• No results found

Okun’s law within the OECD

N/A
N/A
Protected

Academic year: 2021

Share "Okun’s law within the OECD"

Copied!
58
0
0

Loading.... (view fulltext now)

Full text

(1)

André Netzén Örn, Viktor Boman Spring 2017

Bachelor, 15 ECTS Economics

Okun’s law within the OECD

A cross-country comparison

Author: André Netzén Örn & Viktor Boman

Supervisor: Mattias Vesterberg

(2)

I

-Page intentionally left blank-

(3)

II

Acknowledgements!

We would like to pay deep gratitude to our supervisor, Mattias Vesterberg, for his guidance during the writing of this thesis. Also, we would like to thank Humberto Barreto and Frank Howland for their providence of information during the research process. It has been of immense help.

Sincerely

________________________

Viktor Boman

&

André Netzén Örn 06-09-2017

(4)

III

-Page intentionally left blank-

(5)

IV

Table of content

1. Introduction ... - 1 -

1.1. The relationship of output and unemployment ... - 1 -

1.2. Development over time ... - 1 -

1.3. Purpose ... - 3 -

1.4. Delimitation ... - 3 -

2. Theoretical framework ... - 4 -

2.1. Okun’s law ... - 4 -

2.2. The first difference model ... - 5 -

2.3. The Gap version. ... - 6 -

2.4. The Dynamic model ... - 8 -

2.5. Summary of the models ... - 8 -

2.6. Leaving the US ... - 9 -

2.7 Earlier studies ... - 10 -

2.8. Discussion of earlier literature ... - 12 -

3. Methodology ... - 14 -

3.1. Raw data ... - 14 -

3.2. Descriptive statistic ... - 15 -

3.3. The Least Squares Assumptions ... - 17 -

3.4. The estimation of potential output and long-term rate of unemployment ... - 18 -

3.5. Decomposition Procedure ... - 19 -

3.6. Endogeneity ... - 20 -

4. Results ... - 23 -

4.1. The ordinary least squares assumptions ... - 23 -

4.2. Estimation of the Okun coefficient ... - 23 -

4.3. Country interaction model ... - 24 -

4.4. Union Density model ... - 26 -

(6)

V

4.5. The European Union model ... - 27 -

5. Discussion and analysis ... - 28 -

6.Conclusion ... - 32 -

Reference list ... - 33 -

Appendix ... - 37 -

(7)

VI

-Page intentionally left blank-

(8)

VII

Abstract

In the 60’s, the first article identifying the relationship between output growth and unemployment were released, with the purpose of providing a tool for US authorities to estimate the effect of labour policy on output. This article, presented by Arthur Okun, came to lay the foundation for the commonly known empirical relationship, named Okun’s law.

However, since the 60’s, the world has gone through political and economic shocks, such as the oil crisis, fall of the berlin wall, the crisis of the 90’s, the financial crisis and crisis of 2008.

These events open up the question: has the relationship changed?

This study focuses on 21 OECD countries for the time period 1991-2016, with the purpose to identify their respective relationship between output growth and unemployment, namely their Okun coefficient. The test that will be performed calculates the marginal effects of respective country to observe differences. Further, this study aims to give the reader a greater understanding of the complexity underlying the simple model Okun presented in the 60’s.

This is done by investigating whether there are any differences in the coefficient for countries within the EU, compared to those out of the EU. To explain the complexity further we check whether factors that affects labour market rigidity, such as union density, create differences in the Okun coefficient. The results from the study shows that the Okun coefficient differs between different countries. They also show that countries belonging to the European Union has a lower Okun’s coefficient on average. Finally, the results show that countries with a union density of over 75 % have a lower coefficient on average.

Keywords: OECD, Okun’s coefficient, Output growth, Unemployment.

(9)

- 1 -

1. Introduction

___________________________________________________________________________

In this chapter, the reader will be given an understanding about the purpose and objectives of this study. It will give a brief understanding about the relationship between unemployment and output, that is, the Okun’s Law.

___________________________________________________________________________

1.1. The relationship of output and unemployment

In the beginning of the 21st century, many European countries started to recover from the recent recession taking place during the year of 2000. For Sweden, the time-period between 2001 and 2004 the GDP grew by between 1.5 % to 4.3 % annually, while for their neighbour, Germany, the same time period corresponded to a slowdown in their annual growth rate from 1.7 % in 2001 to -0.71 % in 2003. During the same time-period, Sweden and Germany both experienced an increase in the unemployment rate. From macroeconomic theory and reasoning, increasing unemployment should correspond to a slowdown of the GDP growth for both Germany and Sweden, so one could ask: should not these geographically close countries experience similar trends suggested by macroeconomic theory? That is, when slowdowns in GDP occurs, the unemployment rate is expected to increase and vice versa.

In the 1960’s, the relationship between unemployment and output growth where documented by the economist Arthur Okun, who suggested a negative relationship which became known as Okun’s law. National authorities and economist need to consider the relationship between the two variables when they try to build economic models or when they try to forecast the effect from a fiscal policy on unemployment. Hence, the relationship between GDP growth and unemployment has become important when analysing economic forecasts within countries or regions. Due to the macroeconomic relevance of the relationship, Okun’s law has come to been widely studied and over time gained empirical support.

1.2. Development over time

The majority of the data that is used when estimating the Okun’s coefficient come from the US, and the studies has over time come to focus on the dynamics and robustness of the relationship following recessions. Furthermore, one interesting term named jobless recoveries has come to

(10)

- 2 -

been widely discussed and studied among economist. Owyang & Sekhposyan (2012) has suggested that Okun’s law is too simple and that during times following a recession unemployment lagged the recovery of GDP suggesting that the Okun’s original relationship should not be considered as a law.

Nickell (1997), Siebert (1997) and Blanchard and Wolfers (1999) started to direct their attention towards factors that characterized the labour market. Together with new data from other countries than the United States, Guisinger et al. (2015), Palombi et al. (2015) and Sögner and Stiassny (2002) began to map the underlying factors on country and regional level that could help to explain the dynamics of Okun’s coefficient and the difference dynamics between countries. From their result, they could observe that the level of union participation among workers, employment protection indicators and the degree of unemployment benefits contributed to the differences in the Okun coefficient between countries and states.

A common assumption within labour market theory has been that high market rigidities corresponds to high unemployment rate and that relaxed labour market policies should lead to lower unemployment rate(Blanchard & Wolfers, 1999; Nickell, 1997). The assumption relies heavily on the fact that Europe, historically, has come to experience high unemployment rate for a lengthy period, relative to its trans-Atlantic neighbour, namely the United States. Cross- country studies made by Ball et al. (2013) and Nickell (1997)challenges if this assumption is valid and if institutional differences in the labour market has an impact on the unemployment rate. Moreover, the advancements made by recent research mapping the different labour market characteristics of countries, together with their institutional differences, creates an interesting link that could further explain the country differences of Okun’s coefficient. This sets the purpose of our study, namely to investigate the cross-country differences of Okun’s coefficient of 21 OECD countries and try to identify if labour union participation and employment protection could explain these differences. Furthermore, since earlier research base their results comparing Europe and North America, we will perform a regression to see if we can see similar result comparing Europe against the rest of the OECD countries.

(11)

- 3 - 1.3. Purpose

The purpose of this paper is to observe and test if there are any cross-country differences in the Okun relationship between 21 OECD countries, and to test if some of the differences could be explained by the specific country’s grade on union density or by whether it is in the Euopean Union or not. To estimate our models, we will use the Gap-version of Okun’s Law to see how actual output and unemployment behave around their long-term trends. This is done using a band-pass filter called the Hodrick-Prescott (HP) – filter.

1.4. Delimitation

This thesis will be limited to the study of the Okun’s coefficient for 21 OECD countries. The main purpose for dropping out some of the OECD countries is the lack of data in both the unemployment rate and the GDP. For the same reason, we will also limit the study to investigate the period ranging from 1991 to 2016. A further argument to drop out specific countries and hence increase the length of the period investigated is the interest in capturing the recent recessions in 1991, 2001 and 2008. These three recessions have been followed by the so called

“jobless recovery”, which is a phenomenon where the recovery in the unemployment rate lags the recovery in the GDP (Andolfatto & MacDonald, 2004; Ball et.al., 2013; Knotek, 2007;

Koenders & Rogerson, 2005).

The study will aim to investigate whether there are cross country differences in the Okun’s coefficient or not. The study also tries to describe the reason why these potential differences exists. However, the study will not explain the variation in the coefficients by econometrics, rather it will focus on mapping the cross-country differences by inspection, logic and economic theory. It would be of interest to conduct the analysis for more countries and over a longer period. However, due to the lack of time and resources this is not possible.

(12)

- 4 -

2. Theoretical framework

___________________________________________________________________________

This theoretical chapter involves previous research on the subject. The reader gets a deeper understanding about what the Okun’s law is and about its complexities. The chapter will present Okun’s original work, and how the relationship has evolved since then.

___________________________________________________________________________

2.1. Okun’s law

Okun’s law is within the economic field a widely-known empirical relationship between the unemployment and real GDP growth. As mentioned earlier, the purpose of the research done by Okun (1962) on the relationship was to give policy makers in the United States a tool to forecast what impact labour programmes had on output growth. In his original work, based on data for the US between 1947-1960, Okun used two different models to estimate the relationship between unemployment and its impact on GNP. From the estimates, he found a negative relationship between the two variables and suggested that a 1 % increase in unemployment would lower the output growth with approximately 3 %.

Graph 1 Visual presentation of actual and potential GNP (Okun, 1962)

(13)

- 5 -

By using the data from the same time-period Knotek (2007) investigated the finding of Okun’s work. He found that, to keep unemployment constant, a real output growth of 4 % had to be reached. From this he concluded that zero growth of output would correspond to an increase in the unemployment rate of 0.3 %. Hence, Knotek’s research suggest the following model:

∆𝑈𝑡= 0.3 − 0.07∆𝑌𝑡

In the model presented, ∆𝑢 defines the change in the unemployment rate and ∆𝑦𝑡 defines the change of output. The value of Okun’s coefficient suggests that each percentage point of real output growth above 4 % decreases the unemployment rate with 0.07 percentage points.

Okun used two approaches when he estimated the Okun coefficient, the two models are presented below. In the next section, we will present the two models in more detail and shortly introduce a third model used to estimate the coefficient.

First difference version:

∆𝑈𝑡 = 𝛽0+ 𝛽1∆𝑌𝑡+ 𝜀𝑡 𝛽1 < 0 Gap version:

(𝑈𝑡− 𝑈𝑡) = 𝛽2(𝑌𝑡− 𝑌𝑡) + 𝜀𝑡 𝛽 < 0

2.2. The first difference model

The first difference model also called the growth model estimates the Okun coefficient by using a simple linear regression model where the rate of change in unemployment rate is used as the dependent variable.

𝛥𝑈𝑡 = 𝛽0 + 𝛽1𝛥𝑌𝑡 + 𝜀𝑡 Where:

• Given zero growth of unemployment rate, 𝛽0 gives the percentage change in output during the period

• β1 gives the percentage change in Y given a 1 % change in U.

• 𝛥𝑌𝑡 is the change in output between period t-1 and t.

• 𝛥𝑈𝑡 is the percentage change in unemployment rate between period t-1 and t.

• The error term ε is a random term that catches all other effects.

(14)

- 6 -

2.3. The Gap version.

“If, in fact, aggregate demand is lower, part of potential GNP is not produced; there is unrealized potential or a ´gap´ between actual and potential output.” (Okun, 1962)

Compared to the first difference model that relies on accessible data, Okun’s second approach tries to estimate how much output would grow under the scenario where the economy experiences full employment. The second approach, named the Gap-version, measured the gap between actual and potential output to estimate the unemployment level.

According to Chamberlin (2011) potential output can be defined as “an equilibrium level of output where the economy can grow without experiencing inflationary or deflationary pressure”. The problem that arise with this model and which creates uncertainty is that potential output and the level of full unemployment must be estimated and cannot be gathered from statistical sources.

Ball et al. (2013) argues that to fully understand Okun’s reasoning of the underlying variables in the Gap-model, the relationship can be explained through the inflation mechanism. The natural rate of employment, is defined as the equilibrium state within the economy where inflation does not cause unemployment and vice versa. Additionally, Ball et al. (2013) points out that an accelerating inflation in the economy causes a shift in domestic demand that have a negative effect on the competitiveness of the domestic goods, changing the unemployment rate in the economy. From the changes in domestic demand, output starts to fluctuate around the potential output level causing firms to react on these movements. Hence, firms start to layoff or hire workers, and two underlying relationships let us understand the original gap model estimating the Okun coefficient.

𝐸𝑡 – 𝐸𝑡 = 𝛼 (𝑌𝑡– 𝑌𝑡) + 𝜈𝑡 , 𝛼 > 0; (1)

𝑈𝑡 – 𝑈𝑡 = ώ (𝐸𝑡 – 𝐸𝑡) + 𝜌𝑡 , ώ < 0; (2)

Where:

• 𝑈 is the long term unemployment rate

• 𝑌 is the potential output

(15)

- 7 -

• 𝐸𝑡 is the logged value of employment

• 𝑌𝑡 is the logged value of output

• 𝑈𝑡 is the unemployment rate in time t.

• The error term 𝜈𝑡 is a random term that catches all other effects.

• The error term 𝜌𝑡 is a random term that catches all other effects.

As mentioned earlier, changes in output changes employment and according to Ball et al.

(2013), given a value of 𝛼 less than 1.5, it is costly for firms to lay off workers in the short run, so other adjustments are used for firms to handle the short-run fluctuations in the economy.

From the equations above, Okun’s coefficient, namely 𝛽2, is derived from the coefficient α multiplied by ώ from equations (1) and (2) above:

(𝑈𝑡− 𝑈𝑡) = 𝛽2(𝑌𝑡− 𝑌𝑡) + 𝜀𝑡, 𝛽 < 0

Even though Okun in his original work choose to set unemployment as the dependent variable, economist such as Lee (2000) and Guisinger et al. (2015) choose to set output as dependent variable when estimating Okun’s coefficient, arguing that shocks does not affect unemployment but affects output. A deeper discussion about the potential problem with endogeneity in Okun’s law will be discussed later in this paper. The model becomes:

𝑌𝑡 − 𝑌𝑡= 𝛿(𝑈𝑡− 𝑈𝑡) + 𝜀𝑡 , 𝛽 < 0 Where:

• (𝑈𝑡− 𝑈𝑡) is the gap between the actual unemployment rate and the natural unemployment rate in time period t and captures the cyclical level of unemployment rate.

• (𝑌𝑡− 𝑌𝑡) is the gap between logged actual output and the potential output in time period t and captures the cyclical level of output.

• 𝛿 gives the percentage change of the output gap given a 1 % change in the unemployment gap. From economic theory, 𝛿 should take a negative value.

• The error term 𝜀 is a random term that catches all other effects.

(16)

- 8 -

2.4. The Dynamic model

Over time this first difference model has been extended using time lagged variables, the developed model is called the dynamic. According to Knotek (2007), the dynamic version corrects for omitted effects of past output on the unemployment rate, which the first difference model fails to include. However, the drawback of the model compared to the first difference model is that it includes a lagged coefficient and is not as easy to interpret as the first difference model (Knotek, 2007; Stock et.al., 2010; Sögner & Stiassny, 2002). The dynamic model is defined as follows:

𝛥𝑌𝑡= 𝛽0+ 𝛽1𝛥 𝑈𝑡 + 𝜃1 𝛥𝑈𝑡−1+ 𝜀𝑡 Where:

• Where β0 is the constant.

• β1 gives the percentage change in Y given a 1 % change in U in time period t.

• 𝜃1 gives the percentage change in Y given a 1 % change in U in time period t-1.

• 𝛥𝑌𝑡 is the change in output between period t-1 and t.

• 𝛥𝑈𝑡 is the percentage change in unemployment rate between period t-1 and t.

• The error term ε is a random term that catches all other effects.

2.5. Summary of the models

Comparing the three different models we can see both advantages and drawbacks in the process of estimating the Okun coefficient. The two most common models used to estimate the Okun coefficient are the first difference and the gap model, the same two used by Arthur Okun in the 60’s. In our effort to interpret the Okun coefficient and to see potential differences between countries we will use the gap model. This version is well examined in papers investigating differences on national and regional level, which provides guidance in the effort to interpret our results. Furthermore, as Ball et al. (2012) argues, if we are choosing the first difference version we need to assume a constant long run growth rate and a constant unemployment rate. Using the gap-version with its estimated variables of natural unemployment and potential output we do not need to consider this assumption and could avoid a potential error in our results. Since we estimate deviations around a trend, the means for output and unemployment rate will be very close to zero. This means that there is no economic interpretation in the coefficient, and we therefore chose to leave them out of the regression. Hence, we end up estimating the following models:

(17)

- 9 - The Okun coefficient:

(𝑌𝑖− 𝑌𝑖)𝑡= 𝛿(𝑈𝑖 − 𝑈𝑖)𝑡−1+ 𝜀𝑖𝑡 The Okun coefficient with country specific effects:

(𝑌𝑖 − 𝑌𝑖)𝑡= 𝛿(𝑈𝑖− 𝑈𝑖)𝑡−1+ 𝜃𝐷1𝑖(𝑈𝑖− 𝑈𝑖)𝑡−1+ 𝜀𝑖𝑡

The Okun Coefficient with Union Density as interaction:

(𝑌𝑖 − 𝑌𝑖)𝑡 = 𝛿(𝑈𝑖− 𝑈𝑖)𝑡−1+ 𝜃𝐷2𝑖(𝑈𝑖− 𝑈𝑖)𝑡−1+𝜀𝑖𝑡 The Okun Coefficient with EU as interaction:

(𝑌𝑖 − 𝑌𝑖)𝑡 = 𝛿(𝑈𝑖− 𝑈𝑖)𝑡−1+ 𝜃𝐷3𝑖(𝑈𝑖− 𝑈𝑖)𝑡−1+ 𝜀𝑖𝑡

Where:

• 𝑌𝑖is logarithm of actual real GDP for country i.

• 𝑌𝑖is potential GDP for country i.

• (𝑌𝑖− 𝑌𝑖)𝑡 is the output gap at time t.

• 𝑈𝑖 is the actual harmonized unemployment rate for country i

• 𝑈𝑖is the natural rate of unemployment for country i.

• (𝑈𝑖 − 𝑈𝑖)𝑡−1 is the unemployment rate gap at time t-1.

• 𝛿 is the Okun’s coefficient for the reference category.

• 𝜃 is the country-, union- and EU specific effects for each model.

• 𝜀𝑖𝑡 is the error term that catches all other effects.

2.6. Leaving the US

Even if a majority of the research made on the relationship between unemployment and GDP growth is based on US data, later research estimates the relationship based on data on the OECD and EU. Findings from the research made from the OECD countries shows evidence of different coefficients between countries within the OECD, and that the relationship between unemployment and output growth are different in times when the economy experiences a recession or boom (Lee, 2000; Moosa et al, 2004; OECD,2012)

Moreover, economist such as Lee (2000) and Nickell (1997) suggest that countries with relatively high flexibility within their labour market tends to have a higher responsiveness from changes in unemployment on economic growth compared to countries with more strict

(18)

- 10 -

regulations, since it cost firms more to lay-off workers within these countries. Hanusch (2013) suggest that another factor explaining the phenomena of jobless recoveries is that firms entering a recession hoard labour, and later when exiting the recession firms has an excess of workers.

To hire workers includes training them for their task and that is why firm may choose to use this type of labour hoarding, and this could be one factor explaining why the phenomena of jobless recoveries exists.

Blachard & Wolfers (2000), Nickell (1997), Moosa (1997) and Sögner & Stiassny (2002) are presenting studies suggesting that factors within the labour market such as levels of union density, minimum wages and grade of employment protection influences the level of unemployment and the relationship of unemployment and output. Hence, there should exist differences in the Okun coefficient between countries, something that is both interesting and necessary to investigate.

Another thing to account for when studying the relationship is the choice of data for the output growth. Okun did use GNP is his original paper which is GDP plus the total capital gains from abroad, but economist that have examined the dynamics of Okun’s coefficient debate whether output should be measured with GDP or GDI. Some economist, Meyer & Tasci (2012) and Baker & Rosnick (2011) suggests that real GDP represent a closer indicator for the home economy of countries and is less volatile than GNP, while other argues that GNP should be used since it correlates closer to other business cycle indicators. Without going into depth, we assume that GDP is an at least as good economic indicator as GNP to estimate the Okun coefficient for the OECD countries.

2.7 Earlier studies

The robustness of Okun’s law: evidence from the OECD countries.

(Lee, 2000)

In this paper, Lee choose to set output as the dependent variable and uses post-war data from countries within OECD to estimate the robustness of the coefficient between 1955-1996. Lee argues that most of the coefficients are statistically valid for many of the countries within OECD but the result differs depending on the choice of model, which is the purpose of his study. The models he chooses to estimate the coefficient with are the gap and first difference method arguing that a higher value of the coefficient correspond to a more rigid labour market. He further discusses and uses three different bandpass-filters to observe the long-term

(19)

- 11 -

unemployment rate that is required for estimating using the gap-method. In his findings, he argues that some evidence of asymmetric behaviour of the coefficient exist and finds convincing evidence of structural breaks in the 70’s.

How useful is Okun’s law?

(Knotek, 2007)

The author of this paper updates Okun’s law with quarterly data on the US stretching from the second quarter of 1948 until the second quarter of 2007. The author further mimics the same models that Okun used in his original research to observe potential differences compared to the results from Okun’s original paper. He further estimates Okun’s coefficient over different time- periods to observe potential dynamic features of Okun’s coefficient, discussing the different coefficients the Gap and first difference model presents.

The conclusion by Knotek is that Okun’s law has not been stable over time and changes depending on which time-period the data is gathered for both the gap and first difference version. Furthermore, he found that the coefficient was different during times of economic recession or expansions for the US economy and observed that periods after recession experienced an increase in output growth while the unemployment rate did not fall. Knotek called this phenomenon jobless growth. He argued that the first difference method of Okuns law is too simple to capture the dynamics of the relation between output and the labour market following by a recession.

Okun’s law: Fit at fifty?

(Ball et al., 2013)

In this paper, Ball et al. discusses Okun’s law in the United States and 20 advanced countries, where the data begins on the year 1948 for the United States and in 1980 for the advanced economies. They examine the data on quarterly and yearly basis with both the gap model, first difference model and the lagged first difference model and uses the HP filter to generate the long-term trends. They find that for the US the coefficient differs depending on which of the three models used. They further discuss the term jobless recovery which is widely argued among economist and explains that following a recession, employment growth is weak and unemployment becomes higher than Okun suggest in his paper.

(20)

- 12 -

In this paper, the results are estimated for 21 OECD countries and the researchers finds that Okun’s coefficient differs across countries. Additionally, the author of the paper tries to explain the countries with extreme values of their coefficients by looking at their respective labour protection measured by the employment protection (EPL) - index.

Unemployment and labour market rigidities: Europe vs North America Stephen Nickell (1997)

In this paper, the author examines institutional factors that could explain why some countries has consistently higher unemployment than for example the US. The time-period that Nickell choose to examine ranges between 1989-1994, he strengthens his choice by arguing that this period followed a recession. The author examines factors such as wage flexibility, employment protection, benefit replacement rate and trade union density to address which factors that could potentially characterize high unemployment rate.

The result from this paper highlights the fact that countries with high unemployment rate are characterized at some extent by high unionization density, with wages bargained collectively.

Furthermore, the length of the period that the unemployed workers can participate in generous benefit programs, with no pressure to find work, increases the unemployment level. The author also points out low minimum wages for young people and low educational level as factors contributing to high unemployment rate within economies. However, in his result, Nickell (1997) shows that some labour market institutional factors have a significant impact on the unemployment level, while others do not and concludes that there are institutional differences among countries within Europe.

2.8. Discussion of earlier literature

To summarize the four articles presented, the research focus mainly on the differences between the gap and first difference model and the choice of filter when estimating potential output and the long-term unemployment rate for the gap version. The research made in the early days was mainly based on data from the US but over time, with new data available, the research came to focus on the differences between countries, especially between the European economies.

Furthermore, the main conclusion about why the coefficients differ between economies has come to focus on the impact of shocks and institutional factors within the labour market. Ball

(21)

- 13 -

et al.’s and Nickell’s research highlight the fact that different labour market policies, together with institutional differences in the labour market, partly explains the cross-country differences.

With that said, due to limited time and data, our paper will focus on mapping the difference of Okun’s coefficient between Europe and the rest of the world within the OECD. Furthermore, even if we should consider many different institutional factors to estimate Okun’s coefficient for our chosen country our time is limited. The choice of the variable Union Density is based on our limited time-frame, has been used in earlier research when estimating Okun’s coefficient and possesses good data point for our research question.

Also, the results from earlier research of the coefficient differ depending on the time-period and econometric method used to estimate Okun’s relationship. However, one should understand that, based on the literature presented, the estimation of the Okun coefficient has come to take large criticism for being too simple and for not accounting for the dynamics of changing economies.

(22)

- 14 -

3. Methodology

___________________________________________________________________________

This chapter describes and argues for the choice of method. It also explains how we handle the problem with estimating long-term trends for unemployment and output. Further, the parameters used in the study will be explained more in detail. Finally, the chapter also gives a deeper understanding about potential problems in Okun’s law, such as endogeneity, and how they have been handled.

___________________________________________________________________________

3.1. Raw data

The data is gathered from the national accounts and labour database from OECD statistical database. They provide quarterly panel-data for cross-country comparison between our selected economies within OECD from the first quarter of 1991 until the fourth quarter of 2016. The advantages with using quarterly data instead of yearly is that we get a bigger dataset, and that it is more informative. Since quarterly data better represents short term fluctuation, the cyclical deviations from the long-term trend will be more realistic. Data on NAIRU is available from the OECD database but the disadvantage is that the data are annually. Instead we will use OECD harmonized unemployment rate which is gathered quarterly. Furthermore, the period 1991 to 2016 includes the fall of DDR which give us data on Germany and allows us to include three economic crises that the economies within the OECD faced in the years 1991, 2001 and 2008.

The data for trade union density is gathered from the same OECD database as mentioned earlier.

According to OECD, the data measures the ratio of the wage- and salary earners that are trade union members, divided by the total numbers of wage- and salary earners. The data is gathered on annual basis during the time-period between 1991 and 2014. Furthermore, we transform this data into a union index rating between 1-3, where 1 represents countries with a union density up to 25 %, 2 represent countries with a union density above 25 % and up to 75 %, and 3 represent countries with over 75 %.

The data on real GDP uses 2010 as reference year, is in US dollar and is seasonally adjusted by OECD with the X-12-ARIMA method, which smoothen out our quarterly data from its sig-saw shape and makes it easier to analyse the real GDP growth over the given time-period. According to Franses et al. (2005), the X-12-ARIMA procedure is one of the most popular procedures used

(23)

- 15 -

for seasonal adjustment and calendar effects. Also, Ghysels and Osborn (2001) points out that the X-12-ARIMA is based on the procedure called X-11 which is a moving average filter and together with the routine called regARIMA the model is fitted to the data which creates the X- 12-ARIMA procedure. The X-12-ARIMA is standardized among statistic bureaus such as US Census Bureau and Canada Statistic for removing seasonal and calendar effects. Finally, to extract the potential output from the data we will be using the HP-filter. Hodrick & Prescott (1986) views the timeseries as a trend component, a cyclical component and a seasonal component, and argues that the seasonal component must be eliminated for the HP-filter to be consistent. Hence, the X12-ARIMA adjustment to the data is necessary.

As for unemployment, it is gathered from OECD statistical database. The OECD uses a method for making it easier to compare the data between different countries (harmonization). This method is according to OECD (2017) a standardized method for estimating unemployment for international comparison. They lift the fact that the uniform application of the definition makes the data more comparable between countries. OECD defines the unemployed as: people aged 15-64, who are without work, available for work and have taken certain steps to find work.

Where 15 is defined as the minimum age for the labour force for every country except for Spain, UK and US, which uses 16 as minimum age. The unemployment rate is gathered between the time-period 1990 until 2017 and is seasonally adjusted.

3.2. Descriptive statistic

Our variable (𝑈𝑖− 𝑈𝑖) shows how the actual unemployment rate fluctuates around the natural unemployment rate. Here actual unemployment corresponds to variable 𝑈𝑖 while the natural unemployment corresponds to variable 𝑈𝑖. Hence our unemployment gap variable becomes (𝑈𝑖 − 𝑈𝑖).

The variable for output, (𝑌𝑖 − 𝑌𝑖), uses real GDP which is measured in US dollars. To estimate the variable of potential GDP for our model we use the HP-filter on the logarithmic values of our countries real GDP. Here the logarithm of real GDP corresponds to 𝑌𝑖 while potential GDP corresponds to 𝑌𝑖. Our output gap variable becomes (𝑌𝑖− 𝑌𝑖).

(24)

- 16 -

In earlier research done by Adanu (2005), the statistical tests have been done with GDP in logarithmic form and with unemployment rate in its original form. We will use the same method, that is, the GDP will be logged but not the unemployment rate.

Variable Description Database

Actual Real GDP Q1 1991-Q4 2016

Real GDP measured in US dollar on quarterly data gathered from 21 selected OECD countries.

OECD

Potential GDP Q1 1991-Q4 2016

Estimated variable from Real GDP using its logarithm value and HP-filter

OECD

Harmonized

unemployment rate Q1 1991-Q4 2016

Quarterly data for the harmonized unemployment rate from 21 selected OECD countries.

OECD

Natural unemployment rate

Q1 1991-Q4 2016

The natural unemployment rate is estimated from the harmonized unemployment rate.

OECD

Trade Union density 1991-2014

Percentage of total wage- and salary earners participating in a trade union., Rank 1=0-25%, 2=25-75%, 3=75-100%.

OECD

Table 1: descriptive statistic

(25)

- 17 -

The characteristics of the data, such as means, standard deviations etc. are summarized in the table below.

Summary of treated data.

Variable Observations Mean Standard Deviation

Min Max

Real GDP 2184 1680160 2860047 21607.54 1.70e+07 Unemployment

Rate

2184 .0725632 .0370324 .015 .2626667

Logarithmic GDP

2184 13.43305 1.386837 9.980798 16.64982 GDP Gap 2184 -2.46e-12 .0156989 -.0913857 .096339 Unemployment

GAP

2184 -2.46e-12 .0067079 -.0383317 .0350896 Union Density 1764 33.78666 20.78529 7.547659 83.86255 Table 2

From table 2, we see that the unemployment rate takes a mean value of 7.25 %. We can also see that the unemployment rate for the countries takes values between 1.5 % and 26 %, which is not surprising considering the geographical and institutional differences between them. The same reasoning follows considering the union density for the countries. With the lowest value of 7.5 % and the highest of 83.9 % the density of workers in trade unions differ remarkedly and could potentially contribute to explain the cross-country differences among the OECD members.

3.3. The Least Squares Assumptions

For our model to be consistent and for our estimators for the OECD countries to be good, we need to check the five OLS assumption for panel data. This is needed for our economic interpretation, when discussing the differences between countries and to justify our findings.

• The residuals have the mean of zero

We test for normality, that is, if our sample has residuals with mean of zero. Since we have 2184 observation in our data we can rely on the Central Limit Theorem which states that when the number of observations in the sample is large, the sampling distribution of the sample average is approximately normally distributed (Stock & Watson, 2015). To check for this assumption, we perform a kernel density test (see Appendix 3).

(26)

- 18 -

• (𝑿𝒊𝟏,, 𝑿𝒊𝟐, … , 𝑿𝑰𝒕´𝑻,,𝒀𝟏𝒕,𝒀𝟐𝒕, … , 𝒀𝒊𝒕), i = 1,…,n are i.i.d. across entities.

Stock and Watson (2015) argues that this assumption is fulfilled if the data are collected by simple random sampling. Our data should be considered trustworthy and reliable since we collect the raw data from Stat OECD database. The sampling method that is used is well described by the OECD database, and are based on household surveys from most of the National Statistical Institutes, which uses multi-staged stratified random sample design.

However, according to Stock and Watson (2015) we should check for serial correlation. Serial correlation occurs when the value of 𝑋𝑖𝑡 correlates over time and is a pervasive feature of time- series.

• Large outliers are unlikely

Observations with values that are far outside the usual range of the data can negatively influence the OLS estimators of the coefficients in the regression (Stock and Watson, 2015). Since we have a very large dataset, large outliers should have a relatively insignificant effect on our estimators. Hence, there is no risk that this assumption is not fulfilled.

• No perfect multicollinearity

The problem with perfect multicollinearity arises when two or more of the regressors are perfect linear functions of each other (Stock & Watson, 2015). According to Stock and Watson it is impossible to compute the OLS estimator if perfect multicollinearity exists. Stata 14 has a built- in feature that omits regressors that are perfectly correlated, hence, this problem is easily handled in the study.

3.4. The estimation of potential output and long-term rate of unemployment In our model, we are using the Gap-version approach of Okun’s Law. This means that the model involves a measurement of how much the real output and unemployment rate deviates from their respective long-term trend, that is how they deviate from the long-term unemployment rate and the potential GPD. These values are not observable in the economy and are required to be estimated. Adanu (2005) argues that the method of choice to estimate these variables can have a significantly influence on the estimated coefficients. Further Adanu (2005), Baxter &

King (1999), Freeman (2001) and Dennis and Razzak (1999) highlights that when estimating the long-term unemployment rate and the Potential GDP there are several ways to go. Among these are the removal of deterministic trends and the use of first differencing. They also mention

(27)

- 19 -

more sophisticated methods such as band pass filtering and the use of a Hodrick Prescott filter, with the advantage that they remove both high and low frequencies from the time series. In addition to these methods Giorno et al. (1995) also points out the possibility to use a production function for the estimation. Since this is a method that requires a lot of economic data, we find it more suitable to use the HP-filter for the purpose mentioned above. Furthermore, in earlier research (Crespo Cuaresma, 2003; Kim et.al. 2015; Lee, 2000), the HP-filter has been the method of choice for breaking out the cyclical component when estimating Okun’s coefficient using the gap-method.

The purpose of using the HP-filter is the fact that it decomposes the time series into a cyclical and a growth component, that is 𝑦(𝑡) = 𝑔(𝑡) + 𝑐(𝑡), where 𝑦(𝑡) is the natural logarithm of given series, 𝑐(𝑡) is the cyclical component and g(t) is the growth component. (Cogley &

Nason, 1995; Ravn & Uhlig, 2002). Furthermore, the method – unlike other method such as first difference – does not suffer from a loss of data when applied. That is, for a time series y(t) for 𝑡 = 1, … 𝑇, the procedure estimates the cyclical component, 𝑐(𝑡), for 𝑡 = 1, … 𝑇. (Baxter &

King, 1999).

Although there are many advantages with using the HP-filter to de-trend and breaking the cyclical component out of time series, it is important to consider the fact that the method has received criticism over the years. The most important is lifted by the Giorno et al. (1995) and Ball et.al (2013), and is a problem with end-point bias. By using the HP-filter we fit a trendline symmetrically through the data. This brings a problem if the starting- and the ending point of the dataset does not reflect similar points in the cycle. Hence, there is a possibility that the trend is moved either upwards or downwards towards the actual path of the end-points. Ball et.al (2013) has performed a test of robustness to check whether the end-point bias has a considerable influence on the estimation of the Okun’s coefficient or not. However, they find that influence of end-point bias in the decomposition of the time series does not affect the estimation of the coefficient. Hence, we do not find it necessary to address the problem with end-point bias in this study.

3.5. Decomposition Procedure

It is important to state that Hodrick and Prescott (1986) view the time series as a cyclical component and a trend component, as described by Cogley and Nason (1995) and Ravn and Uhlig (2002), but also as a seasonal component. But, as discussed above in the chapter about

(28)

- 20 -

raw data, OECD has smoothened out the seasonal effects by applying a X-12-ARIMA filter to the data. Hence, there is no necessity for us to handle this effect. According to Giorno et al.

(1995), when applying the HP-filter we fit a trend to the data points of the logarithm of the real GDP, and by making the regression coefficients vary over time, the HP-filter takes structural breaks into account.

The variation in the growth component, g(t), is penalized by a smoothing parameter, λ (Ravn and Uhlig, 2002; Hodrick & Prescott, 1997; Dennis & Razzak, 1999). This parameter is described by the Giorno et al. (1995) as a weighted factor that determine to which grade the trendline should be smoothed. Thus, the smoothing parameter will regulate the sensitivity of the trend to short-term fluctuations. That is, with a lower value of λ, the trend will follow the actual output more closely, while a higher value of λ will create a trend that is less sensitive to the short-term fluctuations, and which follows the mean growth rate of the time series more closely. When choosing the value of λ, there are several different methods. For example, it is possible to choose a dynamic value of λ, which adapts depending on which country is estimated (Giorno et al., 1995). Although there are different methods, such as the one described, which could give a more accurate estimation of the Okun’s coefficient, we will choose to go with the approach that is described by the OECD to be the industry standard, e.g. choosing the same λ, which is 1600, for every country. This choice of λ is also strengthened by Baxter & King (1999), Hodrick & Prescott (1986) and Dennis & Razzak (1999), who points out that 𝜆 = 1600, is the common choice when working with quarterly data. For deeper knowledge of the technicalities of the HP-filter we recommend reading Theory Ahead of Business-Cycle Measurement by Edward C. Precott.

There are also several methods for estimating the unemployment rate gap. Earlier researchers (Cogley & Nason 1995; Lee (2000); Adanu (2005) uses the HP filter with at smoothing parameter of 1600 on the data points to estimate the deviations from the long-term unemployment rate. This is the method of choice that will be used in this study as well.

3.6. Endogeneity

When we are estimating economic models, we assume that the causality runs from the dependent variable to the independent variable (Stock & Watson, 2015). One problem that arises in Okun’s law is that there is uncertainty in whether the unemployment rate affects the

(29)

- 21 -

output or if it is the other way around. Stock and Watson (2015) describes this problem as simultaneous causality. Further, according to Stock and Watson this problem causes biases and inconsistency in the OLS estimator. The phenomenon is often referred to as endogeneity.

Barreto and Howland (1993) states that there is no doubt that simultaneity exists in the Okun’s law and that unemployment and output are endogenous. This brings a potential problem to this study. Earlier researchers, among others; Aschoff & Smith (2008), Brinks & Coppedge (2006), Clemens et al. (2011), Cornett et al. (2007) and Green et al. (2005) suggest several methods for handling endogeneity, the most mentioned being the use of instrumental variables and the use of lags in the independent variable. To address this potential problem with biases in the estimators from endogeneity we will go with the approach of lagging the unemployment gap.

By doing this we can regress a model where unemployment gap affects output gap, but where output gap does not affect unemployment gap. That is, unemployment gap, (𝑈 − 𝑈), at time 𝑡 influences output gap, (𝑌 − 𝑌), at time 𝑡 + 1. At the same time, output gap at time 𝑡 + 1 cannot influence unemployment gap at time 𝑡. Furthermore, there is an advantage in the economic interpretation when lagging the unemployment gap. Intuitively the effects of unemployment on output does not happen in an instant, but rather there should exist inertia in the process.

Barreto and Howland (1993) argues that the choice of the dependent variable when regressing the Okun’s coefficient should be made based on the purpose of the study. The main purpose for Okun (1962) when investigating the relationship between unemployment and output was to give policymakers an instrument to measure the impact of labour programmes on output growth. They did this by estimating how output affected unemployment, and then taking the reciprocal of the coefficient to show how unemployment affected output. The problem with this is, according to Barreto and Howland (1993), that a decreasing unemployment are caused only partly by an increasing output. The other way around, an increasing output is caused only partly by a decreasing unemployment. They describe this by pointing out that some of the changes should be attributed to unobserved shocks in productivity and working hours. This means that there exist different biases in the two estimations. They strengthen this argument by using both methods of estimating the coefficient on the same data, and show that they differ. By estimating using unemployment as dependent and take the reciprocal of it tends to overestimate the coefficient compared to using output as the dependent variable. It is obvious that taking the reciprocal of any one of the coefficients give an incorrect measure of the other coefficient. Since

(30)

- 22 -

we are interested in investigating the original Okun’s law, that is how changes in unemployment change output, we find it more justified to use output as the dependent variable in the model.

(31)

- 23 -

4. Results

___________________________________________________________________________

In this chapter, the reader will get a brief presentation of this study’s results and empirical findings. For an easier understanding, the presentation will be divided into different sections for each of the regressions we make.

___________________________________________________________________________.

4.1. The ordinary least squares assumptions

We have checked for the assumptions regarding the ordinary least squares method. As argued above, stata omits variables if they are perfectly multicollinear. We have no omitted variables in our results, hence no multicollinearity. Since we have a very large sample, the influence of outliers is very small, hence, the distortion from outliers is negligible. According to the Central Limit Theorem, the large sample also makes it feasible to assume that the sampling distribution of the sample average is approximately normally distributed (Stock & Watson, 2015). We have also checked for heteroskedasticity and autocorrelation, the tests (see Appendix 6), shows that there exists both autocorrelation and heteroskedasticity. According to Stock and Watson (2015), this problem can be handled by using heteroskedasticity- and autocorrelation-consistent standard errors, which has been done.

4.2. Estimation of the Okun coefficient

We begin with presenting the standard model of the Okun’s coefficient, followed by additional models with different interaction terms. We estimate the following regression for the standard model:

(𝑌𝑖− 𝑌𝑖)𝑡= −1.226(𝑈𝑖− 𝑈𝑖)𝑡−1+ 𝜀𝑖𝑡

We can see that the Okun’s coefficient is estimated to -1.226. This result corresponds with what earlier research has found. Although, one should know that since the method of choice influences the estimated coefficients, a comparison with earlier research is hard to make.

(32)

- 24 -

4.3. Country interaction model

(𝑌𝑖 − 𝑌𝑖)𝑡= 𝛿(𝑈𝑖− 𝑈𝑖)𝑡−1+ 𝜃𝐷1𝑖(𝑈𝑖− 𝑈𝑖)𝑡−1+ 𝜀𝑖𝑡

The model consists of a dependent variable, output gap, an independent variable, unemployment gap and an interaction variable between unemployment gap and a dummy for country, the results are presented in table 3 below. The empirical tests are conducted with Australia as the reference country in the dummy variable, this means that the coefficients are estimates of the difference in slope between a given country and Australia. The results show a negative relationship between output gap and unemployment gap for all countries included in the study. This is a result that is expected. We can also see that there are country-specific effects, on at least a 5 % significance level, for all countries except Belgium.

Okun’s Coefficient

Standard Error

t-value P>t 95 % Confidence Interval

Australia -.255 .131541 -1.94 0.052 -.5133368 .0025859 Country:

Belgium −.367 .2228058 -1.65 0.100 -.8038446 .0700321 Canada −1.231∗∗∗ .203051 -6.06 0.000 -1.629112 -.8327163 Germany −.980∗∗∗ .2896714 -3.38 0.001 -1.548533 -.4123997 Denmark −.862∗∗∗ .2347178 -3.67 0.000 -1.322037 -.4014398 Spain −. 582∗∗∗ .1396495 -4.17 0.000 -.8558213 -.3080958 Finland −.926∗∗∗ .2046208 -4.53 0.000 -1.327475 -.5249226 France −.592∗∗ .2409756 -2.46 0.014 -1.064334 -.119193 Great Britain −1.104∗∗∗ .2802261 -3.94 0.000 -1.653651 -.5545634 Ireland −1.391∗∗∗ .3013152 -4.62 0.000 -1.981867 -.8000647 Italy −1.069∗∗∗ .2174275 -4.92 0.000 -1.495569 -.6427867 Japan −2.062∗∗∗ .5728531 -3.60 0.000 -3.185409 -.9385956 South Korea −1.386∗∗∗ .3221012 -4.30 0.000 -2.017627 -.7542993 Luxemburg −2.211∗∗∗ .5385568 -4.11 0.000 -3.267463 -1.155165 Mexico −2.517∗∗∗ .273609 -9.20 0.000 -3.053835 -1.980701 Netherlands −.914∗∗∗ .2069308 -4.42 0.000 -1.319456 -.5078437 Norway −.947∗∗∗ .3152998 -3.00 0.003 -1.565081 -.3284295 New Zealand −.684∗∗∗ .2191367 -3.12 0.002 -1.114 -.2545138 Portugal −.859∗∗∗ .1867109 -4.60 0.000 -1.225283 -.4929763 Sweden −.937∗∗∗ .2248225 -4.17 0.000 -1.378274 -.4964874 United States −.793∗∗∗ .1679579 -4.72 0.000 -1.122569 -.4638131

Table 3: Regression of OECD countries. ***, ** and * corresponds to 1-, 5-, and 10 percent significant level.

(33)

- 25 -

To calculate the marginal effects, we take the derivative of the model with respect to unemployment rate, and test whether 𝛿𝑖+ 𝜃𝐷1𝑖 ≠ 0. We can see from the results that 𝛿𝐷𝑖, together with 𝛽, are significantly different from zero at a 5 % significant level for all countries in the study. The coefficients that are estimated are mostly in line with what earlier research has found. However, we find a couple of coefficients that have extreme values compared to earlier research. Among those countries are for example Australia, where a 1 % change in unemployment gap leads to a change in output gap by -0.255 %, We can see that Mexico (- 2.773) is the country where changes in the unemployment has the most powerful effect on the output. On the other hand, Australia (-0.255) is the country with the lowest effect from unemployment on output. These results are presented in table 4 below.

𝝏(𝒀𝒊− 𝒀𝒊)

𝝏(𝑼𝒊− 𝑼𝒊)

Standard Error

t-value P>t 95 % Confidence Interval

Country

Australia −.255 .131541 -1.94 0.052 -.513337 .0025859 Belgium −.622∗∗∗ .1798316 -3.46 0.001 -.974945 -.269619 Canada −1.486∗∗∗ .1546825 -9.61 0.000 -1.78963 -1.18295 Germany −1.235∗∗∗ .2580824 -4.79 0.000 -1.74196 -.729724 Denmark −1.117∗∗∗ .194395 -5.75 0.000 -1.49834 -.735891 Spain −.837∗∗∗ .0468929 -17.86 0.000 -.929295 -.745374 Finland −1.182∗∗∗ .1567376 -7.54 0.000 -1.48895 -.874201 France −.847∗∗∗ .2019065 -4.20 0.000 -1.24309 -.451186 Great

Britain

−1.359∗∗∗ .2474341 -5.49 0.000 -1.84472 -.874246 Ireland −1.646∗∗∗ .2710864 -6.07 0.000 -2.17796 -1.11472 Italy −1.325∗∗∗ .1731234 -7.65 0.000 -1.66406 -.985046 Japan −2.317∗∗∗ .5575461 -4.16 0.000 -3.41077 -1.22399 South

Korea

−1.641∗∗∗ .2940173 -5.58 0.000 -2.21793 -1.06475 Luxemburg −2.467∗∗∗ .5222455 -4.72 0.000 -3.49085 -1.44253 Mexico −2.773∗∗∗ .2399143 -11.56 0.000 -3.24313 -2.30216 Netherlands −1.169∗∗∗ .1597415 -7.32 0.000 -1.48229 -.855761 Norway −1.202∗∗∗ .2865501 -4.20 0.000 -1.76408 -.640185 New

Zeeland

−.940∗∗∗ .1752651 -5.36 0.000 -1.28334 -.595925 Portgual −1.115∗∗∗ .1325063 -8.41 0.000 -1.37436 -.854651 Sweden −1.193∗∗∗ .1823242 -6.54 0.000 -1.55031 -.835205 United

States

−1.049∗∗∗ .1044359 -10.04 0.000 -1.25337 -.84376

Table 4: Marginal effects OECD countries. ***, ** and * corresponds to 1-, 5-, and 10 percent significant level.

(34)

- 26 -

4.4. Union Density model

To investigate whether the grade of union density affects the Okun coefficient or not, we do a regression with an interaction term between union density and the unemployment rate:

(𝑌𝑖 − 𝑌𝑖)𝑡 = 𝛿(𝑈𝑖− 𝑈𝑖)𝑡−1+ 𝜃𝐷2𝑖(𝑈𝑖− 𝑈𝑖)𝑡−1+𝜀𝑖𝑡

The results show us that we have a significant effect from the grade of union density on the relationship between unemployment and output. We have significance on a 5 % level for the countries that achieved rank three on the grade of union density, both when compared to those countries that achieved one and those that achieved two. What the results show is that countries that have rank number three on the grade of union density have a lower Okun’s coefficient by 0.386 on average compared to those with rank two, and 0.477 compared to those with rank one (see Appendix 5). Hence, we can see that countries grade of union density affects the Okun coefficient, which is in line with what earlier research suggests. Countries with a high grade of union density have a lower coefficient on average. This result differs from what earlier research shows. They argue that a higher union density should lead to a higher Okun’s coefficient. When we calculate the marginal effects, and check if 𝛿 + 𝜃𝐷2𝑖 ≠ 0, the results show that they differ from zero and that they together influence the relationship.

Union Density

𝝏(𝒀𝒊− 𝒀𝒊)

𝝏(𝑼𝒊− 𝑼𝒊)

Standard Error

t-value P>t 95 % Confidence Interval

Rank 1 −1.235∗∗∗ .0750863 -16.44 0.000 -1.38190 -1.08737 Rank 2 −1.325∗∗∗ .10412 -12.73 0.000 -1.52971 -1.12128 Rank 3 −.849∗∗∗ .0901955 -9.41 0.000 -1.02540 -.671599

Table 5: Marginal effects, union density % 0-100, rank 1=0-25%, 2=25-75, 3=75-100. ***, ** and * corresponds to 1-, 5-, and 10 percent significant level.

(35)

- 27 -

4.5. The European Union model

We also want to check whether belonging to the European Union has any effect on the relationship or not. This is done by regressing the model with an interaction term between if the country is in the EU or not and Unemployment rate:

(𝑌𝑖 − 𝑌𝑖)𝑡 = 𝛿(𝑈𝑖− 𝑈𝑖)𝑡−1+ 𝜃𝐷3𝑖(𝑈𝑖− 𝑈𝑖)𝑡−1+ 𝜀𝑖𝑡

The results from the regression shows that the countries that are a part of the European Union have a lower Okun’s coefficient of 0.311 on average. In table 7, where we present the test for marginal effects, we can see that 𝛿 + 𝐷3𝑖 ≠ 0, hence the interaction effect together with the unemployment gap affects the output gap.

Okun’s coefficent

Standard Error

t-value P>t 95 % Confidence Interval

Non-EU −1.458∗∗∗ .1042318 -13.98 0.000 -1.662071 -1.253263

EU . 311∗∗∗ .1167294 2.67 0.008 .0821887 .5400139

Table 5: EU member vs non-EU. ***, ** and * corresponds to 1-, 5-, and 10 percent significant level. Non-EU as reference.

𝝏(𝒀𝒊− 𝒀𝒊)

𝝏(𝑼𝒊− 𝑼𝒊)

Standard Error

t-value P>t 95 % Confidence Interval

Non-EU −1.458∗∗∗ .1042318 -13.98 0.000 -1.662071 -1.253263 EU −1.147∗∗∗ .0525498 -21.82 0.000 -1.249619 -1.043513

Table 6: Margins-test EU vs non-EU. ***, ** and * corresponds to 1-, 5-, and 10 percent significant level.

References

Related documents

The Drosophila Activity Monitoring System (DAMS) was employed to monitor the sleeping time and the activity of Knocked-down flies and normal flies over 48 hours, and I have seen

xed hypotheses in general linear multivariate models with one or more Gaussian covariates. The method uses a reduced covariance matrix, scaled hypothesis sum of squares, and

During this time the out- come from political interaction between geographically divided groups in society will be non-cooperative in nature, as groups try to grab as large a

In the hatched area, the flow on the torus contains a stable node, a saddle point and an unstable periodic orbit.. This will be briefly discussed in

The arm-trombone system consists of a rod, tilted at a …xed angle, and two bars, connected by a hinge, that represents the arm. The shoulder consists of another hinge and is

If you release the stone, the Earth pulls it downward (does work on it); gravitational potential energy is transformed into kinetic en- ergy.. When the stone strikes the ground,

One of the main objectives of the BYN is to ensure that qualified la- bour manpower is available in the Swedish construction industry. The parties in the construction

Maridea Petrova (MK); Profes- sor Dr Nuria Benach, University of Barcelona (ES); Professor Dr Albert Depagne, Liège University (BE); Dr Raimund Gutmann, Independent