Selected Issues and Analytical Notes

Abstract

Selected Issues and Analytical Notes

Estimating the Growth Effect of Public Infrastructure: Evidence from Canadian Provinces1

A. Introduction

1. Canada has launched an ambitious infrastructure program to boost its long-term growth potential. The federal government is committed to investing Can$ 187 billion (equivalent to 7½ percent of GDP) in public capital, a half which (Can$ 96 billion) is new additional infrastructure spending. This will triple investment in public capital over the next decade. Priority projects include: (i) public transit infrastructure to better connect Canada’s transportation system within the country and to the world to facilitate trade and movement of people; (ii) green infrastructure to create greener energy and cleaner land; and (iii) social infrastructure to provide affordable housing, child cares spaces, and communities centers.

2. The objective of this chapter is to estimate the macroeconomic effects of infrastructure investment. For countries where infrastructure bottlenecks are constraining growth such as in Canada, the gains from alleviating these bottlenecks are likely to be large. Using panel data for 10 Canadian provinces over 1961-2015, we present some evidence that public investment in infrastructure could boost economic growth and complement private investment. Importantly, we find that public infrastructure investment generates positive spillover effects across regions: higher public investment in one province boosts output in the rest of Canada

B. The Size of the Problem

3. There is broad consensus that the infrastructure gap in Canada is sizable, although estimates range widely from as low as Can$150 billion to as high as Can$ 1 trillion (Advisory Council on Economic Growth, 2016). The World Economic Forum Global Competitiveness Index suggests that Canada’s overall ranking in the quality of infrastructure dropped to 21st position most recently, down from 17th position a decade ago, with the decline in the ranking of roads and railroad as the most evident.

Table 1.

Canada: Global Competitiveness Index

(Rank)

article image
Source: World Economic Forum Global Competitiveness Index.

4. In addition, the stock of economic infrastructure stands at a historical low. Economic infrastructure includes railways, ports, airport, pipelines, roads, and energy networks that supports the delivery of public services, connects people to jobs, and facilitates businesses through reduction of costs of doing businesses. Investment in economic infrastructure therefore directly affects economic production. However, the stock of economic infrastructure has declined from about 20 percent of GDP in 1961 to as low as 12 percent in 2007, reflecting a steady decline in public investment for four and a half decades. The downward trend was reversed during the 2008-09 global financial crisis when the government increased infrastructure spending as part of its stimulus efforts, but the stock of economic infrastructure assets remains at historic lows, 5¾ percent of GDP below the level in 1961.

Figure 1.
Figure 1.

Economic Infrastructure Assets and Investment

(Percent of GDP)

Citation: IMF Staff Country Reports 2017, 211; 10.5089/9781484309650.002.A002

Sources: Statistics Canada CANSIM Table 031-0005; and Haver Analytics

C. Policy Considerations

5. Canada’s fiscal federalism presents both advantages and challenges in coordinating infrastructure investment. The bulk of economic infrastructure assets is owned and managed by provincial and local authorities. The federal government’s share of infrastructure assets has halved to 7 percent since 1990, while provincial and local governments have increased their share to more than 90 percent of total assets today.

Figure 2.
Figure 2.

General Government Economic Infrastructure Assets 1/

(Percent share)

Citation: IMF Staff Country Reports 2017, 211; 10.5089/9781484309650.002.A002

Sources: Statistics Canada, CANSIM Table 378-0121; and Haver Analytics1/Book value.
  • Advantages. Lower levels of government (provincial and local governments) are better positioned to identify what local communities need. For example, they are more likely to know which communities would benefit most from local infrastructure projects, such as water supply or urban transit. In this case, decentralization to provincial and local governments is both feasible and desirable for allocative and administrative efficiency (Ahmad, Hewitt, and Ruggiero, 1997).

  • Challenges. Since many infrastructure facilities—including roads, railways, and airports—have connectivity and are part of networks, investing in infrastructure in one province would likely have spillover effects to other provinces. Accordingly, better coordination across jurisdictions in the planning and execution of infrastructure projects could improve economic benefits (in terms of efficiency and growth gains). However, currently, there is neither an established institutional mechanism for federal-provincial coordination nor a national infrastructure strategy that aligns priorities across all levels of government. The federal government provides conditional grants to provincial and local governments to influence their decision making over infrastructure projects, but some question their effectiveness as a i policy tool (Public Policy Forum, 2016).

6. Better coordination. Increasing the efficiency of public investment is critical to reap its full benefits. Canada has a highly efficient public investment framework but there is room for improvement. The federal government should develop coordination mechanisms that identify infrastructure gaps and increase transparency. More specifically, the IMF Public Investment Management Assessment framework suggests merits of reforming the following areas.

  • Specifying capital funding amounts by department in the medium-term budget framework, which could support more detailed sectoral investment strategies and a project pipeline

  • Promoting inter-governmental coordination by consolidating federal and provincial approved projects into a single report updated annually

  • Identifying major public investment projects in budget documents to clarify their contribution to achieving program objectives and protect their funding during budget execution

  • Improving transparency regarding public investment spending by publishing Project Briefs and business cases for approved major projects, total expected financial obligations of ongoing projects, project audits, and PPP risks

  • Establishing and publishing recommended methodologies for appraising proposed projects and evaluating projects upon completion

  • Publishing the total cost of projects at the time of initial funding.

D. Empirics

7. The empirical analysis focuses on estimating the macroeconomic effects of increasing economic infrastructure and asks the following questions.

  • Does infrastructure investment contribute to boost output? If so, how much? 2

  • Does infrastructure investment have spillover effects across Canadian provinces? Since most infrastructure assets are owned and managed by provincial and local authorities, but an infrastructure policy coordination framework has not been fully developed, finding positive spillover effects would argue for establishing a coordinating framework.

  • Does infrastructure investment crowd in private capital investment? This question revisits a long-standing debate on whether public infrastructure crowds in or crowds out private capital (see for example, Afonso, A., and M. St. Aubyn, 2008). On the one hand, an increase in public investment could lead to higher interest rates (if the investment is debt-financed) or to a higher tax burden for the private sector: in either case, demand for private investment would be dampened. On the other hand, an increase in public investment could lead to higher quantity and quality of infrastructure assets, thus contributing to higher productivity of private capital, as costs of doing business fall. In this case, an increase in infrastructure investment would crowd in private investment.

8. To answer these questions, we run panel regressions, using data from 10 Canadian provinces. The sample period is from 1961 to 2015 (annual data), and the panel dataset is balanced (see data description in Annex I). For the average of the full sample period, we observe a relatively strong positive relationship between the annual increase in the stock of infrastructure assets and provincial GDP growth (Figure 3).

Figure 3.
Figure 3.

Growth of Infrastructure and Provincial GDP, 1961-2015

(Period average, percent)

Citation: IMF Staff Country Reports 2017, 211; 10.5089/9781484309650.002.A002

Sources: Statistics Canda CANSIM Table 031-0005; and Haver Analytics
9. A standard Cobb-Douglas production function is estimated. Following Aschauer (1979) and Munnell (1990), we assume that infrastructure capital contributes to output and enters the production function.
yit=ALitβLKitβkGitβGGO¯itβGO(1)

where Yit is real GDP, A captures total factor productivity, Lit is employment, Kit is real private capital stock, Git is real infrastructure stock, GO¯it, real infrastructure stock in the other provinces, for province i and time t. Coefficients βL, βK, βG, and βGO denote elasticities of output with respects to inputs.

We include GO¯it to examine whether output in province i depends on infrastructure in other provinces. In other words, we test whether infrastructure in one province could cause spillover effects to the rest of Canada. We experiment with two different weighting methods to calculate this variable.

  • Weighted by geographical distance. Spillover effects could be greater across nearby provinces. For example, Ontario’s GDP would be affected more by infrastructure in Quebec than in British Columbia. To test this hypothesis, we construct weights based on the distance between pairs of provinces (capital cities).

  • Simple average. Given that economic activity across Canadian provinces is highly integrated, geographical distance may not matter. In this case, Ontario’s GDP would be affected not necessarily by infrastructure in Quebec but by overall infrastructure in all Canadian provinces. To examine this hypothesis, we calculate GO¯it as the simple average of other provinces.

Taking logarithms of equation (1), we estimate the following baseline model.

yit=βiLlit+βiKkit+βiGgit+βiGOgo¯it+uit(2)
uit=αi+γiωt+εit,(3)

where lower case variables denote natural logarithms. Unobservable uit can be interpreted as total factor productivity, and is composed of a province specific intercept, αi a time trend, γiωt, and the white noise error terms εit∼IID(0, 𝜎2). On the time trend, we examine both a linear trend and a time-variant trend (see below).

10. We estimate the above equation using a fixed effects model and pooled mean group estimation. Note that Canadian provinces are diverse, with different settings in structural and economic fundamentals, including natural resource endowments, legal and regulatory frameworks, industry structures, and labor markets. In estimating equation (2), heterogeneity across Canadian provinces should be considered. The fixed effects model is the most restrictive one, where the slope coefficient βi and the time trend coefficient γiωt are assumed to be the same across provinces (βi = β, and γi = 1, for all i), but the intercept term, αi, is different across provinces. Next, we relax the assumption of the fixed coefficient β and allow it to vary across provinces (but the time trend, γiωt, is still assumed to be constant). To this end, we apply the mean group (MG) estimator developed by Pesaran and Smith (1995).3 Finally, we relax the assumption of the constant time trend, and apply a mean group estimator with a common dynamic process (MG-CDP), developed by Eberhardt and Teal (2010).4

Results

11. The regression results are presented in Table 2.

  • In all models, the coefficients on private capital and employment are positive and highly significant.

  • Models 1-3 present the results of fixed-effects models. In Model 1, the coefficient on infrastructure assets is positive but not significant. Model 2 includes other provinces’ infrastructure assets (weighted by geographical distance). This shows that ‘province i’s (for example, Ontario’s) own investment in infrastructure would boost its own GDP, and infrastructure investment in the rest of Canada would boost Ontario’s GDP, too. But this evidence is weak, as their coefficients are not significant. Model 3 includes the simple average of other provinces’ infrastructure assets, but these coefficients (both own and other provinces’) are negative and insignificant.

  • Models 4-6 are based on the MG estimator. The coefficient on ‘province i’s own infrastructure assets is positive but is still not significant in Model 4. The coefficients on other provinces’ infrastructure assets in Model 5 (using simple average GO¯jt) and in Model 6 (using weighted average GO¯jt) are positive but not significant. Moreover, the coefficients on own infrastructure are now negative and insignificant.

  • Models 7-9 are based on the MG-CDP estimator. The coefficient on ‘province i’s own infrastructure assets is now positive and significant (Model 7). Including other provinces’ infrastructure assets, however, Model 8 shows that the coefficient on ‘province i’s own infrastructure assets is positive but not significant, while that on other provinces’ infrastructure assets (weighted) is negative and insignificant. Model 9 performs relatively well. The coefficient on ‘province i’s own infrastructure assets is positive and significant (at 15 percent level), while that on other provinces’ infrastructure assets is positive and highly significant.

Table 2.

Regression Results (Production Function Models) 1/

Dependent variable: Real GDP

article image

Robust standard errors are in parentheses, with *** indicating significance level at 1 percent, ** at 5 percent, * at 10 percent, and + at 15 percent.

Pesaran and Smith (1995) mean group estimator.

Eberthardt and Teal (2010) mean group estimator.

In sum, we can conclude that there is some evidence that infrastructure assets are positively correlated with real GDP and have spillover effects. We judge that the estimates based on the MG-CDP model—which accounts for both a province-specific slope β and a province-specific time dummy γiωi—look most plausible, given that the estimated coefficients are all significant at reasonable levels. Models 7 and 9 suggest that the elasticity of infrastructure assets would be around 0.08 and 0.10, respectively.

Results

11. The regression results are presented in Table 2.

  • In all models, the coefficients on private capital and employment are positive and highly significant.

  • Models 1-3 present the results of fixed-effects models. In Model 1, the coefficient on infrastructure assets is positive but not significant. Model 2 includes other provinces’ infrastructure assets (weighted by geographical distance). This shows that ‘province i’s (for example, Ontario’s) own investment in infrastructure would boost its own GDP, and infrastructure investment in the rest of Canada would boost Ontario’s GDP, too. But this evidence is weak, as their coefficients are not significant. Model 3 includes the simple average of other provinces’ infrastructure assets, but these coefficients (both own and other provinces’) are negative and insignificant.

  • Models 4-6 are based on the MG estimator. The coefficient on ‘province i’s own infrastructure assets is positive but is still not significant in Model 4. The coefficients on other provinces’ infrastructure assets in Model 5 (using simple average GO¯jt) and in Model 6 (using weighted average GO¯jt) are positive but not significant. Moreover, the coefficients on own infrastructure are now negative and insignificant.

  • Models 7-9 are based on the MG-CDP estimator. The coefficient on ‘province i’s own infrastructure assets is now positive and significant (Model 7). Including other provinces’ infrastructure assets, however, Model 8 shows that the coefficient on ‘province i’s own infrastructure assets is positive but not significant, while that on other provinces’ infrastructure assets (weighted) is negative and insignificant. Model 9 performs relatively well. The coefficient on ‘province i’s own infrastructure assets is positive and significant (at 15 percent level), while that on other provinces’ infrastructure assets is positive and highly significant.

In sum, we can conclude that there is some evidence that infrastructure assets are positively correlated with real GDP and have spillover effects. We judge that the estimates based on the MG-CDP model—which accounts for both a province-specific slope β and a province-specific time dummy γiωi—look most plausible, given that the estimated coefficients are all significant at reasonable levels. Models 7 and 9 suggest that the elasticity of infrastructure assets would be around 0.08 and 0.10, respectively.

Results

11. The regression results are presented in Table 2.

  • In all models, the coefficients on private capital and employment are positive and highly significant.

  • Models 1-3 present the results of fixed-effects models. In Model 1, the coefficient on infrastructure assets is positive but not significant. Model 2 includes other provinces’ infrastructure assets (weighted by geographical distance). This shows that ‘province i’s (for example, Ontario’s) own investment in infrastructure would boost its own GDP, and infrastructure investment in the rest of Canada would boost Ontario’s GDP, too. But this evidence is weak, as their coefficients are not significant. Model 3 includes the simple average of other provinces’ infrastructure assets, but these coefficients (both own and other provinces’) are negative and insignificant.

  • Models 4-6 are based on the MG estimator. The coefficient on ‘province i’s own infrastructure assets is positive but is still not significant in Model 4. The coefficients on other provinces’ infrastructure assets in Model 5 (using simple average GO¯jt) and in Model 6 (using weighted average GO¯jt) are positive but not significant. Moreover, the coefficients on own infrastructure are now negative and insignificant.

  • Models 7-9 are based on the MG-CDP estimator. The coefficient on ‘province i’s own infrastructure assets is now positive and significant (Model 7). Including other provinces’ infrastructure assets, however, Model 8 shows that the coefficient on ‘province i’s own infrastructure assets is positive but not significant, while that on other provinces’ infrastructure assets (weighted) is negative and insignificant. Model 9 performs relatively well. The coefficient on ‘province i’s own infrastructure assets is positive and significant (at 15 percent level), while that on other provinces’ infrastructure assets is positive and highly significant.

In sum, we can conclude that there is some evidence that infrastructure assets are positively correlated with real GDP and have spillover effects. We judge that the estimates based on the MG-CDP model—which accounts for both a province-specific slope β and a province-specific time dummy γiωi—look most plausible, given that the estimated coefficients are all significant at reasonable levels. Models 7 and 9 suggest that the elasticity of infrastructure assets would be around 0.08 and 0.10, respectively.

Marginal product of infrastructure assets

12. The regression results permit a calculation of the marginal product of capital (MPK). Using the estimated elasticity of public infrastructure, an MPK can be calculated as the elasticity multiplied by the output to public infrastructure ratio.5 Figure 4 presents the MPK for Canada (based on the estimated elasticity of 0.08, Model 7). Although some caution is needed in interpreting this result, the figure shows a steady increase in the MPK between the early 1960s and mid-2000s, which reflects a decline in the stock of public infrastructure as a percent of GDP.6 As the government changed course and raised investment in public infrastructure after the 2008-09 global financial crisis, the MPK has started to fall. However, the current level of the MPK remains above historical levels. This seems to suggest—together with a fact that the current level of interest rates is much lower than its historical norm—that there is room to further increase public infrastructure while keeping the return of infrastructure investment higher than the cost of funding.

Figure 4.
Figure 4.

Marginal Product of Capital

Citation: IMF Staff Country Reports 2017, 211; 10.5089/9781484309650.002.A002

Sources: Authors’ estimates

Crowding in effects of infrastructure

13. We test whether infrastructure assets crowd in private capital. Whether public capital spending would crowd-in or crowd-out private capital spending has been much debated. We assume that infrastructure capital would affect real GDP through private capital and modify equation (1) as follows.

yit=ALitβLKitβK+βGlnG+βGOlnGO¯(4)

Taking logarithms of equation (4), we estimate the following model.

yit=βiLlit+βiKkit+βiGgitkit+βiGOgo¯itkit+uit(2)
uit=αi+γiωt+εit,(3)

14. The results suggest evidence of crowding-in (Table 3). The estimated coefficient of the cross term between ‘province i’s own infrastructure (g) and private capital assets (k) is highly significant and positive 0.008 (Model 1), while the cross term between other provinces’ infrastructure (go) and “province i”s own private capital assets (k) is also highly significant and positive 0.01 (Model 2).

E. Conclusion

15. This chapter analyzed the macroeconomic effects of public infrastructure. We found some evidence that higher public infrastructure is correlated with higher economic growth and could crowd in private investment. We also found that infrastructure investment has positive spillover effects across provinces. This means that more concerted efforts across different layers of jurisdictions to better coordinate infrastructure planning would maximize the economic benefit of infrastructure investment. In sum, the empirical findings in this chapter provides an economic case for more infrastructure spending. With underinvestment of infrastructure over the past several decades, combined with historically low cost of funding, investing in infrastructure would generate a net positive return.

Table 3.

Testing Complementarity of Public Infrastructure and Private Capital 1/

Dependent variable: Real GDP

Estimation method: a mean group estimator with a common dynamic process (MG-CDP)

article image

Based on E&T Mean Group Estimators. Robust standard errors are in parentheses, with *** indicating significance level at 1 percent, ** at 5 percent, * at 10 percent, and + 15 percent.

Annex I. Data Description

  • Real GDP. In logarithm. Gross domestic product at market prices, chained 2007. 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.

  • Employment. In logarithm. Labor Force Survey, CANSIM Table 282-0089.

  • Real private capital stock. In logarithm. Non-residential buildings plus engineering construction in the non-government sector, geometric end-year net stock, chained 2007, CANSIM Table 031-0005.

  • Real infrastructure stock. In logarithm. Non-residential buildings plus engineering construction in the government sector, geometric end-year net stock, chained 2007, CANSIM Table 031-0005.

References

  • Afonso, A., and M. St. Aubyn, 2008, “Macroeconomic Rates Return of Public and Private Investment Crowding-In and Crowding-Out Effects,European Central Bank, Working Paper Series, No 864 February 2008

    • Search Google Scholar
    • Export Citation
  • Ahmad, E., D. Hewitt, and E. Ruggiero, 1997, “Assessing Expenditure Responsibilities,” Chapter 2 in Fiscal Federalism in Theory and Practice, edited by T. Ter-Minassian, International Monetary Fund, 1997.

    • Search Google Scholar
    • Export Citation
  • Aschauer, A.D., 1989, “Is Public Expenditure Productive?Journal of Monetary Economics, Vol.23, No.2, pp. 177200

  • Eberhardt, M., and S. Bond, 2009, “Cross-section Dependence in Nonstationary Panel Models: A Novel Estimator,MPRA Paper 17692, University Library of Munich

    • Search Google Scholar
    • Export Citation
  • Eberhardt, M., and F. Teal, 2010, “Productivity Analysis in Global Manufacturing Production,Discussion Paper 515, Department of Economics, University of Oxford

    • Search Google Scholar
    • Export Citation
  • International Monetary Fund, 2014, ““World Economic Outlook: Is it Time for an Infrastructure Push The Macroeconomic Effects of Public Investment,October (Washington)

    • Search Google Scholar
    • Export Citation
  • Jordà, Ò., 2005. “Estimation and Inference of Impulse Responses by Local Projections.American Economic Review 95 (1), pp. 16182.

    • Search Google Scholar
    • Export Citation
  • Munnell, H.A., 1990, “How Does Public Infrastructure Affect Regional Economic Performance?New England Economic Review, September/October, pp. 1131

    • Search Google Scholar
    • Export Citation
  • Pesaran, M. H., and R. P. Smith, 1995, “Estimating Long-run Relationships from Dynamic Heterogeneous Panels,Journal of Econometrics Vol.68, pp. 79113

    • Search Google Scholar
    • Export Citation
  • Public Policy Forum, 2016, “Building the Future,http://www.ppforum.ca/publications/building-future-strategic-infrastructure-long-term-growth-interim-report

    • Search Google Scholar
    • Export Citation
1

Prepared by Kotaro Ishi and Rodrigo Mariscal (both WHD) and David Gentry (FAD).

2

IMF (2014) analyzes the macroeconomic impact of public investment shocks in advanced economies, using the local projection methodology proposed by Jordà (2005). The analysis shows that public investment shocks have statistically significant and long-lasting effects on output, with an unanticipated 1 percentage point of GDP increase in investment spending increasing the level of output by about 0.4 percent in the same year and by 1.5 percent four years after the shock.

3

The MG procedure is simple: first estimate equation (2) by OLS for each province, and then take the average.

4

MG-CDP procedures takes the three steps. First, obtain an estimate of the CDP, by estimating a pooled differenced OLS model with time dummy variables. The estimated parameters of these time dummies will represent the CDP. Second, the CDP is then added to the model by either subtracting it from the dependent variable (i.e. y˜it˜ = yitCDP, or by including it in each of the 10 regressions. Third, estimate 10 individual regressions and compute the averages or the individual estimated slopes as in Pesaran and Smith (1995).

5

In equation (2), 𝛽𝐺 = (𝜕𝑌/𝜕𝐺)(𝐺/𝑌) (for exemplification, i is dropped). Hence, 𝜕𝑌/𝜕𝐺 = 𝛽𝐺 (𝑌/𝐺).

6

We assume that the elasticity of public infrastructure is constant at any point in time, but this assumption may not hold if the elasticity fluctuates, for example depending on the cyclical state of the economy. A caution is also needed in interpreting the level of the MPK, as the estimated elasticity varies depending on model specification and estimation methodology.

Canada: Selected Issues and Analytical Notes
Author: International Monetary Fund. Western Hemisphere Dept.