Empirical essays onunemployment and businesscycles Erik Hegelund

Erik Hegelund    Em

pirical essays on u

nem

ploymen

t and bu

siness cycles

Doctoral Thesis in Economic History at Stockholm University, Sweden 2020

Department of Economic History andInternational Relations

ISBN  978-91-7797-821-3

Empirical essays on unemployment and businesscyclesErik Hegelund

Academic dissertation for the Degree of Doctor of Philosophy in Economic History atStockholm University to be publicly defended on Friday 28 February 2020 at 13.00 inNordenskiöldsalen, Geovetenskapens hus, Svante Arrhenius väg 12.

AbstractThis dissertation examines business cycles in Sweden, and the patterns in and driving forces of short- and long-termmovements in unemployment in a selection of high-income countries throughout the 20th century. While this has beenstudied numerous times before, this dissertation starts from the point of view that there is no consensus in social scienceon how to understand these phenomena. This study consists of an introductory chapter and four related but self-containedpapers. One contribution of this thesis is the use of temporal disaggregation methods to estimate more detailed time serieson gross domestic product (GDP) and unemployment. New quarterly estimates of GDP are then used, with the help of theBry-Boschan algorithm, to reconstruct the Swedish business cycle in the period 1913–2014. This identifies a number ofnew patterns not visible in the annual data. A second contribution is different analyses of the extent to which unemploymentcan be explained by macroeconomic indicators such as GDP growth, capital formation and productivity. Different methods,such as band spectrum regression and wavelet analysis, are used to capture longer-term effects. Numerous results arepresented that indicate that macroeconomic performance, notably capital formation, can have medium- to long-term effectson unemployment. This is in line with theoretical models on equilibrium unemployment that take account of the possibilityof persistence in the return to long-run equilibrium, or models that comprise more than one unemployment equilibria. Whilethis is not unknown in previous research, it contradicts several highly influential versions of equilibrium unemploymentmodels, as well as a great body of research on the subject. These contributions have several important implications forfuture research. Historical chronologies should take account of the possibility that data of higher or lower frequency maylead to important differences in results. Empirical and theoretical research on labor markets should continue to investigatemore deeply the possibility that unanticipated short-term events can have long-term effects on labor market outcomes.

Keywords: business cycles, Okun’s law, unemployment, capital formation, temporal disaggregation, band spectrumregression, wavelet analysis.

Stockholm 2020http://urn.kb.se/resolve?urn=urn:nbn:se:su:diva-177231

ISBN 978-91-7797-821-3ISBN 978-91-7797-822-0

Department of Economic History and InternationalRelations

Stockholm University, 106 91 Stockholm

EMPIRICAL ESSAYS ON UNEMPLOYMENT AND BUSINESSCYCLES 

Erik Hegelund

Empirical essays onunemployment and businesscycles 

Erik Hegelund

©Erik Hegelund, Stockholm University 2020 ISBN print  978-91-7797-821-3ISBN PDF 978-91-7797-822-0 Printed in Sweden by Universitetsservice US-AB, Stockholm 2020

Introduction

Erik Hegelund

22nd January 2020

Contents

1 Aim and purpose 2

2 Background and earlier research 3

2.1 Growth and labour markets in the 20th and 21st centuries . . . . 32.2 Theoretical background . . . . . . . . . . . . . . . . . . . . . . . 7

2.2.1 What do we know about business cycles? . . . . . . . . . 72.2.2 What do we know about unemployment? . . . . . . . . . 11

3 Data and source criticism 14

3.1 Data overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143.2 Source criticism . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

4 Methods 18

4.1 Estimating series of higher frequency . . . . . . . . . . . . . . . . 194.2 Analysis on one country across time . . . . . . . . . . . . . . . . 194.3 Analysis on many countries across time . . . . . . . . . . . . . . 214.4 Band spectrum regression and wavelet analysis . . . . . . . . . . 214.5 Caveats regarding causality . . . . . . . . . . . . . . . . . . . . . 23

5 Summary of the papers 24

5.1 Similarities and di�erences in the papers . . . . . . . . . . . . . . 27

6 Concluding remarks 28

6.1 General conclusions from the papers . . . . . . . . . . . . . . . . 286.2 Future research . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

6.2.1 Unemployment theory . . . . . . . . . . . . . . . . . . . . 296.2.2 Analysis of unique historical processes could bene�t from

theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306.2.3 The need for more data on national accounts . . . . . . . 316.2.4 The need for more data on labour market institutions . . 31

1

1 Aim and purpose

This dissertation presents four essays on the subject of macroeconomic trendsand �uctuations in economic activity and unemployment. This introductorychapter frames the empirical and theoretical background to these studies, de-scribes the similarities and di�erences among the papers and provides a sum-mary of the key issues and �ndings.

The overarching aim of this thesis is to investigate the patterns in and driv-ing forces of short- and long-term movements in macroeconomic activity. The�rst contribution consists of new estimates of gross domestic product (GDP)and unemployment data for Sweden, and the annual unemployment rates ofnine other countries. The second contribution is to interrogate the relation-ships between aggregate unemployment and economic activity and policy, assuggested by various theories and earlier research.

Two papers focus on Sweden since the early 1900s (papers 1 and 2), whiletwo papers use data on 101 (paper 4) and 352 (paper 3) high-income countries,respectively. The empirical analyses focus on macroeconomic business cycle�uctuations (paper 1), and the correlation between unemployment and economicactivity (papers 2�4).

The aims for the four papers are set out below:

1. The Business Cycle in Historical Perspective: Reconstructing QuarterlyData on Swedish GDP, 1913�2014. The purpose of the �rst paper isto demonstrate how temporal disaggregation can be used to reconstructnew estimates of quarterly GDP for 1913�2014, and to construct a newchronology of the Swedish business cycle using this new data.

2. Testing Okun's law on di�erent frequencies: Reconstructing monthly un-employment and GDP for Sweden 1913�2014. The purpose of the secondpaper is to investigate Okun's law, the correlation between GDP and un-employment, using data on Sweden for the period 1913�2014, across timeand di�erent frequencies, including new estimates of monthly GDP andunemployment.

3. Can capital formation explain why unemployment has increased since the1970s in the OECD countries? The purpose of the third paper is to invest-igate to what extent the development of unemployment in 35 countries, allmembers of the Organisation for Economic Co-operation and Development(OECD), in 1961�2017 may be explained by capital formation.

4. What determines unemployment in the long run? Band spectrum regres-sion on ten countries 1913�2016. The purpose of the fourth paper is to

1Australia, Belgium, Canada, Denmark, Finland, Netherlands, Norway, Sweden, the UKand the USA

2Australia, Austria, Belgium, Bulgaria, Canada, Croatia, Cyprus, Czech Republic, Den-mark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Japan, Latvia,Lithuania, Luxembourg, Malta, Netherlands, New Zealand, Norway, Poland, Portugal, Ro-mania, Slovakia, Slovenia, Spain, Sweden, Switzerland, the UK and the USA.

2

examine the extent to which unemployment in 10 countries in 1913�2016can be explained by macroeconomic conditions, such as GDP, capital form-ation and productivity, and separate the short- to long-run e�ects, usingwavelet analysis and band-spectrum regression models.

Economic activity, production, and employment are possibly among the moststudied phenomena in the social sciences, and the ideas applied by researcherstoday can be traced, albeit in a di�erent institutional context than the moderneconomy, back to at least the 1700s in texts by Smith (2009, originally 1776) andChydenius (1931, orginally 1765), or even the 1300s, such as in Ibn Khaldun'sbook Muqaddimah (Khaldûn and Lawrence, 2015, originally 1377). Despite thislong tradition, and that all the countries studied in this thesis have experienceda long list of events during the period covered, such as the two World Wars andseveral signi�cant economic crises, there is no complete consensus regarding whythe macroeconomy has developed in the way it has. This lack of consensus isno secret among researchers (Acemoglu, 2009; Vroey, 2016).

The disposition of the rest of this introductory chapter is as follows: Section2 provides general background, a theoretical overview, an outline of existingresearch and a chronology of the key facts from the historical period covered.Section 3 describes the data used and section 4 the primary methods. Section5 summarises the content of each paper. Concluding remarks are discussed insection 6, and section 6.2 describes some possibilities for future studies.

2 Background and earlier research

On a general level, this study departs from the observation that there are im-portant short- and long-term �uctuations in growth and labour market outcomesand the determinants of which are still unclear. Famous historical examples in-clude the Great Depression during the 1930s, as well as the long-term increase inunemployment since the 1970s. This section describes earlier research of relev-ance to this thesis. First, a chronology describes the period in Sweden and othercountries from the early 1900s until the 2010s, focusing on the development ofGDP, unemployment and labour market institutions. There follows some the-oretical background and a more general framing of macroeconomic trends andhow �uctuations can be understood.

2.1 Growth and labour markets in the 20th and 21st cen-

turies

The period covered is 1911�2017 (papers 1, 2, and 4). Among the countriescovered in this thesis, Sweden is included in all the papers and is the primaryfocus of papers 1 and 2 (cf. the historical chronology in paper 1).

Since the late 1800s, the total monetary value of world production has in-creased at an exponential rate. During this long period of growth, economicdevelopment has gone through both short and long periods of stagnation as

3

well as high growth. These �uctuations have often resulted in dramatic shiftsin conditions and standards of living, in industrial transformation and di�erentkinds of crises in the lives of millions. This dissertation aims to measure thesemore closely and to contribute to a better understanding of these developments.

Figure 1 depicts the development of GDP in four of the countries includedin the papers: Australia, Sweden, the United Kingdom and the United States.The top graph shows log GDP per capita and illustrates the positive trendthroughout the period. The bottom graph shows long-term trends in the growthrate of GDP per capita using the trend component from a Hodrick-Prescott (HP)�lter (Hodrick and Prescott, 1997).3

Throughout the 20th century, a massive increase in economic activity andincome took place in many countries throughout the world, and all countriescovered in this thesis, but with important di�erences in growth rates over time.As illustrated in the bottom graph, growth rates have varied over the decades,at the same time as a somewhat similar pattern is seen in most countries overthe period: low, then higher, then lower again.

The early 1900s was a time of economic hardship in many countries, withrecurring �uctuations, booms and busts. World War 2 was followed by decadesof stability and high growth, until a slowdown during the 1970s and 1980s,followed by decades of even slower growth, which has sparked a renewed debateon �secular stagnation� (Hansen, 1938; Baldwin and Teulings, 2014).

Figure 2 shows the development of the unemployment rate, estimated as theannual percentage of the labour force, from paper 4 (Hegelund and Taalbi, 2019).During this period the aggregate unemployment rate in these countries �uctu-ated over both shorter and longer periods, with relatively higher unemploymentduring the 1920s and 1930s, lower levels in 1950�1980, and again higher levelsafter 1980.

The early 1900s, were characterised by both World War 1 as well as sharp�uctuations in GDP and its components, such as capital formation and privateconsumption. These events also coincided with a more extensive food crisis innorthern Europe in 1916�1917, as well as social unrest in Sweden and elsewhere(Montgomery, 1954; Heckscher, 1970). This period is also well-known for boomsand busts in private credit and share prices in Sweden (Schön, 2000; Ahnland,2017), as well as other countries (Schularick and Taylor, 2012; Jordà et al.,2013). Around 1920, both GDP and unemployment �uctuated sharply over afew years, a period famous for �uctuations in both �nancial and credit marketsas well as the aggregate price level (Örtengren, 1979; Lönnborg et al., 2011).

The period up until the end of World War 2 was, in general, tumultuousin many countries (e.g. Holmlund, 2013; Mathy, 2018). During the Great De-pression of the 1930s there was a well-known sharp decline in economic activity,production and other components of GDP (Kindleberger et al., 2005; Galbraith,

3The Hodrick-Prescott �lter minimises the sum of squared deviations in the original datafrom the trend of the same data, with a penalty, often denoted λ, for volatility in the trend (cf.Whittaker, 1922). The choice of λ is more or less arbitrary and here set at 400 (cf. Ravn andUhlig, 2002). The HP-�lter is a common method in analytical work, but also highly debatedin the literature (Hamilton, 2017b,a), as discussed in papers 2 and 3.

4

2009), and a sharp increase in unemployment in Sweden (Schön, 2000; Jonung,2009), as well as many of the other high-income countries (Smiley, 1983; Romer,1986) covered here (cf. Zagorsky, 1998).

While annual data indicate relatively mild �uctuations in Sweden at the time(Edvinsson, 2005; Jordà et al., 2016), we argue in paper 1 that our reconstructionof quarterly GDP indicates that Sweden also experienced a double dip in the1930s.

Macroeconomic instability continued during World War 2 in the early 1940s,with large �uctuations in economic activity, production and trade throughoutEurope and North America, as well as high levels of aggregate unemploymentduring the war. For a description of the Swedish experience during these years,see Johansson (1985), Lundberg (1983) and Jonung (1994).

After the war followed a period renowned for economic prosperity and rel-atively stable growth in many countries. From around 1950 to the late 1970s,and in some countries even longer, both Europe and North America experi-enced low unemployment, growing employment rates, as well as high levels ofeconomic growth and capital formation. For Sweden, the postwar period wasrelatively stable up until 1991. Annual data indicate no apparent signs of �uctu-ations during this period (Edvinsson, 2005) but as we discuss in paper 1, therewere several short-term �uctuations in production in the 1970s, visible when wecompare quarterly data, as well as a longer-term shift towards a slowdown ineconomic growth (cf. Örtengren, 1979).

During these decades, labour market institutions also went through fairlysigni�cant shifts. As pointed out in research on the di�erent welfare state models(Esping-Andersen, 1990), many high-income countries chose somewhat di�erentpolitical paths and designs for their institutions, but as we focus on the longer-term development, several similarities are perhaps just as apparent.

Lundh (2010) describes three phases of development in Sweden: During the�rst period, 1890�1930, Sweden was transformed from a mainly agrarian eco-nomy to one dominated by manufacturing industries. During the same period,Sweden went through an urbanization process, as well as a reconstruction ofthe economy. This was also a period of increasing market concentration, and agrowing share of the labour force was hired by larger companies at larger work-places than before. In earlier stages of industrialization, family businesses werestill typical, as well as patriarchal relations, where workers' professional andprivate lives were tightly controlled. As industrialization and the restructuringof the Swedish economy continued, these patriarchal relations were replaced bymore professional, economic ones. During the same period, trade union mem-bership increased, as well as their in�uence over the labour market. Collectiveagreements increased in popularity and became more common in several keyindustries, where collective agreements covered a growing share of the labourforce.

During the second phase, 1930�1975, industrialization continued, paving theway for an increase in standardized mass production. A growing share of (mar-ried) women entered the labour force (cf. Silenstam, 1970) and during the 1930s,the �rst more general unemployment replacement funds were started (cf. Na-

5

tional Board of Health and Welfare, 1955). During this period, organized capitaland labour pushed for greater centralization of wage negotiations, which at thetime was viewed as favourable to both workers and employers. This resulted inpractices in the labour market which Lundh, as well as many other writers, hasde�ned as �the Swedish Model�. The title refers to key aspects of labour mar-ket relations, which Lundh summarizes as high organizational density amongworkers and employers, consensus among both parties at the local as the wellas national levels. Adding to that, wage negotiations were handled by workersand employers without government interference.

Around 1960, manufacturing industries reached the highest share of the la-bour force employed. According to Lundh, this is also when the Swedish modelof labour market relations was at its peak. During the 1970s and 1980s, themodel was being increasingly questioned by both employers and unions, whichpaved the way for new relations in wage formation. Service industries were em-ploying a growing share of the labour force and from the early 1970s, becamethe dominant sector, compared to manufacturing and agriculture. During thisthird phase of the Swedish economy, 1975�2000, wage bargaining became moredecentralized again and favoured more local, individual, �exible agreements.Lundh describes this as a result of, or at least coinciding with, increasing �glob-alization� and computerization of production, which was making work, as wellas life in general, more individualized (cf. Castells, 1996, 1997).

Patterns of greater or lesser similarities in these overall developments arealso visible when comparing Sweden with other high-income countries � suchas the overall trend in the strength of trade unions, as illustrated by growingunion density. Even if levels vary and there are some exceptions, many countriesfollowed a more or less similar long-term pattern, with growing union densityin the �rst half of the 1900s, and stagnation or long-term decline starting in thelast quarter of the century.

For example, in Sweden, union density grew from below 30% in the early1910s to around 80% in the 1980s and 1990s (Kjellberg, 1983), and after thatdeclined to below 70% in the 2010s (cf. Lundh, 2010, ch. 3). In the UK,union density varied around 20�40% in the 1910s to the early 1930s, increasingafter that until the early 1980s, reaching over 50%, and declining thereafter. InAustralia, union density �uctuated around 30�40% in the 1910s to the 1930sand increased to around 50% in the 1950s. A more long-term decline beganin the late 1980s. The USA is the exception in the sense that average densitywas relatively low compared to most other countries throughout most of thetime period, but the long-term pattern is still similar � increasing union densityfrom around 10% in the early 1900s to around 35% in the 1950s, and decliningthereafter (Kjellberg, 1983; Visser, 2006; OECD, 2019).

Unemployment bene�ts followed a somewhat similar pattern, starting at lowlevels in the early 1900s, and increasing up until around the 1980s in Sweden,as well as several other countries, followed by a substantial decrease. As anillustration we can compare the data on average replacement rates comparedto wages using the Social Policy Indicators database (The Swedish Institute forSocial Research, 2015), and its measures of net replacement rate for the �rst 26

6

weeks of unemployment for workers in single households or families (a similarpattern also visible for other measures). In the USA and the UK, the measuresfor single workers went from being below, or around, 30% in the 1930s and1940s, to around 60% in the 1970s. A similar increase from the 1930s until the1970s took place in Australia and Sweden, where top values were around 37%and around 80%, respectively. Replacement rates for familied workers are oftenhigher but follow a similar long-run pattern. By the 1970s and 1980s, a rathersubstantial decrease had taken place in the UK, and a more mild decrease inmany other countries, such as Australia and Sweden. Exceptions include thenet replacement for familied workers in the USA, which increased somewhat inrecent years, see �gure 3.

More historically focused research on labour market institutions in bothSweden and other countries has also highlighted similar developments, as a po-tential connection to both labour market relations and wage formation (Dims-dale et al., 1989; Alexopoulos and Cohen, 2003), and to a more general discussionon long-term growth in economic activity and total income (Eichengreen, 1994;Vartiainen, 1998).

During the 1970s, GDP growth began to slow down, along with a decline incapital formation. At the same time, the unemployment rate began to show apositive trend. In Sweden, unemployment remained relatively low until a sharpincrease in the early 1990s, after which the unemployment rate remained relat-ively higher compared to earlier decades. This period also coincided with thespread of information technology and structural transformation in both Swedenand other countries (Magnusson and Ottosson, 2003; Castells, 2010). The sharpincrease of Swedish unemployment in the early 1990s also coincides with sim-ilar developments in other countries, such as Finland (Fregert and Pehkonen,2008), as well as short-term �uctuations in house prices, private credit andproduction (Ja�ee, 1994; Söderström, 1995; Finansdepartementet, 2001). This�nancial crisis in Sweden has also been compared to historical events with sim-ilar �uctuations in private credit and housing prices (Larsson and Lönnborg,2014; Ahnland, 2017). Since the increase in the aggregate unemployment ratein 1970s, unemployment has remained at a relatively high level in Sweden andmany other high-income countries. As described above, the research literaturecontains an extensive debate on the causal explanations behind this develop-ment, and these discussions are also central to papers 3 and 4.

2.2 Theoretical background

2.2.1 What do we know about business cycles?

Short-term �uctuations are often referred to as business cycles. Lucas (1980)describes how modern business cycle analysis, for the most part, is a productof the early 20th century and the works of John Maynard Keynes (1936, 1976,originally 1930) and Wesley Mitchell. Based on earlier research, Burns andMitchell (1946) formulated what is probably the most authoritative de�nitionof business cycles as

7

Figure 1: GDP per capita in 2011 US dollars PPP

10000

20000

30000

40000

50000

GD

P p

er

capita, lo

g

1920 1940 1960 1980 2000 2020

Australia

Sweden

UK

USA

GDP per capita per year

−2

0

2

4

6

Perc

ent change

1920 1940 1960 1980 2000 2020

Australia

Sweden

UK

USA

Trend generated with smoothing parameter = 400

GDP per capita growth per yearHP−filter trend−component

Source: Jordà et al. (2016)

8

Figure 2: Unemployment, percent of labour force

0

5

10

15

20

Un

em

plo

ym

en

t ra

te

1920 1940 1960 1980 2000 2020

Australia

Sweden

UK

USA

Source: Hegelund and Taalbi (2019) building on di�erent sources, see article fordetails.

9

Figure 3: Net replacement rates for four countries 1930�2010

0

.5

1

0

.5

1

1900 1950 2000 1900 1950 2000

Australia Sweden

Unied State United KingdomWorker single

Worker familied

Pe

rce

nt

of

wa

ge

Source: The Swedish Institute for Social Research (2015)

10

a type of �uctuation found in the aggregate economic activity ofnations that organize their work mainly in business enterprises: acycle consists of expansions occurring at about the same time inmany economic activities, followed by similarly general recessions,contractions, and revivals which merge into the expansion phase ofthe next cycle; this sequence of changes is recurrent but not periodic;in duration business cycles vary from more than one year to tenor twelve year; they are not divisible into shorter cycles of similarcharacter with amplitudes approximating their own. (p. 3)

This de�nition is still highly in�uential. For instance, both the National Bureaufor Economic Research (NBER) and the Centre for Economic Policy Research(CEPR) use similar de�nitions (cf. Hall et al., 2003; Center for Economic PolicyResearch, 2019).

It is worth noting here how many of the commonly used terms, such asrecession, crisis, or depression, lack a clear de�nition. Several possible de�nitionsof these phenomena are available, none of which are immune from criticism,partly since it is a question of how we theoretically understand �uctuations ineconomic activity in the short to long term.

Burns and Mitchell (1946) explain in their in�uential work on the subjectthat �[h]ow 'general' [business cycle movements] are, what types of activity sharein them and what do not, how the consensus di�ers from one cyclical phaseto another, and from one business cycle to the next, can be learned only byempirical observation�. Empirical work on business cycles then often focuses onseparating the cycles from the long-term trend. One method of achieving this isby using a band-pass �lter on the time series of various economic phenomenon,such as GDP, as is discussed in more detail below.

2.2.2 What do we know about unemployment?

Strong �uctuations are also observable in labour market outcomes, such as inthe short- and long-term movements in the percentage of the labour force that isunemployed. Research on this topic also has a long tradition but both empiricaland theoretical results are still highly debated.

As total production increases in an economy, this often coincides with anincrease in the demand for labour. A famous observation regarding the trendsand �uctuations in economic activity and work was presented by Okun (1962) onthe correlation in the US economy between aggregate production (measured asgross national product, GNP) and unemployment, using di�erent models such asthe correlation between changes in the two variables and the correlation betweendeviations from long-term trends. This collection of observed correlations hassince been referred to as �Okun's law�. As discussed in paper 2, di�erent aspectsof these phenomena are still highly debated, such as how stable this relation isacross countries and time (Silverstone and Harris, 2001; Knotek and II, 2007;Cazes et al., 2013), and whether this is primarily a short-term phenomenon,merely another observation of business cycle �uctuations, or a more long-runrelationship (cf. Gallegati et al., 2014, 2015).

11

Di�erent aspects in connection with Okun's law are also highlighted in re-search on how to understand aggregate unemployment. An illustrative exampleis the academic debate regarding why unemployment in the United Kingdom(UK) in the 1920s and 1930s was relatively high compared to earlier and laterdecades. Benjamin and Kochin (1979) argue that this could be explained bygenerous unemployment bene�ts, which can be expected to push up reservationwages and reduce willingness to work. This is put into question by Broadberry(1983) who, in a comparison between the interwar period and the increase inunemployment in the 1970s, stresses that

the major point we wish to make is that the economy in the interwarperiod showed very little evidence of the kind of self-equilibratingforces relied upon by economists within the Walrasian tradition. Wesee no reason to believe that these self-equilibrating forces are anystronger now; indeed, the huge rise in unemployment and its per-sistence since the late 1970s suggests that these forces are as weakas they were before the Second World War. (p. 483)

Instead, Broadberry argues, unemployment during this period must be under-stood from the perspective that demand for labour can shift for reasons otherthan the long-term design of labour force conditions (Keynes, 1936, 1937): �Wetake the pessimistic view that the interwar economy was subject to demandshocks and was unable to return to full employment without government inter-vention� (cf. Sohn, 2013; Bengtsson and Stockhammer, 2018).

The Keynesian theory of e�ective demand as one of the key drivers of aggreg-ate unemployment are described by Diamond (1982) as a search and matchingmodel of the labour market with multiple equilibria, in which a coordinationfailure among actors may result in prolonged sub-optimal outcomes, i.e. thatthe economy get stuck in a �bad� equilibrium with relatively higher unemploy-ment rate. Fregert (2000) argues that the increase in unemployment in Swedenduring the Great Depression in the 1930s, was prolonged by sticky nominalwages as a result of coordination failures among wage setters. Thus, theoreticaldiscussions on multiple equilibria concern both the development of aggregatedemand as well as that of institutions.

During the 1970s, as unemployment increased in the UK, many other high-income countries experienced a similar pattern of increasing aggregate unem-ployment rates. As described in papers 3 and 4, di�erent explanations for thishave been debated since. Research on the subject may be described as pushingeither an �institutional� story or a �shock� story (cf. Stockhammer, 2008), ora combination of both. Many studies on this subject focus on a selection ofhigh-income OECD countries, partly due to the supply of comparable data.

The institutional story often highlights long-term conditions of labour andcapital that are expected to a�ect wage and price formation, such as tradeunions, legislation and taxes (cf. Stockhammer, 2004, 2008). Empirical analysesof these issues have found no clear consensus. Several empirical studies con�rmthe importance of the structure of long-term institutional conditions (Ljungqvistand Sargent, 1998; Elmeskov et al., 1999) or the combination of institutions in

12

the interaction with shocks (Blanchard and Wolfers, 2000; Bassanini and Duval,2009). However, many empirical results on aggregate institutional measureshave been criticized for being sensitive to research design (Baker et al., 2005;Kim, 2011), which also calls into question the political recommendations basedon these types of results (Howell and Rehm, 2009; Howell, 2011). For instance,Baccaro and Rei (2007) compare several earlier results and methods, and arguethat di�erent analytical methods applied to commonly used aggregate data donot support a strong correlation between aggregate unemployment and measuresof institutional variables, and therefore any simple story of either regulation orderegulation. Instead, they highlight monetary policies and real interest rates asmore important in explaining unemployment, compared to institutional factors.

The shock story instead highlights the importance of unanticipated short-term events, such as exogenous shifts in macroeconomic conditions not explainedby long-term conditions, that for di�erent reasons may a�ect more long-termdevelopments in unemployment. Di�erent factors have been pointed out as po-tential sources of macroeconomic shocks in this context (Blanchard and Wolfers,2000; Ramey, 2016). One strand of this literature focuses on capital formationas an essential factor behind the development of unemployment in Europe andNorth America since the 1960s. Rowthorn (1995) argues that the increase in un-employment after 1970 may be explained by weak capital accumulation, whichin turn may be explained by high interest rates and low pro�ts (cf. Rowthorn,1999a). The more long-term correlation between unemployment and capitalformation is examined by Herbertsson and Zoega (2002). They �nd a strongcorrelation between unemployment and capital, also when compared with a se-lection of institutional factors (cf. Smith and Zoega, 2009). Stockhammer andKlär (2011) compare capital formation, a selection of commonly used variablesand earlier results from other studies (cf. Bassanini and Duval, 2006), and arguethat capital formation and real interest rates can better explain the developmentof unemployment in many high-income countries, compared to the institutionalstory (cf. Stockhammer et al., 2014).

Furthermore, there is no simple theoretical answer to how to understandthe relationship between economic activity, production and unemployment. Asdescribed and discussed in papers 2, 3 and 4, according to in�uential versions oftheoretical models on equilibrium unemployment, such as dynamic search andmatching models (Pissarides, 2000) or static wage- and price-setting models (La-yard et al., 2005), unanticipated shocks to production may have only tempor-ary impacts on unemployment in the short term. However, we may expect morelong-term e�ects if there is persistence/hysteresis in the process by which unem-ployment returns to equilibrium (Drèze and Bean, 1991; Roed, 1997). One mayalso expect a long-run negative correlation between unemployment and growth,or between unemployment and capital formation, with low enough substitutionof elasticity between production factors. Rowthorn (1999b) illustrates this byshowing how the theoretical framework presented in Layard et al. (1991) witha production function with constant elasticity (often denoted σ) of substitution(CES) between production factors below 1, results in a negative long-run rela-tion between unemployment and capital formation (cf. Rowthorn, 1977, 1995).

13

Sigurdsson (2013) shows that a negative long-run relationship between unem-ployment and capital formation may also result from a two-sector search andmatching model, with capital production and CES production, assuming σ < 1.Christodoulakis and Axioglou (2017) derive a negative relation with an overlap-ping generations model with labour market frictions, for both a Cobb-Douglasproduction function and a CES production function with σ < 1. A large body ofresearch seems to support the assumption that σ < 1 (Rowthorn, 1999b; Klumpet al., 2007; Chirinko, 2008; Semieniuk, 2017).

Shifts in macroeconomic conditions and activity, such as capital formation,may also have more long-term e�ects on unemployment if there are multipleunemployment equilibria (Blanchard and Summers, 1988). Several theoreticalarguments for this are well known in the literature, such as exogenous e�ects inthe search and matching process (Diamond, 1982; Pissarides, 1992) or econom-ies of scale in production, which may result in multiple equilibria in search andmatching models (Weitzman, 1982) and overlapping generation models (Cazza-villan and Pintus, 2004; Christodoulakis and Axioglou, 2017), as well as in wageand price-setting models (Manning, 1990, 1992).

Just as the ideas in research on production, employment and business cycleshave a long tradition, so have di�erent theories on unemployment. The equilib-rium unemployment theories presented today can often be traced back to thetheory of a non-accelerating in�ation rate of unemployment (NAIRU) or a nat-ural rate of unemployment (NRU) (Phelps, 1967; Friedman, 1968). But theseideas had a long tradition before that (cf. the introductory model in Bagge,1931). Models with multiple equilibria are often traced back to the theory inKeynes (1937), that e�ective aggregate demand is the primary determinant ofemployment. However, similar arguments were being discussed long before that(Tieben, 2009).

Papers 3 and 4 revisit these debates. We see several motivations for addingmore studies on this subject. Both of the papers include data on capital forma-tion. Earlier studies on the development of unemployment in the OECD coun-tries in this period do in many cases not include any analysis of capital forma-tion. Those studies that do discuss capital formation seldom include many morevariables besides that one (cf. literature review in Stockhammer and Klär, 2011).Paper 3 addresses this by including a selection of institutional variables. Thesevariables have been highlighted as important by several earlier publications onthe subject. In addition, many cross-country panel studies on the subject fo-cus on the period from 1960 due to the availability of data. Paper 4 presentsextended time series on unemployment, which allows a richer analysis.

3 Data and source criticism

3.1 Data overview

Table 1 describes the primary datasets used. The overall research design andchoice of data in the papers are highly in�uenced by earlier research on each

14

topic.While all of the data used in this thesis are compiled and prepared by others,

we have made several adjustments and preparations, such as collecting tables inolder publications, linking newer and older series, and interpolating some miss-ing values. In papers 2 and 4 we estimate a new time series on unemploymentfor +100 years in Sweden and nine other countries. The method is describedin greater detail in these papers. The �rst step is the recollection of earlier ob-servations sampled from di�erent sources, such as Galenson and Zellner (1957),Maddison (1959, 1964), among others (see paper 4). The data taken from Ga-lenson and Zellner (1957) is on unemployment among members of trade unions.Since unemployment were higher among members of trade unions than for thewhole labour force, we adjust this data downwards. To get a consistent timeseries for 1913�2016 we connect the changes in the old series to newer data fromthe OECD database on unemployment, where the commonly used data meas-ures unemployment among 15�74 years old who are unemployed, available forwork and have taken active steps to search for an open position in the last fourweeks. This gives us an estimate of the total annual average unemployment ratein percentage of the labour force for the whole period.

Paper 2 uses this annual estimate of unemployment to reconstruct monthlyunemployment for Sweden, through temporal disaggregation (Chow and Lin,1971). To disaggregate the annual series into monthly data, we use monthlyobservations on unemployment as a percentage of members of trade unionsand unemployment insurance funds. This data is well known (Molinder, 2013;Ericsson and Molinder, 2018) and described in more detail in the paper. We thenconnect this monthly estimate to newer observations on monthly unemploymentas a percentage of the labour force, taken from Statistics Sweden, which resultsin a more extended time series.

In papers 2, 3 and 4 we analysed correlations between unemployment anddi�erent explanatory variables. The dependent variable in paper 2 is the ag-gregate unemployment rate, as described above. Paper 3 applies the HP �lterto separate short-run �uctuations and the long-term trend in unemploymentand use the trend component as the primary dependent variable in the ana-lysis. In in�uential theoretical models of equilibrium unemployment (Pissarides,2000; Layard et al., 2005) there is a unique long-run equilibrium determined byexogenous long-term institutional conditions. Actual unemployment gravitatestowards this equilibrium. According to theory, the long-term trend in observedunemployment should therefore be a good proxy for equilibrium unemployment(cf. Ball and Mankiw, 2002; Flaig and Rottmann, 2013).

3.2 Source criticism

This section discusses the most crucial aspects of source criticism relevant tothis thesis. The primary source-critical question is whether the sources are valid,reliable and relevant for the speci�c purpose of these studies. When analysingseveral countries in the same study, it is essential to choose between nationalsources and sources covering many countries at the same time. Since two of the

15

Table 1: Main data sources used in this thesisData Comment

AMECO: Annual Macro-economicdatabase of the EuropeanCommission's Directorate General forEconomic and Financial A�airshttps:

//ec.europa.eu/economy_finance/

ameco/user/serie/SelectSerie.cfm

Aggregate data onmacroeconomic variables suchas capital, unemployment,among others. Papers 3 and 4.

Bergeaud et al. (2016) Aggregate data on total factorproductivity. Paper 4.

The CEP-OECD Institutions Data Set(Nickell, 2006)http://eprints.lse.ac.uk/19789/

Aggregate data on taxes. Paper3.

Edvinsson et al. (2014) Aggregate GDP. Papers 1 and 2.Groote et al. (1996) Aggregate data on gross �xed

capital formation. Paper 4.ICTWSS v.6: Database onInstitutional Characteristics of TradeUnions, Wage Setting, StateIntervention and Social Pacts, fromAmsterdam Institute for AdvancedLabour Studies (AIAS) (Visser, 2019)

Aggregate data on factorsrelevant to the labour market,such as trade unions. Papers 3and 4.

National Board of Trade (1913, 1954);Statistics Sweden (1963, 1968)

Data on import and exports.Paper 2.

Macrohistory Database (Jordà et al.,2016) http://www.macrohistory.net/data/

Aggregate data on variousmacroeconomic factors such asGDP. Paper 4.

OECD Statisticshttps://stats.oecd.org/

Aggregate data onmacroeconomic variables.Papers 3 and 4.

Sjölund and Wiklander (2003) Indices of industrial production.Papers 1 and 2.

Statistics Swedenhttps://www.scb.se/

Aggregate data on di�erentvariables. Papers 1 and 2.

Vliet et al. 2012 Unemployment replacementrates. Paper 3.

Waldenström (2014) Consumer price index. Paper 2.

16

papers here focus on Sweden, this section also discusses the choice of country,and the quality of Swedish data. Since three papers present new data, it is alsoessential to consider what we can learn, and what we cannot be learned fromthese. Finally, this section highlights the source-critical aspects of the methodswe use in this thesis.

When analysing data that covers several decades and large geographicalareas, such as the 35 countries in paper 3, the data must be comparable acrossboth time and countries. For instance, to be able to make calculations on cross-country variations and correlation, both the dependent variable, the aggregateunemployment rate and the explanatory variables, such as capital formation,must be comparable (cf. Mitchell, 2003). The article uses data on unemploymentfrom the AMECO database, which comprises data from the Eurostat database,which in turn gets data from national statistical agencies, collected according toguidelines from the International Labour Organisation. By linking new and oldseries, AMECO presents time series over several decades, which are comparableacross countries (AMECO, 2019). Both of these, as well as the other databasesused, have also been used in earlier studies.

National sources may be of greater reliability for studies of an individualcountry following the source-critical principle that primary sources are more re-liable than secondary sources, secondary sources are more reliable than tertiarysources, and so on (Ågren, 2017). But the data creation may be based on some-what di�erent methods, making comparisons over time and between countriesproblematic. For instance, paper 4 uses data on GDP taken from Jordà et al.(2016). Comparisons across countries and over time are facilitated by use oftime series from the same source.

If , for instance, we were to use data from each country's statistical o�ce, oneproblem that might occur is that these revise their time series, applying newde�nitions and utilizing additional sources. These updates are then appliedretroactively. Papers 1 and 2 make use of Swedish historical national accounts(Edvinsson et al., 2014), which have been revised extensively in recent years.However, even if revisions of national accounts by statistical o�ces alter totallevels, they seldom involve signi�cant changes in aggregate �uctuations. Forexample, the main focus in paper 2 is on developments over time, rate of changeand relative di�erences, but not absolute levels.

Data reconstructed by economic historians do not always conform to mod-ern de�nitions. Although there are many uncertainties, especially for the periodbefore 1950, Swedish data are among the most reliable in the world, especiallyfor the First and Second World War given that Sweden remained at peace. Inaddition, for Sweden, di�erent series on national accounts exists (Schön andKrantz, 2015), which makes the overall picture less uncertain. The di�erencebetween these series is not essential to the studies in this thesis (cf. discussionin Edvinsson et al., 2014). Some of the series used here come from statistical of-�ces with a long history of data collection. One advantage in these cases is thatthe data generating process may have gone through several improvements alongthe way, which have also been documented. One such example is the data usedin paper 2 on the monetary values of the international trade of goods (National

17

Board of Trade, 1913, 1954; Statistics Sweden, 1963, 1968). Statistics Sweden(1972) describes how the collection of Swedish trade data went through sev-eral improvements during the late 1800s and early 1900s, and how the primarymaterial for this data was adapted to international standards, for instance, onthe use of values of imports at the border (Cost Insurance Freight, CIF) andvalues of exports at the border (Free On Board, FOB) (cf. SOU 2002:118, 2002,bilaga 3). Swedish Customs Services and the General Duty Board collected theprimary material for these data at duty o�ces, post o�ces or railways stations.Schön (2015) describes how historical foreign trade data is one of the sections ofthe Swedish national accounts that is best documented (cf. Lindahl et al., 1937;Johansson, 1967; Edvinsson, 2005).

It is also crucial to be cautious about the extent to which our results can begeneralized. Papers 1 and 2 present a reconstruction of quarterly and monthlydata on GDP and unemployment, applying standard methods of temporal dis-aggregation. While we believe that these estimates o�er a good approximationof the values we are seeking and are useful for these speci�c studies, it is worthhighlighting that these new data series are mere estimates. Other data andmethods might result in signi�cant di�erences. Papers 3 and 4 in this thesismake use of data covering both Sweden and a selection of other high-incomecountries. We have chosen periods and countries according to the availabilityof data. Long time series, comparable across countries and time, and suitablefor statistical analysis are hard to �nd. It is important to note here that theresults do not necessarily hold for other countries and periods.

In papers 1, 2 and 4, we present and make use of data that spans more than100 years. In two of the papers we present analyses of several countries at once.The challenge in these cases is to be careful with comparisons over time. Webelieve that our new estimations of GDP and unemployment are useful for thesestudies. However, it is also important to remember that such periods also spanother signi�cant shifts in the economy. One should be careful of interpretingthe lives of the people represented in those numbers.

Finally, when studying unemployment, it is crucial to understand that this isa measure of the number of people unemployed, divided by the number of peoplein the labour force, which gives a continuous number. However, the calculationof this data is dichotomous: a person is treated as either unemployed or notunemployed. A person is either part of or not part of the labour force. Inreal life, things are often not that simple, and many people are in a grey area,such as discouraged workers, or people in education looking for part-time work.While we believe that the studies presented in this thesis answer importantmacroeconomic questions, it is also possible that a closer comparison of di�erentde�nitions of unemployment would reveal other results.

4 Methods

This section describes the primary methods used in the papers: temporal disag-gregation, regression analysis on one country across time, and panel regression

18

analysis on several countries across time. Section 4.5 also o�ers some commentsregarding causality.

4.1 Estimating series of higher frequency

The two papers on the Swedish economy, papers 1 and 2, present new dataon quarterly GDP (paper 1), and monthly GDP and unemployment (paper 2).The new data is generated by taking known values for annual GDP (Edvinssonet al., 2014), and estimating how we should distribute these over quarters andmonths. To estimate new monthly and quarterly data, we make use of temporaldisaggregation methods. There are several standard methods for temporal dis-aggregation available (for introductions see Eurostat 2013, IMF 2018, and Quilis2018).

The basic idea is that we have a low-frequency data series such as annualGDP, ya, (which is known), and we wish to know GDP per quarter, yq (whichis unknown). Thus, we know that the annual sums of yq are equal to ya.We can then use one (or several) indicator variable(s), pq, to estimate howquarterly GDP data should be distributed over the four quarters in each year.A simple, but rather strict, assumption could be to assume that the growth rateper quarter for y follows the growth rate of p per quarter.

Common methods include Denton and Denton-Cholette (Denton, 1971; Dagumand Cholette, 2006), which use a single indicator variable, and Chow and Lin(1971) and Quilis (2009), which make use of generalized least square (GLS)regression methods (cf. Sax and Steiner, 2013). In paper 1, we present a com-parison of some common methods. Paper 2 focuses on the method presented inChow and Lin (1971) (cf. Eurostat, 2013; IMF, 2018).

In the papers, we use an industrial production index and data on interna-tional trade to perform the temporal disaggregation. This method rests on acorrelation between GDP and these variables in the later decades. It is opento question whether such a relationship was present in the earlier years, for in-stance, because of the larger agricultural sector. However, the co-movement atthe annual level indicates that the method is sound (see paper 1).

4.2 Analysis on one country across time

In papers 2, 3 and 4, we discuss connections between aggregate unemploymentand di�erent variables based on earlier research and theory. We estimate correl-ations between unemployment and explanatory variables to test various theoriesof unemployment.

The overall research design in all three papers is highly in�uenced by theextensive literature on the subject (Bassanini and Duval, 2006; Stockhammerand Klär, 2011), but as described in the introduction to this chapter as well as inthe papers we see several good reasons to revisit these debates, not least becauseof inconclusive earlier results (Baker et al., 2005). At the same time, since we areusing a similar method to earlier research, this facilitates comparison of results.The regression methods used in these papers are relatively basic and follow

19

standard textbook recommendations (e.g. Gujarati and Porter, 2008; Baltagi,2013; Hansen, 2017).

The primary analytical setup in these articles estimates the extent to whichunemployment, uit, in country i at time t is a function of various explanatoryvariables in the same country i during the same year t. In the papers, wealso discuss possible connections across time periods. One way to describethese methods is to start with an analysis of one country. For instance, paper2 focuses on di�erent correlations between unemployment u and real GDP yin Sweden. As part of that study, we present an analysis of the correlationbetween a change in the unemployment rate, ∆u (in percentage points), andthe percentage change in GDP, ∆y. So the change in unemployment at time tis a function of the change in GDP at time t. When we use monthly data, thet is a notation for months. When we use quarterly data, t is instead a notationfor each quarter, such as 1995:3, and so on.

The baseline econometric regression model for this is then

∆ut = α1 + α2∆yt + εt (1)

where the αs are estimated in the regression and εt is the error term at timet. This regression model can then be estimated, for instance using the ordinaryleast squares (OLS) method, which will give us the values of α1 and α2. In atwo-dimensional space this calculation, and the equation 1, give us a straightline through the combined values of u and y, where the two axes measures thevalues for u and y.

For this estimate using OLS to be e�cient, the data must satisfy severalassumptions. For example, the model must be linear, and the error term (ε)must have a constant variance across values of yt (homoscedasticity). Since itis a time series, both u and y must be stationary. Stationarity is a central issuefor regression analysis of time series since if two time series follows some trend,such as a long-term increase, a regression will often indicate a correlation thatis of little practical value. These are mere examples, and standard econometrictextbooks usually discuss these and others (cf. Gujarati and Porter, 2008).

In papers 3 and 4, we study several countries over time and use both panelregression models (see below) and estimate regression models on one countryat a time as a comparison of our panel results. The explanatory variable ofprimary interest in paper 3 is capital formation, k. Since we now have severalcountries, we add notation for countries. Thus, unemployment in country i attime t is now a function of capital formation in the same country i in the sametime period t, which means that the baseline model is now:

uit = β1i + β2ikit + εit (2)

where the β1i and β2i are estimated and εit is the error term for country i attime t. Since we estimate the model for each country separately, we also getone value for β1 and β2 per country i.

20

4.3 Analysis on many countries across time

In paper 3 and 4, we make use of an annual cross-country dataset. Makinguse of data over long periods and several countries, i.e. panel data, can beproblematic since it increases the risk that the data are not comparable (seesection 3.2). The advantage is that panel data can be used to test in�uentialtheories on larger samples, reducing the risk of false conclusions.

To take account of the variations in all countries over time, we estimatepanel regression models, where unemployment in country i at time t, uit, is afunction of capital formation in country i at time t, kit, or in equation form:

uit = γ1 + γ2kit + εit (3)

Using standard OLS method we estimate this panel model, which in this caseis called a pooled OLS model. Estimating a pooled OLS model requires that weassume that there are no unique characteristics in each country of importanceto the development of unemployment. In addition, we assume that there are noglobal events at each point in time that a�ect all countries at the same time.Paper 3 uses a pooled OLS model as a comparison for parts of the analysis.

As discussed in the papers and many earlier publications, it is highly un-likely that such assumptions across countries and time are correct. To take thecountry- and time-persistent e�ects into account, we can add dummy variables.The number of dummies is then I − 1 and T − 1, where I and T is the numberof countries and time periods included in the dataset. This method is called�xed e�ects, and gives us the following baseline two-way �xed e�ects model:

uit = γ1 + γ2kit + γ3Fi + γ4Ft + εit (4)

where Fi and Ft are time-persistent and country-persistent e�ects, which arenot otherwise captured by the model. This means that Fi is used to control forfactors that do not change over time but are constant for each country, suchas national cultures. Ft is used to control for characteristics that are commonacross all countries each year, such as global events. Estimations of γ3 and γ4then give us values, one for each dummy variable, for how substantial the time-and country-consistent e�ects are. Often when using such models, we are notespecially interested in the estimated coe�cients of all the dummies, but themethod is instead primarily used to control for such e�ects and to obtain moree�cient values for other coe�cients, such as γ2.

By estimating the model in equation 2 on data for each country, we calculateintercept and slope for each country, based on the correlation in each countryseparately. By estimating the two-way �xed e�ects model in equation 4 wecalculate the correlation between u and k, for all countries combined, withrespect to the variation in the whole dataset together.

4.4 Band spectrum regression and wavelet analysis

In paper 4, we study the development of unemployment and its relation todi�erent variables, such as GDP and capital formation. We use wavelet analysis

21

and band spectrum regression to separate the short- and long-term e�ects ofthe explanatory variables.

Somewhat simpli�ed, standard versions of commonly used versions of the-oretical models for equilibrium unemployment (Pissarides, 2000; Layard et al.,2005) entail that unemployment follows the long-run equilibrium in the longterm, which is determined by long-term conditions a�ecting wage and priceformation such as labour market institutions, trade unions, taxes, legislationand the conditions for competition in service and product markets (cf. Rogersonet al., 2005). In the short run, unemployment is determined both by equilib-rium unemployment and short-term events, such as unanticipated shocks thata�ect the demand for and supply of labour. If there are multiple unemploymentequilibria, the movement between these equilibria may be a�ected by short-termevents, such as unanticipated shocks or psychological shifts, pushing unemploy-ment between equilibria. Farmer (2010) argues that such psychological shiftsin a theoretical model may be described as a fundamental long-run factor (cf.Farmer, 2012).

Empirical analysis of unemployment often focuses on the relationship betweenexplanatory variables and the long-term development of unemployment, withthe aim of capturing equilibrium unemployment. However, there is no con-sensus on how to do this optimally (see papers 3 and 4).

One method to separate the short- and long-term e�ects of an explanatoryvariable is band spectrum regression, which allows for a separation of the rela-tionship between variables over di�erent horizons (Engle, 1974). For instance,say we wish to test the e�ect of an explanatory variable, x, on unemployment,u, and wish to separate the e�ects that x may have on unemployment in theshort and long run. The regression model would then look like the following:

ut = β1xSRt + β2x

LRt + εt (5)

where ut is unemployment at time t, and xt is divided into short- and long-runcomponents: xSR

t and xLRt . The βs are the coe�cients we wish to estimate, for

instance by OLS regression, and εt is the error term.To be able to estimate equation 5 we must decompose data on x into short-

and long-run components. One such method is the HP �lter, mentioned above.The HP �lter technique decomposes a time series into two components: one�trend� component and one �cyclical� component. The quotation marks areused here to underline that these components are only words � we can observecyclical patterns in the world, but that is not necessarily a proof of some theoryof why an economy moves in cycles.

The estimated HP components are determined by the value chosen for theterm λ. It shapes the smoothness of the trend component. If we set λ highenough, the trend component will become more linear. If we set λ lower, thetrend component will become more like the cyclical component.

However, there is no reason why the economy should be divided into justtwo components, instead of three or more. An alternative to the HP �lter isthen to use wavelet methods. Wavelet transform is a mathematical tool that

22

can be used on a time series to separate it over di�erent time horizons. Unlikethe HP �lter, wavelet transformation can decompose the time series into morethan two components. The cyclical components may change size and shape overtime. The method and its application within social sciences is relatively new,but rests on older ideas such as spectral analysis (for introductions to waveletanalysis see Percival and Walden 2006; Crowley 2007; Andersson 2008, 2016).

In paper 4 we work with discrete time series with one observation per year, soif our time series vector is X it has length N , where N is an integer. If we havea time series that is de�ned over the whole real line, we could, for instance, usea continuous transform. Using a wavelet transform, W, to transform X, we canwrite the transformation as W = WX, where W is a column vector of lengthN , containing a set of discrete wavelet coe�cients and a scaling coe�cient.

There are di�erent functions that de�ne wavelets, and thereby the waveletcoe�cients in the transform matrix (for a more detailed description, see paper4). In the paper we use a wavelet transform that decomposes our time series ytinto 6 components, such that

yt = D1t +D2t +D3t +D4t +D5t + St (6)

where St is a smooth trend. The sum of the components on the right-hand sideof equation 6 is equal to the original time series on the left side of the equation,yt. We now have six components, which we sum into three parts. For the shortterm, we use cycles that cover 2� to 8-year �uctuations = D1t + D2t. For themedium term, we use 8� to 32-year �uctuations = D3t +D4t. For the long termwe use 32+ year �uctuations = D5t + St.

As long as y is stationary, the trend component S will also be stationary. Asdiscussed in paper 4, one of our primary variables of interest (capital formation,k) may not be panel stationary. The cyclical components are, still stationaryhowever, so by excluding the trend component, S, in the long-run component forcapital formation, kLR, we get a panel stationary variable. So for our primaryanalysis, we calculate kLR by using D5t, which contains 32� to 64-year �uctu-ations. The short-, medium- and long-term components can then be used toanalyse how �uctuations in capital formation and other explanatory variablesa�ect unemployment over di�erent time horizons.

4.5 Caveats regarding causality

In papers 2, 3 and 4, we present analyses of correlations between unemploy-ment and other macroeconomic variables and discuss possible interpretations ofthese results. However, it is hard to say whether there is an actual causal rela-tionship between variables. There is no easy solution to this problem. We usesome simple methods to at least approach a discussion on causality. There arealso several other methods for discussing correlations and causal interpretationsavailable in the literature that are not used in this thesis.

One common method is to estimate correlations between variables on laggedvalues. Granger (1969) argues that such regressions would then suggest to what

23

extent a time series may be able to predict another, and decide �the directionof causality�. Even if this is not a test of causation in the literal meaning of theword, this method is often used as a basis for such discussions on time series,so-called Granger causality.

Another method is to use an exogenous instrument variable, which causesvariation in an endogenous explanatory variable but is exogenous to the depend-ent variable. Truly exogenous instruments are however hard to �nd (Imbens andAngrist, 1994; Angrist and Pischke, 2009).

A related method for dynamic panel models (panel models with laggedvariables of the dependent and explanatory variables) is the one proposed inArellano and Bond (1991) for a generalized system of equations. Here, �rstdi�erences can be used to eliminate the individual e�ects, generating valid in-struments for the explanatory variables. However, all estimators of this kindare limited to di�erent types of datasets, while the Arellano-Bond estimator �tsbest for datasets with many countries and few years (big i, small t in equation4) (cf. Baltagi, 2013).

5 Summary of the papers

This section summarizes the analyses in each paper and provides a generaloverview of the results.

Paper 1: �The Business Cycle in Historical Perspective: Reconstruct-

ing Quarterly Data on Swedish GDP, 1913�2014�4 In the �rst paper, westudy the Swedish business cycle in the period 1913�2014. The paper presentsan estimation of quarterly GDP for Sweden by disaggregating annual GDP usinga monthly series on industrial production from Sjölund and Wiklander (2003)and Statistics Sweden, based on common methods for temporal disaggregation(cf. Sax and Steiner, 2013).

While several long historical annual time series have been presented in theexisting literature, more high-frequency data, such as quarterly GDP, is lesscommon. The new quarterly series on GDP is used to present a new chronologyof recessions and expansions over the business cycle in Sweden in the period1913�2014 and compare this to earlier research.

To identify recessions and expansions, peaks and troughs, we apply the Bry-Boschan (Bry and Boschan, 1971) algorithm (cf. Romer, 1994; Watson, 1994).Several �uctuations are identi�ed in the quarterly data that are not visible in theannual series, such as at the outbreak of World War 1 in 1914, a second recessionduring the Great Depression of the 1930s, as well as recessions during the early1960s and 1970s. Most of the recessions identi�ed by this study correspondwell with earlier research using similar methods (Christo�ersen, 2000; Bergman,2011), as well as studies using other methods (Edvinsson, 2005) and on closelyrelated issues, such as economic crises (Jonung, 1994; Holm, 2007).

4Published in The Journal of European Economic History 2018/1: http://www.jeeh.it/

articolo?urn=urn:abi:abi:RIV.JOU:2018;1.33 (Edvinsson and Hegelund, 2018)

24

Paper 2: �Testing Okun's law on di�erent frequencies: Reconstruct-

ing monthly unemployment and GDP for Sweden 1913�2014�5 Thesecond paper analyses Okun's law by comparing monthly data on GDP and un-employment in Sweden in 1913�2014. Monthly GDP and unemployment seriesare estimated using temporal disaggregation, with the method described byChow and Lin (1971) and monthly data on industrial production from Sjölundand Wiklander (2003) (also used in paper 1), data on international trade frompublic records and di�erent measures on unemployment.

While variations in GDP and unemployment are common, relatively fewstudies have been published on monthly data. Using the new estimated data,we test three version of Okun's law, using monthly, quarterly, annual and 5-yearvalues:

1. The di�erence version: change in unemployment as a function of changein GDP, such as

∆ut = α1 + α2∆yt + εt (7)

2. The dynamic version, which is the same as the di�erence version but withthe addition of lagged values of u and y.

3. The gap version: deviations from the long-run unemployment trend as afunction of corresponding deviations from GDP trend, such as

yt − y∗t = β1 + β2 (ut − u∗t ) + εt (8)

While the study �nds a robust negative relationship between unemploymentand GDP growth (di�erence and dynamic version), and also between deviationsfrom trends (the gap version), the correlation coe�cients are much weaker formonthly and quarterly data compared to annual or 5-year values. When com-paring rolling regressions over smaller sub-samples, 13 years at a time (the samelength used in Okun 1962), estimations made using annual data show a statist-ically signi�cant negative correlation over most of the period 1913�2014, whilemost sub-samples of monthly and quarterly data do not result in statisticallysigni�cant coe�cients.

Paper 3: �Can capital formation explain why unemployment has in-

creased since the 1970s in the OECD countries?� The third paper re-visits the empirical debate on why unemployment has increased since the 1970sin many OECD countries. To contribute to this debate, we use data on 35countries for 1961�2017.

While the analysis focuses on capital formation, we also make use of dataon long-term interest rates, terms of trade and a selection of common measuresof labour market institutions, such as unemployment replacement rate and av-erage tax wedge. Research design and choice of variables and data is in�uenced

5Forthcoming in Rodney Edvinsson, Tor Jacobson, and Daniel Waldenström (eds), Histor-ical Monetary and Financial Statistics for Sweden, vol. 3, Sveriges Riksbank.

25

by earlier research (cf. literature reviews in Bassanini and Duval, 2006; Stock-hammer and Klär, 2011). We discuss the robustness of the correlation betweenunemployment and capital formation by exploring variations in the correlationsacross countries and time, as well as interaction e�ects between capital forma-tion and the other explanatory variables.

While several studies on unemployment in the OECD countries have beenpublished in recent years, few of them make use of data on capital formation orcover the period since 2008 (Stockhammer et al., 2014). Those studies that dofocus on capital formation seldom use data on labour market institutions. Whileseveral earlier studies have focused on interaction e�ects between institutions,it is unusual to �nd results on interaction e�ects between capital formation andlabour market institutions and real interest.

The results from this study indicate that three variables were of importancefor the development of unemployment in OECD countries: tax wedge, long-term real interest rates and capital formation. Tax wedge is one of the variablespointed out as important to unemployment by several earlier studies (cf. Scar-petta, 1996; Elmeskov et al., 1999; IMF, 2003; Bassanini and Duval, 2006). Itis interesting to note here the robust results for real interest rates and capitalformation, considering the lack of attention these variables have received in theliterature so far. Neither of them are identi�ed as central in in�uential theoret-ical models on equilibrium unemployment (Pissarides, 2000; Layard et al., 2005).We �nd indications of variations in the correlation between unemployment andcapital formation across time, and some minor variations across countries, butno strong support for interaction e�ects between capital formation and the otherexplanatory variables.

Paper 4: �What determines unemployment in the long run? Band

spectrum regression on ten countries 1913�2016�6 The fourth paper inthis thesis focuses on the long-run impact of macroeconomic conditions on un-employment and makes use of an unbalanced panel dataset on ten countries forthe period 1913�2016. Wavelet analysis and band-spectrum regression is used toseparate short- and long-run e�ects and test the correlations over di�erent timeframes. We construct a new longer time series on annual unemployment and aretherefore able to study a more extended period than several earlier studies. Ourprimary variables of interest are GDP growth, total factor productivity growthand capital formation. While we �nd a robust negative correlation between un-employment and long-term shifts in capital formation, we �nd no reliable resultsfor the other variables. As part of the robustness analysis, we also include aselection of institutional variables. This study is, to the best of our knowledge,the �rst to use cross-country data for this entire time period. Combined withour use of wavelet analysis, we show that there is a long-term relation betweencapital formation and unemployment (cf. Stockhammer and Klär, 2011).

6Published as working paper in Lund Papers in Economic His-tory 2019:203: https://portal.research.lu.se/portal/en/publications/

what-determines-unemployment-in-the-long-run(c32ea6fa-7b88-4a5e-bc3a-feffaf2f687c)

.html (Hegelund and Taalbi, 2019)

26

5.1 Similarities and di�erences in the papers

Papers 1 and 2, illustrate how it is possible to obtain new, signi�cant and moredetailed information on the economy using simple standard methods of temporaldisaggregation to estimate monthly and quarterly data. In paper 1, we comparedi�erent standard methods for doing this but since we only use one indicatorvariable, an index on industrial production, the estimated quarterly data arehighly similar using all methods. Paper 2 instead focuses on the commonlyused method of temporal disaggregation presented in Chow and Lin (1971).

While the methods of temporal disaggregation overlap between these twopapers, paper 1 only uses industrial production to estimate quarterly GDP,while paper 2 uses data on both industrial production and international trade.As part of that work, paper 2 also presents a selection of trade data from publicarchives.

Papers 2, 3 and 4, study the relationship between unemployment and mac-roeconomic activity. All three papers �nd a robust correlation between unem-ployment and macroeconomic activity (such as a negative correlation for realGDP and capital formation). All three papers also �nd variations in the correl-ations across time, and papers 3 and 4 �nd some indication of variations in thecorrelation across countries.

Di�erent variables spanning di�erent countries and periods are used in allthree papers. Paper 2 focuses on Sweden in 1913�2014 but only on real GDP.Paper 3 focuses on 35 high-income countries in the period 1961�2017 and makesuse of data on capital formation, real interest rates, terms of trade and a se-lection of labour market institutions, highlighted by earlier research, such asunion density, unemployment replacement rates and taxes. Paper 4 focuses onten countries in the period 1913�2016 and data on GDP, capital formation andtotal factor productivity. Most of the data used in paper 3 are from o�cialstatistical o�ces. Paper 4 also makes use of data from earlier research by eco-nomic historians and others. The advantage of the underlying data in paper 3 isits higher level of comparability, but the disadvantage is the shorter time spancompared to paper 4, which by itself explains the need for two di�erent papers.

The three papers studying unemployment also di�er in method and researchdesign. Paper 2 focuses on Okun's law, following the research design from Okun(1962), such as testing how a change in unemployment correlates with a changein real GDP, but in this case using monthly data. Paper 3 focuses on annualdata on unemployment and capital formation, testing cross-country time seriespanel models, such as the two-way �xed e�ect models described above. Thedependent variable is the trend component from the HP �lter, used on annualdata on unemployment. This unemployment trend is a proxy for equilibriumunemployment (cf. Ball and Mankiw, 2002). In paper 4, we apply waveletanalysis to decompose our explanatory variables.

27

6 Concluding remarks

6.1 General conclusions from the papers

As discussed in sections 2 and 4.4, it is hard to identify cyclical macroeconomicpatterns given various source-critical considerations of the data and that varioustheoretical or methodological choices could have an impact on the results. Inpaper 1 and 2, we demonstrate how methods of temporal disaggregation can beused to reveal new information about the economic development and businesscycles. Using our new estimation on quarterly GDP, in paper 1 we present anew chronology which we compare with earlier research. While this comparisoncon�rms several earlier results, such as the timing of some recessions (Chris-to�ersen, 2000; Holm, 2007; Bergman, 2011), it also identi�es new recessions(Jonung, 1994; Schön, 1994; Edvinsson, 2005), such as during the 1950s, 1960sand 1970s. Our quarterly data are of higher frequency than those used in sev-eral earlier studies. So while the di�erences in results are natural, they shouldalso be taken into consideration in future research on the economic history ofSweden.

In in�uential versions of theoretical models of unemployment equilibrium(Pissarides, 2000; Layard et al., 2005) shifts in macroeconomic activity, suchas the short- to long-term variations in capital formation (papers 3 and 4) ordevelopments in long-term real interest rates (paper 3), are only considered im-portant as one among several factors through which unanticipated short-termevents may a�ect short- and medium-term unemployment. Deviations froma unique unemployment equilibrium are at most prolonged in the presence ofhysteresis/persistence (Roed, 1997). However, papers 2 and 3 �nd correlationsbetween unemployment and variations in macroeconomic activity (real GDP,capital formation and the long-term real interest rate) that indicate a connec-tion, at least over the medium term. That is, the results seem to indicate arelation that goes beyond short-term business cycle variations. The results inpaper 4 indicate an even more long-term connection between unemploymentand capital formation.

Papers 3 and 4 also estimate correlations for several measures of labourmarket institutions, which are often highlighted as central to the development ofunemployment in empirical (Bassanini and Duval, 2006) and theoretical models(cf. Rogerson et al., 2005). The only �nding to indicate any support for such astory for unemployment in high-income countries in this thesis is the correlationbetween unemployment and the tax wedge (see paper 3). This result is robustand indicates that the tax wedge is vital to our understanding of the developmentof unemployment. However, considering the non-robust results for the othervariables tested in the analyses, the overall picture does not support the highlyin�uential idea that the development of unemployment in these countries inrecent decades, and the overall increase since the 1970s, can be easily explainedby institutional design (cf. Layard et al., 2005).

Policy advice based on the popular theories of unemployment equilibriumand corresponding empirical studies (the institutional story described in section

28

2) often focus on the design of labour market institutions and to some extent onthe conditions for competition in product and service markets (cf. Ljungqvistand Sargent, 1998; Layard et al., 2005). Policy advice based on theoretical mod-els with multiple unemployment equilibria and corresponding empirical studies(the shock story in section 2) instead often focus on the importance of demandmanagement, such as using �scal and monetary policy, or various actions topromote long-term shifts in growth or capital formation (cf. Broadberry, 1983;Stockhammer and Klär, 2011). However, as discussed in papers 2, 3 and 4,several results in this thesis indicate that even if a general pattern, over longperiods and many countries can be observed, that there is a negative correlationbetween unemployment and economic activity such as capital formation, we also�nd variations in this correlation over time and to some extent across countries.While our results are important for our understanding of the economic historystudied here, the results do not support any one-size-�ts-all solution for redu-cing unemployment. In addition, even if we accept that an increase in economicactivity would decrease long-term unemployment in all the countries studied,how this might be achieved through policy measures remains an open question.That is, our results do not show whether the government should turn to ex-pansionary �scal or monetary policy, perhaps combined with budget de�cits, orwhether the most e�ective policy would be to boost private consumption andinvestment by reducing taxes.

6.2 Future research

The four papers in this thesis present new data series on several variables andof di�erent frequencies for Sweden and several other countries, spanning over100 years. The papers also answer a number of empirical questions motivatedby a long tradition of theoretical and empirical debate on issues at the centreof both the historical and the macroeconomic debate regarding macroeconomicactivity and labour market outcomes. This section outlines possible avenues forfuture research.

6.2.1 Unemployment theory

The results in papers 3 and 4 have several potentially signi�cant implicationsfor unemployment theory. A correlation between unemployment and capitalformation, beyond the most short-term time horizon, may �t relatively wellwith models with a unique unemployment equilibrium, if we take into accountthe presence of hysteresis/persistence (Bean, 1994; Roed, 1997). However, wehave to assume very strong hysteresis if we wish to explain results such asthese using the in�uential equilibrium unemployment models (cf. Galí, 2015).This indicates that we need extended versions of these models that better �t thedevelopment of unemployment in OECD countries throughout the 20th century.

What kind of models would be the most useful is beyond the scope of thisthesis, but some examples may su�ce as illustrations. For instance, Galí (2016)proposes a New Keynesian model with insider-outsider labour markets and

29

strong hysteresis. Numerous earlier studies (cf. Broadberry, 1983; Smith andZoega, 2009) have pointed out theoretical models with multiple steady states(Acharya et al., 2018) or equilibria (cf. Manning, 1990; Pissarides, 1992; Farmer,2010) as a general explanation for the correlation between unemployment andeconomic activity.

However, there are numerous di�erent technical arguments for why theremight be multiple equilibria. Research on this issue is in some publicationsframed as a discussion between �heterodox� and �mainstream� economics (Stock-hammer, 2008; Lavoie and Hein, 2015). It is probably more fruitful to describethis as a discussion regarding the assumptions, and therefore the technical de-tails, of theoretical models (cf. Weitzman, 1982; Skott, 1985; Weitzman, 1985;Davidson, 1985a,b). The question can then instead be framed as a scienti�cproblem and tested empirically, rather than as something researchers can as-sume or not assume (cf. Sigurdsson, 2013; Christodoulakis and Axioglou, 2017).

In the papers in this thesis that study the correlation between unemploymentand macroeconomic activity, theoretical models with more than one uniqueequilibrium might be one of several explanations for the empirical results. Anatural opening for future research would thus be to study the characteristics ofthe economies in these aspects, such as the responsiveness of wages and pricesto shocks, which might indicate hysteresis (Bean, 1994), the returns to scale inproduction, which is one of several possible reasons for the existence of multipleequilibria (Manning, 1990), or the behaviour of the search and matching processin labour markets (Diamond, 1982, 2011).

6.2.2 Analysis of unique historical processes could bene�t from the-

ory

The work presented here consists of data collection, data reconstruction andstatistical analysis. In this thesis econometrics are used together with probab-ility theory to test theories. The underlying logic for this is that a selection ofcomparable observations may be used to run probability tests, which may helpto understand whether a correlation is the product of chance. Econometrics issometimes described in opposition to historical methods, which instead recom-mend logic reasoning and narratives (Torstendahl, 1971; Stone, 1979; Gaddis,2004), or as inappropriate for analysis of historical events (Boldizzoni, 2011;McCloskey, 2013). While this thesis makes use of a di�erent perspective on eco-nometrics, there is a possible bridge between the historical approach that seesmost past processes and events as unique and the kind of econometrical analysisspanning long time periods presented in this thesis. For instance, Gaddis (2004,e.g. ch 5) compares the historical method to mathematical systems withoutunique solutions, such as the three-body problem in physics. However, suchtheories are comparable to theoretical models with more than one unique equi-librium and similar theoretical ideas, such as path dependency, i.e. that earlierevents may be important for future outcome and how causal relationships worksat di�erent points in time. For example, Ebbinghaus (2005) discuss whether thedevelopment of institutions can be explained by path dependence and di�erent

30

ways to frame and understand such discussions. Theories and models on mul-tiple equilibria or path dependence has been used not only for labour markets,as discussed in this thesis (Diamond, 1982; Acharya et al., 2018), and economicgrowth (Sachs et al., 2004; Azariadis and Stachurski, 2005) but also in other�elds, such as gender studies (Esping-Andersen et al., 2013) and studies of classcon�ict (Mehrling, 1986). Historical research that focuses on unique processesmight therefore gain from theoretical discussions on multiple equilibria, in thesame way as the essays in this thesis use theories with multiple equilibria tomotivate the empirical analyses.

6.2.3 The need for more data on national accounts

A large share of this thesis explores the development of national accounts inSweden (papers 1 and 2) and also presents new estimates of unemployment innine other countries (paper 4). The choice of Sweden is partly motivated by theknowledge of local statistics, as well as the supply of detailed data on nationalaccounts (cf. Edvinsson, 2005, 2013). However, this thesis does not presentmore detailed high-frequency data on the di�erent components of GDP, suchas consumption, investment, exports and imports, which is a natural subjectfor future studies. Besides uncovering new information about the economy, thiswould also verify the quality of the series presented in this thesis.

The new data in this thesis also open up opportunities for more historicalresearch. For instance, the new data series covers several notorious episodesduring the period, such as the World Wars, the Great Depression and other�nancial crises. A more detailed study of such events using our new data wouldopen up a wide range of issues in each country that our data series covers,as well as comparisons between these countries, such as Sweden and the otherhigh-income countries included in paper 4. For example, with the use of theestimated monthly data on GDP and unemployment it is now possible to studyin more detail the early to mid-20th century, including during events such asthe Great Depression and the World Wars.

6.2.4 The need for more data on labour market institutions

Papers 3 and 4 include an analysis of unemployment, capital formation anda selection of variables which the literature often describes as �labour marketinstitutions�, such as unions and taxes. Earlier studies on the subject in�uencedthe research design and the choice of data. Nevertheless, it is also well-knownin the literature that these measures are coarse measures for all the informationthat might potentially be of interest for unemployment. When using aggregateaverages, we may miss information on variations between geographical regionsand di�erent sectors of the economy. It is also well-known that a lot of this dataprimarily captures legislation, rather than detailed variations in actual labourmarket practices (cf. Eichhorst et al., 2008). In addition, most of the measuresused in this kind of research today seldom stretch further back than 1955 (cf.Allard, 2005; The Swedish Institute for Social Research, 2015).

31

One possibility for future research might then be to collect more data on theinstitutional contexts and factors. Such additions might alter our knowledgeof how the labour market works, as well as economic historical developmentsin the time periods studied in this thesis. This would be of wider interest tohistorians, economic historians and economists more generally.

32

References

Acemoglu, D. (2009). Introduction to Modern Economic Growth. Princeton:Princeton University Press.

Acharya, S., Bengui, J., Dogra, K. andWee, S. L. (2018). Slow recoveriesand unemployment traps: monetary policy in a time of hysteresis.

Ågren, M. (2017). Källkritik. Studentlitteratur AB.

Ahnland, L. (2017). Financialization in Swedish Capitalism : Debt, inequalityand crisis in Sweden, 1900-2013.

Alexopoulos, M. and Cohen, J. (2003). Centralised wage bargaining andstructural change in Sweden. European Review of Economic History, 7 (3),331�363.

Allard, G. (2005). Measuring The Changing Generosity Of UnemploymentBene�ts: Beyond Existing Indicators. Working Papers Economia wp05-18,Instituto de Empresa, Area of Economic Environment.

AMECO (2019). Annual macro-economic database - European Commission.

Andersson, F. N. G. (2008). Wavelet Analysis of Economic Time Series.dissertation, Lund University.

� (2016). Identifying and modelling cycles and long waves in economic timeseries.

Angrist, J. D. and Pischke, J.-S. (2009). Mostly Harmless Econometrics:An Empiricist's Companion. Princeton: Princeton University Press, 1st edn.

Arellano, M. and Bond, S. (1991). Some Tests of Speci�cation for PanelData: Monte Carlo Evidence and an Application to Employment Equations.Review of Economic Studies, 58 (2), 277�297.

Azariadis, C. and Stachurski, J. (2005). Chapter 5 Poverty Traps. InP. Aghion and S. N. Durlauf (eds.), Handbook of Economic Growth, vol. 1,Elsevier, pp. 295�384.

Baccaro, L. and Rei, D. (2007). Institutional determinants of unemploymentin OECD countries: Does the deregulatory view hold water? InternationalOrganization, 61 (03), 527�569.

Bagge, G. (1931). Arbetslöshetsutredningens betänkande. 1, Bilagor, Bd 1, Or-saker till arbetslöshet. Statens o�entliga utredningar, 0375-250X ; 1931:21,Stockholm: Nord. bokh. i distr.

Baker, D., Glyn, A., Howell, D. R. and Schmitt, J. (2005). Labor MarketInstitutions and Unemployment: Assessment of the Cross-Country Evidence.In Fighting Unemployment, Oxford Scholarship Online, pp. 72�118.

33

Baldwin, R. and Teulings, C. (eds.) (2014). Secular Stagnation: Facts,Causes and Cures. CEPR Press.

Ball, L. and Mankiw, N. G. (2002). The NAIRU in theory and practice.Tech. rep., National Bureau of Economic Research.

Baltagi, B. H. (2013). Econometric Analysis of Panel Data. Chichester, WestSussex: Wiley, 5th edn.

Bassanini, A. and Duval, R. (2006). Employment Patterns in OECD Coun-tries: Reassessing the Role of Policies and Institutions. OECD Social, Em-ployment and Migration Working Paper 35, OECD Publishing.

� and � (2009). Unemployment, institutions, and reform complementarities:re-assessing the aggregate evidence for OECD countries. Oxford Review ofEconomic Policy, 25 (1), 40�59.

Bean, C. R. (1994). European Unemployment: A Survey. Journal of EconomicLiterature, 32 (2), 573�619.

Bengtsson, E. and Stockhammer, E. (2018). Wages, income distributionand economic growth in Scandinavia. Lund Papers in Economic History. Gen-eral Issues, (2018:179).

Benjamin, D. K. and Kochin, L. A. (1979). Searching for an Explanationof Unemployment in Interwar Britain. Journal of Political Economy, 87 (3),441�478.

Bergeaud, A., Cette, G. and Lecat, R. (2016). Productivity Trends inAdvanced Countries between 1890 and 2012. Review of Income and Wealth,62 (3), 420�444.

Bergman, U. M. (2011). Tidsbestämning av svensk konjunktur 1970-2010.Tech. rep., Finanspolitiska rådet.

Blanchard, O. and Wolfers, J. (2000). The Role of Shocks and Institu-tions in the Rise of European Unemployment: The Aggregate Evidence. TheEconomic Journal, 110 (462), C1�C33.

Blanchard, O. J. and Summers, L. H. (1988). Beyond the Natural RateHypothesis. The American Economic Review, 78 (2), 182�187.

Boldizzoni, F. (2011). The Poverty of Clio: Resurrecting Economic History.Princeton University Press, 1st edn.

Broadberry, S. N. (1983). Unemployment in Interwar Britain: A Disequilib-rium Approach. Oxford Economic Papers, 35 (3), 463�485.

Bry, G. and Boschan, C. (1971). Cyclical Analysis of Time Series: Selec-ted Procedures and Computer Programs. NBER Books, National Bureau ofEconomic Research, Inc.

34

Burns, A. F. and Mitchell, W. C. (1946). Measuring Business Cycles.NBER Books, National Bureau of Economic Research, Inc.

Castells (1997). The Power of Identity byCastells. Wiley.

Castells, M. (1996). The rise of the network society. Malden, Mass.: BlackwellPublishers, oCLC: 33819751.

� (2010). The rise of the network society. Chichester, West Sussex ;: Wiley-Blackwell.

Cazes, S., Verick, S. and Al Hussami, F. (2013). Why did unemploymentrespond so di�erently to the global �nancial crisis across countries? Insightsfrom Okun's Law. IZA Journal of Labor Policy, 2 (1), 10.

Cazzavillan, G. and Pintus, P. A. (2004). Robustness of multiple equilibriain OLG economies. Review of Economic Dynamics, 7 (2), 456�475.

Center for Economic Policy Research (2019). Business Cycle DatingCommittee Methodology | Centre for Economic Policy Research.

Chirinko, R. S. (2008). σ: The long and short of it. Journal of Macroeconom-ics, 30 (2), 671�686.

Chow, G. C. and Lin, A.-l. (1971). Best Linear Unbiased Interpolation, Dis-tribution, and Extrapolation of Time Series by Related Series. The Review ofEconomics and Statistics, 53 (4), 372�75.

Christodoulakis, N. and Axioglou, C. (2017). Underinvestment and Un-employment: The Double Hazard in the Euro Area. Applied EconomicsQuarterly, 63 (1), 49�80.

Christoffersen, P. (2000). Dating the Turning Points of Nordic BusinessCycles. SSRN Scholarly Paper ID 249613, Social Science Research Network,Rochester, NY.

Chydenius, A. (1931). The national gain,. E. Benn.

Crowley, P. M. (2007). A guide to wavelets for ecnomists. Journal of Eco-nomic Surveys, 21 (2), 207�267.

Dagum, E. B. and Cholette, P. A. (2006). Benchmarking, Temporal Distri-bution, and Reconciliation Methods for Time Series, Lecture Notes in Statist-ics, vol. 186. Springer New York.

Davidson, P. (1985a). Introduction. Journal of Post Keynesian Economics,7 (3), 350�351.

� (1985b). Liquidity and Not Increasing Returns Is the Ultimate Source ofUnemployment Equilibrium. Journal of Post Keynesian Economics, 7 (3),373�384.

35

Denton, F. T. (1971). Adjustment of Monthly or Quarterly Series to An-nual Totals: An Approach Based on Quadratic Minimization. Journal of theAmerican Statistical Association, 66 (333), 99�102.

Diamond, P. A. (1982). Aggregate demand management in search equilibrium.Journal of political Economy, 90 (5), 881�894.

� (2011). Unemployment, vacancies, wages. American Economic Review,101 (4), 1045�72.

Dimsdale, N. H., Nickell, S. J. andHorsewood, N. (1989). Real wages andunemployment in Britain during the 1930s. The Economic Journal, 99 (396),271�292.

Drèze, J. H. and Bean, C. R. (eds.) (1991). Europe's Unemployment Problem.Cambridge, Mass: The MIT Press.

Ebbinghaus, B. (2005). Can Path Dependence Explain Institutional Change?Two Approaches Applied to Welfare State Reform. MPIfG Discussion Paper05/2, Max Planck Institute for the Study of Societies.

Edvinsson, R. (2005). Growth Accumulation Crisis: With New MacroeconomicData for Sweden 1800-2000. Stockholm: Almquiest & Wiksell Intl.

� (2013). New annual estimates of Swedish GDP, 1800�2010. The EconomicHistory Review, 66 (4), 1101�1126.

� and Hegelund, E. (2018). The business cycle in historical perspective:Reconstructing quarterly data on Swedish GDP 1913-2014. The Journal ofEuropean Economic History, p. 28.

�, Jacobson, T. and Waldenström, D. (eds.) (2014). House prices, stockreturns, national accounts and the Riksbank balance sheet, 1620-2012. No.volume II in Historical monetary and �nancial statistics for Sweden, Stock-holm: Ekerlids Förlag Sveriges Riksbank.

Eichengreen, B. (1994). Institutional prerequisites for economic growth:Europe after World War II. European Economic Review, 38 (3-4), 883�890.

Eichhorst, W., Feil, M. and Braun, C. (2008). What have we learned?Assessing labor market institutions and indicators. No. 22/2008 in IAB Dis-cussion Paper: Beiträge zum wissenschaftlichen Dialog aus dem Institut fürArbeitsmarkt- und Berufsforschung, Nürnberg.

Elmeskov, J., Martin, J. P. and Scarpetta, S. (1999). Key Lessons ForLabour Market Reforms: Evidence From OECD Countries' Experience. SSRNScholarly Paper ID 181273, Social Science Research Network, Rochester, NY.

Engle, R. F. (1974). Band Spectrum Regression. International Economic Re-view, 15 (1), 1�11.

36

Ericsson, J. and Molinder, J. (2018). A Workers' Revolution in Sweden?Exploring Economic Growth and Distributional Change with Detailed Dataon Construction Workers' Wages, 1831�1900.

Esping-Andersen, G. (1990). The Three Worlds of Welfare Capitalism. Prin-ceton, N.J: Princeton University Press.

�, Boertien, D., Bonke, J. and Gracia, P. (2013). Couple Specializationin Multiple Equilibria. European Sociological Review, 29 (6), 1280�1294.

Eurostat (2013). Handbook on quarterly national accounts. Eurostat.

Farmer, R. E. (2010). Animal Spirits, Persistent Unemployment and the BeliefFunction. Working Paper 16522, National Bureau of Economic Research.

Farmer, R. E. A. (2012). Con�dence, Crashes and Animal Spirits. The Eco-nomic Journal, 122 (559), 155�172.

Finansdepartementet (2001). Finans- och penningpolitiskt bokslut för 1990-talet : särtryck ur 2001 års ekonomiska vårproposition. Stockholm: Finans-dep., Regeringskansliet.

Flaig, G. and Rottmann, H. (2013). Labour market institutions and unem-ployment: an international panel data analysis. Empirica, 40 (4), 635.

Fregert, K. (2000). The Great Depression in Sweden as a wage coordinationfailure. European Review of Economic History, 4 (3), 341�360.

� and Pehkonen, J. (2008). Causes of structural unemployment in Finlandand Sweden 1990-2004. Tech. Rep. 2008:14, Lund University, Department ofEconomics.

Friedman, M. (1968). The Role of Monetary Policy. American Economic Re-view, 58 (1), 1.

Gaddis, J. L. (2004). The Landscape of History: How Historians Map the Past.Oxford: Oxford University Press, 1st edn.

Galbraith, J. K. (2009). The Great Crash 1929. Boston: Mariner Books, �rstedition edn.

Galenson, W. and Zellner, A. (1957). International Comparison of Unem-ployment Rates. NBER, pp. 439�584.

Galí, J. (2015). Hysteresis and the European unemployment problem revisited.Tech. rep., National Bureau of Economic Research.

Galí, J. (2016). Insider-outsider labor markets, hysteresis and monetarypolicy. Universitat Pompeu Fabra, Department of Economics Working Pa-pers, (1506).

37

Gallegati, M., Gallegati, M., Ramsey, J. B. and Semmler, W. (2014).Does Productivity A�ect Unemployment? A Time-Frequency Analysis forthe US. In M. Gallegati and W. Semmler (eds.), Wavelet Applications inEconomics and Finance, Dynamic Modeling and Econometrics in Economicsand Finance, Cham: Springer International Publishing, pp. 23�46.

�, �, � and � (2015). Productivity and unemployment: a scale-by-scalepanel data analysis for the G7 countries. Studies in Nonlinear Dynamics &Econometrics, 20 (4), 477�493.

Granger, C. W. (1969). Investigating causal relations by econometric mod-els and cross-spectral methods. Econometrica: Journal of the EconometricSociety, pp. 424�438.

Groote, P., Albers, R. and Jong, H. d. (1996). A standardised time seriesof the stock of �xed capital in the Netherlands, 1900-1995. Tech. Rep. 199625,Groningen Growth and Development Centre, University of Groningen.

Gujarati, D. N. and Porter, D. C. (2008). Basic Econometrics. Boston:McGraw-Hill Education, 5th edn.

Hall, R., Feldstein, M., Frankel, J., Gordon, R., Mankiw, N. G.and Zarnowitz, V. (2003). The NBER's Business-Cycle Dating Procedure.Business Cycle Dating Committee, National Bureau of Economic Research.

Hamilton, J. (2017a). Why you should never use the Hodrick-Prescott �lter.

Hamilton, J. D. (2017b). Why You Should Never Use the Hodrick-PrescottFilter. The Review of Economics and Statistics, 100 (5), 831�843.

Hansen, A. H. (1938). Full Recovery or Stagnation? W.W. Norton & Co., 1stedn.

Hansen, B. E. (2017). Econometrics. University of Wisconsin.

Heckscher, E. F. (1970). Industrialismen : den ekonomiska utvecklingensedan 1750 : av Eli F. Heckscher. Stockholm: Raben & Sjögren.

Hegelund, E. and Taalbi, J. (2019). What determines unemployment in thelong run? : Band spectrum regression on ten countries, 1913-2016. LundPapers in Economic History., (2019:203).

Herbertsson, T. T. and Zoega, G. (2002). The Modigliani `puzzle'. Eco-nomics Letters, 76 (3), 437�442.

Hodrick, R. J. and Prescott, E. C. (1997). Postwar U.S. Business Cycles:An Empirical Investigation. Journal of Money, Credit and Banking, 29 (1),1�16.

Holm, L. (2007). A non-stationary perspective on the european and swedishbusiness cycle.

38

Holmlund, B. (2013). Wage and employment determination in volatile times:Sweden 1913�1939. Cliometrica, 7 (2), 131�159.

Howell, D. R. (2011). Institutions, Aggregate Demand and Cross-CountryEmployment Performance: Alternative Theoretical Perspectives and theEvidence. In A Modern Guide to Keynesian Macroeconomics and EconomicPolicies, Edward Elgar Publishing Ltd.

� and Rehm, M. (2009). Unemployment compensation and high Europeanunemployment: a reassessment with new bene�t indicators. Oxford Review ofEconomic Policy, 25 (1), 60�93.

Imbens, G. W. and Angrist, J. D. (1994). Identi�cation and Estimation ofLocal Average Treatment E�ects. Econometrica, 62 (2), 467�475.

IMF (2003). World Economic Outlook, April 2003 � Chapter 3: Growth andInstitutions - chapter3.pdf. Tech. rep., International Monetary Fund.

IMF (2018). Quarterly National Accounts Manual (2017 Edition). InternationalMonetary Fund.

Jaffee, D. M. (1994). The Swedish Real Estate Crisis. eScholarship.

Johansson, M. (1985). Svensk industri 1930-1950 : produktion, produktivitet,sysselsättning. Skrifter utgivna av Ekonomisk-historiska föreningen, 0424-7493 ; 44, Lund: Ekonomisk-historiska fören.

Johansson, Ö. (1967). The gross domestic product of Sweden and its composi-tion 1861-1955. Stockholm economic studies, 0348-3614 ; N.S., 8, Stockholm.

Jonung, L. (1994). 1990-talets ekonomiska kris i historisk belysning: efterskrifttill Erik Lundbergs bok Ekonomiska kriser förr och nu. SNS.

� (2009). Vad säger vår historia om �nanskriser? Ekonomisk Debatt, 37, 73�85.

Jordà, Ò., Schularick, M. and Taylor, A. M. (2013). When credit bitesback. Journal of Money, Credit and Banking, 45 (s2), 3�28.

�, � and Taylor, A. M. (2016). Macro�nancial History and the New Busi-ness Cycle Facts. NBER Macroeconomics Annual 2016, Volume 31, pp. 213�263.

Keynes, J. M. (1936). The general theory of employment, interest and money.New York: Harcourt, Brace, oCLC: 167708.

� (1937). The General Theory of Employment. The Quarterly Journal of Eco-nomics, 51 (2), 209�223.

� (1976). A Treatise on Money. New York: Ams Pr Inc.

39

Khaldûn, I. I. and Lawrence, B. (2015). The Muqaddimah: An Introductionto History - Abridged Edition. Princeton, NJ: Princeton University Press,abridged edition edn.

Kim, J. (2011).Why do Some Studies Show that Generous Unemployment Bene-�ts Increase Unemployment Rates? A Meta-Analysis of Cross-Country Stud-ies. Tech. Rep. 2011:18, Stockholm University, Department of Economics.

Kindleberger, C. P., Aliber, R. and Solow, R. (2005). Manias, Panics,and Crashes: A History of Financial Crises. Hoboken, N.J: Wiley, 5th edn.

Kjellberg, A. (1983). Facklig organisering i tolv länder : [Trade union organ-isation in twelve countries]. Lund: Arkiv.

Klump, R., McAdam, P. and Willman, A. (2007). The long-term sucCESsof the neoclassical growth model. Oxford Review of Economic Policy, 23 (1),94�114.

Knotek, E. S. and II (2007). How useful is Okun's law? Economic Review,(Q IV), 73�103.

Larsson, M. and Lönnborg, M. (2014). Finanskriser i Sverige. Lund: Stu-dentlitteratur.

Lavoie, M. and Hein, E. (2015). Going from a low to a high employment equi-librium. IMK Working Paper 144-2015, IMK at the Hans Boeckler Founda-tion, Macroeconomic Policy Institute.

Layard, R., Nickell, S. and Jackman, R. (1991). Unemployment: Mac-roeconomic Performance and the Labour Market. Oxford, New York: OxfordUniversity Press.

�, � and � (2005). Unemployment: Macroeconomic Performance and theLabour Market. Oxford ; New York: Oxford University Press, 2nd edn.

Lindahl, E., Dahlgren, E. and Kock, K. (1937). Wages, cost of livingand national income in Sweden 1860-1930 Vol. 3 National income of Sweden1861-1930. London :: P.S. King ;.

Ljungqvist, L. and Sargent, T. J. (1998). The European UnemploymentDilemma. Journal of Political Economy, 106 (3), 514�550.

Lönnborg, M., Ögren, A. and Rafferty, M. (2011). Banks and Swedish�nancial crises in the 1920s and 1930s. Business History, 53 (2), 230�248.

Lucas, R. E. (1980). Methods and Problems in Business Cycle Theory. Journalof Money, Credit and Banking, 12 (4), 696�715.

Lundberg, E. (1983). Ekonomiska kriser förr och nu. StudieförbundetNäringsliv och Samhälle.

40

Lundh, C. (2010). Spelets regler : institutioner och lönebildning på den svenskaarbetsmarknaden 1850-2010. Stockholm: SNS förlag.

Maddison, A. (1959). Economic Growth in Western Europe 1870-1957. BancaNazionale Del Lavoro Quarterly Review, XII.

� (1964). Economic growth in the West : comparative experience in Europe andNorth America. A Twentieth Century Fund study, 99-0112347-X, New York:The Twentieth Century Fund.

Magnusson, L. andOttosson, J. (2003). Den tredje industriella revolutionenoch� den nya ekonomin��mellan sken och verklighet. Ute och inne i svensktarbetsliv, p. 57.

Manning, A. (1990). Imperfect Competition, Multiple Equilibria and Unem-ployment Policy. The Economic Journal, 100 (400), 151�162.

� (1992). Multiple equilibria in the British labour market: Some empiricalevidence. European Economic Review, 36 (7), 1333�1365.

Mathy, G. P. (2018). Hysteresis and persistent long-term unemployment: theAmerican Beveridge Curve of the Great Depression and World War II. Clio-metrica, 12 (1), 127�152.

McCloskey, D. N. (2013). The poverty of Boldizzoni: resurrecting the Ger-man historical school. Investigaciones de Historia Económica-Economic His-tory Research, 9 (1), 2�6.

Mehrling, P. G. (1986). A Classical Model of the Class Struggle: A Game-Theoretic Approach. Journal of Political Economy, 94 (6), 1280�1303.

Mitchell, B. R. (2003). International historical statistics : Europe, 1750-2000. New York: Palgrave Macmillan.

Molinder, J. (2013). From Widespread Unemployment to Full Employment -Unemployment, Wages and Productivity on the Path Towards Full Employ-ment, 1935-1948. Master's dissertation, Uppsala university, Uppsala.

Montgomery, A. (1954). Svensk och internationell ekonomi 19131939. Stock-holm: KF:s bokförlag.

National Board of Health and Welfare (1955). Sociala meddelanden.1955: 1-6 (pdf) - Statistiska centralbyrån. Tech. Rep. 1, Kungliga Social-styrelsen, Stockholm.

National Board of Trade (1913). Statistiska meddelanden. Ser. C, Mån-adsstatistik över handeln 1913-1953.

National Board of Trade (1954). Månadsstatistik över handeln 1954-1963.

Nickell, W. (2006). The CEP-OECD institutions data set (1960-2004).

41

OECD (2019). OECD Statistics.

Okun, A. M. (1962). Potential GNP, its measurement and signi�cance. Pro-ceedings of the Business and Economic Statistics Section, pp. 98�104.

Örtengren, J. (1979). Den nuvarande krisen i jämförelse med tidigare kriser.In Teknik och industristruktur � 70-talets ekonomiska kris i historisk belys-ning, Stockholm: Industriens utredningsinstitut, pp. 64�77.

Percival, D. B. andWalden, A. T. (2006). Wavelet methods for time seriesanalysis, vol. 4. Cambridge University Press.

Phelps, E. S. (1967). Phillips curves, expectations of in�ation and optimalunemployment over time. Economica, pp. 254�281.

Pissarides, C. A. (1992). Loss of skill during unemployment and the persist-ence of employment shocks. The Quarterly Journal of Economics, 107 (4),1371�1391.

� (2000). Equilibrium Unemployment Theory - 2nd Edition. Cambridge, Mass:The MIT Press, second edition edition edn.

Quilis, E. M. (2009). Temporal Disaggregation Library - File Exchange - MAT-LAB Central.

� (2018). Temporal disaggregation of economic time series: The view from thetrenches. Statistica Neerlandica, 72 (4), 447�470.

Ramey, V. A. (2016). Chapter 2 - Macroeconomic Shocks and Their Propaga-tion. In J. B. Taylor and H. Uhlig (eds.), Handbook of Macroeconomics, vol. 2,Elsevier, pp. 71�162.

Ravn, M. O. and Uhlig, H. (2002). On Adjusting the Hodrick-Prescott Filterfor the Frequency of Observations. The Review of Economics and Statistics,84 (2), 371�376.

Roed, K. (1997). Hysteresis in Unemployment. Journal of Economic Surveys,11 (4), 389�418.

Rogerson, R., Shimer, R. and Wright, R. (2005). Search-theoretic modelsof the labor market: A survey. Journal of economic literature, 43 (4), 959�988.

Romer, C. (1986). Spurious volatility in historical unemployment data. Journalof Political Economy, 94 (1), 1�37.

Romer, C. D. (1994). Remeasuring Business Cycles. The Journal of EconomicHistory, 54 (03), 573�609.

Rowthorn, R. (1977). Con�ict, in�ation and money. Cambridge Journal ofEconomics, pp. 215�239.

42

� (1995). Capital formation and unemployment. Oxford Review of EconomicPolicy, pp. 26�39.

� (1999a). Unemployment, capital-labor substitution, and economic growth. In-ternational Monetary Fund.

� (1999b). Unemployment, wage bargaining and capital-labour substitution.Cambridge journal of Economics, 23 (4), 413�425.

Sachs, A.-N. J. D., McArthur, J., Schmidt-Traub, G., Kruk, M., Ba-hadur, C., Faye, M. and McCord, G. (2004). Ending Africa's PovertyTrap. Brookings Papers on Economic Activity, 35 (1), 117�240.

Sax, C. and Steiner, P. (2013). Temporal disaggregation of time series.

Scarpetta, S. (1996). Assessing the Role of Labour Market Policies and In-stitutional Settings on Unemployment: A Cross Country Study. OECD Eco-nomic Studies, 26.

Schön, L. (1994). Långtidsutredningen 1995 Bil. 3 Omvandling och obalans :mönster i svensk ekonomisk utveckling. Bilaga till LU, Stockholm: Finansde-partementet.

Schön, L. (2000). En modern svensk ekonomisk historia: tillväxt och omvand-ling under två sekel. SNS förlag.

Schön, L. (2015). Historiska nationalräkenskaper för Sverige 8 Utrikeshandel1800-2000. Lund: Ekonomisk-historiska föreningen i Lund.

Schön, L. and Krantz, O. (2015). New Swedish Historical National Accountssince the 16th Century in Constant and Current Prices. Lund Papers in Eco-nomic History. General Issues, (140).

Schularick, M. and Taylor, A. M. (2012). Credit booms gone bust: Mon-etary policy, leverage cycles, and �nancial crises, 1870-2008. American Eco-nomic Review, 102 (2), 1029�61.

Semieniuk, G. (2017). Piketty's elasticity of substitution: A critique. Reviewof Political Economy, 29 (1), 64�79.

Sigurdsson, J. (2013). Capital Investment and Equilibrium Unemployment.Tech. Rep. wp61, Department of Economics, Central bank of Iceland.

Silenstam, P. (1970). Arbetskraftsutbudets utveckling i Sverige 1870-1965.Stockholm: Almqvist & Wiksell.

Silverstone, B. and Harris, R. (2001). Testing for asymmetry in Okun'slaw: A cross-country comparison. Economics Bulletin, 5 (2), 1�13.

Sjölund, N. and Wiklander, R. (2003). The Swedish Industrial ProductionIndex 1913�2002 - Time Series Analysis. Tech. rep., SCB Department ofEconomic Statistics.

43

Skott, P. (1985). Increasing Returns and Involuntary Unemployment: Is Therea Connection? Journal of Post Keynesian Economics, 7 (3), 395�402.

Smiley, G. (1983). Recent Unemployment Rate Estimates for the 1920s and1930s. The Journal of Economic History, 43 (2), 487�493.

Smith, A. (2009). The Wealth of Nations. Blacksburg, VA: Thrifty Books.

Smith, R. and Zoega, G. (2009). Keynes, investment, unemployment andexpectations. International Review of Applied Economics, 23 (4), 427�444.

Söderström, H. T. (1995). Realräntechock, skuldsanering och budgetunder-skott : en balansräkningsanalys av den svenska depressionen. Ekonomisk de-batt, 23 (3).

Sohn, K. (2013). Did unemployed workers choose not to work in interwar Bri-tain? Evidence from the voices of unemployed workers†. Labor History, 54 (4),377�392.

SOU 2002:118 (2002). Utveckling och förbättring av den ekonomiska stat-istiken : slutbetänkande. Bilaga 3 Beräkningsrutiner för nationalräkenska-perna. Stockholm: Fritzes o�entliga publikationer.

Statistics Sweden (1963). Månadsstatistik över utrikeshandeln 1963-1967.

Statistics Sweden (1968). Utrikeshandel. Månadsstatistik 1968-1976.

Statistics Sweden (1972). Historisk statistik för Sverige D. 3 Utrikeshandel1732-1970 = Foreign trade 1732-1970. Örebro: Statistiska centralbyrån (SCB.

Stockhammer, E. (2004). The Rise of Unemployment in Europe. Books, Ed-ward Elgar.

� (2008). Is the Nairu Theory a Monetarist, New Keynesian, Post Keynesianor a Marxist Theory? Metroeconomica, 59 (3), 479�510.

�, Guschanski, A. and Köhler, K. (2014). Unemployment, capital accu-mulation and labour market institutions in the Great Recession*. EuropeanJournal of Economics and Economic Policies: Intervention, 11 (2), 182�194.

� and Klär, E. (2011). Capital accumulation, labour market institutions andunemployment in the medium run. Cambridge Journal of Economics, 35 (2),437�457.

Stone, L. (1979). The Revival of Narrative: Re�ections on a New Old History.Past & Present, (85), 3�24.

The Swedish Institute for Social Research (2015). The Out-of-WorkBene�ts Dataset (OUTWB). Tech. rep., Stockholm University.

Tieben, B. (2009). The Concept of Equilibrium in Di�erent Economic Tradi-tions: A Historical Investigation. 449, Rozenberg Publishers.

44

Torstendahl, R. (1971). Introduktion till historieforskningen : historia somvetenskap. Stockholm: Natur och kultur.

Vartiainen, J. (1998). Understanding Swedish social democracy: victims ofsuccess? Oxford Review of Economic Policy, 14 (1), 19�39.

Visser, J. (2006). Union Membership Statistics in 24 Countries. Federal Pub-lications.

� (2019). ICTWSS Database. version 6.0. Tech. rep., Amsterdam Institute forAdvanced Labour Studies (AIAS), University of Amsterdam, Amsterdam.

Vliet, V., Olaf and Caminada, K. (2012). Unemployment ReplacementRates Dataset Among 34 Welfare States, 1971-2009: An Update, Extensionand Modi�cation of the Scruggs' Welfare State Entitlements Data Set. SSRNScholarly Paper ID 1991214, Social Science Research Network, Rochester,NY.

Vroey, M. D. (2016). A History of Macroeconomics from Keynes to Lucas andBeyond. New York: Cambridge University Press, 1st edn.

Waldenström, D. (2014). Swedish stock and bond returns, 1856�2012.

Watson, M. W. (1994). Business-Cycle Durations and Postwar Stabilizationof the U.S. Economy. American Economic Review, 84 (1), 24�46.

Weitzman, M. L. (1982). Increasing Returns and the Foundations of Unem-ployment Theory. The Economic Journal, 92 (368), 787�804.

� (1985). Increasing Returns and the Foundations of Unemployment Theory:An Explanation. Journal of Post Keynesian Economics, 7 (3), 403�409.

Whittaker, E. T. (1922). On a New Method of Graduation. Proceedings ofthe Edinburgh Mathematical Society, 41, 63�75.

Zagorsky, J. L. (1998). Was depression era unemployment really less inCanada than the U.S.? Economics Letters, 61 (1), 125�131.

45