Wealth and the economic vote : How assets and liabilities shape election outcomes

(1)

ACTA UNIVERSITATIS

UPSALIENSIS UPPSALA

2021

Digital Comprehensive Summaries of Uppsala Dissertations

from the Faculty of Social Sciences

181

Wealth and the economic vote

How assets and liabilities shape election outcomes

ANTON BRÄNNLUND

ISSN 1652-9030 ISBN 978-91-513-1115-9 urn:nbn:se:uu:diva-432091

(2)

Dissertation presented at Uppsala University to be publicly examined in Brusewitzsalen, Östra Ågatan 19, Uppsala, Monday, 8 March 2021 at 13:15 for the degree of Doctor of Philosophy. The examination will be conducted in Swedish. Faculty examiner: Associate Professor Andreas Bergh (Department of Economics, Lunds University).

Abstract

Brännlund, A. 2021. Wealth and the economic vote. How assets and liabilities shape

election outcomes. Digital Comprehensive Summaries of Uppsala Dissertations from the Faculty of Social Sciences 181. 36 pp. Uppsala: Acta Universitatis Upsaliensis.

ISBN 978-91-513-1115-9.

This thesis contributes to the literature on economic voting, especially the subfield of the electoral impact in relation to wealth. The thesis consists of an introductory chapter and four independent research articles based on data from Sweden. Based on the first article, I find that the support for right-wing parties decreases in areas where voters are heavily invested in financial assets when there is a large amount of volatility in the world markets. Such patterns suggest that voters are responsive to financial risks. Through the second study, I illustrate that voters are sensitive to changes in monetary policies as well. More precisely, I show that voters tend to reward governments for decreases in interest rates. With the third study, I investigate the interplay of markets in a more direct way, estimating the effect of unemployment on voting in relation to household wealth. I find that the Swedish left-wing parties gain electoral advantage when the unemployment rates rise in less wealthy areas but that they lose support where voters are comparatively well-off. Finally, based on the fourth study, I investigate whether wealth has an impact on how voters behave with individual level data. The findings in this study suggest that wealthy citizens vote for right-wing parties to a grater extent. However, the estimated effect is small.

Keywords: Economic voting, wealth, political parties, financial markets, economic inequality Anton Brännlund, Department of Government, Box 514, Uppsala University, SE-75120 Uppsala, Sweden.

(3)

(4)

(5)

List of papers

This thesis is based on the following essays.

1. Brännlund, Anton (2020) Patrimony at risk: Market risk and right-wing voting. Working paper. Manuscript submitted for publication.

2. Brännlund, Anton (2020) Zero per cent accountability? How low inter-est rates save governments from electoral defeats. European Journal of Political Economy. https://doi.org/10.1016/j.ejpoleco.2020.101986 3. Brännlund, Anton (2020) Labor Market Conditions and Partisan Voting:

How Unemployment Hurts the Left. Political Behavior. https://doi.org/10. 1007/s11109-020-09655-w

4. Ahlskog and Brännlund (2020) Uncovering the source of patrimonial voting. Accepted in Political Behavior

(6)

(7)

Acknowledgements

It is ironic that in writing a thesis about wealth, I ended up in debt to a great amount of people. Jokes aside, I owe my gratitude to a great many colleagues at the department because my process was dependent on their support and con-tributions. To begin with, I was not even supposed to attend the PhD program since I only made it to the reserve list. However, Professor Joakim Palme of-fered me a position partly ﬁnanced by the Uppsala Center for Labor Studies, and I am forever grateful for that opportunity. To the others who supported me during this episode, thank you. You know who you are.

I suppose that my work has flown under the radar for most colleagues, given my general absence from the department. But staying at home was important to me as a highly introverted person, since it allowed me to produce something of value. I remember being introduced to all the organizations one can attend as a PhD student in Uppsala, which rendered me extremely nervous – the list of social activities seemed to be endless! Fortunately, later that day, we had a seminar with Professor Li Bennich-Björkman about personality types and writing strategies. I was quite surprised since the standard procedure in my life has often been to treat everybody as extroverts. However, Professor Li Bennich-Björkman stated that it was fine to keep these social activities to a minimum and work from a place where one can find inspiration. That is what I did, and it has been crucial for me and my work. I want to thank the department because few other employers accept that employees work from home for so long, with so little supervision.

Furthermore, I owe the greatest of debts to my supervisors Kalle and Pär – I cannot stress this enough. We have, throughout these years, kept it simple: I provided them with texts, and they gave me tons of comments in return; this cycle carried on until there were no comments left to give and the paper was ready for submission. I learned a lot during this rewarding process, but it was tough from time to time. I joke sometimes about the joy of defending my work at seminars because such events include both praise and not only criticism; that’s a compliment! Criticism is gold in this business, and my work is pretty much built on such comments. In the end, I think of this as a team effort, and I hope that we can work on some common project in the future.

Another person that deserves attention is Linuz Aggeborn, who provided signiﬁcant comments on all chapters in the thesis. I appreciate the effort in making this dissertation better, and I thank him for his contributions. I also want to praise my co-author Rafael Ahlskog, who enabled me to study the relationship between wealth and voting at the individual level. Sven Oskarsson

(8)

deserves to be mentioned here as well, as I made use of data originating from their project. Finally, other important contributions were made by Marcus Österman and Adrian Adermon during the Manus conference – their efforts have not been forgotten. Thank you!

(9)

Introductory essay

The big economic story of the last century was the sharp increase in labour income that took place after the two world wars. Many social scientists, there-fore, became preoccupied with the relationship between labour-market condi-tions and political outcomes. For instance, it was around this time that schol-ars of economic voting began to relate variables such as unemployment and income growth to political behaviour. Consequently, cycles in these indicators were said to affect election outcomes in democratic nations, with the conﬂict between labour and capital generating a clear left/right dimension in politics (Alesina 1987; Alesina and Rosenthal 1995; Fauvelle-Aymar and Stegmaier 2013; Hibbs 1977; Kiewiet and Lewis-Beck 2011).

The economic story of the 21st century has been different, however: a mas-sive but unevenly shared increase in household wealth has been seen. Much of the growth in wealth was driven by the Conservative Revolution, which began with the victories of Margaret Thatcher in the United Kingdom and Ronald Reagan in the United States. Later, events such as the collapse of the Soviet bloc, ﬁnancial deregulation, and globalisation also contributed to the trend. Wealth-to-income ratios are now approaching pre-World War I levels in many advanced economies (Piketty and Goldhammer 2014). The focus of much economic analysis, therefore, has shifted towards asset markets, given that a large share of the population has come to own assets.

The current burgeoning of wealth is now outpacing global income growth. Yet political scientists long paid little attention to this development. This lack of interest is surprising, especially since many of the old models of economic voting are under serious attack. A common claim these days is that political characteristics among voters shape economic perceptions and outcomes, rather than the other way round (Gerber and Huber 2010; Key et al. 2017). A related critique is that economic voting is an uneven phenomenon, in that it appears in some elections but not in others (Anderson 2007). A review of the liter-ature conﬁrms such patterns: some studies suggest that unemployment hurts governing parties (Kiewiet and Lewis-Beck 2011; Lewis-Beck and Stegmaier 2013); others say that right-wing parties suffer from it (Helgason and Mérola 2017; Wright 2012); and ﬁnally, a large group of studies says that left-wing parties lose votes when the economy weakens (Kayser and Grafström 2016). 11

(12)

These varied ﬁndings suggest the effect of the economy on voting can go either way.

I hope to nuance this debate a bit, by showing that some of the problems with the models of economic voting stem from the mistake of not includ-ing wealth as an explanatory variable. Unfortunately, wealth lost its lustre as a variable during the last century, due to the massive rise in labour market-generated income that came with industrialisation, as well as the massive de-struction of wealth resulting from the two world wars. However, as the wealth level increased, so did the exposure of voters to asset markets. And such exposure can differ greatly from that connected with the income-generating process found in labour markets. In this article, I discuss and apply theories of economic voting in connection with voters’ responsiveness to the outcomes produced both by labour markets and by asset markets.

There are many similarities between wealth and labour income, but we should not treat them as substitutes for each other in empirical work. Each concept, namely, has unique features that distinguish it from the other. To be-gin with, income is a ﬂow of funds; wealth is a stock of resources that may produce a yield. The main difference, however, is that income is generated in the labour market, and it requires some personal input; while wealth in most cases is generated over time in asset markets, and it depends to a greater ex-tent on what other people do. Yes, the capacity to acquire wealth increases with income; however, the ability to consume increases with income as well. This implies that a person can have a high income without being wealthy, and vice versa. It implies too that voters can, at one and the same time, experience changes in wealth and in income in opposite directions.

Furthermore, a wealth stock is often needed in order to acquire necessities such as a home (and even, in some countries, education and healthcare). Citi-zens without large savings, however, can take on debt, i.e., they can shift their wealth to negative in order to be able to afford such goods. That is, wealth has a negative as well as a positive component. This is an important driver behind economic inequality. The Baby Boomer generation, for instance, has been described as the wealthiest generation of all time. Their grandchildren, on the other hand, are struggling to a much greater extent with various types of debt, such as student loans, mortgages, car loans, etc. The negative side of the balance sheet, however, is poorly understood, and it deserves greater scholarly attention than it has received. It is easy to imagine, for instance, that indebted voters on the one hand, and voters who have lent them the money on the other, have very different economic interests.

Nevertheless, there have been attempts to incorporate asset markets into studies on economic voting. Some scholars, for example, have found that the performance of asset markets affects the level of voter support for incum-bent governments (Fauvelle-Aymar and Stegmaier 2013; Larsen et al. 2019; Ostrom Jr. et al. 2018; Prechter et al. 2012; Sen and Donduran 2017). How-ever, such studies often focus on market performance rather than on personal 12

(13)

wealth: i.e., they portray citizens as focused on the general condition of the economy, as indicated by changes in stock or housing prices. That is, wealth should inﬂuence voters’ behaviour in elections, because it affects their views on government policy.

An effect of this kind has been noted, in that asset wealth often correlates with how individuals describe their views on issues like government spending in election surveys (e.g. Ansell 2014). However, the general effect of per-sonal wealth on election outcomes has yet to be studied. I have accordingly been guided by the following research question in my dissertation: how does personal wealth affect election outcomes?

One interesting effect of personal wealth on politics is the emergence of dual interests among voters. Governments tend, namely, to tax wealth and labour income very differently. Thus, even voters with a relatively low income can beneﬁt from low taxes on assets if they own their home. However, persons without assets have more to gain from higher asset taxation. This means new political cleavages can emerge which are centred on asset markets rather than on labour markets. This contrast is apparent in the COVID-19 crisis, in that global asset markets have rallied to an all-time high, even as millions of work-ers have lost their jobs. This illustrates how votwork-ers can experience different economic forces based on their differing asset wealth. Scholars of economic voting, therefore, are faced with a new ﬁnancial reality to which they must adjust.

In sum, political scientists have always paid attention to material interests, because these are believed to shape political behaviour. Major political trans-formations throughout history have often been due, at least in part, to mate-rial deprivation and poor economic outcomes (Goldstone 1991; Gurr 1970; Skocpol 1979). Nor is this surprising. Political ideas, after all, rarely materi-alise out of nothing, and they tend to reﬂect the inﬂuence of economic forces, among other things (Blyth 2002). However, different epochs have been shaped by disparate forces.

The question now is how the massive but unequal increase in asset wealth shapes political conflict and voting behaviour across the world’s democracies. Wealth has been understudied for too long. It deserves our attention, because it turns many of the old political conflicts upside down. This thesis is an attempt to study these effects, using empirical data from Sweden. It joins a small but rapidly growing field that looks at how the massive rise in household wealth affects not just our voting behaviour, but our lives at large.

Wealth and Economic Voting

Economic-voting models make certain assumptions about material interests. These do not necessarily involve self-interest, but they do involve material matters. Scarcity forces individuals to devise strategies to ensure that they, or 13

(14)

those they care about, receive their share of available resources. Economic voting is one of these strategies. Much of our understanding of what we call economic voting today began to evolve in the late 1950s, as economists sought increasingly to analyse political problems using formal modelling and applied mathematics. A major motivation behind this lay in the rapid expansion of government spending in the post-war Keynesian era, when such spending be-came one of the largest components of GDP in Western nations. To understand the economy, then, we would have to understand politicians and voters as well. This was also, in a sense, a decade of diversion. Increasingly, that is to say, economists began to distinguish themselves sharply from other social sci-entists, by embracing methods usually found in the natural sciences. Social and political problems were to be solved with mathematical optimisation – something which only becomes possible after we break down the world into a simple model of rational and self-interested agents. Voters are easy to under-stand, economists averred, in that they vote based on the beneﬁt they expect from electing certain candidates to ofﬁce (Downs 1957). Such voting models gained considerable popularity over time, in part because they made neat and simple predictions possible.

This ﬁeld gradually divided into two major areas of investigation: the study of valence voting on the one hand, and of positional voting on the other (also known as retrospective and policy-based voting, respectively). Early models of valence voting portrayed voters as forming expectations on the basis of cur-rent conditions, which generates retrospective behaviour. According to these models, voters choose the candidates whom they regard as the most compe-tent (Fiorina 1981; Key and Cummings 1966; Kramer 1971). The challenge, however, is to do this under conditions where the character and competence of most candidates are concealed to everybody but themselves (Ashworth 2012). The solution to this problem was to use current conditions as a proxy for in-cumbent competence: most voters, namely, agree that a low unemployment rate is better than a high one, and that solid growth is to be preferred over eco-nomic stagnation. Accordingly, candidates who deliver low unemployment and high growth are seen as more appealing (Alesina and Rosenthal 1995; Dorussen and Taylor 2002).

From this perspective, elections are about ﬁnding good candidates that can provide stable output over time. Citizens’ perceptions of the state of the econ-omy, then, are seen as an important driver behind their voting choices. Voters are thought to follow a simple retrospective rule: vote for the government when the economy is strong; otherwise, vote for the opposition (Kiewiet and Lewis-Beck 2011; Lewis-Beck and Stegmaier 2013). These predictions are at-tractive in a sense, since economic growth is said to be conditional upon good governance (Mira and Hammadache 2017). In this view voters act as watch-men, keeping society on an optimal path towards greater prosperity. There is a problem here, however, many researchers aver: this is not how democracy works in reality.

(15)

It is namely far from always the case, according to many scholars, that vot-ers punish governments for macro-economic weakness (Andvot-erson 2007). The main reason for this is that economic perceptions are clouded by partisan bias (Bartels 2002; Evans and Pickup 2010; Gerber and Huber 2010; Ramirez and Erickson 2014). Those who support the government of the day tend to per-ceive the economy as stronger than do those who favour the opposition. Sharp downturns in the economy will break such beliefs, but economic variation within the normal range is not enough to do so (Chzhen et al. 2013). Partisan voters are more likely instead to treat elections as a competitive sport (Miller and Conover 2015), and to express overall life satisfaction when their party is in charge (Di Tella and MacCulloch 2005).

However, while partisanship is problematic in valence models of voting, it serves as a key element in positional models. Voters here are assumed to place themselves along the left/right policy dimension, based upon their economic characteristics. Traditionally, left- and right-wing parties have been seen as representing different groups, with right-wing parties attracting voters from the middle class and the business community, by keeping a lid on inflation; and with parties on the left appealing to working-class voters with promises of full employment, even at the cost of higher inflation (Alesina 1987; Hibbs 1977). This idea dates back to the Keynesian era between 1945 and 1970, during which governments achieved economic stabilisation through fiscal policy and tight control over financial markets.

Positional voting models portray economic voting as a competition where the winner can determine the tax rate and the output of government. Scholars in this tradition claim that macro-economic changes produce partisan rather than incumbent-based voting. Wright (2012), for instance, asserts that un-employment in the US has historically been lower under Democratic incum-bents, and that voters in that country perceive Democratic administrations as being better suited to handle economic slowdowns. He further shows that Democratic candidates gain an electoral advantage when unemployment rises. Similar arguments have also been made based on both cross-national and individual-level data (Dassonneville and Lewis-Beck 2013; Helgason and Mérola 2017). The argument here is simple: voters drift towards left-wing parties when unemployment rises, because the expected value of pro-welfare policies increases with income insecurity.

This argument, however, has been challenged. A common finding, namely, is that left-wing governments are penalised particularly severely for high un-employment (Powell and Whitten 1993; Whitten and Palmer 1999). Other scholars simply assert that right-wing parties benefit from weak economic fig-ures in general, because voters turn to fiscal conservatives when times are bad in order to keep taxes down (Durr 1993; Kayser and Grafström 2016; Kayser 2009; Lindvall 2014; Markussen 2008; Stevenson 2001). Such a diversity of empirical findings suggests that voters react in many different ways to labour-market weakness, and that the effect of unemployment is conditional on the 15

(16)

context. The context has changed, moreover, because voters have become more and more exposed to variations in asset markets, due to the increase in their wealth holdings.

The world is not centred on the labour market anymore, and the exposure of voters to ﬁnancial and asset markets has grown tremendously during recent decades. Attempts have indeed been made, moreover, to incorporate asset markets into economic-voting models. Some studies, for example, indicate that gains in the stock market are correlated with approval for the president in the US (Fauvelle-Aymar and Stegmaier 2013; Ostrom Jr. et al. 2018; Prechter et al. 2012), and with satisfaction with the government in the UK (Sen and Donduran 2017). Housing prices too are said to be correlated with support for the sitting government (Larsen et al. 2019). Such studies are based on the assumption that voters gather information about macro-economic conditions from changes in asset prices. I think it is reasonable, however, to argue that the effect of asset markets on voters’ private ﬁnances should matter even more for voting, given the massive rise in their household wealth.

Figure 1 shows the wealth-to-income ratio in Sweden between 1870 and 2016, and for France, Germany, the UK, and the US (taken together) during the same period. This ratio shows how large the stock of available wealth is in a country compared to what the people of that country earn in terms of income. As can be seen, there was massive wealth destruction during the 20th century, due to the demolition of factories and of real estate in the first and second world wars. It is easy to see why the impact of wealth during that period was far less than that of labour market-related variables. In the mid-1970s, however, this began to change. The emerging era of financial liberation generated new political fissures, as governments moved away from tight control towards the liberalisation of markets.

Now the concern was less with consumer prices than with asset prices, given that ﬁnancial markets and their institutions had become too big to fail during economic downturns. Financial regulation became an important pol-icy tool, but one that involved a clear new trade-off: tougher rules and higher taxes serve to reduce returns (Tanndal and Waldenström 2018), but deregu-lated markets tend to bring price bubbles, necessitating expensive intervention when asset owners seek to protect their returns from collapsing markets (Brun-nermeier and Schnabel 2016; Chwieroth and Walter 2019). Political scientists have noted that asset owners are more likely to express support for parties and candidates that advocate laissez-faire policies in the ﬁnancial sector, because asset prices are likely to rise when such policies are applied (Foucault et al. 2013; Persson and Martinsson 2016; Stubager et al. 2013).

Such ﬁndings tell us an important story: voters also incorporate asset mar-kets into their voting decisions. The problem, however, is that ﬁnancial or asset markets do not always pull voters in the same direction as labour markets. Par-ties that serve voters’ interests in asset markets can go against their interests in labour markets, and vice-versa. In situations where the wealth-to-income 16

(17)

3 4 5 6 7 1850 1900 1950 2000 2050 Year

Sweden Western average

Figure 1. This is a ﬁgure over the wealth to income ratio in Sweden and other western

nations over time.

ratio has greatly grown, a decline in the business cycle can produce different types of behaviour than those suggested by classical models of economic vot-ing. So, to understand the capriciousness of economic voting, as well as the inconsistency that haunts empirical studies in this ﬁeld, we should look for situations in which voters’ interests diverge with respect to asset markets and to labour markets.

An example of this arises when voters have more to lose from falling as-set returns than from unemployment. There is a strong tradition of relating economic-policy preferences to self-interest (Boix 1998; Clegg 2004; Esping-Andersen 1979; Iversen and Soskice 2001). Scholars such as Ansell (2014) argue, for instance, that asset owners become less likely to support policies like social-security spending when they grow wealthier. Such effects arise from their ability to use asset wealth as a buttress during times of ﬁnancial constraint, such as a recession. Hence, when the wealth-to-income ratio is high, right-wing parties can beneﬁt from higher unemployment rates, because asset owners want to protect their savings from taxation. However, asset own-ers may instead respond strongly to the risk of losing their savings – which means that, when there is turbulence in asset markets rather than labour mar-kets, left-wing parties may gain support.

Another important difference emerges when credit markets and labour mar-kets pull voters in different directions. The economic interests of voters can 17

(18)

namely diverge when the level of negative wealth – debt – grows faster than the level of income. What previous research has often missed is that, despite their more secure position in the labour market and their large holdings of real-asset wealth, many middle-class persons still have limited ﬁnancial savings, and live from pay check to pay check. Hence, a sharp rise in debt levels makes interest rates important for voters’ welfare. However, a steep downturn in the business cycle is often followed by a decline in interest rates, which of course beneﬁts the indebted middle class. Thus, while disliking unemployment and weak in-come growth, voters may still refrain from punishing governing parties when they have more to gain from falling interest rates than they stand to lose from a poor labour market.

As these scenarios illustrate, then, the great rise in both wealth and debt over recent decades has endowed voters with dual interests. However, such variables are often absent from standard models of economic voting. Now, when studies and models show inconsistent results, scholars often queue up to criticise them. For my part, however, I plan to show that the old models are not completely wrong, in that voters do indeed care about economic indicators; however, we must shift our attention to wealth and asset markets. That is, the economy is still important, but voters’ interests can diverge when they become either wealthy or indebted. Put simply: as the number of households with assets and liabilities rises, so does the importance of including asset and ﬁnancial markets in our models of voting.

The Shift-Share Design

For over 50 years, social scientists have sought to estimate the impact of eco-nomic variables on voting behaviour and the policy mood of citizens. Method-ologically speaking, however, little has changed over that time. The main tool is still regression analysis, and the most common data comes from surveys of various types. In fact, as economic-voting models grew more popular among social scientists, many surveys adjusted their questions to the needs of schol-ars employing such models. Thus, we often ﬁnd questions on personal per-ceptions of the economy in surveys that ask respondents about their political views. However, problems arise if we recognise the importance of wealth, given that there has been little tradition in earlier surveys of collecting data on wealth. Unfortunately, in fact, there is very little data on household wealth generally.

Social scientists in Sweden are otherwise spoiled, in that they can connect data from surveys with the massive registers at Statistics Sweden. This means we can connect voting choices in survey answers to levels of education, of in-come, and so forth. In the case of wealth, however, such data is only available for the period between 1999 and 2007, because the Swedish Wealth Regis-ter was used for tax collection and it has not been updated since 2007, when 18

(19)

the wealth tax was abolished. Sadly, little effort was made to connect these registers with election surveys, because there was little interest in wealth as a variable at the time. Hence, a lack of data on wealth became a crucial problem for me at the early stages of this dissertation.

A ﬁrst potential reaction to such a problem is to look for wealth data in other places. However, similar data is almost impossible to ﬁnd from other countries as well. There are surveys that include questions on asset ownership, but little data is available on asset values. Instead, a better solution is to deploy a method in which information about past asset wealth across Sweden can be used to draw conclusions about the relationship between wealth and voting later in time. This led me eventually to the shift-share instrumental-variable design.

Traditionally, an instrumental variable is a third variable that is useful in empirical studies where the relationship between the explanatory variable of interest and the outcome at hand is believed, for one reason or another, to be confounded by omitted factors. The idea of the instrumental-variable approach is simple: if we ﬁnd a third variable that is unrelated to all of the possible con-founding factors but which still is related to the independent variable, then we can use that variable as an instrument. In such a strategy, our endogenous independent variable becomes a mechanism between the instrument and the outcome. That is, we restrict the identifying variation to come from the instru-ment alone. However, such a strategy depends heavily on what is termed the exclusion restriction: i.e., the effect of the instrument on the outcome is only allowed to travel through our endogenous independent variable.

The shift-share instrument, which has been used extensively in ﬁelds such as labour economics, starts from a very simple principle: we can guess many economic outcomes based on basic individual characteristics. For instance, we can predict how much a person earns based on information about his/her occupation, or even about the industrial sector in which that person works. The shift-share instrument deploys a similar logic: it predicts economic outcomes at the group level based on differences in initial conditions.

It was noted already in the 1950s, for instance, that regional differences in industrial composition could be used to predict regional differences in in-come across the US (Perloff 1956). Still, the big recognition of the shift-share design came in 1991, when Tim Bartik analysed employment growth across industrial sectors in the US, along with regional industrial composition, to predict changes in regional employment rates. The design came to be asso-ciated so closely with Bartik that it is often referred to as a Bartik-instrument (Goldsmith-Pinkham et al. 2018).

Bartik’s idea was that the national economy consists of regional parts, and these parts have discrete industrial compositions. Industrial dependence at the regional level is often measured by the share of workers employed in a given industrial sector – hence the term share. These shares are distributed unequally across the country: some regions depend on manufacturing, others depend on 19

(20)

services, and so on. By exploiting differences in these shares, Bartik showed, we can predict changes in local employment, by multiplying national trends in industrial employment by the share of workers employed in that industry at the regional level. That is, regions with a heavy dependence on manufacturing suffer more when manufacturing suffers at the national level, while regions which depend on services suffer more when the national demand for services falls.

Bartik’s point, however, was that shifts in demand at the national level were exogenous from all unobserved factors found among the regions. Insights of this kind make it possible to answer many labour market-related questions. Since the early 1990s, however, many scholars have applied similar designs for a large array of other inquiries as well. For instance, it is very com-mon to analyse migration-related research questions with a shift-share design (Goldsmith-Pinkham et al. 2018). The logic in such a case is similar, in that we can predict regional migration outcomes based on the current distribution of nationalities across regions and trends in migration at the national level. This type of prediction works because migrants tend to settle where they ﬁnd citizens with backgrounds similar to their own (Jaeger et al. 2018).

As this method grew more popular, however, more time was devoted to discussing the methodological challenges found in shift-share studies. One possible caveat, for instance, is that the initial conditions in one period de-pend on growth rates in the past. That is, the mixture of nationalities or the shares of workers employed in certain industries in one area depend on pre-vious trends at the national level, too. Such relationships are not desirable in the instrument-variable framework, due to the strict exclusion restriction. One solution to this problem is to set the initial conditions to a base year, because regional conditions in the past cannot depend upon future trends at the national level (Broxterman and Larson 2020).

The ability, or even the recommendation, to use past regional characteristics facilitated my problem with the lack of data. That is, this method allowed me to study the impact from asset markets on voting across Sweden despite having only had actual wealth data from one election. The general key to my research design is to compare electoral districts but restrict their ﬁnancial characteristics to the year 1999. More precisely, my approach is based on the idea that those districts that owned more assets in 1999 are more exposed to the risk and rewards in asset markets in later years compared to those districts that owned less.

For the intuition of this strategy, consider areas with large holdings of hous-ing wealth. Houshous-ing prices has increased more than 100 percent on average in Sweden during the last 20 years. What I assume is that districts where voters have more money invested into housing grow wealthier when housing prices rise on a national basis compared to areas where people rent their homes to a greater degree. This relationship not only helps me to overcome the problem of data shortages, but will reduce the bias in the estimates as well given that 20

(21)

the national rise in house prices should be exogenous from changes in other important district characteristics.

Data Description and Applications

My identifying strategy in this study has some resemblance to a shift-share in-strumental variable design, in the sense that I allow sub-national units to react differently to changes in higher national trends based on their initial distribu-tion of either assets or liabilities. This strategy reduces the need for wealth data over longer time periods but places demand on the data that I have, as I require considerable variation in the allocation of wealth and debt across the geographical units during the initial years. Therefore, in most of the studies, I use voting data from electoral districts, which are freely available from Statis-tics Sweden. Electoral districts in Sweden are small areas nested within larger local jurisdictions, known as municipalities, usually with 1,000–2,000 eligible voters.

I link these districts over time with their identiﬁcation number. However, a general problem is that the number of districts varies between elections; I have only used those that remain unchanged throughout the entire period. On the other hand, data on assets and liabilities are available from the Swedish wealth register and I have collected data from three years: 1999, 2002, and 2006. This data includes the number of owners plus the total value of stocks, mutual fund shares, housing and apartment wealth, and the total value of all household debt. Computing wealth per voter in the districts reveals that the wealth distribution is highly skewed. The average wealth per voter is 600,000 SEK (Swedish Krona) while the highest recording is over 70,000,000 SEK.

I use this data to construct initial financial characteristics in 1999, which I multiply with higher-level trends in later years. In the first study, I investigate, for instance, if exposure to financial risks affects voting behaviour. Previous studies state that asset wealth pushes owners towards right-wing parties due to their low tax and slimmed welfare policy platforms, but the effect does not seem to appear in every election (Hellwig 2008; Quinlan and Okolikj 2019). Therefore, I aim to determine whether the risk of financial losses decreases the support for the conservative parties. This is a question that has not been investigated in previous research. The problem is, of course, the lack of cred-ible financial data over time. It is necessary to have longer panels since the risk associated with asset ownership changes between election years. I deploy a type of shift-share design to address this problem.

In my design, I make use of the share of financial-to-total-assets in electoral districts during 1999 as an initial financial characteristic. That is, I divide the value of the financial assets with the value of all assets in the districts and I analyse this share together with volatility in world markets over time. The basic idea follows a simple logic: Districts that held several financial assets 21

(22)

in 1999 should be more likely to vote based on market volatility in later years compared to districts with little or no ﬁnancial wealth during that initial year. My strategy is heavily inﬂuenced by the studies that estimate the impact of labour market risks on voting and social policy preferences.

These studies suggest that individuals who find themselves in occupations with very high rates of unemployment are more likely to support generous welfare spending and, thus, left-wing parties (Cusack et al. 2006; Helgason and Mérola 2017; Milita et al. 2019; Rehm 2011; Vlandas 2019). In my study, I argue that individuals with a large share of assets with risky returns become more likely to defect from the right when market volatility rises. However, since we lack individual-level data, I investigate this question on the district level. I find that districts with relatively large holdings of financial assets, compared to real wealth, are more likely to abandon liberal and conservative parties when there is uncertainty in the market. This result suggests that voters are responsive to not only labour but also asset market risks.

In my second study, I investigate why citizens are slow to change their voting behaviour even in the midst of a recession. For instance, several po-litical scientists expected rapid change after the financial crisis, which was the sharpest economic downturn since the Great Depression. But political life, to the contrary, quickly returned to business as usual in most countries. Therefore, to investigate why voters are slow to react to significant economic changes, I look at mitigating forces. Moreover, most governments try to fend off slow economic growth and weak inflation rates with monetary stimulus. The purpose of my second study is to investigate if such policies have an elec-toral impact as well.

The empirical strategy is straightforward. I exploit the fact that electoral districts in Sweden have an unequal distribution of liabilities and use this di-versity in debt as an initial condition. I argue that the districts that had a high level of debt in 1999 must devote relatively more or fewer funds towards debt service in later years when interest rates rise or fall, respectively. However, it should be noted here that I do not observe actual expenditures but rely instead on the studies that have already been conducted with household-level data. For instance, Flodén et al. (2017) relate changes in the Swedish policy rate, better known as the repo-rate, to household debt-to-income ratios in a sample from Sweden. They do so to estimate the effect of interest rate expenditures on consumption among indebted households. What they ﬁnd is that expenditures increase among indebted households as the repo-rate rises, which forces these households to cut back on consumption.

In my study, I aggregate similar data up to the district level. This aggrega-tion allows me to estimate an electoral response from a change in the repo-rate using the variation in debt across districts in 1999. Using this design, I ﬁnd that districts with large holdings of debts become relatively more supportive of the parties in government when interest rates fall. In other words, districts with relatively large debts become less likely to turn against the incumbent when 22

(23)

interest rates fall. Governments usually lose support between the elections and this is known as the cost of ruling (Paldam and Skott 1995). My results suggest that this cost becomes lower when interest rates fall in indebted areas. This is important since interest rates are lowered during economic hardship and can explain why governments survive the grimmest recessions.

In the third study, I investigate the interplay of markets in a more direct manner. I look at an interactive effect between changes in wealth and unem-ployment. In order to estimate the effect from wealth, I deploy a shift-share instrumental design, where I explore an interaction between the local owner-ship structure of assets in 1999 and trends in asset values found nationally. Some scholars have estimated the response in social policy preferences with individual-level data using changes in house prices and homeownership status (e.g. Ansel 2014). This strategy is similar to my design and is based on the premise that homeowners grow wealthier when house prices rise. Areas in which citizens own a large share of mutual funds are assumed, on the other hand, to grow wealthier when the stock market rises and so on.

I was able to investigate the reliability of this assumption more carefully given that I can estimate how well my independent variable predicts changes in actual district wealth during the available years. It is revealed that there is a strong linear relationship between changes in my independent variable and changes in district wealth per voter. The main result in this study suggests that Swedish left-wing parties gain an electoral advantage when the unemployment rate rises in less wealthy areas, but they lose support when unemployment rises in richer districts. I argue that the wealthy have less need for social insurance and thus vote for conservative parties in order to put a cap on social spending when the unemployment rate rises. By contrast, asset-less voters opt for the left, with an eye towards preserving their entitlements.

Finally, in the fourth study, I turn, together with Rafael Ahlskog, to individual-level data. In greater detail, we compare survey answers on voting choices and attitudes within twin pairs, connecting these answers to their personal hold-ings of wealth. The political variables come from the SALTY survey that was conducted to investigate the link between inheritance, health and behaviour in Sweden. Data on personal wealth was collected from the wealth register. This design has some advantages over my three earlier studies because we can more precisely estimate the size of the causal effect of wealth on voting. How-ever, such a design also constitutes an extremely tough test, given that there is very little variation left to exploit within each twin pair. We manage, however, to register some effects from asset wealth on behaviour and attitudes. Even if the effects are small, we ﬁnd that owners of large real-estate-based wealth holdings are more prone to vote for the right and resist asset taxation.

(24)

Empirical Caveats

An important strength of my research design is that it allows me to study the relationship between wealth and voting even if such data are scarce. This strat-egy works better however, under the condition that there is a strong positive correlation in the districts ﬁnancial characteristics over time. In other words, districts that held considerable debts or assets in 1999 should do so in later years as well for the design to work properly. To evaluate the credibility of this assumption, I turn to the data that I have and the correlation between these different shares in the years 1999, 2002 and 2006. The point is to determine whether these correlations drop rapidly during this short period because such patterns would raise questions about the relevance of my independent vari-ables in later years.

Fortunately, the correlations are strong. For instance, the correlation coef-ficient is 0,85 for the share of risky assets between 1999 and 2002 and 0,75 between 1999 and 2006. If we take a look at the debt-to-income ratio instead, we find that the correlation coefficient is 0,96 between the share in 1999 and 2002 and 0,94 between the share in 1999 and 2006. We finally find a simi-lar pattern in the share used in the third study as well, in that the correlation coefficient is 0,93 for wealth per voter between 1999 and 2002 and 0,90 be-tween 1999 and 2006. These strong correlations suggest that my independent variables should have relevance in the years where I lack wealth data.

However, using data from districts implies that I model micro-level rela-tionships using aggregated data. Hence, one obvious problem in this case is the ecological fallacy and what economists refer to as aggregation bias (King 1997). In other words, it is hard to draw conclusions about how individuals behave based on group-level correlations because correlations found in aggre-gated data can be much smaller or even non-existent among individuals. Still, electoral districts are the lowest unit of analysis for which we can study actual election outcomes before we must turn to survey data. Therefore, the level of aggregation bias should be lower here in contrast to a study on a higher level involving, say, municipalities, counties or even countries. Nevertheless, it re-mains difﬁcult to say that the effect is driven by the suggested mechanisms.

Still, we should remember that individual-level data are only preferable if we can specify the individual-level model better than the aggregated one. This problem was described in the early studies on economic voting. For instance, Kramer (1983) suggested that aggregated time-series rather than survey-based studies was preferable given that individual-level estimates are biased by omit-ted variables and measurement error. Today, cross-sectional survey based studies are criticized because they give scholars little ability to adjust the es-timates for reverse causality (Evans and Pickup 2010). The great advantage of using aggregated data is that I can make my models more robust against such errors. More precisely, changes in wealth should not be driven by reverse causality if they are based on time-invariant shares at the district level and

(25)

ogenous national trends over time. However, this is not to say that the analysis is without problems.

One important caveat with the analysis is the assumption that the typical voter in a wealthy district is wealthy because the level of wealth at the district level can be driven by one or two ultra-wealthy households. We ﬁnd for in-stance, great inequality in stock-based wealth but this could be true for other asset classes as well. We simply cannot know from the data who the owners are and how much they own. This inability can become problematic since both individual wealth and the context of living in wealthy districts can have inde-pendent effects on voting. One may consider a theoretical example in order to understand why this is an important issue.

Let us assume that we want to know how changes in asset wealth affect voting using data at the district level. In this case, I test whether increasing asset values in the district increases district support for conservative candi-dates, assuming that this tells us about the effect of individual asset value on individual voting behaviour. But it is easy to imagine a competing story. One may suppose that people who are not better off, because, for example, they do not own assets, would see others in their neighbourhoods who are better off and become disaffected and therefore vote to express this dissatisfaction. For instance, as housing prices rise, homeowners become better off, but the group of renters remains worse off and the voting behaviour of both groups could be affected. If there was a positive rise in district house prices and 10 percent of the voters who are homeowners switched from the left to the right because they became better off, 10 percent of the voters who are renters could, on the other hand, switch to the left because they would become relatively worse off from this house price increase. The net change in voting shares, however, would be zero at the district level and an aggregated analysis would not help us understand voting behaviour in this case.

The magnitude of this problem depends on how transparent changes in wealth are for non-asset owners. Changes in the price of ﬁnancial assets are for instance, available to everyone since all citizens can follow the stock market in real-time. However, how much ﬁnancial assets your neighbor own remains a secret. Homeownership is harder to conceal but changes in house prices where on the other hand less transparent during this period. Property owners has an advantage since taxes and fees are based on the assessed value of their prop-erties. Renters are to a greater degree dependent on anecdotal evidence. Thus, voting driven by relative differences in wealth within districts will be based on highly speculative guesses. Still, it is not possible to rule out this competing explanation with aggregated data.

I have mentioned in my studies different ways of addressing this problem, the main strategy being, of course, to introduce individual-level data when it is possible. For instance, in the third study, I collected political variables from the SALTY survey and personal wealth data from the wealth register to verify that my argument holds among individuals. I also devote the fourth study to 25

(26)

investigate the relationships with individual-level data. This study does not contradict my main ﬁndings but suggests that wealth does affect voting at the individual level as well. Hence, the fourth study complements the aggregated studies well, but it should be noted that I do not claim with these tests that the problem with the main studies has been solved. What I do recognise, however, is that this is the extent of what is possible with available data. Thus, it is crucial that questions about household wealth and liabilities are included in the surveys that track political opinion and voting behaviour for this research ﬁeld going forward.

The Swedish Case

Sweden is known for its egalitarian welfare-state policies and for its even distribution of income. However, while the country has among the highest marginal taxes on income, it treats wealthy asset owners in a very different manner. This makes the country an interesting case to study. My sample cov-ers both local and national elections between 1998 and 2014. A Social Demo-cratic government held ofﬁce at the national level between 1998 and 2006; a liberal-conservative one between 2006 and 2014.

In 2007 and 2008, respectively, the wealth tax and the property tax were abolished by the right-wing government. The Social Democratic government, for its part, had abolished the inheritance tax and the gift tax. Government rev-enues from capital income declined further after the introduction of a special account in 2012 for ﬁnancial savings. Corporate taxes were lowered in several steps. This was a bullish period for asset owners, and the national wealth-to-income ratio more than doubled over that in the 1990s (Waldenström 2017).

It is easy to spot a pattern whereby governments of both political colours have encouraged households to accumulate assets. However, asset wealth is distributed less evenly than labour income in most economies, and this is true for Sweden as well. The median-income citizen today has less than half the savings of the average citizen, indicating that a small group with large savings pushes the average higher, while a typical Swede has considerably smaller savings. One explanation for this pattern is that a significant share of wealth in Sweden is inherited – up to 50 per cent of all private wealth, according to some estimates (Ohlsson et al. 2019). Today, the wealthiest 10 per cent of households own 75 per cent of all wealth, whereas the common Swede has very little in the way of financial assets. However, most households do save in mutual-fund shares, which also entails exposure to financial markets.

Housing wealth, on the other hand, is more evenly distributed, although treating property valuations as a perfect substitute for wealth is problematic, since home equity is harder to realise than are ﬁnancial assets. Nevertheless, homeowners can realise real-estate equity by taking out loans; or they can reduce their current savings and increase current cosumtion if they expect to 26

(27)

spend their housing wealth after retirement. Homeowners thereby have an in-terest in loose credit regulation and in low asset taxation. Yet again Sweden proves to be an interesting case, inasmuch as it takes a top position among Western nations when it comes to rising housing prices. However, high prices affect the level of debt as well, given that most citizens ﬁnance their homes with a mortgage. Such behaviour has made Swedish households heavily in-debted – among the most heavily in the world. Record-high housing prices and debt levels ensure that households are affected by changes in both asset prices and interest rates. The question is how far these results can travel.

The current literature does not give a straight answer to that question. Stubager et al. (2013) argue that voters in countries with small welfare states tend to fo-cus more on macro-economic indicators than on their own private economic affairs, given that the government has relatively little inﬂuence on personal ﬁ-nances. Voters in generous welfare states, on the other hand, should be more inclined to vote based on their self-interest, because the government plays a larger role in their lives. However, while Sweden has a generous welfare state, it has very different tax rates on wealth and on labour income, with the former being much lower. Inasmuch, then, as Sweden does not tax asset wealth or capital income very severely, it becomes hard to draw any conclusions from this perspective. Other studies suggest the Swedish case constitutes an arduous test for economic-voting models, and in particular for a wealth-based hypoth-esis. This is because a stronger effect of asset wealth on voting has been found in countries with a more liberal welfare system, such as the UK (Quinlan and Okolikj 2019).

However, Sweden constitutes a most likely case from a third perspective, because asset owners are said to become more likely to vote based on eco-nomic self-interest when there is a sharp policy divergence between the polit-ical parties (Hellwig and McAllister 2019). And during the period under here, there were indeed clear policy differences between the Swedish parties. For instance, the right-wing parties promised before the 2006 election to remove asset taxes if they were elected; in 2010, the left-wing parties promised to reintroduce them. Other important policy differences during this period con-cerned a significant reduction in income taxes, an increase in unemployment-compensation coverage fees, and reduced access to long-term sick-leave ben-efits. In the 2010 election, these policy reforms clearly divided the political parties along an economic left/right scale. My findings thus derive from a pe-riod when there was a clear economic left–right dimension in Swedish politics. It may not seem altogether clear, consequently, that my findings are pertinent to less polarised contexts.

To ease such concerns, I carried out some of my studies on the local rather than the national level. Local governments in Sweden have very little inﬂuence over either the national economy or asset taxation, but they are responsible for many welfare services. Municipalities provide two thirds of the welfare services in the country, and this of course generates local political divisions. 27

(28)

Thus, since I estimate effects both at the national and at the local level (which is a more neutral setting), it should be possible to generalise my results to other advanced economies.

Implications of the Findings

Recent research has questioned the validity of work on economic voting – and even the very existence of the phenomenon (Andersson 2008). This thesis suggests, on the other hand, that economic voting is actually flourishing, but that scholars need to focus more on personal wealth and less on labour market-related variables. For instance, my results indicate that voters can react to changes in their own pocketbook brought on by financial markets. Scholars should accordingly think carefully about different economic interests. This is important, because interests can diverge or even conflict in both valence and positional models of voting.

Some researchers on valence voting have incorporated asset markets into their studies. One example is (Larsen et al. 2019), who show that housing prices can predict incumbent support in Denmark; another is (Adler and Ansell 2020), who find that home prices can be used to predict support for populist parties. However, while these scholars connect housing prices with the gen-eral view of the economy held by citizens, my findings suggest that financial forces influence voters’ behaviour because changes in asset and credit markets affect their personal welfare. That is, changes in asset markets determine how much money individuals have to spend, which in the end affects their voting behaviour. Such contrasts should be seen in the light of the longstanding de-bate in political science on the factors that make voters act as they do. A large proportion of scholars within the field of economic voting have suggested that voters mainly react to changes in national conditions, rather than in their own personal circumstances. My results clearly show, however, that personal fi-nances are of equal importance here.

Where positional models are concerned, my results suggest that type of vot-ing exists as well, but that voters are far more complex than we had previously believed. For instance, voters who become wealthy due to rising asset values can simultaneously grow fearful of future losses if there is uncertainty in the market. Such uncertainty can reduce support for right-wing parties even if asset prices are rising. Moreover, while economic resources such as wealth can determine whether a given policy will make voters winners or losers, their inﬂuence over election outcomes may be weaker if non-economic issues are perceived as salient. Wealth may, for example, have a small impact on the average voter in a typical election, yet function more powerfully as a cue for left/right sorting when voters face a clear trade-off (if, for example, social spending is drastically increasing as a result of higher unemployment). In such a situation, the well-off will tend to favour the right, in the hope of seeing 28

(29)

government spending curtailed. However, my results also suggest that clear and visible threats to asset owners – taxes on wealth and property, for example – will give rise to more direct reactions.

Moreover, since voters may refrain from reacting due to the dual nature of their interests, economic variables are more likely to affect election outcomes when different economic forces are pushing voters in the same direction. For example, we can expect a greater impact on wealthy voters when asset prices are falling at the same time that unemployment is rising. It becomes harder, however, to predict election outcomes based on the economy when there are different trends in these variables. Future studies should try to determine how voters prioritise between different sides of the economy when economic trends diverge.

I will end this thesis with a shorter reflection on the bigger picture, i.e. how my results relate to the greater political discourse today. We find ourselves at the beginning of the 21stcentury, but one cannot help but note that current con-ditions are very similar to how things evolved hundred years ago, with grow-ing economic inequality and increasgrow-ing political tensions. However, current trends in Western politics represent a stark contrast to how inequality in the past often mobilised disadvantaged citizens into labour unions and left-wing parties (Boix 1998; Esping-Andersen 1979). Today, economic discontent is often blamed as the mechanism of interest behind the rise of the radical right, in that the losers of globalisation are said to flock around populistic policy platforms as a form of protest (e.g. Adler and Ansell 2020; Autor et al. 2020). The rise of Trump and Brexit has become especially central to this perspec-tive that exists both within and beyond academia. However, while this narra-tive can be comfortable since it links increasing political polarisation to fuzzy exogenous forces such as globalisation or technological progress, it overlooks the fact that much of the inequality has been generated at home by political de-cisions. As, today, much of the inequality is driven by changes in wealth rather than the income distribution and an important explanation behind this pattern is that wealth is not as heavily taxed (Piketty and Goldhammer 2014). This should perhaps be expected given that the wealthy do not like wealth taxes and politicians are more likely to appease the wishes of the affluent rather than the typical voter (Gilens and Page 2014; Powell and Grimmer 2016).

Future research must strive to investigate how to solve increasing tensions between the wealthy and the less affluent. It is noteworthy that much has happened since I started to write this thesis in 2016. Ansell (2020) has, for instance, linked relative differences in housing prices to growing populism in both the UK and France. Other studies have related financial variables to the rise of right-wing populism in both Poland (Ahlquist et al. 2020) and Hungary (Gyongyosi and Verner 2018) or even to the rise of the Nazi party in Ger-many (Doerr et al. 2019). Thus, we find a growing interest in wealth-related and financial variables in political science, but the need for further studies is urgent. A neglect of this topic can prove costly since greater polarisation is 29

(30)

linked to legislative gridlocks (Jones 2001), instability in the ruling cabinets (Warwick 1992) and even democratic breakdowns (Dalton 2008). One prob-lem, however, is that the wealthy and the asset-less seem to act in two different economies.

Asset prices are soaring towards new all-time highs as I write this ﬁnal reﬂection, but the real economy is struggling with record unemployment. The long food lines seen in both the US and Europe also suggests that the gap between the have and have nots has reached massive levels and that voters face different economic forces. Political tensions are set to rise if voters are guided by an economic rationale as suggested in my studies. Thus, scholars on economic voting and social policy preferences need to have access to better data at the individual level in order to fully understand the mechanisms behind political polarization. Still, much can be done at the country level while we wait for more reliable individual-level data.

Comparing countries can on the one hand, tell us less about the causal mechanisms of voting behaviour. It can, however, teach us about the policy implications from rising wealth inequality. It is crucial here to understand when and how governments can impose asset taxation without generating large political conflicts. There is also plenty of data on the wealth distribu-tion among western nadistribu-tions, especially the global wealth report provided by Credit Suisse can be of value. This report covers the level and distribution of wealth in over 200 countries during the 2000–2020 period. It is in the end, crucial for us to understand the source of the rising political conflicts. In fact, I believe that political scientists have an obligation to do what we can to deflate these tensions.

(31)

Bibliography

Adão, Rodrigo, Michal Kolesar, and Eduardo Morales (Aug. 2019). “Shift-Share Designs: Theory and Inference”. The Quarterly Journal of Economics 134.4, pp. 1949–2010.

Adler, David and Ben Ansell (2020). “Housing and populism”. West European

Politics 43.2, pp. 344–365.

Ahlquist, John, Mark Copelovitch, and Stefanie Walter (2020). “The Politi-cal Consequences of External Economic Shocks: Evidence from Poland”.

American Journal of Political Science 64.4, pp. 904–920.

Alesina, Alberto (1987). “Macroeconomic Policy in a Two-Party System as a Repeated Game*”. The Quarterly Journal of Economics 102.3, pp. 651– 678.

Alesina, Alberto and Howard Rosenthal (1995). Partisan politics, divided

gov-ernment, and the economy. Cambridge: Cambridge University Press.

Anderson, Christopher J. (2007). “The End of Economic Voting? Contingency Dilemmas and the Limits of Democratic Accountability”. Annual review of

political science 10.1, pp. 271–296.

Ansell, Ben (2014). “The Political Economy of Ownership: Housing Markets and the Welfare State”. American Political Science Review 108.2, pp. 383– 402.

Ashworth, Scott (2012). “Electoral Accountability: Recent Theoretical and Empirical Work”. Annual review of political science 15, pp. 183–201. Autor, David, David Dorn, Gordon Hanson, and Kaveh Majlesi (2020).

“Im-porting Political Polarization? The Electoral Consequences of Rising Trade Exposure”. American Economic Review 110.10, pp. 3139–83.

Bartels, Larry M. (2002). “Beyond the Running Tally: Partisan Bias in Politi-cal Perceptions”. PolitiPoliti-cal Behavior 24.2, pp. 117–150.

Blyth, Mark (2002). Great transformations: economic ideas and institutional

change in the twentieth century. New York: Cambridge University Press.

Boix, Carles (1998). Political Parties, Growth and Equality: Conservative and

Social Democratic Economic Strategies in the World Economy. Cambridge

Studies in Comparative Politics. Cambridge University Press.

Broxterman, Daniel A. and William D. Larson (2020). “An empirical ex-amination of shift-share instruments”. Journal of Regional Science 60.4, pp. 677–711.

Brunnermeier, Markus K. and Isabel Schnabel (2016). “Bubbles and Central Banks: Historical Perspectives”. Central Banks at a Crossroads: What Can

We Learn from History? Cambridge, UK: Cambridge University Press.

(32)

Chwieroth, Jeffrey M. and Andrew Walter (2019). “The Wealth Effect”. The

Wealth Effect: How the Great Expectations of the Middle Class Have Changed the Politics of Banking Crises. Cambridge University Press.

Chzhen, Kat, Kat Chzhen, Geoffrey Evans, Geoffrey Evans, Mark Pickup, and Mark Pickup (2013). “When do Economic Perceptions Matter for Party Ap-proval?: Examining the Endogeneity of Economic Perceptions Before and During the Economic Downturn”. Political Behavior 36.2, pp. 291–313. Clegg, Daniel (2004). “The Politics of Social Risk: Business and Welfare State

Development”. Journal of Social Policy 33.3, pp. 510–511.

Cusack, Thomas, Torben Iversen, and Philipp Rehm (2006). “The demand and supply sides of government redistribution”. Oxford Review of Economic

Policy 22.3, pp. 365–389.

Dalton, Russell J. (2008). “The Quantity and the Quality of Party Systems: Party System Polarization, Its Measurement, and Its Consequences”.

Com-parative Political Studies 41.7, pp. 899–920.

Dassonneville, Ruth and Michael S. Lewis-Beck (2013). “Economic Policy Voting and Incumbency: Unemployment in Western Europe”. Political

Sci-ence Research and Methods 1.1, pp. 53–66.

Di Tella, Rafael and Robert MacCulloch (Apr. 2005). “Partisan Social Happi-ness”. The Review of Economic Studies 72.2, pp. 367–393.

Doerr, Sebastian, Stefan Gissler, José-Luis Peydró, and Hans-Joachim Voth (2019). From Finance to Fascism. Working Papers 1092. Barcelona Gradu-ate School of Economics.

Downs, Anthony (1957). An economic theory of democracy. New York: Harper & Row.

Durr, Robert H (1993). “What Moves Policy Sentiment?” American Poltical

Science Review 87.1, pp. 158–170.

Esping-Andersen, Gösta (1979). “The Working Class in Welfare Capitalism: Work, Unions and Politics in Sweden. By Walter Korpi.” American Political

Science Review 73.3, pp. 905–907.

Evans, Geoffrey and Mark Pickup (2010). “Reversing the Causal Arrow: The Political Conditioning of Economic Perceptions in the 2000–2004 U.S. Presidential Election Cycle”. The Journal of Politics 72.4, pp. 1236–1251. Fauvelle-Aymar, Christine and Mary Stegmaier (2013). “The stock market and

U.S. presidential approval”. Electoral Studies 32.3, pp. 411 –417.

Fiorina, Morris P. (1981). Retrospective voting in American national elections. New Haven: Yale U.P.

Flodén, Martin, Matilda Kilström, Jósef Sigurdsson, and Roine Vestman (2017).

Household debt and monetary policy: Revealing the cash-ﬂow channel.

eng. Sveriges Riksbank Working Paper Series 342. Stockholm.

Foucault, Martial, Richard Nadeau, and Michael Lewis-Beck (2013). “Patri-monial voting: Reﬁning the measures”. Electoral Studies 32.3, pp. 557– 562.