Unemployment effects of government spending cuts : a panel data approach among 17 OECD countries from 1978-2009

(1)

1

Unemployment effects of government spending cuts:

A panel data approach among 17 OECD countries from

1978-2009.

Abstract:

This paper examines the unemployment impact resulting from government spending cuts.

A sample of 17 OECD countries is constructed, covering a time period from 1978 to 2009.

A deficit-driven fiscal consolidation measure from Devries et al. (2011) is used as proxy for

exogeneous government spending cuts. The empirical findings indicate a positive

government spending cut-unemployment relation, consistent with most existing research

on this topic. However, endogeneity problems remain present in all 8 estimated models.

Boudewijn Pragt

Track: Economics

Studentnumber: 10389601

Field: Macroeconomics

(2)

2

Statement of Originality

This document is written by Student Boudewijn Pragt who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the

(3)

3

Statement of Originality 2 1. Introduction 4 2. Literature review 7 2.1. Theoretical insights 7 2.2. Labor market institutions 7 2.2.1. Institutional structure of wage determination 7 2.2.2. Employment protection policy 8 2.2.3. Wage rigidity and Minimum wages 8 2.2.4. Labor Tax 8 2.3. Institutional interactions 8 2.4. Shocks 9 2.4.1. Aggregate demand shocks 9 2.4.2. Aggregate supply shocks 9

2.4.3. (Ex-post) real interest rate 9 2.5. Interactions between shocks and labor market institutions 10 2.6. Unemployment effects of government spending cuts 10 2.6.1. Purely empirical papers 10 2.6.2. Theoretical and partially empirical papers 11 3. Data and Methodology 13

3.1 Baseline model of Hooker & Knetter (1994) 13

3.2. Modifications of baseline model 13

3.2.1. Different sample countries relative to the baseline model 13

3.2.2. Different main independent variable 13 3.2.3. Disregarding quintile sub-sampling 14 3.3. Variable parameterization 14

3.3.1. Dependent variable (U it) 14

3.3.2. Independent variables 14 3.4. Key statistics 15 3.5. Panel Data Fixed Effects models 17 3.6. Fixed Effects Regression Assumptions 18 3.7. Hypotheses 19 3.8. Empirical Results 20 4. Conclusion 23

4.1. Suggestions for future research 23 References 24

Appendices Appendix A: Test Diagnostics 27

Appendix B: Residuals vs Independent variables for model 4 30

Appendix C: Goodness of fit measures 32

(4)

4

1. Introduction

During the 2008 financial crisis European governments faced increasing social security expenditures, since unemployment rose in nearly all countries. This dynamic is consistent with results of Prasad & Gerecke (2010), who found that specifically advanced countries exhibit substantial countercyclical social security expenditures. Figure 1.A shows substantial unemployment increases in the first crisis years for 13 European countries. Reinhart & Rogoff (2009) showed that together with bank bailouts and collapsed tax revenues, these aspects were responsible for a major increase in government debt as percentage of GDP. In Europe the European Growth and Stability Pact forced countries to cut back on spending and decrease debt levels. Therefore, several countries including the Netherlands

implemented austerity oriented measures. These were designed to ensure compliance with the European Stability and Growth Pact. The Stability and Growth Pact aims at budgetary discipline and debt control (Beetsma & Uhlig, 1999). It is designed to mitigate governments’ propensity to not internalize negative externalities of excessive debt. However, the effectiveness of some

implemented austerity policy measures is contested amongst economists. Jacobs (2016) estimated the unemployment increase of the Dutch austerity policy to be around 365.000 jobs.1_{This estimate}

involves uncertainty, but indicates the potential magnitude of unemployment effects stemming from government spending cuts.

Figure 1.A

Note: Data stems from the World Bank 2017 (WDI).

1_Source: https://basjacobs.wordpress.com/2016/09/17/ing-heeft-gelijk-overheidsbeleid-in-de-periode-2011-2017-kostte-volgens-het-cpb-ongeveer-365-000-banen/ 0 5 10 15 20 25

FIN NLD ITA IRL DEU BEL DNK FRA PRT ESP SWE GBR AUT

Harmonized Unemployment rates

(5)

5 Figure 1.B shows a positive correlation of 0.31 between unemployment and government spending cuts. This co-movement proofs no causal relation, but does indicate a potential relationship between government spending and unemployment. Unemployment is persistent and relatively rigid, which implies no immediate contemporaneous response resulting from government cuts. Therefore, this correlation coefficient is modest enough to continue the analysis. To assess if the unemployment-government cut relation holds more generally, also non-crisis periods shall be taken into account. We are not interested in cyclical changes in the macro environment driving government cuts, but in deficit-driven fiscal consolidation. Using such an exogeneous government spending cut approach theoretically justifies a causal inference. The research question of this paper is:

“What are the unemployment effects of government spending cuts?”

A clear-cut answer to this question is important since it addresses relevant policy implications. Government’s job creation goals are intended to be reached via fiscal stimulation packages (Brückner & Pappa, 2012). These packages are financed using taxpayers’ money. Therefore, an efficiency assessment of these programs is relevant for society as a whole. Conducting research on how government spending cuts influence unemployment is therefore important. In order to answer the research question, the methodological setup of Hooker & Knetter (1994) will be used. A sample of 17 OECD countries is constructed, covering a time period from 1978 to 2009. The sample choice is based on data availability of the main independent variable, as constructed by Devries et al. (2011).

Figure 1.B

Note: the figure plots data for all 17 OECD countries in the sample ranging from 1978 to 2009. Vertical black lines indicate a separation of country data. The government cut measure from Devries et al. (2011) is used as government spending cut proxy. It is expressed as government spending cuts as % of GDP. Furthermore, harmonized unemployment rate data from The World Bank is used.

(6)

6 This paper is structured as follows: section two identifies potential control variables by examining several unemployment determinants. Also literature regarding the unemployment effect of government spending cuts will be explored. In section three, the methodological setup, a variable parameterization, the respective hypotheses and regression output will be presented. The empirical findings indicate a positive government spending cut-unemployment relation, consistent with most existing research on this topic. Finally, the last section comprises the conclusion and suggestions for future research.

(7)

7

2. Literature review

This section discusses the existing literature about the government spending-unemployment relationship. In the research community no consensus exists about the impact of government spending cuts on unemployment. The magnitude and significance of the unemployment effect seems to depend on which model assumptions and specifications are being used by the researchers. Firstly, relevant unemployment determinants other than government spending cuts and channels influencing the ultimate unemployment effect shall be discussed. By doing so, potential control variables for the empirical part can be identified.

2.1. Theoretical insights

This paper analyzes the impact on the actual unemployment rate. Actual unemployment contains both equilibrium and cyclical parts. Therefore, factors determining equilibrium unemployment as well as factors affecting short-run deviations from this equilibrium rate will be discussed.

Most studies look at factors influencing the equilibrium unemployment rate. According to Rogerson (1997) nearly all modern models regarding unemployment include so called matching elements, to accommodate in- and outflow of workers into the labor market. Because finding a match between labor demand and supply takes time and costs resources, unemployment arises. Next to matching functions, the models also have equations for real wage determination. The resulting bargaining between workers and entrepreneurs ultimately determines the wage outcome. 2.2. Labor market institutions

Factors influencing the real wage outcome via the matching process and therefore indirectly affecting unemployment, are captured rather well by labor market institutions.

Mortensen (1994) used such a matching model to analyze the impact of several labor market institutions on unemployment. Excessive real wage demands, employment protection policy and other institutions occurring regularly in the existing literature are discussed below.

2.2.1. Institutional structure of wage determination

The process of wage formation differs significantly across industries and across countries. Nickell et al. (2005) showed the ultimate wage outcome to be dependent on union power in the bargaining process, union coverage and the extent of coordination. The first two mentioned aspects cause positive wage pressure, whereas coordination might mitigate this wage pressure. The resulting wage determines the equilibrium unemployment. Higher wages induce higher equilibrium unemployment, in line with the conventional neoclassical labor model (Borjas, 2012).

Layard et al. (2005) found a similar relation in their empirical study. Higher degrees of collective bargaining coverage coincide with higher unemployment. However, if the wage bargaining process is more coordinated and centralized, the unemployment effect is mitigated.

Nickell et al. (2005, p.8) define coordination as: “Mechanisms whereby the aggregate employment implications of wage determination are taken into account when wage bargains are struck”. According to them, centralization refers to the level at which bargaining takes place.

Mortensen (1994) showed an increase in workers bargaining power to have an ambiguous unemployment effect. Unemployment duration increases with bargaining power but changes in unemployment level are ambiguous. No unemployment level effects emerge because changes in equilibrium reservation productivity are also ambiguous.

Oswald (1997) analyzed a potential relationship between the degree of unionization across countries and unemployment. No positive effect is found. Therefore, according to Oswald’s findings, the proportion of workers in the country who are members of trade unions does not seem to coincide with unemployment.

(8)

8 2.2.2. Employment protection policy

Tougher unemployment protection laws make firms less inclined to hire new employees. Mortensen (1994) used taxation on worker layoffs as a measure for employment protection. Essentially this entitles a firing tax, which forces employers to pay a fine to the government if they fire employees. According to Mortensen’s model both job creation and destruction decrease, as a result of increases in this layoff tax. However, empirical evidence is ambiguous about the ultimate unemployment effect.

According to Nickell et al. (2005), employment protection laws decrease the propensity of firms to hire new workers, therefore negatively influencing job matching efficiency. Concurrently, employment laws can increase job matching efficiency. Nickell et al. found an ambiguous net unemployment effect resulting from employment protection regulation.

2.2.3. Wage rigidity and Minimum wages

Wage rigidity is often stated to be responsible for higher and persistent unemployment rates. Blanchflower & Oswald (1994) found evidence inconsistent with this conventional theory. Their finding is that wage flexibility is more similar across countries then stated by conventional theories on this matter. They found a wage-unemployment relationship persistent over time and across some countries. Furthermore, their estimated unemployment elasticity of pay of -0.1 is in line with other research, suggesting similar wage-unemployment dynamics between countries. According to Blanchflower & Oswald, wage rigidity is therefore no important determinant of unemployment.

According to the neoclassical labor model of Borjas (2012), a minimum wage set above the market equilibrium wage outcome increases unemployment. However, Card and Krueger (1994) found evidence that an increase in minimum wage increased employment. The unemployment effect of minimum wage is therefore not clear cut. Gertler & Trigari (2009) state that wage rigidity is responsible for increasing the unemployment impact of shocks in technology. As opposed to

(Monacelli et al. 2010), who found real wage rigidity to have a mitigating effect on unemployment caused by government spending fluctuations. Therefore, the minimum wage-unemployment relationship is contested amongst researchers.

2.2.4. Labor Tax

Taxes are considered to be institutional determinants of unemployment by Bassanini & Duval (2006). Nickell (1997) also found taxes to influence unemployment rates. Nickell identified the total tax rate (sum of the average payroll, consumption and income tax rates) to be the correct measure of tax impacts on unemployment. Nickell identified a decrease in labor supply stemming from labor tax increases, since the higher tax burden discourages people to work. Both two papers identified this inter-temporal substitution effect as main driver for a positive relation between unemployment and labor taxes. According to Hooker & Knetter (1994), persistent deviations from the long run

equilibrium unemployment rate might occur because of prevalent labor market rigidities, including tax distortions. Layard & Nickell (1986), found a modest significant unemployment increase due to raised employer taxes. Oswald (1997) specifically focused on payroll taxes, as opposed to a mix of taxes used by Nickell. Oswald showed this payroll tax rate to be uncorrelated with unemployment dynamics across a sample of OECD countries.

2.3. Institutional Interactions

Belot & van Ours (2004) analyzed unemployment patterns by looking at changes in institutions. These institutional reforms are analyzed by using interactions between respective institutions. According to them union bargaining power, the labor tax rate and unemployment benefits all increase unemployment. Their “Right-to-manage” model of wage bargaining shows the

interrelatedness of institutional components, implying possible complementarities between them. For most countries institutional interaction effects amplify direct effects of institutional change. The significant impact of institutional interaction terms implies that similar institutional change has

(9)

9 heterogeneous net effects across countries. This last remark is in line with findings of Freeman (1998), who emphasised that effectiveness of labor market reforms is conditional on the overall institutional framework in a country. Nickell & van Ours (2000) found such a positive institutional interaction effect to be partially responsible for the successful Dutch structural reforms since the 1980’s. These respective institutional reforms were: changes in wage bargaining structures, the tax benefit system and active labor market policies. According to them, gradual changes in these three institutional parameters are responsible for decreasing the Dutch equilibrium unemployment rate substantially since the 1980’s.

2.4. Shocks

Nickell et al. (2005) integrated several shocks as proxy for short-run deviations from the long run equilibrium unemployment rate. They defined shocks as: “exogenous shifts in the macroeconomic environment”. Productivity shocks, demand shocks and real interest rate fluctuations will be explored in more detail:

2.4.1. Aggregate demand shocks

Hoon & Phelps (1992) provided a general equilibrium model in which the unemployment effect of shocks is assessed. According to them real aggregate demand shocks arising from fiscal stimulus will increase unemployment. This unemployment effect is reached via increasing real interest rates, necessary to equilibrate the goods market after the demand shock. This higher real interest rate ultimately negatively changes firms hiring decisions, hence unemployment increases. Though, this result describes effects for the medium-term. The researchers leave open the possibility of short-run Keynesian effects, being able to mitigate the negative employment effect. These Keynesian effects are price rigidities inducing the extensive margin to be inflexible.

2.4.2. Aggregate supply shocks (Productivity shock):

Permanent productivity improvements have no effect on equilibrium unemployment. Only tentative shocks in stationary variables, which revert back to their mean rather quickly seem to have lasting unemployment effects according to Hoon & Phelps (1992). Hoon & Phelps furthermore studied impacts of temporary productivity shocks. Temporary productivity increases seem to diminish firms’ investments in new employees. Wage rigidities disallow contemporaneous wage increases to match the productivity increase. This causes unemployment to increase provisionally. Unemployment increases because more workers quit their job as a result of them not being compensated with higher wages for temporary productivity increases. In other words, the opportunity cost of working increases, which makes the non-working option more attractive for labor market insiders. However, in the new steady state situation, both real interest rate and unemployment dynamics return to their initial pre-shock values.

Nickell et al. (2005) on the other hand, found a decrease in productivity to cause temporary unemployment if wages are not readjusted for this productivity downfall. Their finding is

inconsistent with results of Hoon & Phelps, who identified exactly the opposite effect. 2.4.3. (Ex-post) real interest rate

Both Blanchard & Wolfers (2000) and Phelps (1994) found evidence for a significant impact of real interest rates on unemployment.

Hoon & Phelps (1992) studied effects of permanent increases in the real interest rates, for the case of an open economy. Firms start hiring less employees, since hiring becomes more costly. Assuming this real interest rate increase is persistent, unemployment increases as well.

(10)

10 2.5. Interactions between shocks and labor market institutions

Blanchard & Wolfers (2000) modified the approach of Nickell et al. (2005) by focusing on potential variable interactions. They state labor market institutions to be unable to completely explain

European unemployment increases in past decades, since these institutions were quite similar in the 1960s as they are today. They therefore analyzed the impact on European unemployment resulting from the interaction between labor market institutions and shocks. They found that specifically this interaction is able to explain the unemployment pattern over time and across countries very well. Their concept implies that shocks are responsible for unemployment, but the magnitude of the ultimate unemployment effect is determined by labor market institutions in the respective geographical entity (country). Blanchard & Wolfers identify three shocks determining equilibrium unemployment in Europe. These shocks are: a decline in Total Factor Productivity (TFP) growth, real interest rate fluctuations and shifts in labor demand. Their findings imply that unemployment effects of variables of interest have higher magnitudes and are more persistent conditional on what type of institutional structure a country has.

A different model structure is applied by Ljungqvist & Sargent (1998). They analyzed the interaction of shocks and institutions. Given stable institutions, they investigate if different degrees of turbulence affect unemployment magnitudes. The calibrated model is able to partially justify some dynamics in unemployment time-series of Europe and the United States. However, Nickell et al. (2005) qualify this approach to be unable topass some necessary tests and therefore question the validity of the resulting inference. More specifically, Nickell et al. find no evidence for the assumed increase in turbulence. Furthermore, they also state that Ljungqvist & Sargent give no satisfactory explanation for heterogeneous patterns in European unemployment.

Hooker & Knetter (1994) also include oil and exchange rate variables in their model. These two variables are likely to have a differential impact on unemployment across European countries in our paper as well due to differences in cross-country competitiveness.

2.6. Unemployment effects of government spending cuts

The previous part of the literature review identified important unemployment determinants. The following sub-section will specify papers regarding unemployment effects of government spending cuts. In doing so, major channels through which the government cuts run will be analyzed.

Blanchard and Leigh (2013) show unemployment multipliers to be high, especially when government cuts coincide with economic downturns. Therefore the timing of fiscal policy interventions seems to impact the effectiveness of these interventions. They also show that the OECD and IMF underestimated unemployment effects of fiscal consolidation in the 2008 crisis period as a result of these errors. These underestimated fiscal multipliers provided policymakers with wrong inputs. Initiated fiscal consolidation based on these underestimated multipliers has potential for economic damage. Therefore, minimizing forecast errors is important.

An important notion from Monacelli et al. (2010) states that variable responses (including unemployment) to government spending fluctuations can vary significantly depending on which model specifications are being used. The following analysis will divide research into two parts: purely empirical papers and secondly theoretical and partial empirical papers. This dichotomy is used in order to determine whether empirical or theoretically oriented papers show contrasting

unemployment results. 2.6.1. Purely empirical papers

Pappa (2009) used a restricted VAR model, using New-Keynesian sticky price characteristics. In their flexible price variant, where labor demand is fixed, government consumption increases financed with debt issuance cause labor supply to increase and real wage to decrease. The sticky price model has similar dynamics but also increases labor demand. In this case the net effect causes real wages to increase. Both model variants found that an increase in government consumption and investment raised employment. Estimated unemployment elasticities range between 0.10 and 0.25.

(11)

11 Brückner & Pappa (2012) used a New-Keynesian model with matching frictions. By including a labor force participation choice, they showed that government spending increases can increase both unemployment and employment. This seemingly contradictory result is robust and significant for several OECD countries in the sample. A likely explanation is that conventional NK-models assume the labor market participation rate to be constant, whereas Brückner & Pappa endogenize a participation margin. Government spending increases induce increases in labor force participation via negative wealth effects. Assumed price rigidity and worker heterogeneity lead to labor demand increases. The participation increase dependents on the magnitude of labor supply elasticity. Ultimately all three labor market indicators: employment, unemployment and labor force

participation might increase, conditional on certain model modifications. This result of Brückner & Pappa is inconsistent with previous papers shown, but is certainly not unique.

Tagkalakis (2013) also found the possibility for unemployment increases resulting from government spending increases in standard real business cycle models. More specifically in his 2006 research, Tagkalakis identified a negative employment effect resulting from government spending increases. This negative effect is mainly driven by the wage bill aspect of the spending shock. The wage bill refers to compensation for civil servants. Wage bill fluctuations influenced unemployment the most out of all five government spending components analyzed.

Monacelli et al. (2010) used an augmented neoclassical model with search and matching frictions. They examined fiscal policy effects on the US labor market. Increases in government spending are estimated to decrease the reservation wage of consumers, because expected non-labor benefits diminish. Non-non-labor benefits diminish for the model variant which assumes that the leisure value of unemployment only includes home production. This creates a decrease in the bargained wage and ultimately materializes into a lower real wage. The baseline model estimates an unemployment multiplier of -0.6 percentage points. Adding Unemployment Insurance benefits to the model decreases the magnitude of this unemployment multiplier. This last notion is not in line with the literature discussed in section 2.2.3, where introducing UI systems were thought to increase unemployment. Furthermore also higher degrees of real wage rigidity are estimated to decrease unemployment multipliers. However, if debt financing and distortionary taxation are allowed for, taxation significantly alters the relative work-non-work payoff. Dynamic unemployment responses also change. In this situation government spending increases cause tax increases necessary to finance government debt yields. This leads to an aggregate unemployment increase after 1,5 years. Therefore inconsistencies in the literature are partially explainable by introducing debt financing and distortionary taxes.

Hooker & Knetter (1994) investigated the impact of (military) procurement spending

fluctuations on state unemployment in the United States. By using an error components model, they found a positive relation between procurement spending cuts and unemployment.

2.6.2. Theoretical and partially empirical papers

In their New-Keynesian model variant Nekarda & Ramey (2011) described labor market reactions of government spending fluctuations. Total federal government spending is used as proxy variable. Their results show that government spending increases create outward shifts in both the labor demand and supply curves. This leads to an increase in hours worked, since the estimated elasticity of production worker hours is 0.84. Furthermore, the real wage effect is ambiguous.

However, Nekarda & Ramey state that some transmission elements of government spending are potentially missed by conventional NK- and Neoclassical models. This is so because increasing returns to scale are not accounted for in both standard versions. Devereux et al. (1996) indicated that unemployment effects stemming from government spending increases induced different results due to an introduced increasing returns to scale assumption. This increasing returns to scale

assumption enabled total factor productivity increases, as opposed to conventional NK- and Neoclassical models. These productivity changes indirectly induced alternative unemployment dynamics.

(12)

12 Nekarda & Ramey (2011) also analyzed which dynamics government spending increases create in a Neoclassical setting. In this case the hours worked still increased, following a labor supply expansion. Real wage on the other hand declined.

Rotemberg & Woodford (1992) analyzed the impact of aggregate demand shocks on

economic indicators. For exogeneity purposes, they used military purchases as proxy for government purchases. Increases in these purchases were found to increase hours worked and raise real wages. This finding is in line with results of Nekarda & Ramey (2011) regarding hours worked.

Baxter & King (1993) used a restricted neoclassical model. Their results showed that increased government spending creates a negative wealth effect for consumers. This ultimately increases labor supply and decreases real wages. They furthermore found that the public financing decision of government spending influences expenditure effectiveness. Baxter and King conclude output to decrease after tax financed government spending increases. This last notion is in line with empirical findings of Monacelli et al. (2010).

Ramey & Shapiro (1998) constructed a theoretical two-sector neoclassical model of government spending, which they also tested empirically. Adding costly capital mobility created a friction that influenced the magnitude of effects of aggregate government spending. Their empirical evidence shows that employment might decrease due to labor market imperfections and sectoral shifts. This effect is reached via a decrease in real wages, which is in line with Keynesian model predictions.

Layard & Nickell (1986) used a model with both NK- and neoclassical features. They decompose sources of unemployment changes into several “wage pressure variables” to estimate their causal impact on unemployment. Demand shocks (including government spending fluctuations) decreased unemployment. The wage pressure variables ( labor market institutions) responsible for higher unemployment outcomes are according to them: labor taxes and employment protection legislation. This last finding is in line with results presented in section 2.2.

Burnside, Eichenbaum & Fisher (2004) used exogenous changes in military purchases as proxy for government spending shocks. An increase in military purchases is estimated to induce an increase in hours worked, hence a decrease in unemployment. Furthermore, real wages decreased. After including habit formation and investment adjustment costs in the standard neoclassical model, quantitative theoretical predictions fit their empirical findings well.

Theoretical models differ in their assumptions, by which their output is sometimes difficult to compare against other model predictions. Empirical models, which are based on these theoretical predictions therefore don’t tend to show unanimous results. It cannot be stated that either empirical or theoretical model types inherently outperform the other type. Each model analyzed has its own advantages and disadvantages. Overall, the papers presented here tend to agree upon a positive unemployment-government cut relationship.

(13)

13

3. Data and methodology

In this section the methodological setup, a variable parameterization and the respective hypotheses will be presented. Furthermore, the empirical results and the panel data conditions shall be

examined.

3.1 Baseline model of Hooker & Knetter (1994)

To answer the question what the unemployment effects of government spending cuts are, an empirical regression model shall be tested. The panel data model as discussed by Hooker & Knetter (1994) will be used as a baseline model. However, several alterations to their approach shall be made. Hooker & Knetter (1994) examined the unemployment impact of exogenous changes in military spending for all US states, using the following model:

𝑈

𝑖𝑡

= 𝛩

𝑡

+ 𝛬

𝑖

+ ∑

𝜌

𝑚

𝑈𝑖

𝑡−𝑚

+ ∑

5𝑘=1

∑

3𝑗=0

𝛾

𝑗𝑘 4

𝑚=1

∆𝑀𝐼𝐿

𝑘_{𝑖𝑡−𝑗}

+ 𝜑

𝑖

𝑂𝐼𝐿

𝑖𝑡

+ 𝜀

𝑖𝑡

In this equation “𝛩𝑡 “ represents time effects. The subscript “t” refers to a specific year in the

dataset. The variable: ”𝛬𝑖” indicates state effects. The “i” subscript refers to a specific US state in the

sample. ”𝑈𝑖𝑡−𝑚” denotes an unemployment lag up to four periods. The main variable of interest:

“∆𝑀𝐼𝐿𝑘_{𝑖𝑡−𝑗}” represents exogenous changes in military procurement spending at the state level. They used percentage changes in procurement spending, because this approach resulted in more

significant results relative to level effects. “OIL” represents oil price fluctuations.

Subscript “j” represents lags, allowing state specific estimates to capture cross-state procurement spending variations. Subscript “k” signifies quintile groupings based on procurement spending per capita. This enables them to analyse the unemployment effect conditional on the procurement spending level in specific states.

3.2. Modifications of baseline model

Several alterations from the baseline model of Hooker & Knetter are being made to end up with our own empirical setup.

3.2.1. Different sample countries relative to the baseline model

Instead of analysing unemployment effects at US state levels, this paper examines aggregate unemployment effects for a sample of 17 OECD countries. Table 1 below shows the list of countries included in our sample.

3.2.2. Different main independent variable

Whereas Hooker & Knetter used exogenous changes in military spending, the main independent variable in our paper is government spending cuts. However, unemployment and government spending are serially correlated. Hooker & Knetter identified exogenous spending cuts, which is crucial to mitigate this reverse causality. We use data from 1978 till 2009. The sample countries and

(14)

14 time interval are based on an IMF dataset regarding deficit-driven fiscal consolidation. In our paper this fiscal consolidation measure is used as proxy for government spending cuts.

3.2.3. Disregarding quintile sub-sampling

Resampling countries on the basis of degree of labor market institutional rigidity would allow analysing unemployment influences of these institutional parameters. Sub-sample groupings are omitted from our model due to space constraints.

3.3. Variable parameterization

Furthermore, additional control variables proxying labor market institutions are added. Also a variable parameterization shall be provided below.

3.3.1. Dependent variable (U it)

The dependent variable in all regression models in this paper is the unemployment rate. Data from the World Bank is used. Their unemployment measure is unemployment as percentage of the total labor force. The unemployment data is harmonized to correct for differences in national

unemployment definitions and data collection methodologies. Therefore, this measure is suitable for international comparisons.

3.3.2. Independent variables

For all independent variables explained below the subscript “i” represents a specific country in the sample and “t” a respective year.

3.3.3. Government spending cuts (Govcuts it)

Government spending is able to affect unemployment. However, the opposite holds too. Anti-cyclical fiscal policy is evidence for this opposite relation. Too high unemployment figures can trigger Active Labor Market Policies. Because of this reverse causality, regressing government spending on unemployment would create endogeneity problems. To address the endogeneity issue, data from Devries, et al.( 2011) will be used. They provide a deficit-driven fiscal consolidation measure for 17 OECD countries. The authors state themselves: “our discretionary changes in government spending are primarily motivated by a desire to reduce the budget deficit and not by a response to

prospective economic conditions.” So, cyclical changes in government spending due to fluctuations in the macroeconomic environment are corrected for. Their measure for changes in government spending can therefore be considered to be exogenous.

The fiscal consolidation data is measured as a percentage of contemporaneous GDP. Positive values represent spending cuts, whereas negative values indicate spending increases relative to GDP. 3.3.4. OIL prices (OIL it)

The variable “OIL” is added to capture shocks hitting the labor market. Hooker & Knetter (1994) argue that fixed effects partially capture cyclical movements due to oil price volatility. They however emphasize that resulting shocks have heterogeneous effects across states. This same argument holds for our own analysis on a country scale. Oil dependent industries in oil importing countries are affected more severely by oil price increases than the sample average. For this reason controlling for oil, besides fixed effects variables seems justified. Regarding the direction of the relationship: Carruth et al. (1994) found a positive correlation between oil prices and unemployment for the UK and Canada. In our paper the oil variable enters respective regressions as year on year growth rates. World Bank oil price data is used and measured as the average spot price for crude oil expressed in real 2010 dollars.

(15)

15 3.3.5. Unemployment lags (U it-m)

Unemployment is very persistent. This implies that unemployment rates in previous periods are a good predictor for future unemployment. Therefore, following the approach of Hooker & Knetter, unemployment time lags are included as variables. The number of lags necessary to capture the persistent dynamics of unemployment is however ambiguous. A maximum of three unemployment lags shall be included and their significance will be tested with robustness checks, just as regular coefficients. The subscript m in (U it-m) therefore takes the value 3.

3.3.6. Labor tax (tax it)

The implicit labor tax is used. Two reports from the European Commission are being used to compute it. Both datasets combined have available tax data from 1991-2007 for 13 countries in our sample.

3.3.7. Real interest rate (r it)

Real interest rates at time t for country i are calculated using World Bank data. The measure is constructed as (i - P) / (1 + P). Where P represents the inflation rate (GDP deflator) and i stands for the nominal lending interest rate.

3.3.8. Employment Protection Laws (EPL it)

The OECD measure for employment protection is used as proxy for EPL’s. Their indicator captures the strictness of regulation on dismissals and the use of temporary contracts. A six point scale ranging from “zero” for no protection whatsoever up to “Six” for very strict employment protection regulation is computed.

3.3.9. Unionization (Union it)

Trade union density is used as a proxy for the system of wage determination. OECD data calculates the union density measure by dividing working union members by the total number of workers. 3.3.10. Minimum Wage Ratio (MWR it)

OECD data regarding the minimum wage ratio is used as proxy for minimum wage presence and its magnitude. It is computed as minimum wage relative to median wage. Federal minimum wage entitlements are used for this MWR measure. Estimation errors therefore are prevalent because within country minimum wage differences are not accounted for. Median wages instead of average wages are used to enable more precise international comparisons. This corrects for discrepancies in earnings distributions across countries (OECD library).

3.3.11. TFP growth (TFP it)

The Conference Board Total Economy Database is consulted for TFP growth estimates. These estimates are made using a Tornqvist index. It comprises an indirect approach in which TFP estimates result from calculated residuals in growth accounting equations.

3.4. Key statistics

Table 2 below shows some key statistics for all regressors. The 3 unemployment lags had roughly similar magnitudes and variation dynamics. Therefore only U it-1 is shown in this table.

Furthermore, oil price changes are computed as world averages. Therefore cross-country variation in oil prices are not present and enter the table as zero. One should also notice significant data gaps for some variables. This data availability problem is especially present for labor taxes.

Most variation in EPL stems from between variation (cross-country differences) in EPL. The within EPL variation is relatively small, which indicates EPL systems to be persistent for individual countries. The same argument can be made about Unionization. This finding is in line with the notion of Nickell et al. (2005), who identified labor market institutions to be relatively rigid.

(16)

16 The correlation matrix in table 3 is constructed based on 160 observations in the sample. The tax variable is omitted from this table in order to increase the number of observations from 70 to 160. This is done to increase the results' reliability. The version with 70 observations resulted in tax correlations with a maximum of 0.55. Furthermore, only one unemployment lag is shown. The second and third lag showed very similar dynamics as the first lag and generated correlation coefficients that were structurally lower in absolute value terms relative to the first lag. High correlations between unemployment and its lags is trivial. Lags only have the purpose to control for unemployment persistence. Besides unemployment lags, other regressors show relatively low correlation between one another. Therefore multicollinearity problems are less prevalent.

𝑖𝑡

+

𝛽

₇

𝑀𝑊𝑅

_𝑖𝑡

+ 𝛽

₈

𝑇𝐹𝑃

_𝑖𝑡

+ 𝛾

₂

𝐶

₂

+ ⋯ + 𝛾

₆

𝐶

₆

+ 𝜀

_𝑖𝑡

In this equation C2 till C6 represent sample countries included. Omitted countries are disregarded

due to incomplete data for control variables. These binary regressors capture individual (country) effects. The drivers of these regressors stem from time-invariant effects that differ across countries.

The Modified Wald test shows evidence for the presence of heteroskedasticity in model 1. Furthermore, the Wooldrigde test concludes serial correlation to be present as well. Presence of heteroskedasticity and serial correlation invalidate usual standard errors. Both these tests are shown in Appendix A. To mitigate these two effects, subsequent models will incorporate robust standard errors. Model 2 adds three unemployment lags to capture persistence in unemployment.

𝑖𝑡

₆

+

𝛿

₉₂

𝑇

₉₂

+ … + 𝛿

₂₀₀₆

𝑇

₂₀₀₆

+ 𝜀

_𝑖𝑡

Overlapping data series for all six included countries in model 4 are determined for the time interval 1992 till 2006. Data points outside this interval are omitted from the regression. Whether time-fixed effects are needed is determined by testing if all time dummies are jointly equal to zero. The test outcome is shown in appendix A, table 7. Significance of time-fixed effects is found and they therefore remain inclusive for consecutive models. Furthermore, the Hausman test is run to distinguish if a random effects model is more suitable than their fixed effects counterpart. The Hausman test is provided in table 8 of appendix A. Fixed rather than random effects models are preferred. Model 5 omits taxes, because the tax coefficient was insignificant in previous models.

(18)

2

𝐶

2

+ ⋯

+ 𝛾

17

𝐶

17

+ 𝛿

91

𝑇

91

+ … + 𝛿

2009

𝑇

2009

+ 𝜀

𝑖𝑡

In this equation

𝐷. 𝑈

_𝑖𝑡−1

,

refers to the first lag of first difference unemployment. Besides solving the unit root problem, respecifying the regression in first difference form also mitigates omitted variable bias from unobserved time-invariant sources (Stock& Watson, p.401).

3.6. Fixed Effects Regression Assumptions

There are four critical assumptions that should hold for fixed effects models, which are summarized as follows by Stock & Watson (p. 412):

have nonzero finite fourth moments. 4. There is no perfect multicollinearity.

3.6.1. Conditional mean zero

The conditional mean zero assumption is tested by plotting residuals against the independent variables. No clear patterns arise from these plots presented in appendix B. This indicates that variation in residuals seems unconditional on magnitudes of independent variables.

However, figure 12 in the appendix C shows some serious deviation from the normal distribution Also the standard normal probability plot and quintile plot shown in figure 13 and 14 respectively confirm this nonnormality issue.

3.6.2. i.i.d. condition

The data is drawn separately for each sample country, which is in line with the ( i.i.d.) condition. However, 𝑋𝑖𝑡 has a nonzero correlation with the error term for all 8 models and therefore creates

serial correlation. Furthermore, our panel data approach assumes that unemployment is linear in the independent variables. Figure 15 plots residuals against the fitted values. No distinctive pattern can be identified, which implies no linearity violation.

3.6.3. Large outliers are unlikely

To decrease the likeliness of outliers, robust standard errors are being used for six out of eighth models.

3.6.4. No perfect multicollinearity

All model estimates are computed by Stata, which indicates that no regressor is a perfect linear function of another regressor. Perfect multicollinearity is therefore not present in either of the models.

(19)

19 3.7. Hypotheses

For some independent variables discussed, no unanimity exists among scholars about the

unemployment relation of these respective variables. If the relation is ambiguous or conditional on specifications, a two-sided alternative hypothesis is constructed for respective variables. The alternative hypotheses are as follows:

𝛽𝐺𝑜𝑣𝑐𝑢𝑡𝑠 > 0

As discussed in the literature review, most papers identified a positive relation between unemployment and government spending cuts. This paper embraces this positive relation. 𝛽𝑂𝐼𝐿 > 0

In line with Carruth et al. (1994) oil price increases are expected to increase unemployment, assuming an average oil dependent and non-oil exporting country.

𝛽𝑟 > 0

Assuming persistent increases in real interest rates, results of Hoon & Phelps (1992) are expected. This implies increasing unemployment due to changes in firms hiring behavior.

𝛽𝐸𝑃𝐿 > 0

Firmer employment protection regulation makes firms less inclined to hire new people. Therefore the frictional unemployment component increases. Therefore, EPL is expected to have a positive relation with unemployment.

𝛽𝑈𝑛𝑖𝑜𝑛 > 0

Unionization is used as a measure for collective wage bargaining systems. Unionization increases positive wage pressure. If this materializes into higher real wages, unemployment is expected to increase. Increased hiring costs decrease firms propensity to hire new employees.

𝛽𝑈𝑖𝑡−𝑚 > 0

Unemployment lags are good predictors of future unemployment. Because yearly data is being used, the three lagged variables describe unemployment up to three years in the past. The first

unemployment lag is expected to have a positive relation with current unemployment. The signs of both the second and third lag are assumed to be similar to the first lag. 𝛽𝑀𝑊𝑅 > 0

Classical arguments predict a minimum wage set above the market equilibrium wage level, to increase unemployment. This expectation also translates into our MWR hypothesis.

𝛽𝑇𝑎𝑥 > 0

Labor tax increases will diminish the net payoff of work. Therefore Employees are disincentivised to work and labor force participation might decrease. In such a scenario unemployment increases. 𝛽𝑇𝐹𝑃 ≠ 0

Evidence for this TFP hypothesis only exists for the short run. Productivity increases cause contemporaneous unemployment increases if wage rigidities are present. However, Nickell et al. (2005) found a decrease in productivity to cause temporary unemployment if wages are not readjusted for this productivity downfall. Therefore a two-sided alternative hypothesis for TFP is used.

(20)

20 3.8. Empirical Results

This section discusses the estimated results. Column (1) till (8) in table 4 represent all the 8 models. Column (1) in table 4 presents the empirical results of the first model. Including additional

unemployment variables reduces the estimated coefficient of government cuts from 2.878 in column (1) to 0.104 in column (2). This major decrease in the coefficient’s magnitude implies that leaving out unemployment lags in regression (1) creates omitted variable bias due to persistence in unemployment. The overall 𝑅2 jumps from 0.0042 to 0.8010 when lags are included; apparently unemployment lags capture a significant amount of unemployment variation. Moreover, labor taxes turn only significant in model 2. It’s magnitude drops after omitting unemployment lags and

introducing time-fixed effects. This finding is in line with table 2 which indicates tax variance to stem mainly from between variation. The cross country decreasing tax tendency over time is captured already by time-fixed effects.

Although the third unemployment lag is significant at the 10% level in model 2, it is left out in subsequent models. This decision decreases the government cut coefficient’s magnitude by half and results in highly significant second unemployment lag coefficients thereafter.

Model 4 includes time-fixed effects, which induces a slightly negative insignificant

government cut coefficient. Furthermore, both MWR and Unionization show negative coefficients, although not significant. Non-significant unionization coefficients were predicted by Oswald (1997). Negative MWR coefficients are however conflicting with findings of Borjas (2012) and Monacelli et al. (2010), who found tougher labor market institutional parameters (minimum wages) to increase unemployment.

As stated, a tradeoff exists between the number of included control variables and the number of observations on which the regression is being run. Model 5 drops both the oil and tax variables, thereby increasing the number of observations to 160 and included countries to 10. The government cut coefficient turns positive again, but remains insignificant. Additionally both Unionization and MWR turn significant now.

Model 6 drops MWR and the real interest rate. This is done since real interest rates never turned positive in either of the 5 previous models. Non-significant coefficients are inconsistent with findings of Blanchard & Wolfers (2000) and Hoon & Phelps (1992), who found a significant positive real interest-unemployment relation. Furthermore, MWR is dropped because it had the most data gaps besides the tax variable and it was moderately correlated with Unionization. This last

correlation notion could induce multicollinearity. Model 6 and subsequent models are based on 334 observations and data from all 17 sample countries are inclusive. The variable EPL turns out highly significant until MWR and real interest rate are dropped as variables in column (6). Table 3 shows low correlations for EPL with all other variables, so multicollinearity cannot be the cause. The sudden drop in EPL’s significance might stem from a duplication of observations and countries on which regression (6) is being run. Table 10 in appendix A shows that countries with no MWR data are excluded from model (5) and had higher average EPL values than the overall EPL sample average. Table 11 shows that the unemployment rate for 4 of these added countries are higher relative to the overall sample average. It might be that average unemployment dynamics are changed thereby. These potentially different unemployment dynamics could explain why EPL no longer determines unemployment as well as in previous models. Furthermore, in model 6 the government cut coefficient now turns significant at the 10% significance level.

Model 7 excludes time-fixed effects as robustness check. Relative to model 6 the implications are quite similar. Several variable coefficients change both in magnitude and in significance. Government cuts for example becomes significant at the 5% level and Unionization even at the 1 % level. Furthermore, from model 7 onwards TFP growth is significant and is estimated to have a negative relation with unemployment. This negative unemployment relation is in line with findings of Nickell et al. (2005), but conflicting with predictions of Hoon & Phelps (1992). These differences are driven by alternative assumptions about whether productivity shocks are permanent or temporary. The methodology describing our TFP data is ambiguous about how long lasting

(21)

21 productivity shocks are. Interpreting why TFP coefficients in our paper are inconsistent with our alternative TFP hypothesis is therefore not feasible.

Model 8 drops the second unemployment lag, although it is significant. This is done in order to bring the model specification more in line with Hooker & Knetter’s initial setup.

Furthermore, Unionization turns insignificant. The notion explained in the key statistics that labor institutional variables are rigid, implies that they are approximately constant over time within a specific country. Country fixed effects in the estimated panel data models already control for fully time-invariant variables. It is therefore no surprise that when examining table 4 most labor market institutional variables turn out insignificant, since their unemployment impact is partially captured by these country fixed effects. As discussed in section 3.5, unit root presence drives the model respecification. Therefore, the dependent variable enters model 8 as first difference unemployment.

Although controlling for time and country fixed effects, sources of endogeneity are still present. This is partially deducible from the Wooldridge test for serial correlation. Because of this endogeneity problem, giving the empirical results a causal interpretation would result in an

inference that is not justified. Therefore, focussing on the coefficient’s sign and not on its magnitude creates more reliable results. The 95% confidence interval for government cuts in model 8 is (-0.030, 0.435). It mainly includes positive values, implying that a positive government cut-unemployment relation seems more valid than to embrace the exact point estimate, which is probably biased. Positive government cut coefficients are in line with findings of most empirical papers. Monacelli et al. (2010) identified a negative unemployment fiscal multiplier. Pappa (2009) found government consumption and investment increases to diminish unemployment. Our empirical results are not in line with Brückner & Pappa (2012), who determined government spending increases to be able to increase unemployment. These contrasted findings stem from differences in data, statistical model specifications and conditions.

The joint hypothesis that all coefficients of regressors are zero is rejected at the 1%

significance level for the 3 models for which it is defined. This is indicated by the F-statistic, which is computed for 3 out of 8 models. The F-statistic is undefined for model 2 till 6 because these estimations contain fewer clusters than parameters.

Furthermore, the overall 𝑅2 is a conventional goodness of fit measure. It decreases

considerably for model 8 relative to model 6. However, model 8 corrects for unit root presence and it’s estimates are therefore more reliable despite the lower overall 𝑅2. After all, higher 𝑅2 are no proof of a more adequate model, nor can it identify biased estimators.

Rho is calculated for every model and represents the percentage of variation that is explained by individual specific effects. It is rather high for most models in our paper, which indicates that most variation is not of idiosyncratic nature. These individual specific effects are exogenous time-invariant unemployment determinants. Individual specific effects are calculated for model 6 and 8 and respective figures are integrated in Appendix D. Figure 16 and 17 show that variation in individual specific effects is much higher for model 6 relative to model 8. Less unobserved heterogeneity across countries would suggest that model 8 fits the data better than model 6. To find out if this is true, unemployment rates are predicted for model 6 and 8 and plotted in figures 18 and 19. Figures 22 and 23 together deduce that model 8 has a considerably higher percentage deviation from the “real” data regarding their respective dependent variable. Model 6 fits the actual unemployment data better. Despite these findings, endogeneity is still present in model 6 and 8, shown by the Wooldrigde test for serial correlation in table 6. Alternative model specifications and regression techniques might solve this problem. “Best” therefore is a relative qualification.

(22)

(23)

23 4. Conclusion

The research question this paper intends to answer is: “What is the unemployment effect of

government spending cuts?”. Firstly, an extensive literature review on the topic has been conducted. To answer the research question an existing panel data approach by Hooker & Knetter (1994) has been used and expanded. The sample countries and chosen time interval are based on a study of Devries et al. (2011), who’s exogeneous measure for government spending cuts is used as main independent variable. In total 8 fixed effects models are estimated. A major concern regarding estimates’ reliability was the trade-off between included control variables and number of observations on which the regressions were being run. To generate more reliable results, most control variables were dropped from the final models. Model 6 and 8 are the most complete

empirical models tested in this paper. Although their dependent variable differs, the implications are the same. Whether measured in first difference or in level effects, unemployment is estimated to increase following an increase in government spending cuts. This positive government spending cut-unemployment relation is in line with Monacelli et al. (2010), who identified a negative

unemployment fiscal multiplier. Pappa (2009) found government consumption and investment increases to diminish unemployment. Our empirical results are however not in line with Brückner & Pappa (2012). These contrasted findings stem from differences in data, statistical model

specifications and conditions. Endogeneity problems remain present in all 8 models, so a causal interpretation would result in an inference that is not justified. Therefore, this paper focused on the coefficient’s sign and not on its magnitude. The 95% confidence interval for government cuts in model 8 is (-0.030, 0.435). It mainly includes positive values, which implies that a positive government cut-unemployment relation is more valid than embracing the exact point estimate, which is probably biased.

4.1. Suggestions for future research

Nearly all empirical papers discussed in the literature review used a VAR model setup to examine the unemployment effect of government spending cuts. VAR approaches have the advantage that no presumptions about variable interrelatedness need to be made on an ex-ante basis. Therefore, serial correlation is less of a problem with VAR methods.

As pointed out, nonnormality and endogeneity are present in all our estimated models. A causal interpretation of the empirical results is therefore not justified. In other words: the results of this paper are not internally valid.

Empirical results in this paper are based on data of 17 OECD countries. These results should be interpreted with caution when intending to generalize the findings to non-sample countries. Developing countries with different economic structures and transmission mechanisms might induce alternative unemployment-government spending cut relations. Endorsing our empirical output for these other country types could therefore potentially result in an unjustified inference. External validity in this paper is therefore difficult to achieve.

Even after introducing deficit-driven fiscal consolidation measures, as defined by Devries et al. (2011), the nature of spending cuts can potentially influence the unemployment result.

Government cuts could for example be part of a restructuring plan to make the economy more competitive in the long run. Therefore government cuts can ultimately influence equilibrium

unemployment indirectly via the nature of the cut. A model setup which also corrects for the nature of government cuts would therefore induce a more complete and reliable estimation.

Future research could additionally analyze the unemployment effects conditional on labor market institutional rigidity. Such a setup requires a sample decomposition based on rigidity of labor market institutional proxies, and estimating fixed effects models for each subgroup. Such an

approach would allow us to analyze if countries possessing more rigid labor market institutions show different unemployment effects of government cuts. Finally, unemployment dynamics and

transmission mechanisms discussed in this paper can to some extent contribute to an understanding of general unemployment effects stemming from deficit-driven fiscal consolidation.

(24)

24

References

Bassanini, A., & Duval, R. (2006). The determinants of unemployment across OECD countries: Reassessing the role of policies and institutions. OECD Economic Studies, 42(1), 7-86. Beetsma, R., Uhlig, H. (1999). An Analysis of the Stability and Growth Pact. The Economic Journal,

109(458), 546-571.

Belot, M., & van Ours, J., C. (2004). Does the Recent Success of Some OECD Countries in Lowering their Unemployment Rates Lie in the Clever Design of their Labour Market Reforms? Oxford Economic Papers, 56(4).

Blanchard, O., & Leigh, D. (2013). Growth Forecast Errors and Fiscal Multipliers. American Economic Review, 103(3), 117-129.

Blanchard, O., & Wolfers, J. (2000). The Role of Shocks and Institutions in the Rise of European Unemployment: The Aggregate Evidence. The Economic Journal, 110(462), 1-33. Blanchflower, D., G. & Oswald, A., J. (1994). The Wage Curve, MIT Press: Cambridge, MA. Borjas, G.J. (2012). Labor Economics. McGraw-Hill, New York, International student edition. Brückner, M., & Pappa, E. (2012). Fiscal expansions, unemployment, and labor force participation

theory and evidence. International Economic Review, 53(4), 1205-1228.

Burnside, C., Eichenbaum, M., Fisher, J.D.M., (2004). Fiscal shocks and their consequences. Journal of Economic Theory, 115(1), 89–117.

Card, D., & Krueger, A. (1994) Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania. American Economic Review.

Carruth, A., Hooker, M., & Oswald, A. (1994). Unemployment, oil prices and the real interest rate: evidence from Canada and the UK (No. dp0188). Centre for Economic Performance, LSE.

Devereux, Michael B., Allen C. Head, and Beverly J. Lapham, (1996). Monopolistic Competition, Increasing Returns, and the Effects of Government Spending. Journal of Money, Credit & Banking, 28(2), 233–54.

Freeman, R. N. (1998). War of the models: Which labour market institutions for the 21stcentury?. Labour Economics, 5(1), 1-24.

Gertler, M., & Trigari, A. (2009). Unemployment fluctuations with staggered Nash wage bargaining. Journal of political Economy, 117(1), 38-86.

Hooker, M., & Knetter, M. (1994). Unemployment effects of military spending: evidence from a panel of states (No. w4889). National Bureau of Economic Research.

Hoon, H., T. & Phelps, E.S. (1992). Macroeconomic Shocks in a Dynamized Model of the Natural Rate of Unemployment. American Economic Review, 82(4), 889-900.

(25)

25

Layard, P. R. G., Nickell, S. J., & Jackman, R. (2005). Unemployment: macroeconomic performance and the labour market. Oxford University Press on Demand.

Monacelli, T., Perotti, R., & Trigari, A. (2010). Unemployment fiscal multipliers. Journal of Monetary Economics, 57(5), 531-553.

Mortensen, D. T. (1994). Reducing supply-side disincentives to job creation. Reducing unemployment: Current issues and policy options, 189-220.

Nekarda, C. J., & Ramey, V. A. (2011). Industry evidence on the effects of government spending. American Economic Journal: Macroeconomics, 3(1), 36-59.

Nickell, S. (1997). Unemployment and Labor Market Rigidities: Europe versus North America. The Journal of Economic Perspectives. 11(3), 55-74.

Nickell, S., Nunziata, L., & Ochel, W. (2005). Unemployment in the OECD since the 1960s. What do we know?. The Economic Journal, 115(500), 1-27.

Nickell, S., & Van Ours, J. (2000). Why has Unemployment in the Netherlands and the United Kingdom fallen so much?. Canadian Public Policy/Analyse de Politiques, S201-S220. Oswald, A. J. (1997). The missing piece of the unemployment puzzle.

Pappa, E. (2009). The effects of fiscal shocks on employment and the real wage. International Economic Review, 50(1), 217-244.

Perman, R., & Tavera, C. (2005). A cross-country analysis of the Okun's law coefficient convergence in Europe. Applied Economics, 37(21), 2501-2513.

Phelps, E. S. (1994). Structural slumps: The modern equilibrium theory of unemployment, interest, and assets. Harvard University Press.

Political Economics blog Bas Jacobs, 19-09-2016. ING heeft gelijk: overheidsbeleid in de periode 2011- 2017 kostte volgens het CPB ongeveer 365.000 banen:

https://basjacobs.wordpress.com/2016/09/17/ing-heeft-gelijk-overheidsbeleid-in-de-periode-2011-2017-kostte-volgens-het-cpb-ongeveer-365-000-banen/

Prasad, N., & Gerecke, M. (2010). Social security spending in times of crisis. Global Social Policy, 10(2), 218-247.

Ramey, V. A., & Shapiro, M. D. (1998, June). Costly capital reallocation and the effects of government spending. In Carnegie-Rochester Conference Series on Public Policy (Vol. 48, pp. 145-194). North-Holland.

Reinhart, C. M., & Rogoff, K. S. (2009). The aftermath of financial crises (No. w14656). National Bureau of Economic Research.

Rogerson, R. (1997). Theory ahead of language in the economics of unemployment. The Journal of Economic Perspectives, 11(1), 73-92.

(26)

26 Rotemberg, J. J., & Woodford, M. (1992). Oligopolistic pricing and the effects of aggregate demand

on economic activity. Journal of political Economy, 1153-1207.

Stock, J. H., & Watson , M. M..’’ Introduction to Econometrics’’. Pearson Education Limited: Harlow, England (2012).

Tagkalakis, A. O. (2013). The unemployment effects of fiscal policy: recent evidence from Greece. IZA Journal of European Labor Studies, 2(1), 1.

Trigari, A. (2009). Equilibrium unemployment, job flows, and inflation dynamics. Journal of money, credit and banking, 41(1), 1-33.

Data

European Commission (2000). Structures of Taxation Systems in the European Union 1970-1997. European Commission (2009). Taxation trends in the European Union. Data for the Eu member

states and Norway.

IMF World Economic Outlook Database, April 2016

OECD (2016). Institutional Characteristics of Trade Unions.

OECD (2016). Strictness of employment protection – individual and collective dismissals. OECD Library Indicators (2016). Minimum Wage Ratio.

Devries, P., Guajardo, J., Leigh, D., & Pescatori, A. (2011). A new action-based dataset of fiscal consolidation. IMF working papers, 1-90.

The Conference Board (2015). The Conference Board Total Economy Database™, May 2015, http://www.conference-board.org/data/economydatabase/

(27)

27

Appendix A: Test diagnostics

Table 6: Wooldridge test for autocorrelation in panel data

X it Coef. Std. Err. t P>t Govcuts D1. 0.2475 0.569 0.44 0.682 OIL D1. 0.002 0.003 0.76 0.482 r D1. -0.001 0.069 -0.01 0.993 Tax D1. 0.020 0.030 0.67 0.534 Union D1. 0.183 0.259 0.71 0.511 EPL D1. 0.342 0.459 0.75 0.489 MWR D1. 0.097 2.795 0.03 0.974 TFP D1. 0.074 0.063 1.18 0.292 F (1, 5) = 246.313 Prob>F = 0.0000

Note: D1. refers to first differences. Hence, the null hypothesis is rejected implies the presence of first order autocorrelation

Conclusion

Homoskedasticity 1763.94 0.0000 reject

Note: the null hypothesis is rejected, which indicates the presence of hetroskedasticity

Table 5: Modified Wald test for groupwise heteroskedasticity

𝐶 𝑖2 _{𝑃𝑟𝑜 𝐶 𝑖}2

Unemployment effects of government spending cuts : a panel data approach among 17 OECD countries from 1978-2009