UNIVERSITY OF GRONINGEN // FACULTY OF ECONOMICS AND BUSINESS
A Retrospective Look at the Patterns of the
Canadian Labor Market
Gerben Jansen
Supervisor: T.M. Harchaoui
MSc International Economics and Business Track: International Capital and Globalization
2
Abstract
The Canadian unemployment rate during the Great Recession reported a mild increase that stands in sharp contrast with that of other developed nations. This development addresses the need to understand the factors behind this development. Both demographic factors and economic growth shed valuable light on this matter, though the latter offers a richer perspective that recently gained consideration in literature. The use of a growth accounting exercise confirms the strong interplay between economic growth and unemployment, leaving little room to other factors. We then take the analysis one step further by exploiting a full-fledged dynamic panel model that confirms the robustness and validity of Okun’s law.
4
Content
Abstract ... 2
1. Introduction ... 5
2. Literature Review ... 6
2.1 Growth Accounting Approach ... 6
2.2 Okun’s Law... 7
2.3 National vs. Sub-National Analysis ... 8
2.4 Summary ... 8
3. Methodology ... 9
3.1 Growth Accounting Approach ... 9
3.2 Okun’s Law... 10
3.3 National vs. Sub-National Approach ... 10
4. Canada’s Labor market during recessions ... 12
4.1 Data ... 12
4.1.1 Provincial Data ... 12
4.1.2 Labor Market and Recessions ... 13
4.2 Broad patterns of the unemployment rate ... 14
4.3 Unemployment rate and demographic factors ... 16
4.4 Growth Accounting Approach ... 18
4.5 Unemployment rate and economic activity ... 20
4.6 A dynamic version of Okun’s law ... 22
4.7 Concluding Remarks and Caveats ... 22
5. The Dynamics of Okun’s law ... 23
5.1 Background ... 23
5.2 Results ... 24
5.2.1 Sub-periods ... 24
5.2.2 Full sample and “Have” “Have Not” results ... 25
5.2.3 Maritimes, West and the Centre ... 27
5.3 Concluding remarks ... 28
6. Conclusions ... 29
7. References ... 30
5
1. Introduction
The Great Recession started out in the US with the bust of the housing bubble. Banks that had invested in highly opaque Mortgage-Backed Securities were forced into bankruptcy. Starting with the Lehman Brothers in the fall of 2008, others soon faced the same problems when the inter-bank loan market dried up. Governments were forced to bail out banks globally as a result of a worldwide lack of liquidity access. In the US, the economy Shrunk by 2.7 percent and employment was nearly 6 percent lower by the end of 2009 (Verick and Islam, 2010 and Herkenhoff and Ohanian, 2011). In other advanced countries the economy on average shrunk even further. Between OECD countries the differences were vast. Guichard and Rusticelli (2010) projected an almost 10 percent increase in the Spanish and Ireland unemployment rate by the end of the crisis, whereas Germany, Norway and Korea were expected to stay below a 2 percent increase. In retrospect we know that these forecasted differences were not in the least overestimated. In other countries like Greece and Spain unemployment was soaring, reaching over 25 percent (Cazes, Verick and Al Houssami, 2013).
Canada’s performance during the recent crisis is nothing short of remarkable to the point that it deserves a reflection – both on the way the labor market performed recently, as well as in comparison to the past recessionary episodes (Zmitrowicz and Kahn, 2014). Our work seeks to investigate the following two questions: What are the patterns of the Canadian labor market during the Great Recession and what contrasts do the previous recessionary episodes offer? What role may demographic factors and economic activity have played, and has their relative importance been altered over time? Ultimately we want to find the answer as to why Canada was less hard hit in terms of unemployment rate changes during the past recession as compared to other countries.
6 activity in the changing patterns of unemployment. We investigated these patterns through the lens of Okun’s law, which has enjoyed renewed attention in a recent strand of literature. We will first develop a static model of the relationship between the real GDP growth rate and change in the unemployment rate, the way Okun constructed his model over 50 years ago. Secondly, we will push this model to allow for more dynamics. We will rewrite the formula to allow for differences in short-term and long-term effects. Next we investigate the stability of the Okun coefficient. Finally we advance to a Pooled Mean Group (PMG) estimator of Okun’s coefficient.
Our results underscore that unemployment levels were kept within acceptable margins during the last crisis. We find evidence that the relation between output and unemployment – Okun’s law – differs between times of economic recession and expansion. During recessions the short-term effects are prevalent. In contrast, the long-term effects are of greater significance during expansionary episodes. The drop in output that Okun holds responsible for unemployment developments does not tell the entire story. During the past recession for example we can see that decreases in labor productivity, as well as average hours worked moderated increases in the unemployment rate. The provincial analysis using the PMG approach provides us with solid proof of a long-term relation between output and the unemployment rate, that reconfirms the validity of Okun’s law. Furthermore we witness provincial differences that are masked when only national data is taken into account.
2. Literature Review
We start this paper with a brief overview of the relevant literature next to shortly mentioning the hypotheses that we will be addressing. This review is sliced up in the following items. First of all we will discuss a growth accounting approach. Second, we discuss the history of Okun’s law. Third, the Pooled Mean Group estimator is discussed in our discussion of a national vs. sub-national analysis.
2.1 Growth Accounting Approach
7
standard Okun equation (Chamberlin, 2011). Nonetheless, it is interesting to look at what the different variables have contributed to output growth. When Okun first developed his framework he based his ideas on that more labor was needed to produce more, something he considered a fact. However, now we know that the unemployment rate is not the only determinant of the amount of labor that is used in the economy, and that the way labor is used is important as well. The interplay between output and employment, described by labor productivity, is influenced by the contribution of capital and production efficiency. Employment inputs are determined by average working hours, participation rates, and the size of the working population (Chamberlin 2011).
The accounting identity approach used by Chamberlin suits our ambition of providing readily interpretable results in our paper, and is comparable to methods used by Daly et al. (2013) and Gordon (2010). We use this analysis to assess whether an analysis of unemployment and output is indeed sufficient without causing an omitted variable bias.
2.2 Okun’s Law
At the heart of this writing are the foundations of Okun’s law. Okun’s 1962 paper “Potential GNP: Its Measurement and Significance” was a – for its time – modern approach to the supply side of macroeconomic modelling. The main conclusion Okun drew was that a one-percentage point reduction in the unemployment rate would establish an approximately 3% higher output. Okun believed that it was important to stress the significance of this large output gain to policymakers. He was of the opinion that policymakers viewed a one-percent point decrease of unemployment as constituting to a raise in the economy’s grade of one-percent point, whereas in reality the rewards are much larger (Okun, 1962).
8 2010). We suspect that the Okun relation will vary over time (specifically during recessions) and geography.
2.3 National vs. Sub-National Analysis
The national analysis of Okun’s law provides us with some primary insights but needs to be extended for a number or reasons. The data used in the existing studies is limited and underscores the need for a more extensive approach. Secondly it can be interesting to exploit the heterogeneity between the Canadian provinces. Moreover, thorough literature on provincial analyses of the Okun relation in Canada is scarce and the analyses performed by other authors such as Adanu (2005) and Beaton (2010) are not without limitations.
In contrast to other studies with limited data, Huang and Yeh (2013) propose to use extensive panel datasets to assess the validity of Okun’s law. Next to extending data to 2 panel datasets (one for countries and one for the U.S. states) they adopt the Pooled Mean Group estimator approach first used by Pesaran, Shin and Smith (1999). The model extends prior research further.
First of all, research that investigates whether Okun’s law is stable over time is increasing but their results are ambiguous. Ball, Leigh and Loungani (2013) for example find that the Okun relationship is strong and stable over time, while Sögner and Stiassny (2002) find that there is no strong evidence of the Okun relation changing over time. Beaton (2010) concludes the opposite claiming that Okun’s relationship “exhibits structural instability” finding evidence for structural breaks in the relationship. Secondly we know that literature investigating the long-term relationship of output and unemployment is scarce and our research may contribute to this field.
2.4 Summary
9
have extended Okun’s original model to a more sophisticated approach that includes e.g. short- and long-term effects. The most promising and recent contribution was provided by Huang and Yeh (2013). Their approach is worth pursuing because there has been little refined research on Okun’s law on a sub-national level.
3. Methodology
3.1 Growth Accounting Approach
To examine the factors that contribute to output growth we use a decomposition method based on a mathematical identity that – in various formations – has been used by many authors (Gordon, 2010, Daly et al., 2013, and Chamberlin, 2011). The Identity is based on the different components that amount to GDP growth.
𝐺𝐷𝑃 = (𝐺𝐷𝑃 𝐻 ) × ( 𝐻 𝐸) × ( 𝐸 𝐿𝐹) × ( 𝐿𝐹 𝑃𝑂𝑃) × 𝑃𝑂𝑃 Where the various components represent labor productivity (𝐺𝐷𝑃
𝐻 ), hours worked per employee (𝐻𝐸), the employment rate (𝐿𝐹𝐸), the labour force participation rate (𝑃𝑂𝑃𝐿𝐹), and population (𝑃𝑂𝑃). To explain that the employment rate can be used as a surrogate for the unemployment rate we show that the unemployment rate is the inverse of the employment rate. 𝐺𝐷𝑃 =𝐺𝐷𝑃 𝐻 × 𝐻 𝐸× ( 𝐿𝐹 − 𝑈𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑒𝑑 𝐿𝐹 ) × 𝐿𝐹 𝑃𝑂𝑃× 𝑃𝑂𝑃 𝐺𝐷𝑃 =𝐺𝐷𝑃 𝐻 × 𝐻 𝐸 × (1 − 𝑢𝑟) × 𝐿𝐹 𝑃𝑂𝑃× 𝑃𝑂𝑃
10 relationship to convert the expression to an additive form. In this manner the relative impacts are not altered and the actual figures represent real values.
3.2 Okun’s Law
Okun (1962) proposed that the magnitude of this relationship could be easily considered by regressing the growth rate of GDP on change in the unemployment rate. Using quarterly data from the US ranging from 1947-1960 he estimated that a 3% increase in GDP will result in roughly a 1 percent decline in unemployment. The basic Okun relation is thus estimated by equation (1):
Δ𝑢𝑡 = 𝛼 + 𝛽1Δ𝑦𝑡+ 𝜖𝑡 (1)
Where Δ𝑢𝑡 is the annual percent point change in the unemployment rate, and Δ𝑦𝑡 is the annual growth rate of real GDP.
That Okun’s coefficient differs per country has long been established (Lee, 2000). Moreover several authors have in the past expressed their concern about the presumed stability of Okun’s relation (Sögner and Stiassny, 2002, Beaton, 2010). To assess the stability of Okun’s coefficient, Beaton (2010) as well as Sögner and Stiassny (2002) propose to add additional dynamics to the standard Okun estimation.
3.3 National vs. Sub-National Approach
The sub-national approach has many advantages. Frist of all, many studies on the Okun relationship rely on filtering techniques. For example the Hodrick-Prescott filter (Hodrick-Prescott, 1997) or the quadratic trend – both used by Adanu (2005) – to decompose data series into trend and cyclical components. It is “unlikely that filtering methods can generate distinct cyclical unemployment and output, result in dissimilar Okun’s estimates, and reach different conclusions” (Huang and Yeh, 2013). The PMG approach allows us to directly estimate Okun’s relationship without the necessity of filtering our data.
11
To estimate the long-term connection between variables Pesaran and Shin (1999) and Pesaran et al. (2001) propose an autoregressive distributed lag model. Suppose that the we can depict the long-run relationship between unemployment (uit) and output (yit) is given by
𝑢𝑖𝑡 = 𝛼𝑖+ 𝜃𝑦𝑖𝑡 + 𝑒𝑖𝑡
where αi is the fixed effects, i = sub-national economies and t = 1, 2, ... , T. The next step Pesaran
et al. (1999) suggest is to put Equation 1 in a general ARDL(p,q) specification which can be written as: 𝑢𝑖𝑡 = 𝛼𝑖 + ∑ 𝜆𝑖𝑗𝑢𝑖𝑡−𝑗 𝑝 𝑗=𝑖 + ∑ 𝛿𝑖𝑗𝑦𝑖𝑡−𝑗 𝑞 𝑗=𝑖 + 𝑒𝑖𝑡
In line with Huang and Yeh (2013) We consider an ARDL lag order of (2,1) to be optimal. Next, a re-parameterization is needed to write this equation as an error correction form:
𝛥𝑢𝑖𝑡 = 𝛼𝑖 + 𝜙𝑖𝑢𝑖,𝑡−1+ 𝛽𝑖𝑦𝑖𝑡+ ∑ 𝜆𝑖𝑗∗ Δ𝑢𝑖,𝑡−𝑗 𝑝−1 𝑗=1 + ∑ 𝛿𝑖𝑗∗Δ𝑦𝑖,𝑡−𝑗 𝑞−1 𝑗=0 + 𝑒𝑖𝑡 = 𝛼𝑖+ 𝜙𝑖[𝑢𝑖,𝑡−1− 𝜃𝑖𝑦𝑖𝑡] + ∑ 𝜆∗𝑖𝑗Δ𝑢𝑖,𝑡−𝑗 𝑝−1 𝑗=1 + ∑ 𝛿𝑖𝑗∗Δ𝑦𝑖,𝑡−𝑗 𝑞−1 𝑗=0 + 𝑒𝑖𝑡 Where 𝜙𝑖 = −(1 − ∑ 𝜆𝑖𝑗), 𝛽𝑖 = ∑ 𝛿𝑖𝑗, 𝜆𝑖𝑗∗ = − ∑𝑝𝑚=𝑗+1𝜆𝑖𝑚, 𝛿𝑖𝑗∗ = − ∑𝑞𝑚=𝑗+1𝛿𝑖𝑚 , 𝑞 𝑗=0 𝑝 𝑗=1
and 𝑗 = 1,2, … , 𝑞 − 1. The long-run relationship between unemployment (
u
it) and output (y
it)is determined by the coefficient 𝜃𝑖 = −𝛽𝑖/𝜙𝑖. The short run coefficient is determined by 𝛿𝑖𝑗∗.
The error correction coefficient 𝜙𝑖 measures the adjustment speed of
u
it toward its long-run equilibrium following a change iny
it,. Having 𝜙𝑖 < 0 provides that a long-term relationship exist. If we find a negative and significant error correction coefficient we can therefore view this as proof of a long-term relation between unemployment and output exists.12 rejected. Pesaran et al. (1999) propose a Hausman (1978)-type test1 applied to the difference between the MG and the PMG estimates because the MG estimator’s estimates of the long-term coefficient means are inefficient if slope homogeneity holds. Based on this Hausman-type test we cannot reject the null hypothesis of long-run coefficient homogeneity and consider the PMG approach to be more consistent and efficient.
4. Canada’s Labor market during recessions
4.1 Data
We examine the patterns displayed by the unemployment rate during the four decades that started in 1976. To the extent possible, our investigation goes all the way back to this period using seasonally-adjusted monthly series. We will consider this period for the following reasons. First of all it is the first year when Statistics Canada started to produce a consistent set of labor market data at both the national and sub-national levels. Secondly this period comprises three major recessions, providing us with enough data to draw meaningful conclusions on the differences between recessionary episodes and expansion.
When combined with real gross domestic product (GDP) series (which Statistics Canada provides from 1981 onwards on a quarterly basis), the labor market activity data need to be altered accordingly in a way to ensure analytical consistency. This means that we had to collapse our monthly employment figures to quarterly data. To extend our real GDP data sample to 1976 we used the real GDP growth rate to calculate figures for Real GDP in the pre-1981 years. We used the average growth rate of real GDP for January and February 2016 as a proxy for the real GDP growth rate of the 1st quarter of 2016. We will consider this period
4.1.1 Provincial Data
We use a panel of cross-province output and unemployment rate data for Canada’s ten provinces (Alberta, British Columbia, Manitoba, New Brunswick, Newfoundland and Labrador, Nova Scotia, Ontario, Prince Edward Island, Quebec, and Saskatchewan). The sample ranges from 1981 to 2014 amounting to a panel of 340 observations. Apart from an increased number
1 The results of the Mean Group estimator and the Hausman-type test are available upon request but are left out
13
of observations this dataset allows us to “control for region specific effects and mitigate omitted variables problems” (Huang and Yeh, 2013).
4.1.2 Labor Market and Recessions
Since the 1990s Canadas fiscal position is the strongest of the G-10 countries, but the Canadian labor market has been turbulent during recessions (Bordo and Meissner, 2015). This paper intents to investigate fluctuations of the Canadian unemployment rate, and specifically the impact of recessions is examined. To do so we need to first determine during which periods in our sample Canada was experiencing an economic downturn. One option would be to examine the periods in our sample that show prolonged rising unemployment. However, unemployment alone does not determine when a country is experiencing a recession.
A more usual approach for dating turning points in the business cycle would be to use data on the GDP, which generally gives slightly different dates for peaks and troughs (see Leamer 2008). To do so we turned to work by Cross and Bergevin (2012) who identify the behavior of the Canadian business cycle based on output and unemployment fluctuations. A recession is a fall in aggregate demand that causes output to drop. Cross and Bergevin identify 3 periods of recession with a subsequent drop in unemployment within our sample. The first major recession starts in June 1981 and lasts until October 1982. The second period of economic downturn was from March 1990 through April 1992. The last period of recession began in October 2008 and ended in May 2009.
Table 4.1: Historical Chronology of Canadian Recessions since 1976
Monthly Peak (Quarterly) Monthly Trough (Quarterly) Category
January 1980 (1979:Q4) June 1981 (1981:Q2) March 1990 (1990:Q1) October 2008 (2008:Q3) June 1980 (1980:Q2) October 1982 (1982:Q4) April 1992 (1992:Q2) May 2009 (2009:Q2) 1 4 4 4
Notes: Category 1: Short, mild drop in real GDP and no decline in quarterly employment. Category 4: Substantial decline in both real GDP and employment, usually for a period of
14
4.2 Broad patterns of the unemployment rate
Figure 4.1 displays the patterns of real GDP and employment rate over the entire historical period considered in this paper, while the recessions identified by Cross and Bergevin (2012) are indicated in grey bars. It shows clearly the co-movement of the annual changes in the employment rate and the growth rate of real GDP. In essence the employment rate is the inverse of the unemployment rate. Would the unemployment rate have been featured here, a strongly inverse relationship would be visible.
Figure 4.1: Real GDP growth rate and annual changes in the employment rate
The unemployment rate itself ranged from a low of 5.9% in January 2008 to a high of 13.1% in December 1982. The pattern shows a steady decline from 1982 onwards accompanied by upturns during the 1990 recession and the Great Recession along with some milder fluctuations. During the Great Recession the unemployment rate “only” increased to 8.6 percent. This is an increase of 2.7 percentage points as compared to the all-time low of 2008, but this is still a considerably lower increase than in other countries.
Another important feature of the unemployment in Canada is its wide variation across provinces and its persistence over the business cycle with some important regularities and important changes. In particular, commodity-rich provinces (Manitoba, Saskatchewan and Alberta) report systematically a below average unemployment rate. The Great Recession shows a major reversal for Ontario, the traditional heartland of manufacturing. For the first time ever, it reports an above-average unemployment rate, a clear indication that the commodity-based
-6,00% -4,00% -2,00% 0,00% 2,00% 4,00% 6,00% -3,5 -2,5 -1,5 -0,5 0,5 1,5 2,5 Q1 19 77 Q3 19 78 Q1 19 80 Q3 19 81 Q1 19 83 Q3 19 84 Q1 19 86 Q3 19 87 Q1 19 89 Q3 19 90 Q1 19 92 Q3 19 93 Q1 19 95 Q3 19 96 Q1 19 98 Q3 19 99 Q1 20 01 Q3 20 02 Q1 20 04 Q3 20 05 Q1 20 07 Q3 20 08 Q1 20 10 Q3 20 11 Q1 20 13 Q3 20 14 Q1 20 16
15
industries were not as hard hit during the crisis as compared to manufacturing and services. This also serves as an explanation as to why Canada did not feel the past recession as hard as other countries, something we will be discussing over the course of this section.
Table 4.1: Ratios of unemployment by province
Ratios of unemployment to average
Jun-81 Mar-90 Oct-08 Nov-82 May-92 Oct-09
Sex Oct-82 Apr-92 May-09 Feb-90 Sep-08 Apr-16
Region
Newfoundland and
Labrador 1.59 1.85 1.95 1.87 2.07 1.76
Prince Edward Island 1.26 1.65 1.62 1.33 1.62 1.50
Nova Scotia 1.21 1.22 1.13 1.26 1.26 1.23 New Brunswick 1.38 1.31 1.16 1.41 1.30 1.32 Quebec 1.33 1.20 1.09 1.18 1.20 1.06 Ontario 0.87 0.88 1.09 0.75 0.94 1.05 Manitoba 0.77 0.86 0.63 0.84 0.73 0.74 Saskatchewan 0.57 0.75 0.59 0.79 0.71 0.64 Alberta 0.62 0.83 0.68 0.98 0.71 0.75 British Columbia 1.04 0.98 0.87 1.28 0.93 0.93
Note: The periods of recession are indicated in grey.
16 Figure 4.2: Rate of unemployment, Canada, Development during recessions starting …
4.3 Unemployment rate and demographic factors
Understanding the sources of the unemployment rate across various downturns is important. Changes in the unemployment rate can be driven by changes in the portion of the population that is employed, and the fraction of the population that is participating in the labor force. Figure 4.3 shows that a declining employment to population ratio is responsible for the largest share in rising unemployment rates. Furthermore figure 4.3 demonstrates the severity of the earlier recessions in terms of rising unemployment. The behavior of Labor Force Participation and the employment to population rates are quite similar between the three recessions.
Figure 4.3: Cumulative percent point changes in Labor Force Participation and Employment rates since start of recessions
-1 0 1 2 3 4 5 6 0 2 4 6 8 10 12 14 16 18 20 22 24 Perc an t p o in t ch an ge in r at e o f u n em p loy m en t
Months since start of recession
June 1981 March 1990 October 2008 -5% -4% -3% -2% -1% 0% 1% 1 3 5 7 9 11 13 15 17 19 21 23 25 Perc en t p o in t ch an ge
Months from start of recession
17
In table 4.2 we decomposed the change in the unemployment rate into Male and Female Full-Time employment and Part-time employment, as well as Labor Force Participation (Borland 2014). From this table we see that over time the increased participation of women in the labor force has substantially influenced the periods of recovery from 1982 to 1990 and from 1992 to 2008. Furthermore we see that during the periods of recession it is a decline in male and female full-time employment that is responsible for the majority of the rise in unemployment.
The table can be read as follows. Over the recession period of 2008-2009, there has been a 2.6 percent point rise in unemployment. A decline in Male Full-Time Employment has been responsible for a 2.4 percent point rise in unemployment. A decline in the Female Full-Time employment accounted for 1 percent of the increase in unemployment, while Labor Force participation was responsible for a negative influence on the unemployment rate. In the recovery periods we see that Female Full-Time- Employment has played a substantial role in pushing unemployment rates down.
Table 4.2: Sources of changes in the rate of unemployment in Canada
Effect of: Change in Rate of UE Males - FTE/POP Males PTE/POP Males LF/POP Females FTE/POP Females PTE/POP Females LF/POP Recession Jun 1981 - Oct 1982 +6.3 +5.6 -0.2 -0.9 +2.1 -0.2 -0.1 Mar 1990 - Apr 1992 +3.8 +4.8 -0.2 -2.0 +1.8 -0.3 -0.3 Oct 2008 - May 2009 +2.6 +2.4 +0.0 -0.5 +1.0 -0.2 -0.1 Recovery/Expansion Nov 1982 - Feb 1990 -5.9 -2.9 -0.6 -0.2 -5.3 -1.5 +5.2 May 1992 - Sep 2008 -5.3 -1.4 -0.9 -0.8 -4.9 -0.8 +3.8 Jun 2009 - Apr 2016 -1.7 -0.0 -0.1 -1.1 -0.3 +0.8 -0.9
Note: The decomposition is derived from: ur ≈ –ln[αmt ((FTE/POP)mt · (POP/LF)mt) + αmt ((PTE/POP)mt · (POP/LF)mt) +
(1 – αmt)((FTE/POP)ft · (POP/LF)ft) + (1 – αmt)((PTE/POP)ft · (POP/LF)ft)] where ur is the rate of unemployment, αmt =
proportion of males in labor force at time t,(FTE/POP)mt and (PTE/POP)mt are the full-time and part-time
employment-to-population rates for males, and (POP/LF)mt is the inverse of the labor force participation rate for males. The decomposition
18
4.4 Growth Accounting Approach
Changes in the unemployment rate are not without economic consequences. When less people are working, a nations’ output will most likely decrease. Before we dig further into the relationship between real GDP growth rate and unemployment is we will first examine which factors contribute to output growth. We want to identify the role that unemployment rate changes play in output fluctuations as compared to other factors. Hence, we next consider a production function to relate changes in output to more inputs than employment alone. It is important to mention that because labor productivity is determined by an array of factors it functions as a residual in this approach since it includes factors that are unmeasured.
Figure 4.4 shows the contributions of labor productivity (output per hour), hours worked per employee (average hours), the employment rate (1-ur), labor force participation, and population to the annual growth of output during the periods of recession/expansion in our sample. Several things are worth noticing. First of all it shows that labor productivity growth varies considerably over time. Through 1992 Q3 to 2008 Q2 it was the most important factor next to population growth which is a stable variable over the different time periods. Average hours have been a relatively unimportant variable. Over the total sample average hours would have reflected a negative impact on output growth. Over the subsamples first
19
subsample this contribution is slightly positive, following a rise in average hours worked in the last 5 years of the 1980s. Similarly, concerning the last subsample, average hours have inclined after reaching an all-time low in the second quarter of 2009. Rising employment rates have not had a colossal impact on output growth but the relative impact has remained stable over the subsamples. This can be viewed as proof of a stable relationship between output growth and employment/unemployment. Concerning labor force participation we see a negative impact on output growth in the period past the Great Recession.
Since we have hypothesized that the relationship between employment and output is largest during periods of economic recession it looks promising to review the sub periods of recessionary episodes. Figure 4.5 shows the contributions to output for these periods. Labor force productivity growth has been a moderating factor in the 1981 and 1990 recession, but contributed to the decline in output during the Great Recession. We know that the unemployment rate has not risen as violently in the last downturn. One explanation could be that employment has remained relatively more stable at the cost of using labor effectively, therefore explaining the negative effect of labor productivity in the latest recession.
As opposed to the expansion subsamples we see that the average hours worked have dropped over all three recessionary periods. This could be an indication that employers have reduced hours to maintain the level of employment. We have established before that unemployment and output show a larger correlation during economic downturns. Once again
20 this is demonstrated in the chart. During all subsample periods we see that output largely co-moves with the employment rate. Labor force participation has mildly dropped during all of our recessionary periods.
The most important take-away from this is that we have determined that – especially during recessions – it is employment that determines the largest part of output fluctuations. Moreover, we know that the similar cyclical behavioral patterns behind unemployment fluctuations are reflected by the other factors that we have examined in this section, reducing the chance of omitted variable bias. This leads us to believe that an analysis of output and unemployment is sufficient in the next part of this writing.
4.5 Unemployment rate and economic activity
Changes in the unemployment rate may also be driven by changes in economic growth in accordance to the so-called Okun’s law. Figure 4.6 reveals a close negative relationship between the two series. Next to highlighting the observations for the past recessionary episode, the line of best fit that a simple OLS regression of GDP growth on unemployment brings about is indicated in the figure as well. We see that half of the data points for the 2008 downturn are not expressing rising unemployment, nor declining GDP. This is an interesting observation that brings us closer to understanding why Canada was less hard hit than other countries in the past crisis.
21
Figure 4.6: Scatter of change in the unemployment rate and the growth rate of GDP
Figure 4.7 shows the growth of GDP consistent with a stable unemployment rate, or in other words, the rate of GDP growth that is needed to maintain the unemployment rate (Chamberlin, 2011). This ratio is based on dividing (minus) the constant by the coefficient for GDP growth, commonly referred to as Okun’s coefficient. The figure below clearly shows that during periods of recession a much higher rate of GDP growth is needed to maintain the unemployment rate than during periods of expansion. The intercept as well as the Okun coefficient are higher during recessions, indicating a higher unemployment growth at zero GDP growth and a steeper inverse relation between GDP and unemployment.
Figure 4.7: Growth of GDP Consistent with a Stable Unemployment Rate
-6,0 -4,0 -2,0 0,0 2,0 4,0 6,0 8,0 -3,0 -2,0 -1,0 0,0 1,0 2,0 3,0 4,0 5,0 Annual rate of growth in Real GDP
22
4.6 A dynamic version of Okun’s law
The unemployment rate does not change immediately due to output effects. Labor market rigidities are caused the costs of changing the workforce, the so-called hiring and firing costs. Taking on extra workers during an upturn creates liability in case the number of workers needs to be brought down in the future. The labor market rigidity is not reflected in the static version that we have examined so far. In line with other authors we add additional dynamics to the standard Okun regression. This enables us to extract from the total Okun effect the Short-term effects.
Since this analysis concerns a somewhat naïve extension of the standard Okun regression we have decided to explain the details of the model, as well as the full results in the appendix. The results shown below demonstrate that during periods of expansion the short-term effects are negligible. In contrast we see that during recessions the short-short-term effects are by far the strongest.
Figure 4.8: Growth of GDP Consistent with a Stable Unemployment Rate: Short-term and Total Effect
Note: Short-run effect for the Expansions not statistically significant
4.7 Concluding Remarks and Caveats
The past analysis has taught us that Canadian unemployment levels showed significant differences over the course of the past 40 years, especially when we compare the recessionary episodes that lie in our sample. The source of these changes can be found mainly in job losses,
23
rather than changes in labor force participation rates. We have seen that the factors that contribute to output fluctuations show the patterns that one would expect. Specifically during recessions, declines in Canadian real GDP are primarily driven by a rising unemployment rate. This leads us to believe that pursuing our analysis of the relationship between the unemployment rate and real GDP is not only sufficient, but also valuable. For this reason we progress to a more advanced model that analyzes unemployment output relation.
It is highly likely that a long series of economic data as the one we are using contains a number of structural breaks. The simple model that we have used so far will just average all these breaks. To assess the stability of Okun’s relation we use a rolling regression based on a window of 40 quarters. We use the 40 quarter window because a smaller window could possibly be too sensitive to structural instability. Figure 1 in the appendix shows that indeed Okun’s law can mask economic changes. On the other hand the results have to be viewed with some degree of caution. Using a 10 year rolling window does provide us with a proper test of stability, but in Canada this does mean that for a large portion of the rolling regressions there is not an economic downturn incorporated in the 10 year window. This is visible in the long-term Okun effect that decreases significantly toward the end of the last decade.
5. The Dynamics of Okun’s law
5.1 Background
To further establish the stability of Okun’s relation an approach is adopted on a panel of data on Canada’s ten provinces2. The analysis performed so far emphasized considerable variation in the relation between economic activity and unemployment across the different phases of the business cycle. However, it remains conspicuously silent on provincial heterogeneity, which has the potential to uncover important differences in the responsiveness of labor markets to changes in output. These differences may arise from differences in economic structures, as well as demographic characteristics (e.g. participation rate, employment rate, etc.). Performing the analysis of Okun’s law at the provincial level has the
2 Canada’s 10 provinces are Alberta, British Columbia, Manitoba, New Brunswick, Newfoundland and Labrador,
24 merit of controlling for the effect of monetary and fiscal policies that apply uniformly to all provinces, while isolating the effects of economic structures and demographic factors.
For example, literature suggests that the more industrialized provinces (e.g. Ontario) experience higher unemployment effects resulting from declining GDP than those that are less industrialized, like the Maritime provinces of Prince Edward Island, New Brunswick, and Nova Scotia (Adanu, 2005). Third, we use our model to estimate differences between the “have” and “have not” provinces, counterintuitively referring to the provinces that do not receive net equalization payments, and those that do receive equalization payments on net, respectively. Later in this chapter we will briefly discuss the equalization payments system further.
To account for these considerations we apply the Pooled Mean Group (PMG) method to the 10 provinces. This method allows for short-run heterogeneous dynamics but imposes a long-run homogeneous relationship. In particular, the method allows the intercepts, short-run coefficients, and error variances to be different across groups, while the long-run coefficients are constrained to be homogeneous. We believe that there are good reasons to believe that the long-run equilibrium relationship amongst variables should remain identical across groups, while the short-run dynamics are heterogeneous. This dynamic estimator is more likely to capture the true nature of the relationship between output and unemployment.
5.2 Results
5.2.1 Sub-periods
Concerning the results of column (1) in table 5.1 for the sample period of 1981-1991 it is interesting to note that there is a positive, but small and insignificant, result for the long run output coefficient. When confronted with similar results for Non-OECD countries, Huang and Yeh (2013) conclude that for these countries there is no tradeoff between unemployment and output in the long-run.
25
are omitted from the regression the long-term coefficient again takes on a strongly negative form. Similar results are expected to surface would we be able to add observations for the years prior to 1981.
The long-term output coefficient in column (2) is strongly negative with a -0.140 sign. A possible explanation is that this is a sign of the Canadian economy gaining momentum after the 1992 recession. The signs for the error coefficient are still negative and significant, with low standard errors, once more providing evidence for a strong long-term relationship between output and unemployment. The autoregressive coefficients for unemployment are again positive and significant at the 10 percent level or more. Column (3) shows a long-term output sign that is comparable to that of the full sample. However, the short-term coefficient for output is considerably larger than those of any subsample we have discussed so far, but we should consider the fact that the last subsample of the years 2008-2014 has only 70 observations.
Table 5.1: Okun’s law in Canada during 1981-1991. 1992-2007, and 2007-2014
Subsamples VARIABLES 81-91 (1) 92-07 (2) 07-14 (3) Long-run coefficient Yit 0.0176 (0.0413) -0.140*** (0.00659) -0.0708*** (0.0115) Error correction coefficient
ϕ -0.318*** (0.0546) -0.417*** (0.0825) -0.366** (0.164) Short run coefficient
Δui,t-1 0.490*** (0.0363) 0.228*** (0.0621) 0.402* (0.218) Δyi,t -13.11*** (4.322) -11.50*** (3.881) -0.391*** (0.119) Constant -1.760*** (0.659) 65.44*** (11.61) 29.90** (13.31) Observations 90 160 70
Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1
5.2.2 Full sample and “Have” “Have Not” results
26 with less than average income3. The program is funded by general tax revenues to improve the fiscal capability of the provinces involved. It is aimed at reducing regional inequalities without threatening provincial autonomy.
As stated before, we show only the results of the PMG approach, since the null hypothesis of identical long-run parameters between the PMG and MG approach examined with a Hausman-type test cannot be rejected at any significant level. The PMG results in table 5.2 show negative error coefficients for the full sample, as well as both subsamples. These results are significant at the 1% level and suggest that there is strong proof of the existence of a long-term relation between output and unemployment. Furthermore, from column (1) we see that the long-term Okun coefficient for the full sample is -0.0691 and significant at the 1% level.
Table 5.2: Okun’s law in Canada and “Have” “Have Not” Provinces
Subsamples
VARIABLES Full (1) Have (2) Have not (3)
Long-run coefficient
Yit -0.0691*** -0.0614*** -0.0938***
(0.00463) (0.00486) (0.00715)
Error correction coefficient
ϕ -0.271*** -0.329*** -0.275***
(0.0398) (0.0790) (0.0768)
Short run coefficient
Δui,t-1 -0.290*** 0.355*** 0.244*** (0.0370) (0.0523) (0.0488) Δyi,t -0.153*** -0.148** -0.162*** (0.0323) (0.0601) (0.0401) Constant 23.04*** 25.70*** 30.01*** (3.0349) (4.934) (8.007) Number of provinces 10 4 6 Observations 320 128 192
Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1
The coefficient implies that if Canada’s GDP is increased by one percent, the unemployment level is dropped by 6.9 percent points. Column (2), for the “Have” province, show a comparable coefficient, of -0.0614. In contrast, the “Have Not” provinces (3) have a
3 The “Have” provinces are Newfoundland and Labrador, Ontario, Saskatchewan, and Alberta. The “Have Not”
27
long run output coefficient of -0.0938, meaning that a rising output has a much larger influence on unemployment for these 6 provinces. This could be interpreted as evidence for the importance of the equalization payments. The nature of the PMG approach does not allow us to take conclusions on a single pooled coefficient for the short-term dynamics. It does, however, provide the mean of the coefficient across provinces which can be interpreted as an average effect.
From the autoregressive lagged unemployment parameters we can conclude that there is a highly significant, but small, and ambiguous relationship underscoring a small persistence of the dynamics of the unemployment rate for the two subsamples. The estimates of the short-term coefficient of output (-0.153 for the full sample, -0.148, -0.162 for the two subsamples) are strongly negative and significant at the 1% level, providing support for the existence of a robust and strong Okun relationship across Canada’s provinces.
5.2.3 Maritimes, West and the Centre
Our final analysis concerns a subsampling of the 4 most western provinces (British Columbia, Alberta, Saskatchewan, and Manitoba) the 2 provinces in the Centre (Quebec and Ontario) and the West provinces, also referred to as the Maritimes (New Brunswick, Nova Scotia, and Prince Edward Island, and NFLD). The results are shown in table 5.3. The error correction coefficient is again negative and highly significant for all subsamples, indicating the existence of a long-term relationship between output and unemployment. The long run coefficients for output (
Y
it ) are again negative and provide proof of Okun’s law for Canada andit’s provinces. The signs provide us with proof that Okun’s relation is significantly different in the Eastern provinces. The signs of the autoregressive part of the equation concerning unemployment are still small and significantly positive, once more indicating persistence of the unemployment rate dynamics. Lastly the short-term effect of output is large and significantly negative.
Table 5.3: Okun’s law in the Western-, Eastern-, and Maritime provinces
Subsamples
28 Long-run coefficient
Yit -0.0973*** -0.0665*** -0.0600***
(0.00770) (0.0214) (0.00495) Error correction coefficient
ϕ -0.376*** -0.144*** -0.291***
(0.0880) (0.0156) (0.0804) Short run coefficient
Δui,t-1 0.272*** 0.235*** 0.326*** (0.0729) (0.0666) (0.0492) Δyi,t -0.0891*** -0.291*** -0.158*** (0.0224) (0.00333) (0.0506) Constant 40.95*** 14.05*** 21.87*** (10.07) (1.732) (5.293) Number of provinces 4 2 4 Observations 128 64 128
Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1
5.3 Concluding remarks
The provincial analysis utilizes an extensive panel dataset which, however annual, provides us with more observations than the quarterly national dataset on unemployment and output. The PMG approach which does not require unit root testing, and allows for co-integration between unemployment and output had produced the following interesting results.
29
6. Conclusions
Our analysis has shown how the Canadian unemployment rate has developed over the past 40 year and provided some valuable insights. Where other countries have yet failed to bring unemployment levels down after the Great Recession, Canadian employment has quickly recovered. From exploring the patterns of Canadian unemployment our research progressed to a sophisticated approach on Okun’s law: the rule that explains the inverse relation between changes in the unemployment rate, and the real GDP growth rate.
In our growth accounting approach we have analyzed the factors that contribute to changes in output. From this we can conclude that although this is an interesting practice, the unemployment rate is the main determinant of output fluctuations. Our analysis of Okun’s law has brought about the following. We see dramatic differences in the relation between the growth rate of output and the change in the unemployment rate during recessions and expansionary episodes. The effects seem to be larger during recessions. One explanation for this could be found in the role of systemic uncertainty, which causes the negative output growth effects of an economic recession to be magnified.
30
7. References
Adanu, K. (2005). A cross-province comparison of Okun's coefficient for Canada. Applied Economics, 37(5), 561-570.
Ball, L. M., Leigh, D., and Loungani, P. (2013). Okun's Law: Fit at Fifty? (No. w18668). National Bureau of Economic Research.
Bartlett, R and Lao, H. (2014). Labour Market Assessment 2014. Office of the Parliamentary Budget Officer .
Bordo, M. D., Meissner, C.M. (2015). Growing up to financial stability, Financial Globalization, Financial Development and Financial Crises (No. w21287). National Bureau of Economic Research.
Borland, J. (2014). Recent Unemployment Experience in New Zealand (No. 14/01). New Zealand Treasury Working Paper. The Treasury, Wellington.
Beaton, K. (2010). Time variation in Okun's law: A Canada and US comparison(No. 2010, 7). Bank of Canada Working Paper.
Cazes, S., Verick, S., and Al Hussami, F. (2013). Why did unemployment respond so differently to the global financial crisis across countries? Insights from Okun’s Law. IZA Journal of Labor Policy, 2(1), 1-18.
Campolieti, M. (2012). The Canada-US unemployment rate gap: A new look with a new decomposition for cross-country differences in unemployment rates. Canadian Public Policy, 38(3), 411-435.
Chamberlin, G. (2011). Okun’s Law revisited. Economic and Labour Market Review, 5(2), 104-132.
Cross, P., and Bergevin, P. (2012). Turning points: business cycles in Canada since 1926. CD Howe Institute, 366.
Daly, M. C., Hobijn, B., Şahin, A., and Valletta, R. G. (2012). A search and matching approach to labor markets: Did the natural rate of unemployment rise?. The Journal of Economic Perspectives, 26(3), 3-26.
Demers, F., and Macdonald, R. (2007). The Canadian business cycle: a comparison of models (No. 2007, 38). Bank of Canada Working Paper.
Elsby, M., Hobijn, B., and Sahin, A. (2008). Unemployment Dynamics in the OECD (2). National Bureau of Economic Research.
31
Elsby, M. W., Michaels, R., and Solon, G. (2009). The ins and outs of cyclical unemployment. American Economic Journal: Macroeconomics, 1(1), 84-110.
Gordon, R. J. (2010). Okun's law and productivity innovations. The American Economic Review, 100(2), 11-15.
Guichard, S., and Rusticelli, E. (2010). Assessing the impact of the financial crisis on structural unemployment in OECD countries.
Hausman, J. A. (1978). Specification tests in econometrics. Econometrica: Journal of the Econometric Society, 1251-1271.
Herkenhoff, K. F., and Ohanian, L. E. (2011). Labor market dysfunction during the Great Recession (No. w17313). National Bureau of Economic Research.
Hodrick, R. J., and Prescott, E. C. (1997). Postwar US business cycles: an empirical investigation. Journal of Money, credit, and Banking, 1-16.
Holmes, M. J., and Silverstone, B. (2006). Okun's law, asymmetries and jobless recoveries in the United States: A Markov-switching approach. Economics Letters, 92(2), 293-299.
Huang, H. C., and Yeh, C. C. (2013). Okun's law in panels of countries and states. Applied Economics, 45(2), 191-199.
Leamer, E. E. (2008). What's a recession, anyway? (No. w14221). National Bureau of Economic Research.
Lecours, A., and Béland, D. (2010). Federalism and fiscal policy: The politics of equalization in Canada. Publius: The Journal of Federalism, 40(4), 569-596.
Lee, J. (2000). The robustness of Okun's law: Evidence from OECD countries.Journal of macroeconomics, 22(2), 331-356.
Macklem, T., and Barillas, F. (2005). Recent developments in the Canada-US unemployment rate gap: changing patterns in unemployment incidence and duration. Canadian Public Policy/Analyse de Politiques, 31(1), 101-107.
Okun, A. M. (1962, September). Potential GNP: its measurement and significance. In Proceedings of the business and economic statistics section of the American Statistical Association (Vol. 7, pp. 89-104).
Pesaran, M. H., Shin, Y., and Smith, R. P. (1999). Pooled mean group estimation of dynamic heterogeneous panels. Journal of the American Statistical Association, 94(446), 621-634.
Pesaran, M. H., Shin, Y. (1999). An autoregressive distributed lag modelling approach to cointegration analysis. Econometrics and Economic Theory in the 20th Century: The Ragnar
32 Pesaran, M. H., Shin, Y., and Smith, R. J. (2001). Bounds testing approaches to the analysis of level relationships. Journal of applied econometrics, 16(3), 289-326.
Riddell, W. C., and Sharpe, A. (1998). The Canada-US Unemployment Rate Gap: An Introduction and Overview. Canadian Public Policy, 24(s1), 1-37.
Shimer, Robert (2007). Reassessing the Ins and Outs of Unemployment, (No. w13421). National Bureau of Economic Research.
Sögner, L., and Stiassny, A. (2002). An analysis on the structural stability of Okun's law--a cross-country study. Applied Economics, 34(14), 1775-1787.
Verick, S., and Islam, I. (2010). The Great Recession of 2008-2009: causes, consequences and policy responses.
Virén, M. (2001). The Okun curve is non-linear. Economics letters, 70(2), 253-257.
33
Appendix
1: Okun’s law: Static version
Table: Okun’s relation during recovery/expansion and recession
(1) (2) (3)
VARIABLES All Recession Expansion
Growth in GDP -0.390*** -0.495*** -0.291*** (0.0338) (0.0766) (0.0435) Constant 0.979*** 1.228*** 0.635*** (0.112) (0.153) (0.153) Observations 157 21 136 R-squared 0.636 0.777 0.401
Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1 regression results using annual rates of change in GDP and difference in the rate of unemployment, with robust standard errors, 1976q1-1977q1
34
Appendix
2: Okun’s law: Dynamic Version
To examine whether unemployment reacts with a delay to changes in output growth, Beaton (2010) adds a lagged output variable to the standard Okun equation (1).
Δ𝑢𝑡 = 𝛼 + 𝛽1Δ𝑦𝑡+ 𝜖𝑡 (1)
Resulting in the relation in regression (1) based on:
Δ𝑢𝑡 = 𝛼 + 𝛽1 𝑌𝑡 + 𝛽2𝑦𝑡−1 + 𝜖𝑡 (2)
Which can be rewritten (regression (2) ) as:
Δ𝑢𝑡 = 𝛼 + 𝛽1𝛥𝑌𝑡 + (𝛽1 + 𝛽2)𝑦𝑡−1 + 𝜖𝑡 (3)
Where Δ𝑢𝑡 is the annual change in unemployment of the same quarter one year before and
Y
t is the annual growth rate of real GDP from the same quarter one year before. The short term-effect corresponds to β1, and the total Okun effect is captured in β1 + β2. We have learned thatthe addition of this variable is preferable because the lagged output variable helps “reduce the bias for the total effect of output on unemployment when output growth is positively auto correlated” (Sögner and Stiassny, 2002). In particular, the bias in the estimated short-run effect (β1), is reduced by an opposite bias in (β2)” (Beaton, 2010).
Table: Okun’s relation in short-term and total effect during recession and expansion
COEFFICIENTS All Recession Expansion
Short term Okun Effect (β 1) -0.236*** -0.425*** -0.0683
(0.0595) (0.0903) (0.0667) Total Okun effect (β 1 + β 2) -0.417*** -0.510*** -0.313***
(0.0341) (0.0888) (0.0370)
Constant (α) 1.041*** 1.274*** 0.671***
(0.110) (0.168) (0.122)
Observations 156 21 135
R-squared 0.681 0.783 0.523
35
Appendix
3: Rolling Regressions of Okun’s Coefficient: Short-term and Long-term effects
Figure: Rolling regression of Short-term effect (β 1)
Figure 2: Rolling regression of Long-term effect (β 2)
-. 5 -. 4 -. 3 -. 2 -. 1 0 .1 .2 1985q1 1990q1 1995q1 2000q1 2005q1 2010q1 2015q1 end
Short Term Okun Effect Lowerbound/Upperbound
-. 6 -. 5 -. 4 -. 3 -. 2 -. 1 0 .1 .2 1985q1 1990q1 1995q1 2000q1 2005q1 2010q1 2015q1 end
