The Decoupling of the Canada-‐US Business Cycle

(1)

The Decoupling of the Canada-‐US Business Cycle

University of Groningen

Faculty of Economics & Business

Master Thesis International Economics & Business

Name Student: Henk (H.J.) Slot Student ID number: S1995537

Student E-‐mail: H.Slot@student.rug.nl Date: 1st of July 2015

(2)

Abstract

The aim of this paper is to investigate the divergence between U.S. and Canadian business cycles. Using aggregate and industry decompositions it documents two fundamental changes that appeared since the mid 1980s. U.S. output volatility dropped while Canadian output volatility did not and U.S labor productivity turned countercyclical while the opposite happened in Canada. This paper argues the changes in these business cycle indicators are related and are due to a more flexible U.S. labor market. It finds stark contrasts in the institutional protection of employees and labor market reallocation abilities that explain both output volatility, as well as cyclical labor productivity differences. Minor econometric evidence only provides evidence of similar employment fluctuations.

(3)

Figure 1: Relative GDP per capita and Labor Productivity 6 Figure 2: 10-‐year forward rolling standard deviation of output 11 Figure 3: U.S. cyclical output and productivity 17 Figure 4: Canada cyclical output and productivity 18 Figure 5: Rolling correlation output and productivity 19 Figure 6: Trade union density 23 Figure 7: Long-‐term unemployment, 6 months or more 25 Figure 8: Job finding probability 26 Figure 9: Job separation probability 26 Figure 10: U.S. and Canada permanent and temporary unemployment rates 27

Table of Tables

Table 1: Business Sector Growth U.S. and Canada 12 Table 2: Growth and variance decomposition from industry data U.S. and Canada 16 Table 3: Correlation U.S. and Canada 19 Table 4: Productivity-‐output correlation largest industries 20 Table 5: ADF test on variables and their first differences 33 Table 6: ADF test on residuals 33

Table of Appendices

Table 1: List of industries 40 Table 2: Productivity-‐Output correlations at industry level 41

Figure 1: Unemployment rates 42

Figure 2: Employment rates 42

Figure 3: Participation rates 42

Stata Output 43

(5)

I. Introduction

There is a renewed interest in the literature on the changing behaviour of labor productivity over the business cycle. Historically, labor productivity tends to move procylical, increasing in economic booms and decreasing during recessions. Economists in the 1960s (Oi, 1962; Solow, 1962) developed the theory, also termed labor hoarding, as an explanation for the profit maximizing behaviour of firms. They were reluctant to fire staff during recessions caused by the existence of fixed costs of hiring, firing and training workers and therefore preferred to keep them employed (Biddle, 2013). It caused labor productivity to fall during recessions and increase when output increased, leading to a positive correlation between labor productivity and output. As for example outlined by Gali & van Rens (2014), the U.S. correlation of productivity with output used to be strongly positive up until the 1980s, around 0.8.

Since the mid 1980s however, this pattern has changed considerably. The estimates of the productivity and output correlation dropped to a level close to zero (Stiroh, 2009; McGrattan & Prescott, 2012; Gali & van Rens, 2014) and differences with periods before the 1980s are highly significant. Especially in recent years of recession, research has found a break with the historical procyclicality in U.S. labor productivity. It increased sharply during the 2008–2009 recession while GDP plummeted. According the early theories productivity would have to fall during these episodes of recessions. As actually the opposite has happened, and labor productivity has grown while total output has fallen, debate has arisen on the existing relationship between labor productivity and output and whether labor productivity has become anti cyclical.

(6)

2005) and financial sector innovation (Dynan, Elmendorf & Sichel, 2006). While its interpretation is still controversial, there is widespread consensus among macroeconomists on the existence and rough timing of the Great Moderation. The change took place in the mid 1980s. Around the same time the correlation between output and labor productivity dropped, suggesting that there could be more than mere temporal coincidence linking them.

These and a significant amount of other papers have all focused on the U.S., but recent research has lacked comparison with similar developed nations. Canada in this case, as a neighbouring country, is an interesting example. The Canadian and U.S. economies are highly integrated through international trade in goods and assets. Given their integrated nature, the two countries have highly correlated business cycles, and cyclical fluctuations in the U.S. tend to have an impact on Canadian economic activity (Bernard & Usalcas, 2014). For instance, the increased trade dependence between the two countries has made Canadian unemployment levels highly vulnerable to cyclical behaviour and recessions in the U.S. (Campolieti, 2011). Comparisons between U.S. and Canadian business cycles and labor markets therefore generate a high level of interest.

Figure 1: Relative GDP per capita and labor productivity (U.S. = 1)

Source: The Conference Board. Figure shows the relative GDP per capita and labor productivity growth of Canada and the U.S using GDP in 2014 prices and labor productivity per person employed in 2014 prices, from 1950 up until 2013.

Figure 1 makes a first comparison by outlining the relative development of GDP per capita and labor productivity. It displays a distinct change appearing in relative labor

(7)

productivity during the mid 1980´s. The Canadian relative labor productivity embarks on a significant decline showing no clear sign of returning to pre-‐1980s levels.

In light of the U.S. decline in the procyclicality of labor productivity (van Zandweghe, 2010; Gali & van Rens, 2014; Wang, 2014) the declining volatility of output (Galί & Gambetti, 2008; Stiroh, 2009) and the clear decline in the relative Canadian labor productivity that all appeared since the mid 1980s, this paper attempts to form a unifying explanation based on structural flexibility differences in U.S. and Canadian labor markets. While the business cycle changes in the U.S. have been increasingly recognized and analysed, similar research in a closely integrated nation has been lacking. A primary motivation for this research therefore is to explore the changes in U.S. business cycle dynamics and to compare them to Canadian business cycles.

This paper adds to the debate by proposing a single explanation for all of the above observations; a structural change in labor market dynamics. This approach hypothesises that structural changes have led to greater flexibility in the American labor market. These changes have primarily come in the form of a reduction in labor adjustment costs, like hiring and firing costs. The adjustment costs act as frictions limiting the ability of firms to fire and hire employees at will. A reduction therefore allows firms to more easily adjust their labor force in response to changes in output. In times of recession, this could lead to sharper adjustments in employment than reductions in output leading to higher labor productivity. Moreover increased flexibility allows firms to more easily reallocate productive resources in response to different shocks, reducing the volatility of output (Cuñat & Melitz, 2012).

(8)

II. Literature Review

Previous empirical work has already recognized the influence of a flexible labor market on the volatility of output and the cyclical behaviour of productivity. A number of papers have investigated the facts and proposed mechanisms that may explain the observed changes. Stiroh (2009) investigates the declining volatility of U.S. output growth and provides an internally consistent framework dubbed volatility accounting. His framework tracks which of the inputs; hours, productivity and covariates, contribute most to the volatility of output. His evidence points to a volatility decline due to smaller covariances between sectors as well as a significant decline in both labor productivity growth and hours growth. These results speak in favour of the presence of common factors across sectors. Among this explanations are underlying changes in labor market flexibility that support and facilitate increased reallocations.

Similarly, Gali & Gambetti (2008) find a significant decline in the correlations between output, hours and labor productivity in their time series from the 1960s up until the 2000s using 1984 as a breaking point. They argue the correlation change is due to more flexible labor markets, which made firms more likely instead of reluctant to change employment in reaction to output changes. Gordon (2010) notices the last three recessions (1990-‐91, 2001 and 2007-‐2009) have been followed by ´jobless recoveries´. In these recoveries output growth was accompanied by a burst in labor productivity, although employment declined. A leading argument in his analysis links the shift to jobless recoveries to the greater flexibility in American labor markets.

(9)

productivity. The results show that labor productivity is strongly procyclical in terms of its correlation with output in the model with a frictional labor market but its procyclicality reduces substantially as labor market frictions are reduced.

Van Zandweghe (2010) assesses two explanations for the shift in the behaviour of U.S. labor productivity; structural changes and supply shocks. Supply shocks are shocks that drive both output and productivity in the same direction, such as an oil price increase. His findings however suggest the importance of these shocks has been stable over time, and so the shift in the business cycle behaviour of labor productivity is due to structural changes in the labor market. These changes include a decline in labor adjustment costs as a result of a decline in hiring and firing costs, and an intensified reallocation across industries to more effectively absorb the existing labor costs.

Wang (2014) differs in relation to the aforementioned papers by using a cross-‐ industry dimension as an alternative method to the much-‐used aggregate time series, to gain a better understanding of the forces underlying the diminished procyclicality of aggregate productivity. The main findings are in line with the hypothesis that both internal and external labor markets have become more flexible, reducing the cost of adjusting and reallocating existing employment.

The above-‐mentioned literature all share a common finding: a reduction in labor market frictions that enhanced U.S. labor market flexibility. With the integrated nature of both countries it would be interesting to investigate how Canadian labor markets have developed. Given the declining relative labor productivity, the vanishing procyclicality of U.S. labor productivity and the decline in U.S. output volatility, it seems the two countries are diverging. This leads to the formulation of the hypothesis that the United States and Canada have diverged in their structural labor market flexibility since the mid-‐1980s.

(10)

III. Assessing how the business cycle in the U.S. and Canada have

evolved

This section presents the evidence for the key changes in the U.S. and Canadian business cycles that the paper seeks to explore. The goal of the paper is to understand these changes as related phenomena that explain the structural change in labor market flexibilities. While most of these changes have been previously documented, a joint explanation is still lacking.

III.1 Source data

Unless otherwise noted the primary source of data for both the U.S. and Canada is the WORLD KLEMS dataset for the period 1961 up until 2007. A major advantage is that it supplies a rich panel dataset covering 27 sectors for both the U.S. economy and its Canadian counterpart. The database has been set up to promote the analysis of growth and productivity patterns based on a growth accounting framework, within which the interaction between variables can be analysed. Specifically, it looks at the relationships between productivity and output, and productivity and labor inputs measured as hours. In these comparisons output is defined as total value added within a country, hours worked is total hours worked by persons employed and productivity is measured as output per hour. These indicators fluctuate over time due to business cycle fluctuations and long-‐term trends. These long-‐term trends include gradual increases in capital stock, technology and populations. Therefore, in several of the calculations, these influences are removed by using the statistical Hodrick and Prescott filter (HP filter), which allows for the gradual movement in trend growth rates over time.

(11)

III.2 The great moderation

The first well-‐documented fact is the declining volatility of U.S. output. This decline is often termed the Great Moderation (van Zandweghe, 2010; Garin, Pries & Sims, 2013; Gali & van Rens; 2014) and simply means that on average, yearly output growth has become more stable; growth has shown less steep peaks and troughs opposed to earlier periods.

Macro evidence

Figure 2 plots the 10-‐year forward rolling standard deviation of output for both the U.S. and Canada. The figure displays very well the U.S. drop in volatility that occurred in the mid 1980s. Also when comparing the two periods on average, the volatility of output post-‐1984 remains well below the heightened levels of the 1970s. This is not necessarily true for Canada, where output volatility shows only a mild decline beginning in the 1980s, and even increases until the recession in the beginning of the 1990s.

Figure 2: 10-‐year forward rolling standard deviation of output

Source: WORLD KLEMS database. Figure shows standard deviations of output for both the US and Canada, from 1961 – 2007.

A complementary method to analyse the declining volatility of output growth in both countries, is to compose a growth and variance decomposition comparing the differences in pre-‐ and post-‐1984 periods (Stiroh, 2009). This method allows a straightforward quantification of the different sources of volatility. Firms can increase output either by producing more output per hour worked, which is by increasing

0.00 0.50 1.00 1.50 2.00 2.50 3.00 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006

Rolling output volatility

(12)

productivity, or by increasing total hours worked. As an identity, aggregate output (Y) equals the product of hours (H) and average labor productivity (ALP), defined as output per hour worked (y = ALP = Y/H). The growth decomposition is the following:

𝑌 = 𝐻 + 𝑦 (1)

where dots over the individual letters signify growth rates.

Next the variance decomposition shows the breakdown of the variance of output growth in the sum of the variance of hours (H), the variance of average labor productivity (y) and twice the covariance. It indicates to what extent output in between the periods is influenced by productivity and hours worked.

𝑉(𝑌) = 𝑉 𝐻 + 𝑉 𝑦 + 2𝐶(𝐻, 𝑦), (2)

Where V(⋅) is the variance and C(⋅,⋅) is the covariance of the variables.

Table 1: Business sector growth U.S. and Canada

_{Growth
Decomposition} _{Variance
Decomposition}

Pre-‐1984 Post-‐1984 Change Pre-‐1984 Post-‐1984 Change

_U.S. Output 3.23 3.04 -‐0.19 6.01 1.92 -‐4.09*** Hours 1.27 1.36 -‐0.09 3.92 2.11 -‐1.81 ALP 1.96 1.68 -‐0.28 2.14 0.96 -‐1.18* 2xCov(Hours, ALP) -‐0.05 -‐1.15 -‐1.10 Corr(Hours, ALP) -‐0.01 -‐0.42 -‐0.41 Canada Output 4.06 2.99 -‐1.07* 4.25 3.02 -‐1.23 Hours 1.78 1.65 -‐0.14 4.09 2.26 -‐1.83 ALP 2.28 1.34 -‐0.94** 2.25 1.05 -‐1.20* 2xCov(Hours, ALP) _-‐2.09 -‐0.30 1.79 Corr(Hours, ALP) -‐0.36 -‐0.10 0.26

(13)

Table 1 compares these growth and variance decompositions for both the U.S. and Canada, before and after the 1984 breaking point. The first thing to note for the U.S. is that there is no clear evidence of changes in mean growth rates of output, hours or labor productivity after 1984. The results show that U.S. output growth diminished by -‐0.19 between the years prior to 1984 and the years after. Canadian output growth does show a sharper decline, by -‐1.07, in which a decline in labor productivity by -‐0.94 contributes the most. The second part of the table shows the variance decompositions of output. In each column, the sum of the individual variances and covariance is the variance of output given in the top row. For the U.S. the variance of output has steeply declined since 1984. It describes the ‘Great Moderation ’and simply means that on average, yearly output growth has become more stable. It reflects roughly equal drops in the volatility of hours, from 3.92 to 2.11, the volatility of labor productivity, from 2.14 to 0.96, and the covariance between them, from -‐0.05 to -‐1.15.

The results for Canada show that output volatility has decreased. Both hours and labor productivity contribute similar declines as in the U.S., -‐1.83 and -‐1.20 respectively. Judging from these components Canada has turned into a much more stable economy with less fluctuations in the post-‐1984 period. Counterbalancing these developments however is the covariance between productivity and hours, which increases by 1.79. It proves that compared to the pre-‐1984 period, productivity and hours increasingly show co-‐movements, and means productivity will fall with hours during recessions. It stands in sharp contrast with U.S. results in which the covariance went in the opposite direction and has turned more negative in the post-‐1984 period.

Lastly, the tables also display the correlations between hours and labor productivity. The Canadian correlation positively changed by 0.26 when comparing the pre-‐1984 and post-‐1984 periods. The U.S. correlation on the other hand went from essential zero at -‐0.01 in the first period to -‐0.42 in the second. This indicates the U.S. and Canada differ in the linear association between hours and productivity over time. In the U.S., with a decrease in hours, labor productivity will move in the opposite direction, positively contributing to a drop in output volatility.

Sectoral variation

(14)

sources in the decline of aggregate volatility (Stiroh, 2009). It allows an identification of the importance of the variance declines within industries, so due to specific changes in one industry, and the importance of variance declines between industries, so general changes that lead to smaller volatilities among sets of industries.

According this composition, aggregate output 𝑌 can be defined as a growth function of industry value added 𝑉𝑖. Growth in aggregate value-‐added is the weighted sum of industry value added growth:

𝑌 = !𝑣!,!𝑉! (3)

where 𝑣!,!, is the two period average nominal share of industry value-‐added in aggregate value-‐added. Each factor in the summation is the industry’s contribution to aggregate value-‐added growth. The output factor can be decomposed into a contribution from hours and labor productivity, but now for each industry. Labor productivity can then be defined as υi = Vi/Hi. The industry aggregate output growth becomes:

𝑌 = (𝑣_! _!,!𝐻_!) + _!𝑣_!,!𝑣_! (4)

Similar to table 1, each decomposition of aggregate output and labor productivity growth comes with a variance decomposition. It is the sum of the direct variance contributions from individual industries and the sum of the covariances between industries and is defined as:

𝑉(𝑌) = _!𝑉(𝑣_!,!𝑉_!) + _! _!!!!!2𝐶(𝑣_!,!𝑉_!, 𝑣_!,!𝑉_!) (5)

The industry labor productivity decomposition is defined similarly as:

𝑉(𝑌) = _!𝑉(𝑣_!,!𝐻_!) + _!𝑉(𝑣_!,!𝑣_!)

+ _! _!!!!!2𝐶 𝑣_!,!𝐻_!,𝑣_!,!𝐻_! ,+ _! _!!!!!2𝐶 𝑣_!,!𝐻_!,𝑣_!,!𝑣_! , (6) + _! _!!!!!2𝐶 𝑣_!,!𝑣_!,𝑣_!,!𝑣_! ,

(15)

all industry pairs. In this decomposition, the hours/hours and productivity/productivity relationships are strictly between industry effects. To make a further distinction, the hours/productivity covariance can be split into a within component that reflects covariances of hours and productivity for a given industry and a between component that reflects the covariances across different pairs of industries as:

_! _!!!2𝐶 𝑣_!,!𝐻_!,𝑣_!,!𝑣_! ,= _!2𝐶(𝑣_!,!𝐻_!, 𝑣_!,!𝑣_!)

+ _! _!!!!!2𝐶 𝑣_!,!𝐻_!,𝑣_!,!𝑣_! , (7) = Within + Between

Table 2 displays the results. The top parts report the growth decomposition columns. The variance decomposition columns, which consist of the sum of the variances and covariances, are shown in the bottom parts of the table. For the variance decomposition, the first line shows the total within effect for each period, and the second the between effect.

(16)

Table 2: Growth and Variance decomposition from industry data U.S. and Canada

_{Growth
Decomposition} _{Variance
Decomposition}

Pre-‐1984 Post-‐1984 Change Pre-‐1984 Post-‐1984 Change

_U.S.

Output 3.23 2.63 -‐0.59 10.37 4.16 -‐6.21

Industry output decomposition

𝑉 𝑣!,!𝑉!, = Within 1.86 0.81 -‐1.05

2𝐶 𝑣!,!𝑉!,𝑣!,!𝑉! = Between 8.51 3.35 -‐5.16

Industry labor productivity decomposition

𝑉 𝑣!,!𝐻!, 1.64 0.84 -‐0.79 0.81 0.33 -‐0.48 𝑉 𝑣!,!𝐴𝐿𝑃!, 1.59 1.79 0.20 1.73 0.83 -‐0.91 2𝐶 𝑣!,!𝐻!,𝑣!,!𝐻! 4.93 2.47 -‐2.46 2𝐶 𝑣!,!𝐻!,𝑣!,!𝐴𝐿𝑃! = Within -‐0.68 -‐0.35 0.33 2𝐶 𝑣_!,!𝐻_!,𝑣_!,!𝐴𝐿𝑃_! = Between _-‐0.47 _-‐0.94 _-‐0.48 2𝐶 𝑣!,!𝐴𝐿𝑃!,𝑣!,!𝐴𝐿𝑃! 4.05 1.82 -‐2.23 Canada Output 4.05 2.77 -‐1.28 7.02 6.54 -‐0.48

Industry output decomposition

𝑉 𝑣!,!𝑉!, = Within 0.82 0.54 -‐0.27

2𝐶 𝑣!,!𝑉!,𝑣!,!𝑉! = Between 6.20 6.00 -‐0.20

Industry labor productivity decomposition

𝑉 𝑣_!,!𝐻_!, 2.02 1.22 -‐0.80 0.71 1.19 0.48 𝑉 𝑣!,!𝐴𝐿𝑃!, 2.04 1.55 -‐0.48 1.05 1.26 0.21 2𝐶 𝑣!,!𝐻!,𝑣!,!𝐻! 5.07 4.35 -‐0.72 2𝐶 𝑣_!,!𝐻_!,𝑣_!,!𝐴𝐿𝑃_! = Within _-‐0.94 _-‐1.91 _-‐0.97 2𝐶 𝑣_!,!𝐻_!,𝑣_!,!𝐴𝐿𝑃_! = Between _-‐2.39 _-‐1.78 _0.61 2𝐶 𝑣_!,!𝐴𝐿𝑃_!,𝑣_!,!𝐴𝐿𝑃_! _3.52 _3.42 _-‐0.09

Note: Decomposition of aggregate output growth for 27 industries. Growth Decomposition shows alternative decompositions of aggregate output. Variance decomposition shows alternative decompositions of aggregate output variance. For industry sums, V(-‐) indicates the variance of weighted growth rates, while C(-‐) indicates the covariance of the weighted growth rate. All growth rates are log differences of annual data multiplied by 100, from 1961 – 1983 and 1984 – 2007.

(17)

individual contribution of the hours covariance (-‐0.72) and the labor productivity covariance (-‐0.09), the between effect only decreases by -‐0.20. This is due to the fact that the covariance between hours and labor productivity actually increased in the same period (0.61). These results strengthen earlier obtained findings in the aggregate Canadian decomposition; the Canadian drop in volatility has been less severe and the covariance between hours and labor productivity display opposing behaviours.

III.3 The shift away from the procyclical behaviour of productivity

The second changing factor in U.S. business cycle fluctuations that points to structural changes in the individual labor markets is the cyclical behaviour of labor productivity. Until the mid-‐1980s productivity growth rose and fell with output growth and is said to have moved procyclical. Since then however, average labor productivity has moved countercyclical.

Macro evidence

Figure 3 plots the cyclical behaviour of output and labor productivity for the U.S. Observation shows that labor productivity has indeed behaved different in recent recessions from the 1980s onwards when compared to earlier decades.

Figure 3: U.S. cyclical output and productivity

Source: WORLD KLEMS database. Trends are removed with the Hodrick and Prescott filter, from 1961 – 2007.

(18)

productivity fell as output dropped and during recoveries both output and productivity rose. In sharp contrast stands the period after the mid 1980s, in which cyclical productivity and output increasingly began to move in opposite directions. Things appear different however in Canada. Figure 4 does not display the same switch in the cyclicality of labor productivity. Despite two exceptions in the 1980s and the 1990s where productivity did not fall with output, the two keep tracking each other in their movements.

Figure 4: Canada cyclical output and productivity

Source: WORLD KLEMS database. Trends are removed with the Hodrick and Prescott filter, from 1961 – 2007.

In addition, to show the distinct difference in correlation movements, both U.S. and Canadian developments can be tracked in a single correlations panel. Figure 5 plots the 10-‐year rolling correlations between output and productivity. It distinctively visualises the difference in the changing behaviour of productivity between the two countries. For the U.S. the two series are positively correlated until the mid 1980s. Afterwards this correlation drops significantly before turning slightly negative during the 1990s. This pattern is contrastingly different for Canada where the correlation increases during the same time period. This marks the fact that productivity has began to move countercyclical in the U.S. while moving more procyclical for Canada.

(19)

Figure 5: Rolling correlation output and productivity

Source: WORLD KLEMS database. The panel plots the 10-‐year forward rolling correlation between output and productivity,

The relationship between productivity, output as well as hours can be examined in more detail. Table 3 displays the correlations between these variables for both Canada and the U.S., both pre-‐ and post-‐1984. The results confirm earlier findings that the correlation between productivity and output has improved for Canada in the period after 1984. The correlation rose from 0.38 in the former period to 0.50 in the latter. The rise in the output-‐productivity correlation reflects a rise in the procyclical behaviour of productivity. In addition, the correlations between productivity and hours, and output and hours, showed similar patterns.

Table 3: Correlations U.S. and Canada

Correlations

Pre-‐1984 Post-‐1984 Change

U.S. Productivity-‐Output 0.59*** 0.27 -‐0.32 Productivity-‐Hours -‐0.01 -‐0.42** -‐0.41 Output-‐hours 0.80*** 0.76*** -‐0.04 Canada Productivity-‐Output 0.38* 0.50** 0.13 Productivity-‐Hours -‐0.36* -‐0.10 0.26 Output-‐Hours 0.73*** 0.81*** 0.08

Note: The variables are expressed in log growth rates and approximated by first difference. The correlation between two variables is based on a Pearson’s correlations test. Significance is denoted by * at the 10% level, ** at 5% and *** at the 1% level.

(20)

The U.S. contradicts in its results as these correlations have all weakened over the same time period. The largest decline is obtained for the relationship between productivity and hours. It dropped from -‐0.01 to -‐0.41. Similarly, the correlation between productivity and output has decreased from 0.59 to 0.27. This indicates that there is evidence that the cyclical behaviour of U.S. productivity is changing in the opposite direction and highlights another stark difference between the two countries. To support these findings, the productivity-‐output correlations from table 5 can be extended to include a productivity-‐output correlation at industry levels.

Sectoral evidence

Table 4 shows the five largest industries in terms of value added for both the U.S. and Canada based on value added in the post-‐1984 period. Four out of five of the biggest industries in terms of value added in both countries, all contribute to the cyclical trend of labor productivity, countercyclical in the U.S. and procyclical in Canada. In total these industries add up to 48% of value added in the U.S. and 41% in Canada. This proves that the cyclical behaviour of labor productivity in each country is not due to one or two industries, rather a movement found across industries. Table 2 in the appendix provides an extensive overview of the productivity-‐output correlations per industry.

Table 4: Productivity-‐output correlation largest industries

_U.S. _Canada

Pre-‐1984 Post-‐1984 Pre-‐1984 Post1-‐984

Industry Corr (1) VA Corr (2) VA Change (2)–(1)

(21)

IV. Labor market flexibility

So far this paper has provided evidence on two business cycle indicators that changed around the mid 1980s. Output volatility has dropped for the U.S. while it only slightly decreased for Canada. Additionally, labor productivity in both countries has shown opposing results, moving countercyclical in the U.S. while moving procyclical in Canada. The main purpose in what follows is to investigate the relation between these developments and structural changes in labor market flexibility. Although there are many factors having an influence on the business cycle developments, such as an increase in the use of capital (Pessoa & van Reenen, 2014) and technology (Gali & Gambetti, 2008), the average share of labor in output throughout the entire period considered in this paper is well above 60%.1_{Therefore
it
remains
interesting
to} investigate the role of labor in output and productivity fluctuations.

Flexibility in the context of labor markets can have multiple interpretations. This paper makes a distinction between macro and micro flexibility as in Eamets & Paas (2007). In the macroeconomic context, flexibility concerns institutional flexibility to show to what extent state institutions are involved in the regulation of labor markets (Eamets & Paas, 2007). It covers labor market regulations and trade union activity. Micro flexibility is the labor market’s ability to carry out the reallocation of workers needed for productivity growth. It involves labor flows indicating worker flows between different labor market states in terms of employment, and job flows characterised by job creation and job destruction (Eamets & Paas, 2007). This section outlines the main structural differences that caused the diverging gap in U.S. and Canadian labor markets.

IV.1 Institutional environment

The first explanation relates to the institutional framework that creates labor market regulations and policies. Legal restrictions exist on adjusting the level and composition of the workforce to adapt to changing demand. It prohibits firms from sudden hiring and firing of employees during cyclical downturns, with the aim of preserving individual job security. The legislative protection therefore has important consequences for the flexibility and freedom with which it allows employers to adjust employment for changes in demand. (Lammam & Macintyre, 2014). According OECD statistics the

(22)

Canadian and U.S. institutional framework can be described as flexible and very similar, especially when compared to other OECD nations (Gordon, 2011; OECD Employment Outlook, 2013). A separation can be made however between flexibility concerning employment protection legislation and the involvement of trade unions.

Employment protection legislation

Employment protections legislations (EPL) govern the rules concerning the hiring and firing of workers (OECD Employment Outlook, 2013). These rules have been designed to protect jobs and increase job stability for employees. Hence, EPL’s raise the costs of workforce adjustments by distorting the optimal composition of employment and pushes firms to use resources less efficiently. Recent research on the labor market impact of employment protection has found that overly strict regulations can reduce job flows and hinder productivity and economic growth (Bassanini, Nunziata & Venn, 2009; Martin & Scarpetta, 2012).

Measured on a scale from 1 to 6, both the U.S. and Canada are characterized by a system that combines little protection for individual workers in comparison with OECD averages. For the protection of permanent workers, 0.26 for the U.S. and 0.92 for Canada, as well as the protection of temporary workers, 0.25 in both countries, the strictness is well below OECD averages of 2.04 and 2.29 respectively (OECD stat extracts, 2013a; OECD stat extracts, 2013b). Further EPL indicators as notification procedures and severance pay are similar to the dismissal indicators in that they are well below OECD averages for both the U.S. and Canada, but also in that they have not changed since the mid-‐1980s. Hence, this might indicate the U.S. and Canada are more flexible in comparison with other developed nations, but a relative lack of change between the U.S. and Canada can therefore hardly be presented as evidence in favour of structural flexibility divergence.

Trade unions

(23)

labor market and stronger reductions in the procyclicality of productivity. A decline in union membership would therefore signify a loss of union power, which had to allow employment layoffs more readily in recent times. The increase in layoffs explains the countercyclical behaviour of U.S. labor productivity.

As can be seen from figure 6, trade union density has dropped in the U.S. in comparison with Canada. Trade union density is measured as the percentage of wage and salary workers who are union members, divided by the total number of salary earners. While trade union density for both countries was almost equal in the beginning of the 1970s, the rates in 2007 were just over 10% for the U.S and almost three times as high for Canada. Additionally, in a direct comparison between the U.S. and Canada, Lammam & Macintyre (2014) show that Canadian labor relation laws are much less balanced and flexible than their U.S. counterparts. These laws are included in union contracts to balance the needs of employers, employees and unions but thereby hinder the proper functioning of a labor market and its performance in terms of productivity.

Figure 6: Trade union density (in % of total employees)

Source: OECD stat extracts 2013c, from 1960– 2007.

IV.2 Labor market reallocation

The second explanation for the diverging flexibility gap relates to the reallocation of workers and jobs across industries. The functioning of labor markets involves a continuous shifting of labor across firms and sectors. It however involves costs for firms, as it is a resource-‐intensive and costly process to open and fill new vacancies. Similarly,

(24)

switching from and searching for new jobs is costly and time consuming for workers (Martin & Scarpetta, 2012). Therefore, rates of worker-‐ and job flows indicate the ease with which workers are able to reallocate, and the ease with which firms are able to adjust their labor force. If workers are more easily reallocated from declining to expanding firms and industries it is an important factor influencing productivity and output growth (Eamets & Paas, 2007; Martin & Scarpetta, 2012; Garin et al, 2013).

Worker flows

Worker flows relate to all transitions between the various states of the labor market: employment, unemployment and the total labor force (Eamets & Paas, 2007). The transitions are measured through the unemployment, employment and labor force participation rates as given by the U.S. and Canadian household surveys. An overview of these rates indicates to what extent the labor market is capable of employing all workers, especially during downturns. A key feature of downturns is an increased mismatch between workers and jobs. Reallocation helps to solve this mismatch and so plays a key role in the recovery from a recession and in understanding the flexibility of a labor market (Garin et al., 2013).

The development of these three rates over the post-‐1984 period points to two important findings (Appendix; figures 1, 2 and 3). The first is that the higher U.S. employment and participation rates signify more labor market flexibility. The traditional explanation comes from Lazear (1990) who argues that countries with less flexible labor markets simply have less suitable employment for workers who become unemployed. As a consequence workers drop out of the labor force, leading to lower participation rates. Di Tella & MacCulloch (2005) support this theory by arguing higher participation rates are reached through a lack of an insurance effect. If less regulation means more layoffs, the risk of losing income becomes greater. The lack of such wage insurance makes workers in flexible labor markets more inclined to participate.

(25)

Additional evidence on the flexibility of worker flows is based on the share of long-‐term unemployment in total unemployment. As stricter labor markets are associated with lower job flows in and out of employment (Eamets & Paas, 2007) unemployed periods tend to last longer (OECD employment outlook, 1996). This has an impact on worker reallocation, as long-‐term unemployed are different from the rest of the unemployed labor force. First, they have a lower chance of exiting unemployment, and second the chance of exiting unemployment declines with the duration of being unemployed (Hornstein, 2012). A larger share of long-‐term unemployed workers therefore negatively impacts the ease of reallocation from declining to growing industries. If U.S. workers circulate through existing jobs faster, the share of long term unemployed in total unemployment should be lower.

Figure 7 displays the share of long-‐term unemployment in total unemployment for the period 1984 up until 2007. Long-‐term unemployment is defined as being unemployed for 6 months or more. It displays the U.S. long-‐term unemployment share is consistently lower than its Canadian equivalent for a majority of the post-‐1984 time period. This indicates larger shares of the Canadian unemployed have more difficulty in finding new jobs and suggests less flexibility.

Figure 7: Long-‐term unemployment (in % of total)

Source: U.S. (BLS) and Canadian (LFS) household survey, from 1984 – 2007

(26)

Job flows

Another way of accounting for reallocation flexibility is by looking at job flows. Greater job flows are associated with more flexibility, as it indicates firms are more easily firing workers, and workers are more easily finding jobs (Eamets & Paas, 2007). Following research by Shimer (2012), two indicators can be identified to support this argument; the rate of job creation and the rate of job destruction. Job creation and job destruction can be measured as the job finding and job separation probabilities.

Figure 8: Job finding probability

Source: U.S. (BLS) and Canadian (LFS) household survey, from 1976 – 2007.

Figure 9: Job separation probability

(27)

Figure 8 and 9 display both probabilities for the U.S. and Canada. It reveals the U.S. job finding probability has been much higher throughout the period. Although the Canadian job finding probability is on a clear upward trend, it only converges with the U.S. rate during the early 2000s. A contrasting result from the separation probability is that it is actually higher for Canada in a majority of the time period. Shimer (2012) however finds that up to 90% of employment fluctuations were a consequence of movements in the job finding probability.

Further evidence on job flows derives from the nature of layoffs. Layoffs can be temporary and permanent. Temporary layoffs account for furloughed employees who stay at home because an employer does not have enough work (Groshen & Potter, 2003). Permanent layoffs however measure the amount of employees who have lost their jobs and are not returning to the same firm. An increase in the relative importance of permanent layoffs is consistent with the idea that it became easier to fire workers.

Figure 10: U.S. and Canada permanent and temporary unemployment rates (in %)

Source: U.S. (BLS) and Canadian (LFS) household survey, from 1976 – 2007.

Figure 10 displays the contribution of both types of layoffs to each of the total unemployment rates. A distinctive characteristic for the U.S. is the disappearing share of temporary layoffs in the post-‐1984 period. Before 1984 both the temporary and permanent unemployment rates rose during recessions. After 1984 temporary layoffs barely increased, while the burst of permanent layoffs remained large. This may suggest U.S. firms have increasingly started to rely on permanent layoffs. In Canada on the other

The Decoupling of the Canada-­‐US Business Cycle