University of Economics, Prague
Bachelor’s Thesis
2019 Azamat Bodeneyev
University of Economics, Prague
Faculty of Economics
Title of the Bachelor´s Thesis:
Okun’s Law in the Eurozone in the Post- Crisis Era
Author: Azamat Bodeneyev
Supervisor: Dr. Ing. Martin Slany
Declaration of Authenticity
I hereby declare that the Bachelor´s Thesis presented herein is my own work, or fully and specifically acknowledged wherever adapted from
other sources. This work has not been published
elsewhere for the requirement of a bachelor’s degree programme. The used literature and sources are stated in the attached list of references.
In Prague on 10.05.2019 Signature: Azamat Bodeneyev
I am grateful to all people I have had the pleasure to work with during my studies in Prague and Barcelona. Each professor has provided me extensive personal and professional knowledge and assisted me with acquiring amazing skills. I would especially like to thank Dr. Martin Slany, the supervisor of mine. He has given me space to write the thesis in the way I wanted it to be and has guided me throughout writing. A very special gratitude goes out to the University of Economics for providing the scholarships to me and inspiring me to be a better person. I am also thankful to the following university professors: Karel Helman and Kamil Kovar for their assistance with analyzing the data and for their passion and motivation that made me fall in love with statistics and economics.
More importantly, this work would not have been possible without the financial support of my parents. I would like to thank my ultimate role models, mother and father. Nobody has been more important to me in the pursuit of the goals than my parents, whose love and support are with me in whatever I chase…
Finally, I am grateful to all my friends for supporting me spiritually throughout writing the thesis. I would like to thank my team deputy, Ksenija Pourová, at DXC Technology, who has given me the flexibility so that I could write the thesis.
Abstract
Eurozóna zažila v letech 2008-2013 dvě tragické finanční krize. Během krize se prakticky všechny země eurozóny setkaly s vysokou mírou nezaměstnanosti a z obecného hlediska regulátorů je nízký hospodářský růst spouštěcím mechanismem vysoké nezaměstnanosti. Tato práce se snaží zjistit, zda je vztah mezi reálným hrubým domácím produktem a nezaměstnaností pro země eurozóny v období 2002-2018 významný. Čtvrtletní analýza panelových dat je použita pro zjištění asociace mezi růstem a nezaměstnaností ve třech obdobích (2002:1-2008:3; 2008:4-2018:4; 2002:1-2018:4) pro devatenáct zemí eurozóny. Hodrick-Prescottův filtr je aplikován na několik makroekonomických časových řad, aby byl ilustrován obraz vlivu globální recese a krize eurozóny. Provádí se také pečlivá analýza a popis souborů údajů. Různé metody jsou použity, aby lépe pochopit teorii a porovnat výsledky s původními zjištěními. Klíčovým zjištěním tohoto dokumentu je, že hospodářský růst má významnější dopad na nezaměstnanost mužů než na nezaměstnanost žen v eurozóně. Okunův koeficent pro postkrizovou éru je navíc mnohem významnější než pro předkrizovou éru.
Klíčová slova:
Nezaměstnanost, hospodářský růst, eurozóna, analýza panelových dat, HP filtr, Okunův zákon.
JEL klasifikace:
C23, C22, E24, E61, F43, J21.
Abstract:
Eurozone has experienced two tragic financial crises during 2008-2013. When the global financial crisis hit, it exposed all the inherent flaws of the eurozone. During the crisis virtually all countries of the euro area has been facing high unemployment rate and from the general perspective of regulators, low economic growth is a trigger mechanism for high unemployment. This paper seeks to investigate whether the relationship between real gross domestic product and unemployment is significant for the eurozone countries for the period 2002-2018. Panel quarterly data analysis is employed to ascertain the association among growth and unemployment in the three periods (2002:1-2008:3; 2008:4-2018:4; 2002:1-2018:4) for the nineteen countries of the euro area. The Hodrick-Prescott filter is applied to several macroeconomic time series in order to illustrate the picture of the effect of Global Recession and Eurozone crisis. The meticulous analysis and description of the datasets are provided as well. Different methods are employed to better understand the theory and compare the results to the original findings. The key finding of this paper is that economic growth has more significant impact on the male unemployment than on the female unemployment in the eurozone. Moreover, Okun’s coefficient for the post-crisis era is found to be much more significant than for the pre-crisis era.
Key words:
Unemployment, Economic growth, Eurozone, Panel Analysis, HP Filter, Okun’s Law.
JEL Classification:
C23, C22, E24, E61, F43, J21.
Table of Contents
INTRODUCTION……….. 5
CHAPTER 1 - CRISES & OKUN’S LAW...……….. 8
FINANCIAL CRISIS & EUROZONE CRISIS………..………... 8
LITERATURE REVIEW………... 13
THEORY OF OKUN’S LAW……… 16
CHAPTER 2 - PANEL ANALYSIS OF OKUN’S LAW………. 18
DATA - SUMMARY STATISTICS………....…. 19
THE EFFECTS OF GREAT RECESSION………...………... 19
BEFORE & AFTER CRISIS………. 28
PRE-CRISIS ERA……….. 28
POST-CRISIS ERA………. 31
GENDER ASYMMETRY IN OKUN’S LAW………. 34
MALE UNEMPLOYMENT……….. 35
FEMALE UNEMPLOYMENT………. 36
CHAPTER 3 - GAP VERSION OF OKUN’S LAW FOR EUROZONE COUNTRIES….. 38
CONCLUSION……….. 71
LIST OF ABBREVIATIONS……… 75
LIST OF FIGURES……… 76
REFERENCES……… 78
ANNEXES……….. 80
THE MAP OF THE EUROZONE……….. ……….. 80
Introduction
Unemployment rate is one of the most important indicators for measuring the performance of any economy. Ergo, there are efforts being made by, especially powerful countries, to tackle the uprising issue of high unemployment. The problem is a cause of concern to all governments of the eurozone and the world for various reasons. The most important one is a negative impact on a society’s well-being which consequently leads to other social and economic problems.
Employment is always thought of as an intermediary between economic growth and poverty reduction because it influences the human development. Generally, from the regulators’
perspective, it could be said that the main explanation behind the high unemployment issue is low economic growth that simply generates less jobs. It is considered that the most crucial variable affecting the level of unemployment is real output, measured by real GDP. Fluctuations in gross domestic product are associated with an inverse reaction in unemployment. The statistical output- unemployment relationship is known as Okun’s law, which was discovered in 1962 by Arthur Okun. It is one of the most famous theories in macroeconomics and it has been found to be hold for a large number of countries and regions. Even until this day, it remains a point of reference in many academic papers and it is used as a rule of thumb despite the fact that it has been a long time since the study has been published. The relationship between economic growth and unemployment will always be fundamental in policymaking and analysis of economic performance.
Understanding Okun’s relationship is imperative in order to construct and promote an efficient economic and social policy, in particular in the crisis period.
Arthur Okun proposed two empirical relationships correlating unemployment rate to real output in his paper, both of the regressions are not complex and have been used as a rule of thumb since then. The author modeled this causality as a simple model in which the growth rate of GNP depends on the fluctuations in unemployment rate in the period 1946-1960 for the US economy.
According to his original findings, a 1% increase in unemployment leads to roughly more than 3%
loss in GDP growth. The responsiveness of unemployment to changes in GDP vary across countries and periods. However, overall the effect of GDP on unemployment is considered to be statistically significant. Typically, the constant growth requires additional labor and growth slowdowns are associated with rising unemployment. The situation changes when crisis hits the economy. Firms react more sensitively to decreases in economic growth and the reaction is
typically quicker and without the delay. As many other theories, Okun’s law has been subjected to many replications and the way in which economic growth affects unemployment rate has been debated over many years. It is important to point out that it is not only a vital issue for big influential countries but also for the countries with growing economies that face jobless growth.
Therefore, considering all the aforementioned factors, my aim is to study the economic relationship between economic growth and unemployment through econometric and economic analysis and to define whether low economic growth is an actual excuse for high unemployment rate. Another goal of the paper is to investigate whether the reaction of unemployment to changes in real gross domestic product is more significant in the post-crisis era than in the pre-crisis era by comparing Okun’s coefficient.
Eurozone members will enter into the long-lasting debate over a new president of ECB on 31 October 20191. The responsibility will be on a newly elected president to bring up the economic challenges of the economically integrated area. The European Central Bank is in charge of controlling the money supply of the euro according to the Article 16 of the Statute of the European system of Central banks and of the European Central Bank2. It also sets the monetary policy in the entire euro area, which consists of nineteen members of the European Union. What’s more, it is de facto the largest monetary zone in the world that decided to replace their national currencies with one single currency - euro. The eurozone was created by eleven member states on 1 January 1999 and then two years after Greece joined them in 2001, followed by Slovenia in 2007, Cyprus and Malta in 2008, Slovakia in 2009, Estonia in 2011, Latvia in 2014 and Lithuania in 20153. However, only in 2002 the euro physically entered into circulation as a legal tender. Many of the papers studying Okun’s law are however focused on elucidating mostly the empirical part and describing the methods. Conversely, this thesis focuses on the corollaries of the crisis on the single currency union more deeply and provides the meticulous economic description of the data.
My paper investigates the adjustment of employment to economic growth change in nineteen countries over the period 2002-2018 using the quarterly data. Most of the countries have experienced dramatic shocks during the period due to the Great Recession and the eurozone debt crisis. Hence, the way the labor markets function in the countries of euro area is addressed.
1 European Central Bank (2019).
2 Protocol on the statute of the European system of central banks and of the European central bank. Article 16:
Banks notes. Official Journal C 191, 29/07/1992 p.72.
3 European Commission (2019). What is the euro area?
However, the observed differences in Okun’s coefficient may be caused by local labor market institutions and other conditions. The paper attempts to shed a light on the different patterns in unemployment within euro area countries applying Okun’s law as a framework. It focuses on testing the validity of this framework for euro area for the pre-crisis and post-crisis periods.
The studies of Okun’s law had so many divergent variations and it has been widely adopted in the field of macroeconomics. This paper specifically focuses on the unemployment-on-output version. Hence, univariate model can be fruitfully used because the regression runs in the correct direction for the question at hand. For this purpose, this thesis considers unemployment as the main subject of its study. What is more, there was no such research in the literature combining all the euro-area countries and the eurozone as a whole. One of the objectives is to improve the previous research on the euro area by expanding the period and analyzing it with more methods.
More importantly, the thesis documents the differences between male and female unemployment rates in response to the changes in real gross domestic product across the countries of the eurozone. In addition, it tests whether the economic growth in the euro area has more significant impact on the male unemployment than female unemployment and whether there is actually a significant effect of economic growth on the female unemployment.
The first chapter includes the brief history background of the financial crisis and the eurozone debt crisis, describing the important milestones. Then, the review of literature follows including the description of the methods and results that authors obtained while observing different countries and periods. The third point in Chapter one starts with a brief theoretical review on Okun’s proposed empirical relationships: the “first-difference model” and the “gap model”. The second chapter describes the panel data analysis of Okun’s law and shows how the growth- unemployment relationship has varied over time describing the pre-crisis stage and the aftermath.
In addition, it expands the original findings by analyzing the effect of economic growth on male and female unemployment, essentially showing the gender asymmetry in Okun’s law. The aim of the third chapter is to show consequences of the crisis for every single country by analyzing the effect of change in gross domestic product on change in unemployment using the gap version of Okun’s law.
Chapter 1 - Crises & Okun’s Law Financial Crisis & Eurozone Crisis
First of all, in order to understand the consequences of the crisis as well as the impact on Okun’s coefficient, the history of the global financial crisis must be unraveled briefly. It all started in 2007, when Lewis Ranieri4 came up with the idea of creating mortgage-backed securities and when the US mortgage market let borrowers borrow money despite their poor credit histories.
These mortgages received the name “ninja” loans, which stand for “no income, no job, no assets”.5 These loans were bundled up and repackaged along with other normal mortgages and then sold out to investors. The key question is: If the risk was eliminated, or better say spread, what could go wrong? In the short-term it did solve the existing housing price boom, however in the long- term it led to catastrophic, severe consequences. Moreover, in order to better describe the situation at that time of how delusional the system was, an argument in the movie “The Big Short” can better explain it, in which Bill Miller faced Steve Eisman in the real confrontation: “Only that in the entire history of Wall Street, no investment bank has ever failed unless caught in criminal activities”.6 What went wrong is that the number of borrowers who were not able to pay back were proliferating and that is when it all started. The housing market began to fall. The early signs, at least in the US, started appearing in April 2007. Following it, a British bank, Northern Rock informed regulators that it was facing funding troubles. Bank of England had to bailout the bank directly. Consequently, the news of the rescue leak led to bank run on Northern Rock on September 14, 2007.7 The panic has subsided, but the US housing market kept deteriorating and prices of mortgage-backed securities kept decreasing. In 2008, several events that impinged the European labor market took place. First of all, the most vulnerable of the investment banks, Bear Stearns was on the verge of bankruptcy. The bank was heavily invested in MBS and reported huge losses on its balance sheet. Fed decided that the bank was “too interconnected to fail”. As a result, JP Morgan was a single contender to purchase Bear Stearns, but only in case the Fed would buy $30
4 Cezary, Podkul (September 6, 2018). The Regrets of Lewis Ranieri. The Wall Street Journal.
5 Edmund, Conway (March 3, 2008). “Ninja” loans explode on sub-prime frontline. The Telegraph.
6 The Big Short movie. The quote is taken from the episode (1:53:22-1:53:29). Film directed by A. McKay and written by A. McKay and C. Randolph, based on the book “The Big Short: Inside the Doomsday Machine” by M.
Lewis.
7 David, Brett (September 8, 2017). The global financial crisis 10 years on: six charts that tell the story. Schroders.
billions of bad MBS from their portfolio [17]. The second event is when the Fed decides to bail out two largest federal agencies, Fannie Mae and Freddie Mac that initially purchased mortgage- backed securities from banks. Finally, the most crucial event in the entire history of investment banking happened on September 15, 2008 (2008:3), when one of the biggest investment banks, Lehman Brothers, filed for bankruptcy, posting a $3.93 bln. third-quarter loss after $5.6 bln. of writedowns on “toxic” mortgages.8 According to the Daily Telegraph, it was the first loss in 14 years.
Following the unfortunate collapse, banks and big corporations faced huge losses. It became a global event. The consequences of its collapse were reflected on MMFs sector, other investment banks (Merrill Lynch was sold to Bank of America, and other two investment banks transformed themselves into standard banks) and on one of the largest insurance companies, AIG [23]. These are just the consequences on the US. After the Lehman Brothers’ collapse the financial crisis became massive in its size. Before the collapse, the economies of the rest of the world were more or less untouched. Since the paper focuses on the eurozone, it is more relevant to describe the consequences on the euro area countries. First of all, one of biggest banks in Belgium, Fortis bank filed for bankruptcy, later was bailed out by Benelux governments. Another Belgian bank that suffered from the collapse was Dexia, which relied heavily on short-term financing. On September 29, 2008 it was recapitalized by French and Belgian government. The capital injection was not enough, and the Belgian bank obtained funds through ELA. On October 9, it was broken up into multiple pieces. The next bank to fail was German bank, Hypo Re which was later rescued by Bundesbank which provided immediate funding. Ireland however responded differently to the collapse by issuing guarantees of all Irish big banks until specific date. During the crisis, European Central Bank was providing euro liquidity to non-eurozone banks through central bank swaps. [17]
Notwithstanding the quick reaction of ECB, nobody at that time anticipated the magnitude of the crisis:
“However, judged in relation to the size of global financial markets, prospective subprime losses were clearly not large enough on their own to account for the magnitude of the crisis” – Ben S.
Bernanke.9
8 The Daily Telegraph (June 10, 2008). Lehman unveils first loss since going public.
9 Statement by Ben Bernanke before the Financial Crisis Inquiry Commission Washington, D.C. (September 02,2010).
Causes of the Recent Financial and Economic Crisis.
The most important question is: what was the cause of the financial crisis? Unequivocally, many famous and infamous economies suggested a lot of theories on what and how. However, there is not just a single factor but multiple factors explaining it and they can just guess because it is very hard to find the exact roots of the crisis. First of all, banks let uncreditworthy borrowers take money virtually for “free” because they were trying to sell these “un-risky” mortgages to investors. And in turn, investors were purchasing these assets because of their traders’ interest in buying risk assets with high credit rating due to its high return. The risk managers of those investors could not hinder their purchases because it basically brought them huge short-term profits as well.
On the other side, managers of banks could not stop it either because of the herd behavior at the stock market. And obviously there is no reason for stockholders to prevent such trades because it was profiting their bank accounts [17].
While the financial crisis hit the eurozone in 2008, it still lasted throughout 2009 and for many more years for several countries. The crisis actually showed how weak the institutional structure of the eurozone appeared to be. There are multiple kicks-off of the euro area crisis: Greek bailout, Greek announcement of extremely high deficit, or when government bond spreads spiked during the global financial crisis. Some, however, address the beginning of the crisis as when the contagion from Greece started spreading to the PIGS countries [16].
The Greek social-democratic party, PASOK, while in charge of the government passes two fiscal packages in February and March 2009. In April, bond spreads increased from 3.5% to 5%
in just two weeks. On April 23, the former Prime Minister of Greece, George Papandreou asks for bail-out from the EU. Later the conditions were discussed under the supervision of the Troika (EC, ECB, and IMF). On May 2, the bailout was agreed but not on the conditions that are in favor of Greek population, such as cuts in salaries and pensions, reduction in number of public workers, increases in retirement age and etc. Greece had to comply because default would have most likely lead to a drastic austerity [16].
However, why did the eurozone agree on bailing out Greece? Letting such a big economy default could trigger another financial crisis in the euro area. It was technically in the interest of eurozone members. On May 9 European Central Bank announced the SMP which allowed it to buy governments bonds. Two month later, the EFSF was created in order to address the issue of non-existing emergency rescue fund. Later on, there was a creation of EFSM which was supported
by the budget of the EU. After Greek bailout, in 2010 Ireland and Portugal had faced troubles.
Irish biggest bank was nationalized in September 2010 because bank creditors refused to finance it due to the expiration of the bank debt guarantee. There was a big political chaos during 2010 because eurozone members were pushing Ireland to ask for a bailout but for obvious reasons in the past, they tried to resist it. Irish government had to accept it on November 21 because ECB threatened them with taking out ELA from them. On April 6, Portuguese government filed for bailout due to the political crisis as well. Due to the existed myriads of economic problems in the eurozone, the EFSF was replaced with ESM, the new permanent reliable rescue fund [16].
The fact is that Germany and all other eurozone members must have the same monetary policy make it even worse. Due to that, on April 13 and July 13, 2010 notwithstanding the fact that PIGS economies deteriorated, ECB decided to increase interest rates. Those actions led to severe consequences for several countries, in particular periphery. As a consequence, it became more costly for all governments to borrow money and it made the real value of government debts higher.
It seemed like it was unlikely for ECB to adjust its monetary policy to serve governments, such as Portugal, Spain and Greece in crisis. On July 21, 2012 Greek requested a second bail-out plan [16].
During spring and summer 2011 the Greek contagion went on to Spain and Italy.
Throughout 2009 and 2010, Spanish government imposed multiple reforms and fiscal packages.
In Italy the government passed 4-year austerity implementations. In early August 2011 Italian and Spanish bond yields continued to increase and then ECB in order to tackle it buys the bonds again.
During November 2011, the euro area was on the verge of total collapse. Throughout the month, Belgian and French spreads kept increasing and the bond yields for all countries besides Germany were going up. The new technocratic government was established in Italy due to Berlusconi’s resignation. After replacement of Trichet with Draghi as a new president of ECB, the refinancing rate was lowered, which is the opposite of policy that was implemented in spring [16].
However, even after the LTRO program, the eurozone enters recession in the end of 2011, which continues throughout 2012. Unsurprisingly, the recession was strongest in periphery countries, however core countries also faced insignificant economic slowdowns. Spanish housing market began collapsing, fiscal situation became severe and recession intensified. One of the consequences of the crisis is nationalization of Bankia. Multiple Spanish banks were downgraded and lost many deposits. On June 9, 2011 Spain was bailed out [16].
Bond spreads were still skyrocketing. As a consequence, eurozone tried to unify the states by asking for a banking integration, which would include direct recapitalization from ESM. That meant single supervision of all euro area banks by European Central Bank. As a matter of fact, it did help it but with the help of the speech by Mario Draghi who said that ECB would “do whatever it takes to preserve the euro”. On August 2 European Central Bank announces the creation of the program, so-called OMT. According to the program, it allowed ECB to purchase short-term governmental bonds in unlimited quantity [16].
Nevertheless, there is still aftermath after the eurozone crisis. The panic had ended in 2012, however the crisis was still present in countries individually. Cyprus was heavily exposed to Greece and they suffered losses on government bonds. In addition, the biggest power station in Cyprus exploded, which led to the downgrading of the government debt. In October 2011, Cyprus got a loan with relatively high interest from Russia. The government requests bailout from eurozone in July 2012 when multiple banks were downgraded. The bailout agreement is reached on March 16, but the European Parliament rejects it. In despair, Cyprus tried to get another loan from Russia, but ECB threatened them with termination of ELA. During the process, two largest banks were completely restructured and Laiki bank was nationalized by the Bank of Cyprus [16].
The eurozone crisis included three fundamental aspects: banking crisis, economic growth crisis and sovereign debt crisis [16]. All three of them made the situation worse in the euro area.
The main problem with eurozone was that bank rescues were done at national and not at the eurozone level. Throughout the eurozone crisis, economic growth is unequivocally unevenly distributed, meaning that German and Austrian economy experienced strong and quick recovery, whereas it was weak in Italy and almost non-existent in Spain and Greece. Great Recession hit strongly the periphery countries with very large declines in GDP. On the other hand, core countries faced only few quarters of GDP decline. Greek gross domestic product had been contracting for six years…10
Literature review
Low unemployment rate is a desirable target that small and big governments try to achieve in order to satisfy its constituents. However, at the same time they have to heed increasing economic growth. So, getting these two things done at the same time is very hard for politicians
10 According to the Eurostat database on the Greek GDP.
and policymakers. What’s more, it is debatable that economic growth has a statistically significant impact on unemployment. Because there is a term like jobless growth, which is currently experienced by India.11 My assumption is that the labor market in India is dysfunctional so Okun’s law might in fact would not work for the country. The general view however is that countries should increase output to increase employment, but as shown the estimates of Okun’s coefficient for obvious reasons are significantly different for every country and even region. For the long time, people have been aware of an inverse relationship between changes in output and the change in unemployment rates. This relationship is an obvious feature of macroeconomics: economic growth generates new jobs. However, the actual quantifiable research appeared only in 1962 by Okun and it received its name “Okun’s Law”. The results obtained by Okun have become an important point of references for many academics analyzing the relationship between unemployment and economic growth. As any other statistical relationship, it is always subject to revisions in a changing macroeconomy. Ergo, there is a numerous amount of papers, reports and articles devoted to investigating the relationship between employment and economic growth and its number using different methods has proliferated. This thesis documents numerous key contributions made to the field of macroeconomics related to Okun’s law. Some of the studies explaining the relationship between growth and unemployment are present below:
Weber (1995) analyzed the 1948:1-1988:4 period for the post-war US economy by using several methods: static OLS, the cointegrating regression, dynamic OLS and VAR method. As a result, the author found that the 13 out of 18 estimates are less than -0.3 and only two exceed -0.4.
Nevertheless, he found Okun’s coefficient overall significant and hence, valid despite the small values of coefficients. He also stated that estimates of the Okun’s coefficient are sensitive to the method used to estimate cyclical output and unemployment. Prachowny (1993) analyzing the US economy in the 1967:2-1986:2 period and in the 1975:1-1988:1 period showed that Okun’s coefficient is volatile over time (the coefficient ranged from -0.04 to 1.31). Based on the obtained results, he proposed to rename Okun’s Law as “Okun’s theory” because of its instability. Knotek II (2007) tested the data for the United States for the 1948-2007 period with the cointegration analysis and the rolling regression technique to show the stability of the relationship. It has been proved again that it has not been stable over time due to the state of the business cycle. Knotek tested Okun’s law over the very long period (1948-2007) and the short period (1948-1960). Given
11 World Bank (2018). South Asia Economic Focus Spring 2018. Jobless Growth.
the findings, the coefficients are quite similar, and his research supports the original findings.
Finally, Knotek pointed out that Okun’s law is only a rule of thumb and not a law, and especially not a structural feature of the macroeconomy. His paper has been used as a point of reference for the further researches as it is comprehensive and easy to read. Makun and Azu (2015) analyzed the relationship between economic growth and unemployment for the Fiji economy over the 1982- 2012 years using a cointegration analysis. Their research confirmed the evidence of negative correlation between unemployment and economic growth in the long run. Besides the standard regression, the authors included the investment as an independent variable and found that it has a positive impact on economic growth, meaning that low level investment could be a possible reason for low economic growth problem in Fiji. Their conclusion was that investment in Fiji is one of the crucial and essential elements in reduction of unemployment and therefore, economic growth.
Freeman (2001) investigated Okun's law for the US national level and eight American regions. For the study, he used the 1958-1998 quarterly and 1977-1997 annual data applying the bandpass filter.
On contrary to the previous papers, it has been found by him that Okun’s coefficient is constant and that “Okun’s relationship continues to be perhaps the closest thing to a law that macroeconomics has”. Apergis and Rezitis (2003) estimated the Okun's coefficient for certain regional areas in Greece using the 1960-1997 annual data by applying the Hodrick-Prescott and the band-pass filters. As a result, the empirical analysis does not show substantial differences between regions except for two Greek islands. The results reveal that Okun’s relation has structural variation in 1981. After this year, unemployment is less responsive to economic growth in all regional areas. They used data for Greece’s regional economies in order to get evidence on regional differences in the responsiveness of labor markets to output changes. It showed that all the coefficients for all regions are statistically significant. VIllaverde and Maza (2008) analyzed Okun’s Law for Spain and its seventeen regions using data from 1980 to 2004 annually. The results showed an inverse relationship between unemployment and output for most of the Spanish regions and for the whole country. Nevertheless, they found that Okun’s coefficient changes according to the region. Okun’s law holds for most regions and in particular, for the country as a whole. It can be considered as “a rule of thumb” to be used with some caveats. They applied three different detrending techniques: a quadratic trend, the Hodrick-Prescott (HP) filter and the Baxter-King (BK) filter. Dritsaki and Dritsakis (2009) estimated Okun’s coefficient for four Mediterranean countries which are members of the European Union and the eurozone by using the Hodrick-
Prescott filter. In the study conducted for the periods 1961-2002 (annual), the Okun’s coefficient is -0.024 for Italy, -0.017 for Spain, -0.016 for Portugal and -0.007 for Greece and for the EU-15 is -0.12. The Italian coefficient represents the larger coefficient while the Greek coefficient represents the smaller coefficient because Italy is regarded as an industrial country whereas in Greece there is no heavy industry. They determined that coefficients are stable and statistically significant for all countries except Greece. I. Kitov (2011) investigated the relationship between employment and GDP in his study of the biggest developed countries: the US, France, UK, Australia, Canada and Spain. As a result of this study, he has found that low economic growth rates affect high unemployment figures on a significant level. His results suggest the absence of structural unemployment in the aforementioned countries. The data was used between 1958 and 2010. As a conclusion, Okun’s law demonstrates extraordinary predictive power for the biggest developed countries. Herman (2012) in her paper came to the conclusion that there is an impact of economic growth on employment analyzing the relationship in Romania for the period 1990-2010.
She analyzed the effect of economic growth on employment (or the employment intensity of economic growth) through employment elasticity in relation to the economic growth. Herman proved that the relationship between economic growth and employment is strong and positive for the 1990-2010 period. Ruxandra (2015) reviewed the relationship between economic growth and unemployment for Romania as well for the 2007-2013 post-crisis period using the Hodrick- Prescott filter. It has been found that Okun’s Law is valid for the Romanian economy. The coefficient was -0.61. It has been found that in fact there is a negative relationship between GDP and the unemployment rate between 2007 and 2013. Fouquau (2008) could not identify a linear relationship for the twenty OECD countries for the period 1970-2004 representing the law with panel data models. They rejected the hypothesis of a linear relationship between cyclical output and cyclical unemployment and showed that Okun’s law cannot be linear. Harris and Silverstone (2001) analyzed unemployment and output levels relation using the data for the 1978-1999 period testing for cointegration using Engle-Granger and Johansen models and found that there is a cointegration. In this study, 7 OECD countries were examined, and the empirical evidences show that there is no long-run relationship between two variables. Although they emphasized that the results would have been different if they tried to estimate Okun coefficients based on a symmetric approach. The long-run Okun coefficient for most of countries lies between -0.39 and -0.5. (Japan and UK are outliers). Kreishan (2010) analyzed the relationship between economic growth and
unemployment for the Jordanian economy for the 1970-2007 period. As a result, he found that Okun’s law is not valid for Jordan and in fact, the empirical evidences showed that the two variables are unrelated. Also, they found out that the economic policies pertained to demand management would not have a significant effect in reducing unemployment in Jordan. However, economic policies more oriented to structural changes and reform in labor market would be more related considering the Jordanian economy and political structure. Their study showed that the unemployment in Jordan and other Arab countries is structural, not cyclical. In this case, the authors suggest economic growth cannot reduce unemployment. Gordon (1984) stated that the relationship has been used in macroeconomic analysis both because it has been sufficiently stable and reliable in “the past two decades” to deserve being labeled a law and also because it short- circuits the rather complex identity that links output and unemployment. Nevertheless, many other papers do not find that relationship quite so stable (Prachowny, 1993; Knotek II, 2007; …).
The results of research can be divided into two streams: one of them suggests that economic growth and unemployment are not negatively correlated and the other strongly believes in existence of a negative relationship. My paper investigates the relationship between growth and unemployment in the euro area. When the literature is scrutinized, similar studies that scrutinize the relationship in the eurozone in the particular period from 2002 to 2018 cannot be found. It can be said that this study is original and will contribute to the literature. In particular because it analyzes the gender asymmetry in Okun’s law for the eurozone which has never been done before.
Theory of Okun’s Law.
There are two types of approaches (versions) towards how Okun’s law can be interpreted:
the first is very simple, named the first-difference model and the second one is the “gap” model.12 While these two models stood for a long time as a useful empirical estimation, reproductions of both the size and the different interpretation of Okun’s law have been proliferating since later.
The first-difference model merely describes the relationship between changes in the unemployment rate and real GDP growth. By definition, the regression is constructed as follows:
∆Ut = 𝛽1 + 𝛽2 x∆Yt + 𝜀t
Equation 1. First-difference model
12 Okun, Arthur (1962). Potential GNP: Its Measurement and Significance. Yale University.
∆Ut represents the quarterly percentage point change in the unemployment rate and ∆Y measures the quarterly change in real gross domestic product.𝛽1 is an intercept coefficient, showing the effects on unemployment rate when there is no economic growth (mathematically it is when 𝛽2 is equal to zero). 𝛽2 is so-called “Okun’s coefficient” measuring by how much changes in the economic growth rate produce changes in unemployment rate and finally, 𝜀t is an error term. The equation demonstrates the contemporaneous correlation between GDP growth and fluctuations in the unemployment rate, i.e. how GDP growth changes simultaneously with changes in the unemployment rate. When dysfunctional labor markets are not present, the Okun coefficient is negative in most of the cases because according to the general view, a higher rate of output growth is pertained to a decline in the unemployment rate. So, logically rapid economic growth causes a falling unemployment rate, and slow economic growth produces a rising unemployment rate.
The second version relates the level of unemployment rate with the output gap, which is a difference between actual and potential GDP.13 Okun tried to identify the value of the potential output and it is very difficult considering that it is how much the economy would produce under conditions of full employment. The regression takes form as follows:
Ut = 𝛽1 + 𝛽2 x output gap + 𝜀t
Equation 2. Gap version
The equation stems from the original one: Ugapt=𝛼 x Ygapt (Knotek II, 2007). Ut has the same interpretation as before. Output gap is the difference between the observed and potential real GDP growth (Yt-Y*t) and it captures the cyclical level of output. Unemployment gap is the difference between the observed and natural unemployment (Ut-U*t) and it represents the cyclical level of unemployment. And Y*t and U*t are equilibrium values, i.e. potential output and natural unemployment. The equation above postulates the output and unemployment gap which are unobservable indicators. As before, 𝛽2 still represents Okun’s coefficient and it is expected to be negative and by definition, it is associated with the unemployment rate that is consistent with full employment. 𝛽1 is the unemployment rate under the full employment. According to Knotek
13 Knotek II, Edward (2007). How Useful is Okun’s Law? Economic Review, 92(4), 73-103.
(2007), 𝛽2/𝛽1 is the level of economic growth that a country would ideally need on average in order to have stable unemployment. It is the rate of economic growth that policy makers will need to achieve in order to avoid rising unemployment.
Unfortunately, the methods are not without their problems. The primary problem with this model is that both full employment and potential output are not observable variables. Okun had an assumption in his model that full employment occurred when unemployment was 4%. However, to overcome this issue, over time economists came up with methods to estimate these variables by filtering techniques: the Hodrick-Prescott filter14 (Used by Apergis and Resitis, 2003; VIllaverde and Maza, 2008; Dritsaki and Dritsakis,2009; Ruxandra, 2015), and the Baxter-King filter15 (Used by Villaverde and Maza, 2008; Freedman, 2001). One could also estimate these variables by using a production function approach.16 Conventional production function requires that output growth is dependent on a combination of labor, capital, and technology. By taking all of these regressors into account with capital and technology, researchers have a clearer picture of what impacts output.
However, in a common sense it is rather hard to measure inputs like capital and technology.
Chapter 2 - Analysis of Okun’s law
This chapter focuses on the estimation of Okun’s coefficient with panel data analysis for three samples (2002:1-2018:4, 2002:1-2008:3, 2008:4-2018:4) and also for two same samples which have different statistics on unemployment rate (male/female unemployment rates). First, the panel of countries is analyzed in order to obtain results for a huge geographical area. The fixed- effects model was employed obtaining the estimates for a dynamic version and a static version of the first-difference model of Okun’s law for the eurozone. Then, the hypothesis of whether the coefficient is more significant in the post-crisis era is carried out by applying the panel data techniques. The last point in this chapter tests whether economic growth has a stronger effect on a male unemployment stronger than on a female unemployment.
14 Hodrick, Robert and Prescott, Edward (1997). Post-war U.S. Business Cycles: an Empirical Investigation. Journal of Money, Credit and Banking. 29(1).
15 Guay, Alain and St. Amant, Pierre (2005). Do the Hodrick-Prescott and Baxter-King Filters Provide a Good Approximation of Business Cycles? Annales d’Economie et de Statistique. Volume 77. 133-155 pp.
16 Knotek II, Edward (2007). How Useful is Okun’s Law? Economic Review, 92(4), 73-103 pp.
Data - summary statistics
The quantitative analysis is employed using quarterly data on real gross domestic product and unemployment rate investigating the period from 2002:1 to 2018:4 for nineteen eurozone countries. 1292 observations have been employed for the entire period. The mean of the dependent variable is -0.0212. The standard deviation is 0.56, consequently the variance is 0.314. The change in unemployment rate ranges from -2.4 to 3.5 percentage points. The mean of the economic growth is 0.51, which has 1.54 as a standard deviation and the variance is around 2.36. It ranges from - 15.09 and 22.61 percent. Data on the unemployment and the real gross domestic product for the nineteen eurozone members was gleaned from the statistical office Eurostat, which provides quality statistics for the European countries.
The effects of the Great Recession
The unemployment rate has varied a lot across the euro area during the crisis. The graph 1 specifically shows the countries that had a troublesome path during the pre-crisis era. The highest unemployment increase during the pre-crisis era was by 2.1 percentage points for Estonia on the pre-crisis stage in the third quarter of 2008, followed by many countries such as Ireland (1.5 percentage points), Latvia (1.2 percentage points). However, if the year 2008 is excluded from the dataset, Belgian labor market faced challenges in the third quarter of 2004 (increase of 1.2 percentage points), shadowed by Malta in third quarter of 2003 (increase of 1 percentage point), and by Greece in the first quarter of 2004 (increase of 1 percentage point). On other hand, many eurozone countries have experienced positive changes. For instance, the highest unemployment decrease is by 2.4 percentage points for Lithuania in the second quarter of 2002, followed by its neighbor with a decrease of 2 percentage points. It can be seen from the graph 1 that unemployment rates for Lithuania and Latvia fluctuated a lot during the pre-crisis period relatively to other eurozone members. The next biggest negative shift is for Slovakia by just one percentage point in the first and second quarter of 2007, shadowed by Belgium with a decrease of 0.9 percentage points in the third quarter of 2007. Moreover, it can be seen in the dataset that the unemployment rates in ten countries were actually on the way to the improvement and started falling. Three countries have experienced increases in its unemployment rate, such as Ireland (from 4.4% to 7.4%), Luxembourg (from 2.1% to 5.1%), Portugal (from 5.3% to 8.9%) in the pre-crisis period. Finally,
five countries have relatively stable change in its unemployment rate: Belgium (from 7.3% to 7.5%), Spain (from 11.2% to 11.5%), Cyprus (from 3.1% to 3.6%), the Netherlands (from 3.3%
to 3.6%) and Austria (from 4.5% to 4%).
2002:1 2008:3 2008:4
Belgium 7.3 7.5 6.8
Germany 8.2 7.1 7.1
Estonia 11.5 6.3 8.0
Ireland 4.5 7.4 8.6
Greece 10.7 7.6 8.0
Spain 11.2 11.5 13.8
France 8.2 7.5 7.8
Italy 8.5 6.7 6.9
Cyprus 3.1 3.6 3.7
Latvia 12.6 7.7 10.5
Lithuania 15.9 6.4 8.3
Luxembourg 2.1 5.1 5.2
Malta 7.7 6.0 6.1
Netherlands 3.3 3.6 3.6
Austria 4.5 4.0 4.4
Portugal 5.3 8.9 8.9
Romania 8.6 5.5 5.6
Slovakia 19.0 9.1 8.9
Table 1. Unemployment rate (in %) in the eurozone countries for the three quarters: 2002:1 (beginning of the analyzed period), 2008:3 (end of the pre-crisis stage), 2008:4 (beginning of the
post-crisis stage)17
17 Source: Eurostat database.
Regarding the economic growth, not taking the pre-crisis stage, dramatic shift was for Ireland that had faced a 2.24% decrease in its real GDP growth in the third quarter of 2003. It can be illustrated on the graph that first, second and third quarter of 2008 were showing that the numbers were all leading to crisis, showing small negative changes in unemployment and real GDP in 2007 and 2008 when the US and the UK started showing small signs for the beginning of the crisis. Estonia in the first quarter has experienced a decrease of 3.88% in economic growth and Latvian economy in the third quarter has shrunken down by 3.791% in 2008. Concerning the positive changes, Slovenian economy grew by 6.563% in the first quarter of 2002 and Slovakian GDP increased by 6.12% in the fourth quarter of 2007.
Graph 1. The plot of change in the unemployment rate and the real output for the pre-crisis period (2002:1-2008:3) 18
18 Constructed in STATA using Eurostat database for nineteen countries the 2002:1-2008:4 period. Source: Own construction.
Graph 2. The plot of change in the unemployment rate and the real output for the post-crisis period (2008:4-2018:4)19
Once the Great Recession hit the euro area all numbers turned around, and the majority of countries have suffered deterioration in their labor markets. The unemployment rate widely differed across the eurozone. Let’s start with the beginning of the crisis, in particular the fourth quarter of 2008 which is a reflection of the severe consequences of the Great Recession. Only two countries had a decrease in their unemployment rate: Belgium (from 7.5% to 6.8%) and Slovakia (from 9.1% to 8.9%). All other countries faced a huge increase in unemployment rate, as it can be shown in the table above. Regarding the economic growth, obviously all countries have experienced only positive changes in real GDP, but there are some in Baltic area that had a huge increase in its real GDP: Lithuania grew by 60.8% from 2002:1 to 2008:3, Latvia grew by 57.7%
and Estonia grew by 48%. At the beginning of the crisis, all countries had negative numbers in their statistics, in particular Estonia experiencing 9.18% decrease in its real GDP and Slovenia with 3.8% decrease. Even right after this quarter, Estonia again was shaken by 4.77% decrease in real GDP. The Great Recession even hit Germany temporarily, real gross domestic product decreased by 4.48% in the first quarter of 2009. In the same quarter, a lot of countries have had
19 Constructed in STATA using Eurostat database for nineteen countries for the 2008:4-2018:4 period. Source: Own construction
tremendous changes in its economic growth: Malta (-3%), Slovenia (-4.4834%), Slovakia (-9.082), Finland (-6.842), Portugal (-2.3%), the Netherlands (-3.5725%), Lithuania (-13.1084%), Latvia (- 2.7819%), Greece (-4.7662%).
There was a second unanticipated challenge, or better say shock coming up for the eurozone – the sovereign debt crisis. The highest unemployment rate in the third quarter of 2013 is 27.7% in Greece and the second highest is 26.2% in Spain, whereas Austrian unemployment rate was only at 5.4% and even lower rate was for Germany – 5.2%. It says a lot about the unreadiness of the European integration towards the ability to sustain during negative economic shocks.
The first method is a panel regression technique applied to the Equation 1. (the first- difference model), which ignores both short- and long-run dynamics but keeps the strict contemporaneous correlation between the variables, as in the original paper of A. Okun. The advantage of using a panel data is that it has a control over time-invariant factors, especially using a FE model.
The advantage of this model is its objectivity and simplicity. Besides that, there is no debate involved over the best suitable approach to estimate the output gap. Despite the simplicity, the first-difference version of Okun’s law omits the relevant variables and it does not completely explain a lot of variation in the unemployment. Thus, my second model is a dynamic version of Okun’s law, where past changes in the unemployment rate, and both current and past changes in real GDP growth are independent variables. However, it deviates from the original paper and discoveries because it is essentially very distinct from the original first-difference version of Okun’s law since it does not show the contemporaneous correlation. The same method was used by Evans (1989) but with time series data estimating the lagged effects of output growth on the unemployment rate. The advantage of this model is that it does not have strong assumptions concerning the definition and calculations of potential output and full employment. For these models, the quarterly data from the first quarter of 2000 through the fourth quarter of 2018 is used.
Panel dataset includes data on nineteen current eurozone members who use the euro as a legal tender: Austria, Belgium, Cyprus, Latvia, Italy, Greece, Germany, France, Malta, Estonia, Finland, Slovakia, Slovenia, Spain, Malta, Netherlands, Portugal, Lithuania.20 My choice was to include all the countries despite late unification, as in the case of Estonia (2011), Malta (2008), Slovakia
20 European Commission (2019). What is the euro area?
(2009), Slovenia (2007), Lithuania (2015) and Latvia (2014). Having this panel regression technique, I do not try to explain why the euro has an impact on the short-run relationship between real GDP growth and unemployment. It is essential to note that the first chapter investigates the growth-employment relationship for the whole eurozone.
Before conducting a panel regression, it has to be decided what effects the regression should account for, either Random (RE) or Fixed (FE). To choose, a regression with multiple variables (change in the unemployment, change in real GDP and dummy variable for all years) using a Wald test is conducted in order to define which model is more appropriate (Equation 3).
The null hypothesis is that the preferred model has random effects and the alternative is fixed effects model should be picked. In a basic idea, it tests whether the random errors in the regression are correlated with the regressor and the null hypothesis is that they are not.
∆Uit =𝛽1 +𝛽2 x ∆Yit + 𝛽3 x Year2003 + 𝛽4 x Year2004 + 𝛽5 x Year2005 + 𝛽6 x Year2006 + +𝛽7 xYear2007 + 𝛽8 x Year2008 + 𝛽9 x Year2009 + 𝛽10 x Year2010 + 𝛽11 x Year2011 + 𝛽12 x
xYear2012 + 𝛽13 x Year2013 + 𝛽14 x Year2014 + 𝛽15 x Year2015 + 𝛽16 x Year2016 + 𝛽17 x xYear2017 + 𝛽18 x Year2018 + 𝜀it
Equation 3. Fixed-effects (within) regression with dummy variables for the period 2002-2018 21
∆Uit = 0.044- 0.098x ∆Yit + 0.0527 x Year2003 +0.052 x Year2004 -0.105 x Year2005 -0.1765 x Year2006 -0.1037 x Year2007 +0.137 x Year2008 +0.652 x Year2009 +0.055 x Year2010 +
+0.03 x Year2011 +0.172 x Year2012 -0.017 x Year2013 -0.154 x Year2014 - 0.18 x xYear2015-0.178 x Year2016 -0.268 x Year2017 -0.221 x Year2018
Equation 4. Estimated fixed-effects (within) regression with dummy variables for the period 2002-2018 22
Wald test:
H0: The coefficients on all years’ dummies are jointly equal to zero H1: Fixed effects model should be used 23
21 23 Oscar Torres-Reyna (2007). Panel Data Analysis Fixed and Random Effects using Stata (v.4.2). Lecture at Princeton University.
22 Obtained using Eurostat database for nineteen countries. Source: Own calculations.
As a result of Wald test, the F-statistic (16, 1256) is equal to 13.70 and the p-value is zero.
Since it is lower than 0.05, the null hypothesis that the coefficients for all years are jointly equal to zero is rejected, i.e. a fixed effects model is more appropriate for the chosen dataset.
Theoretically it also makes perfect sense because control for fixed effects removes the unobserved time-constant factors within the model that vary across eurozone countries. Therefore, the estimated coefficients of the fixed-effects models cannot be biased because of omitted time- invariant characteristics. Another big advantage of using fixed-effects models is that it is particularly aimed at studying the causes of changes within a country. To double check that the chosen model is appropriate, there is also another way of testing it which is so-called Breusch- Pagan Lagrange multiplier (LM) test. The null hypothesis states that variances across the countries are zero. First, the random effects model is constructed and then, the LM test for random effects is performed. As a result, the p-value is 0.4453 and that means the null hypothesis cannot be rejected at any significance level, i.e. the FE model is more appropriate to be used.
The next thing that must be checked is autocorrelation (serial correlation) within the panel data on unemployment and economic growth. One approach to solve this problem involves the use of the fixed-effects model because due to the time-invariant error the impact of autocorrelation is remedied with fixed effects transformations. After confirming what model should be chosen, the regression looks as follows:
∆Uit = 𝛽1 + 𝛽2 x∆Yit +uit Equation 5. Fixed effects model 24
The coefficients 𝛽1 and 𝛽2 have the same interpretations as before, and uit is the error term.
The upside of panel regressing change in unemployment on economic growth is that variable 𝑢it captures the unobserved time invariant factors within the model, such as labor market conditions, cultures, and etc. which do not change over time. If these characteristics were not controlled for, there could be a potential issue with spurious regressions. Also, the advantage of using panel data and a fixed-effects model is that it shows the fraction of variance due to the error term, it is known as intraclass correlation. In this case the coefficient is 0.0146, which means that 1.46% of the variance is due to differences across panels.
24 25 Obtained using Eurostat database for nineteen countries. Source: Own calculations.
∆Uit = 0.05249 - 0.144 x ∆Yit
Equation 6. Estimated fixed effects model.25
Regressing unemployment on real GDP growth using 1292 observations, Okun’s coefficient for the eurozone in the post-crisis era is -0.144 (Equation 6). The R-squared of the value 0.1559 shows that 15.59% of variance of the unemployment rate is explained by the real economic growth for the period 2002:1-2018:4. The low value of R-squared is associated with the high volatility in the dataset due to the economic shock. The null hypothesis that the coefficient on the real GDP growth is not significant is rejected at 5% significance level due to the very low p-value that is lower than 0.05. As expected, the real GDP growth has a significant influence on the change in the unemployment rate, in fact, at any significance level. And it is negative, so as economic growth increases unemployment rate falls. The research shows that on average the eurozone would require the level of economic growth rates to be at 2.74% in order to maintain a stable unemployment rate. The found coefficient shows how fast the eurozone’s economy would need to grow to maintain a given level of unemployment, it is so-called “employment threshold”26. It essentially points out the rate of economic growth regulators would need to accomplish in order to avoid rising unemployment. The big con of using static OLS instead of VAR analysis is that it is a strictly contemporaneous relationship.
Regarding the dynamic model, the panel analysis technique is applied again. In order to have more relevant variables which were omitted in the first model, both current and past GDP growth and the past changes in the unemployment rate are included in the panel regression as the independent variables. The number of lags of the independent variable is decided upon by AIC.
Then, Levin-Lin-Chu test is run, it performs an Augmented Dickey-Fuller regression for each panel separately. Levin, Lin, and Chu recommended to select ten lags to estimate the long-run variance of the series. The constructed model needs to be expressed in terms of current and past values of the independent variable, in which the error term includes the effect of omitted determinants of the change in unemployment. The Levin-Lin-Chu bias adjusted t-statistic is -
26 Michael Christl, Monika Koppl-Turyna, Denes Kucsera (2017). Okun’s Law in Austria. DANUBE: Law and Economic Review, 8(2), 97-110
5.6498 and the p-value is zero which is significant at all significance levels. Thus, the null hypothesis of unit roots is rejected, meaning that the panel series are stationary, as it was hypothesized. The appropriate number of lags that should be used in the ADL regression is two according to AIC. In order to handle the lagged reaction of output on the labor market, the lags variables are included in the model. In addition, Knotek II (2007) pointed out that the dynamic first-difference version is actually considered to be the preferred model. The estimation is done using the data from 2002:1 through 2018:4. Using 19 panels Okun’s coefficient is obtained in the table 2:
Regressions Okun’s coefficient R-squared
ADL (2,1) -0.148 0.4263
ADL (2,2) -0.176 0.4295
ADL (2,3) -0.178 0.4342
Table 2. The dynamic version of Okun’s law - autoregressive distributed lag models with different number of lags of independent variable. Estimates of Okun’s coefficients for the quarterly 2002-2018
period27
In this dynamic first-difference version of Okun’s law, the dynamic effect of real economic growth on unemployment in ADL (2,2) is -0.176. This coefficient describes the effect of a change in the economic growth on current and future values of unemployment rate. It is of particular interest to investigate the effect of change in output on unemployment, which depends not only on the current values of GDP growth but also on the past and expectations of future change in real GDP. The autoregressive distributed lag model captures lingering effects of change in economic growth on unemployment, basically showing the dynamic causal effect, which is interpreted as Okun’s coefficient. The long-run cumulative dynamic multiplier is the sum of all individual dynamic multipliers, so 𝛽1 + 𝛽2 +…+ 𝛽n.28 As a matter of fact, virtually all the coefficients on lagged values of real economic growth and unemployment are statistically significant at any significance level because p-values are very low and close to zero. There is an exception for the
27 Obtained using Eurostat database for nineteen countries. Source: Own calculations.
28 Segura, Anna (November 2018). Econometrics II: Estimation of Dynamic Causal Effects. Lecture at the University of Pompeu Fabra.
ADL (2,3), where the third lag of independent variable is insignificant (the p-value is 0.6 and very low t-statistic -0.52).
The only thing left to check for is multicollinearity. It can be revealed by regressing the lags of independent variables on each other and it shows that there is no multicollinearity problem.
Overall, the effect of a unit change in the real GDP growth on the change in unemployment is negative. The employment threshold in this case for ADL (2,2) is 2.1% in the long-run relation between 2002:1 and 2018:4. It has to be noted that the coefficient has increased by 0.04 compared to the static first-difference model. More importantly, the obtained R squared is 0.4263 and that is a quite substantial increase on the R-squared value of 0.1602 in the static model.
Before & After Crisis
My major second hypothesis tests whether the relationship between the unemployment and real GPD growth is more significant in the aftermath stage in the eurozone than in the pre-crisis stage. Thus, it is relevant to analyze data from the first quarter of 2002 (when the euro was officially introduced as legal tender in twelve countries) to the third quarter of 2008 (the last quarter in which the Lehman Brothers collapse, however the reactions of eurozone banks followed only in October) and then, analyze the consequences after the crisis investigating the period from the fourth quarter of 2008 to the last quarter of 2018. Not all countries have joined the eurozone in 2002 and due to that, analysis of the consequences of unification and its features on the economy is not investigated. Instead, this particular section is focused more on the effects of the Great Recession. In order to see the changes, two panel regressions are conducted and then the significance of Okun’s coefficients between two different periods is compared. The analysis is carried out using two different periods: the pre-crisis era is the period between first quarter of 2002 and third quarter of 2008, and the post-crisis era is the period between last quarter of 2008 and last quarter of 2018. The periods were chosen this way because eurozone has suffered the most in the last quarter of 2008, experiencing an accumulated negative effect of 42.55% on the real GDP.
Pre-Crisis era
Let’s start analyzing the pre-crisis period. The regression should account for either Random (RE) or Fixed (FE) effects. To choose, the regression with yearly dummy variables is conducted
and then, in accordance with Wald test the null hypothesis of random effects (RE) is tested (Equation 4).
∆Uit = -0.0043475- 0.0343762x ∆Yit +0.047519 x Year2003 +0.0447575 x Year2004 -0.123 x xYear2005 -0.2097378 x Year2006 -0.125 x Year2007 +0.1467 x Year2008
Equation 4. Estimated fixed-effects (within) regression with dummy variables for years.29
As a result of Wald test, the F-statistic is equal to 30.81 and the p-value is 0.0007. Since the p-value is lower than 0.05, the null hypothesis that the coefficients for all years are jointly equal to zero, is rejected, i.e. a fixed effects model is more appropriate for this dataset.
Besides Wald test and LM test, there is also another way of checking if the model was correctly chosen. It is so-called “Hausman test”, which tests the null hypothesis that states the random effects model is more suitable. Fundamentally the idea is that it tests whether the errors are not correlated with the regressors. As a result of the test, the p-value is extremely low (0.0008), hence the FE model should be more appropriate. To summarize, the hypothesis is confirmed and checked twice. And as before, the danger of autocorrelation is eliminated by regressing the dataset using the fixed-effects model.30 The first panel regression using 513 observations for the 2002:1- 2008:3 period is constructed as follows:
∆Ut = -0.017 -0.0646 x ∆Yt Equation 7. Estimated fixed effects model31
Okun’s coefficient for the pre-crisis era is -0.0646, which is much lower than the coefficient obtained in the previous estimated model for the full period. Nevertheless, the result is still statistically significant at any significance level because the null hypothesis is rejected due to the low p-value. Real GDP growth has a significant impact on the change in the unemployment rate in the pre-crisis stage. As expected, it is negative, meaning that when real output decreases,
29 Obtained using Eurostat database for nineteen countries. Source: Own calculations.
30 Garavan, Stephen (2017). Okun’s Law: An Empirical Investigation into Eurozone Growth and Unemployment. The Student Economic Review vol. XXXI.
31 Obtained using Eurostat database for nineteen countries. Source: Own calculations.
