Archived — Working Paper Number 18: Regional Disparities in Canada: Characterization, Trends and Lessons for Economic Policy
by Serge Coulombe, Department of Economics, University of Ottawa, November 1997
Since the early 1990s, regional studies have assumed considerable importance in economic research. Their renewed popularity is linked partly to the strong comeback of growth theory, which increasingly is the preferred tool for evaluating economic policies and blurs the traditional distinction between macro- and micro-economic frameworks for analysis. But regional studies also owe their current popularity to the fact that many growth problems at the end of the twentieth century are regional in impact, for example, the collapse of the communist bloc in Eastern Europe, German reunification, European integration and the emergence of trading blocs.
Starting with studies by Barro and Sala-i-Martin (1991, 1992 and 1995) and Mankiw, Romer and Weil (1992), the question of convergence — that is, the catching up of poor economies with rich economies, whether at the regional or international level — has received considerable attention from researchers. Significant progress has been made in dynamic modelling under the neoclassical model and endogenous growth models. Recent analyses stress features such as the dynamics of the adjustment of physical and human capital, migrations, economic and political integration, the stability and effectiveness of public institutions, economic policy and the spread of technology. In Canada, several empirical studies — Helliwell and Chung (1991), Helliwell (1994), Lefebvre (1994), Coulombe and Lee (1993, 1995 and 1996) and Lee and Coulombe (1995) — have made it possible to draw up a statistical picture of the situation showing that regional disparities in per capita income and production, as well as in worker productivity, have tended to diminish since the Second World War. This movement toward convergence, however, has slowed down since the late 1970s, a phenomenon that can also be seen within all the industrialized countries (Sala-i-Martin 1995).
As we shall show later, in 1950 regional disparities in per capita production and income were far greater within Canada than in the 12 U.S. states bordering on Canada. For the present study, these bordering U.S. states — that is, Washington, Idaho, Montana, North Dakota, Minnesota, Michigan, Ohio, Pennsylvania, New York, Vermont, New Hampshire and Maine — together serve as a control group for purposes of comparison since they closely resemble adjacent regions of Canada in terms of economic geography. (The Yukon, the Northwest Territories and Alaska are excluded from the study.) In 1950, the relative dispersion index of per capita personal income was almost three times higher in the different regions of Canada than what could then (and can still) be observed in the case of the bordering U.S. states. The deviation has narrowed since then because the convergence observed north of the 49th parallel in the postwar years had the effect of bringing closer per capita income disparity levels on both sides of the border. Nevertheless, there are still large differences between the different regions of Canada in terms of per capita production. Referring to a relative index of per capita production differences for the most recent regional data, we shall show that regional disparities in Canada today are nearly twice as great as in the bordering U.S. states. Part of the per capita income convergence between Canadian regions can thus be attributed to inter-regional redistribution via "fiscal federalism" and the taxation system.
The persistence of large regional disparities creates major problems for the management of economic policy in a federation such as Canada: its central government has always pursued regional development policies and, since the Second World War, it has set up a vast structure for inter-regional redistribution. More recently, in 1987 Ottawa established agencies to plan and promote regional economic development in the Atlantic provinces (Atlantic Canada Opportunities Agency), Northern Ontario (FedNor) and the Western provinces (Western Economic Diversification), and it did the same for Quebec (Federal Office of Regional Development — Québec) in 1991. The ability to support these programs and policies is directly linked with the persistence of regional economic disparities. This question is especially pertinent at a time when government faces growing financial constraints.
The aim of the present study is to examine the history of Canada's regional disparities through empirical analysis in line with the new wave of studies focusing on convergence and economic growth. The empirical part of the study will be conducted in two phases. First we shall undertake a comparative analysis of the history of regional disparities in per capita production, in both Canada and the bordering U.S. states. After identifying certain statistical facts, we shall then examine three components — the dispersion of productivity, the dispersion of employment rates (via unemployment rates) and the dispersion of the participation rate — in order to study the history of the deviation between Canada and the bordering U.S. states with respect to dispersion of per capita production. This decomposition will be revealing to the extent that the neoclassical adjustment based on the law of diminishing returns and the accumulation (and mobility) of physical and human capital results in a convergence of the productivity of factors. The two other determinants of dispersion of per capita production (dispersion of the unemployment rate and of the participation rate) are linked to the functioning of the labour market and household decisions on the balance to be struck between work and recreation.
In the second phase of the empirical analysis, we make use of a method for estimating ß convergence based on a sample group-type process recently developed by Coulombe and Day (1996) to assess the speed of convergence of productivity and other economic indicators between Canada's regions. This method builds on that used by Coulombe and Lee (1993, 1995) and Lee and Coulombe (1995) in their studies on regional convergence. Compared with the conventional approach, which is based on estimates by transversal section (see Barro and Sala-i-Martin, for example), the method proposed here allows better use to be made of all the data concerning regional growth profiles since it integrates as well the information resulting from the yearly movement of series of regional economic indicators. By integrating data on longitudinal and transversal sections, the sample group method can be viewed as an attempt to modify the approach of Barro and Sala-i-Martin to take into account the criticism of Quah (1993).2 Sample group estimation offers the advantage of producing estimates of the speed of ß convergence; combined with estimate residuals, these allow us to create a dynamic simulation of movements in the variance of regional disparities (convergence). Dynamic analysis highlights the concept of the stationary level of variance, i.e., the long-term balance of regional disparities. The steady-state disparity is determined by the interaction between the strength of the convergence resulting from faster accumulation of physical and human capital in poor regions and, on the other hand, the variance in regional disturbances forcing regional economies to temporarily stray from the path of convergence leading them toward long-term equilibrium.
The proposed analysis goes beyond statistical representation since the level of decomposition, its predictive nature, the theoretical basis and the dynamic simulations are such as to clarify the making of decisions about regional development policies. For example, from the movement of differences in worker productivity in relation to its steady state and to differences in the control group (the bordering U.S. states), we can gain insight into the subject of the dynamics of physical and human capital accumulation and the operation of market forces. It can nevertheless happen that the political and institutional context undermines convergence of worker productivity. In this case, regional development policies should therefore target the mobility of physical and human capital and the spread of technology. Further, if productivity differences remain excessively large, it can happen that they are eliminated automatically, at least in part, since the neoclassical convergence mechanism will start functioning once the hindrances to physical/human capital mobility and to the spread of technology will have disappeared or lost their importance. If per capita production differences between Canada and the United States are related mainly to differences in the unemployment and participation rates, regional development policies should instead target worker mobility and the balancing of work and recreation. Last, through the analysis of variances and co-variances, we can recognize the interaction of different factors and offer an interpretation suitable for guiding regional development actions.
The balance of the study is structured as follows: First, the next section describes the concepts of ß and convergence as well as the theoretical relationship that exists between the two within the framework of the hypothesis of unconditional convergence. It should be noted that this relationship underlies the simulation undertaken later in the study. The following section presents statistical data and compares movements in regional disparities in production and per capita income in Canada and the bordering U.S. states. This comparative analysis is then supported by decomposition of the per capita production variance in terms of the variance and co-variances of its three components (productivity, unemployment rate and participation rate). The study next turns to the sample group-type econometric approach, which is used to estimate ß convergence, and we explain the links between the estimate of parameter ß, the estimate residual and the concept of convergence. The estimate results for ß and the stationary level of variance of regional disparities are presented and analyzed. The dynamic simulation of the path of productivity variance is compared with the path observed. The section presenting conclusions highlights the implications of our analysis for regional development policy issues in Canada.
- Date modified: