motivated by some growing concerns that there are not so many globally competitive firms in Canada. For example, in the list of Fortune 500 (2017), only 11 Canadian firms made it to the list, of which 6 companies are from the financial or the energy sector. Many also argued that Canada’s cooperate sector is dominated by small-and medium-sized enterprises, which are less competitive and less capital intensive than big firms (Canadian Chamber of Commerce, 2016).

See Cao et al (2017) and St-Amant, P., and D. Tessier (2018). Possible factors discussed in the literature are: increased industrial concentration, like the entry of a big retailer such as Walmart, which operates on a large and increasing scale, discouraging new entry; and less interest in entrepreneurship among highly educated young people.

They estimated that two-thirds of the decline in Canada’s share of the U.S. non-energy import market (between 2002 and 2014) is attributable to reduced competitiveness, and 16 percent of the decline is attributable to changes in the composition of U.S. import demand. By sector, almost three-quarters of Canada’s total loss in market share was concentrated in two sectors: motor vehicles and parts (reflecting a reduction in U.S. demand for large cars) and forestry products and building and packaging materials (reflecting the collapse in the U.S. housing market).

Entrants and incumbents can create new products and displace the products of competitors. How exactly the process of innovation works has been much debated since Schumpeter (1939). Most recently, Acemoglu et al (2017) argue that policies to encourage exit of less productive firms would help free up resources and improve economic growth and welfare. In contrast, Gracia-Macia, Hsieh, and Klenow (2018) show that innovation comes more from incumbents, using U.S. firm-level data.

The 10 Canadian provinces are Alberta (AB), British Columbia (BC), Manitoba (MB), New Brunswick (NB), Nova Scotia (NS), Newfoundland and Labrador (NL), Ontario (ON), Prince Edward Island (PE), Quebec (QC), and Saskatchewan (SK).

Trade and productivity linkages could arise through several channels. For example, Keller (2000) stresses the technology diffusion channel through intermediate goods imports from technological leaders, while Blalock and Veloso (2007) present evidence that vertical supply relationships are the channel through which import-driven technology transfer occurs. Bloom, Draca, and Reenen (2016) also show some evidence of import competition and technical advancement.

Note that our GMM models do not fully take account of endogeneity problems arising from omitted variables. For example, Canada’s labor productivity could rise due to an increase in more competitive US imports, leading to exits of less productive Canadian firms.

Foreign ownership is restricted in some sectors (e.g., network industry), and foreign suppliers may not be able to fully participate in public procurement contracts in selected sectors (e.g. transportation services, computer services, construction, telecom, and professional services).

Other Resources Citing This Publication