1.1 Overview

This Element will establish the crucial role of the 2008 financial crisis for the populist wave in Europe. We focus first on how the crisis affected voters across different occupations. And then on the consequent structural transformation of political parties and movements. Our work contributes to several strands of literature: how old and new parties compete in terms of policy platforms as a function of changes in voters’ preferences and beliefs; what determines the preferences of different occupational groups on economic and cultural dimensions of politics; and what shapes their beliefs about the value of different policies. While focusing simultaneously on both voters and parties is not unique to this Element, our conceptual framework offers a novel interpretation by emphasizing the crisis’s effects on beliefs about the feasibility and desirability of different policies. This focus provides the key to our interpretation and helps us maintain a coherent view of the broader picture.

Economic crises tend to generate economic insecurity and erode trust in existing institutions – both market and governmental. However, while the economic threats posed by immigration, globalization, and automation primarily impact low-skilled workers and their communities, financial crisis tend to have distinct effects on groups that are more exposed to dependent from financial markets, such as the middle class. We illustrate this empirically using a novel methodological approach. The extension of distrust to include the middle class is a critical development, as it signals to politicians and parties that winning over this broader group of disillusioned voters can be decisive for electoral success. Consistent with this intuition, we show that the transformation of the political supply, in terms of new and transformed parties and manifestos, happened right after the financial crisis. Moreover, the transformation and the increase in populist voting happened more evidently in countries plagued by low fiscal space, where the feasibility of inclusive welfare redistributive policies is lower. This, in turn, increases the appeal of exclusionary policy commitments, which are more easily championed by populist parties on the extreme right of the ideological spectrum.Footnote ¹

The shrinking fiscal space in many European countries made the financial crisis particularly conducive to the rise of right-wing populist parties. The reason is straightforward: Identity-based protection policies typically require limited public spending, whereas the redistributive welfare policies advocated by left-wing populists became increasingly infeasible and less credible – despite remaining ideologically appealing to a significant portion of the population. As a result, the shift toward right-leaning economic policy preferences can be partly attributed to the perceived decline in the feasibility of more inclusive welfare policies.

Shifts in beliefs about the relative feasibility of left- and right-leaning economic policies have also influenced cultural-political cleavages. Macroeconomic and institutional constraints on fiscal policy have altered perceptions of self-interest about public policies, contributing to polarization along the cultural dimension. When individuals lose faith in the feasibility of inclusive welfare policies – particularly those benefiting minorities – or perceive such policies as misaligned with their own self-interest, they are more likely to support exclusionary measures that heighten cultural tensions. As belief grows that anti-immigration and nationalist identity-protection policies are the most effective means of safeguarding personal interests, support increases for politicians who promote anti-pluralism and oppose values such as diversity, equity, inclusion, and minority rights.

1.2 Some Preliminary Look at the Different Crises

We view the 2008 financial crisis and the resulting great recession as a major factor in the spread of populism in Europe during the twenty-first century. It happened before: One of the first examples of a populist policy in history is that of Urukagina, who seized power in Lagash – a city-state of ancient Mesopotamia – by dethroning his corrupt predecessor Lugalanda. This ruler promised to restore the power of ordinary people during an economic downturn while highlighting his predecessor’s lack of credibility. To gain consensus, Urukagina forgave all loans, relieving debtors while demonizing financiers. He lumped usurers with criminals to be cast out of the city (see Goetzmann, Reference Goetzmann2015).

Much of the economic literature on populism has sought to examine the effect of immigration, globalization, and automation on job destruction, primarily among low-skilled workers. This is thought to have created voter disillusionment in liberal democracies, gradually transforming the demand for policies. In this context, the financial crisis has been treated as yet another factor that enhanced voter appetite for populist policies as a response to economic insecurity. However, this approach has tended to ignore the peculiar effects and consequences of the financial crisis on the middle class.

The effects of globalization and automation on some segments of society are not uniformly negative: There are losers, but undoubtedly there are also winners. Apart from the inequality generated by job destruction and lower wages among workers affected by foreign competition or automation, consumers as a whole have benefited from lower prices of final goods and firms’ intermediate inputs.

This cannot be said of financial crises. It is difficult to think of a beneficial effect that a financial crisis-induced recession might have. Thus, most people, across the entire spectrum of the voter population, lose out. Income losses tend to be deep and universal. Hence, the discontent fostered by the resulting economic insecurity tends to be more pervasive and thus more politically relevant. Moreover, the first thing to note about the financial crisis is that the economic insecurity following the onset of the crisis in 2008 spread to segments of the population that were less affected by the globalization shock. Figure 1, panel (a) shows the share of blue-collar and non-blue-collar workers in the top quartile of economic insecurity in each year of our sample. Prior to the financial crisis, when globalization was the main source of economic insecurity, blue-collar workers were those predominantly experiencing a high level of insecurity (66% on average); during the years after 2008, the share of non-blue-collar increases substantially (by more than 8 percentage points) compared to the years prior to the financial crisis.

Figure 1

Insecurity among blue-collar workers and members of the middle class

Figure 1 Long description

Figure 1 contains two line graphs on economic insecurity. Panel (a), “Blue Collar,” plots the share of blue-collar and non-blue-collar workers in the top quartile of economic insecurity between 2002 and 2016. The blue-collar line remains consistently higher, fluctuating around 0.56–0.68 and declining slightly after 2008. The non-blue-collar line ranges around 0.31–0.44, rising gradually after 2008 and narrowing the gap with blue-collar workers. Panel (b), “Middle Class,” shows the share of people in the middle 50% of the income distribution experiencing insecurity. Values rise from about 26 in 2004 to a peak above 35 around 2010, then gradually decline to about 32 by 2016.

Notes: Panel (a) plots the share of blue-collar and non-blue-collar workers in the top quartile of economic insecurity during the sample period. Panel (b) shows the share of people in the mid-50% of the distribution of income in each country-wave.

Panel (b) shows that the financial crisis also increased economic insecurity among the middle class, as shown by the average share of people in the mid-50% of the distribution of income in each country-year. The share of middle-class voters suffering from serious economic insecurity, that is, in the top quartile of insecurity, climbs during the years of the Great Recession. Thus, the financial crisis not only increased insecurity among social strata that were already distressed by globalization and other processes before the crisis (namely, low-skilled workers at the bottom of the income distribution), but it also spread insecurity among segments of the population that had been relatively sheltered from globalization.

In advanced Western countries, where both firms and households are heavily dependent on financing, a financial collapse can represent an existential threat. Obtaining credit becomes more difficult in a financial crisis, as markets stop working smoothly and financial constraints become tighter. In addition, the accompanying fall in asset prices reduces the value of precautionary savings, thus limiting people’s ability to cope with economic insecurity. On the contrary, up until the financial crisis, financial markets were intact and credit was abundant, implying that some of the people hit by the first wave of globalization threats could use borrowing or their own savings when asset prices were still high. The dramatic spike in the bond spread during the years of the European sovereign debt crisis demonstrates quite clearly how difficult it became for governments to secure funding to run their programs during the financial crisis relative to the globalization years.

The different composition of the population feeling economic insecurity is only one effect. Another factor – more difficult to measure but intuitively central to our argument – is the level of confidence in the existing institutional framework and in the credibility of government objectives and traditional policy commitments. Any rational agent applying standard Bayesian updating must be revising downward such a confidence when a large crisis such as the Great Recession finds the status quo elites unprepared (or complicit). Our novel theoretical framework will allow us to explicitly consider these additional differences, and the theory will guide an empirical analysis that will add important nuances to the simple first observation of Figure 1.

1.3 Introducing the Theoretical Framework

As explained in Bellodi et al. (Reference Bellodi, Morelli, Nicolo and Roberti2025), there are clear reasons – from standard principal-agent logic to endogenous changes in accountability – to expect that a decline in trust in governments and politicians determines what they call a “shift to commitment politics,” whereby the trustee model of political agency is replaced by agents proposing implicit contracts based on simple commitments made ex ante.Footnote ²

The shift to commitment politics generates as an epiphenomenon the strategic use of anti-elite rhetoric and identity protection rhetoric for the people, which are the essential parts of observable political strategies that constitute the definition of populist vs non-populist parties in the political science literature. Moreover, as argued formally in Bellodi et al. (Reference Bellodi, Morelli, Nicolo and Roberti2025), it is also intuitive that parties and politicians shifting more and more to policy commitments desire to weaken the checks and balances, bureaucrats, judges, experts, and intellectuals: because they all could constitute obstacles to the implementation of commitments when proved to violate constitutional or international laws (like in the case of deportations).

The theory presented in Section 2 of this Element develops an important feature of the populist commitments: they are almost always “exclusionary commitments” of protection – protectionism on markets, protection from immigrants, identity protection, nationalism, protection from external influences, and even from international organizations and laws.

The financial crisis has played a crucial role in reshaping political preferences. For individuals weighing a policy agenda centered on open borders, open markets, and inclusive policies against one favoring closed borders, protected markets, and the rollback of inclusion efforts, the crisis disproportionately undermined confidence in the former. In contrast, the latter – an agenda of closure – can come to be seen as not only viable but necessary, particularly when the crisis is perceived as stemming from external forces and internal responses are constrained by limited fiscal space.

We capture all the above complex logical connections with a simple theoretical model that clarifies the determinants of both voters’ policy preferences between open market status quo and various types of protectionist commitments as well as voters’ participation decisions given their abilities and ideologies. The simple model will clarify why one should expect the financial crisis to induce differential turnout effects between left and right and thereby cause the entry and success of right-wing populist protectionist parties.

To anticipate some important details of the theory, we assume that individuals are heterogeneous in ability (and hence type of occupations they can obtain), cost of voting, and ideology. The expected utility of an individual if the open market status quo is kept depends both on the probability of being a winner from the open market and on the probability of being bailed out or compensated by institutions when happening to be a loser. Hence, this expected utility (which we also call economic security) can be different across occupations. Ideology enters the picture when considering an alternative policy commitment to protect from the open market, which typically is a bundle of closures, from closed markets to closed borders and beyond. Such closure policy commitments typically give a higher expected utility to right-wing individuals, since left-wing individuals are typically closer to the moral universalism description of Enke et al. (Reference Enke, Rodríguez-Padilla and Zimmermann2023) and hence have some moral disutility from policies that benefit a subset of the population (the “citizens”) while excluding others (the “immigrants”). Thus, when a crisis arrives and reduces the appeal of the open status quo, right-wing individuals can more readily embrace populist protection commitments, whereas left-wing individuals equally disappointed by the status quo may instead choose to abstain. This differential incentive to participate in elections is what triggers the increased probability of winning of right-wing populist parties after the financial crisis.

1.4 Empirical Methodology

On the voters’ side, we use the European Social Survey (ESS), which is conducted in repeated cross-section waves. The data is used to produce an instrumented pseudo-panel analysis of the economic and financial drivers of change in political attitudes and voter behavior in Europe during the financial crisis and subsequently. Each element of the pseudo-panel is a synthetic cohort of individuals and can be associated with a distribution of occupations. This survey data will be complemented by labor market data that will allow us to construct a methodology for identifying the causal effects of the financial crisis, exploiting heterogeneity in the distribution of occupations across cohorts.

In order to establish a causal link between the financial crisis and economic insecurity, trust in politics, and voter abstention, we build a novel instrumental variable that leverages the idea that financial crises are most damaging to individuals who depend more on borrowing to buffer income shocks and thus to manage economic insecurity. Dependence on borrowing is a function of the steepness of an individual’s age-earnings profile: people with a steeper profile must rely more on borrowing to smooth consumption, which makes them more vulnerable to financial shocks.

The cohorts of respondents to the ESS over time and across countries vary in their distribution by occupation. We find marked differences between occupations in the steepness of the age-earnings profiles. Thus, there will be heterogeneity in the sensitivity to a financial crisis across occupations and hence cohorts. We construct a shift-share instrument where the shifter is the aggregate economic shock affecting a country, the share is the weighted average sensitivity/dependence on borrowing by cohort, and the weights are the shares of each occupation within the cohort.

On the supply side, we document in multiple ways the death of traditional parties, birth of new parties, and transformation of the manifestos of surviving parties, giving a comprehensive picture of the transformation of the whole landscape of political platforms after the financial crisis.

In order to define a European party as populist in a given year, we rely on the PopuList proposed by Rooduijn et al. (Reference Rooduijn, Van Kessel, Froio, Pirro, De Lange, Halikiopoulou, Lewis, Mudde and Taggart2019), which is available at www.popu-list.org. This classification is consistent with both the ideational approach and the political strategic approach to defining populism.Footnote ³ Accordingly, the people-vs-elite rhetoric is a straightforward manifestation of populism, which is also the most common measure used in empirical work (Pauwels, Reference Pauwels2011; Hawkins and Littvay, Reference Hawkins and Littvay2019; Gennaro et al., Reference Gennaro, Lecce and Morelli2024).

To give a first sense, just looking crudely at the number of coded populist parties entering, we can already get a sense of the change: Figure 2 plots the average number of populist parties participating in elections up to 2008 and in subsequent years. It is clear that the Great Recession marks a watershed in terms of the supply of populist parties competing for voters.

Figure 2

Populist parties

Figure 2 Long description

Before 2008, the blue line fluctuates between 1 and 1.7 with no clear trend, and the red average line sits near 1.6. After 2008, the number rises sharply, spiking to about 3.3 around 2012 before settling between 2 and 2.5 in subsequent years. The red average line steps up to approximately 2.2, indicating a structural break in the supply of populist parties after the financial crisis.

Notes: The graph shows the number of populist parties (in blue) and the average number of populist parties (red line) actively participating in elections.

Up until 2008 an average of 1.7 populist parties took part in an election with no clear trend over time. In the years following 2008, the average number of participating populist parties jumps to 2.4 – a 33% increase compared to the precrisis mean – with a spike in the 2012 elections. The financial crisis appears to mark a structural break in the supply of populist parties. And this is just looking at quantity, but, as we will see, content also changes in directions consistent with our arguments within long-lived parties.

The Manifesto Project dataset (Volkens et al., Reference Volkens, Burst, Krause, Lehmann, Matthieß, Regel, Weßels, Zehnter and Sozialforschung2018) will be useful because with the use of Lasso regressions we will be able to trace the most important topic shifts and position shifts directly from the texts. We will complement the analysis on party manifestos over time with the insights from expert surveys, which are useful especially to detect major items of convergence or divergence between populist and non-populist parties on both sides of the ideological spectrum.

1.5 Main Empirical Findings

Our main empirical finding is that economic insecurity triggered by the financial crisis had a causal effect on turnout and on voting choices. Financial shocks specifically hamper the ability to borrow, which particularly affects segments of the middle class characterized by steep age-earnings profiles. This, in turn, significantly expands the pool of voters seeking economic support and who are at the same time losing confidence in the ability of the government to deliver such support. In view of this disappointment in traditional politics, there is an increase in voter abstention, especially among those who had not previously been seriously affected by globalization. Meanwhile, those who were subject to previous crises as well are the most likely to vote for a populist party, especially if ideologically not opposed to exclusionary policy commitments.

The results show that the instrument has a strong predictive power on both the self-reported as well as wage-based measures of economic insecurity. In turn, in IV regressions (that also control for cohort fixed effects as well as for country and time fixed effects), shocks to instrumented economic insecurity tend to cause a reduction in voter turnout, especially among left-wing voters.Footnote ⁴ The shock also increased populist voting (conditional on turnout). The effects on voter behavior along all three dimensions are significant in magnitude: an increase of one standard deviation in economic insecurity increases populist voting by 7 percentage points, which is about 94% of the sample mean, and it lowers turnout by more than 8 percentage points (about 10% of the sample mean).Footnote ⁵ We will show that indeed back-of-the-envelope calculations support our financial crisis mechanism, since roughly 70% of the change in voting for populist parties comes from the middle class.

The above numbers suggest that the financial crisis may indeed have been the moment of maximum entry and transformation of parties on the supply side, which in turn gave disillusioned voters new hope. To investigate this hypothesis, we conduct a novel analysis of the dynamics of the supply of populist parties in Europe. We first examine the manifestos of all European parties, distinguishing between long-lived parties (existing both before and after the financial crisis) from the parties that disappeared or emerged at the time of the crisis. The quantitative structural break documented already in Figure 2 is corroborated by a number of findings on contextual interactions and textual content.

In terms of contexts, the party structure transformation phenomenon occurred particularly in countries with a shrinking fiscal space, which is in line with the expectations given our theory and the considerations of credibility of policies mentioned above. A shrinking fiscal space has dramatically different effects on left-oriented versus right-oriented parties. It also reduces the feasibility and thus the credibility of protection policies that are focused on the supply of expensive public goods and on redistributive spending. Hence, the traditional leftist parties suffered the most in terms of credibility, leading to the exit of traditionally leftist parties and the entry of new parties such as the Five Star Movement in Italy and Podemos in Spain. The nationalism and identity protection policies championed by right-wing parties did not need to be altered because they were less dependent on the level of public spending. Thus, nationalist or far-right ideologies do not need to change in a financial and/or fiscal crisis; in contrast, such crises force a shift on the left from (unaffordable) redistributive platforms towards protection rhetoric and pre-distribution. This creates an advantage for right-leaning parties, which are able to advocate the same policies and measures that depend less on public spending.

In terms of textual content, we corroborate the insights on party transformation using expert survey data on topics and salience shifts to see whether non-populist parties also modified their platform and in which direction. We show that there is a postcrisis convergence, with non-populist parties modifying their platforms in the direction of populist ones.

1.6 Contribution to Comparative Politics Literature

Our contribution to the comparative politics literature is on multiple levels. First, we contribute to the understanding of the effects of economic shocks of different nature on the political preferences of voters (see Margalit (Reference Margalit2019) for a review); second, we contribute to the documentation of changes over time in the political attitudes of occupational groups and the corresponding evolution of parties in the electoral space in the last decades in Europe, thereby complementing the work in the Element by Hall et al. (Reference Hall, Evans and Kim2023); third, we emphasize the key role of the financial crisis in the determination of the success of populists, also due to the strong incentives of political entrepreneurs to politicize this issue that captured and interested the middle class – a point well illustrated by De Vries et al. (Reference De Vries, Hobolt and Walter2021). It is useful to zoom separately on these three contributions.

1.7 Contribution to the Political Economy of Populism Literature

The conceptual framework developed in Bellodi et al. (Reference Bellodi, Morelli, Nicolo and Roberti2025) and the theory proposed in Section 2 of this Element, which adapts the logic of commitment politics to the differences we want to explain, together push the view that populist policy commitments are a political strategy. The typical features of populist platforms can be traced back to the shift to commitment politics, which in turn emerges naturally at a time of increasing distrust, which is itself shown to be endogenous.

Technological change, and in particular the use of the internet and social media for political communication, have endogenously reduced trust in political agents, and this reduction in trust has increased the use of simple policy commitments; in turn, once political campaigns become more and more focused on policy commitments to voters, this determines the rational use of typical populist strategies such as anti-elite rhetoric, misinformation, and rhetoric and actions against judges, bureaucrats, and all checks and balances to make the executive as free as possible to implement the promised policies.Footnote ⁹

Our theory presented below makes this general logic more specific to the financial crisis by zooming in on the particular type of policy commitments that we can call commitments to protect. The financial crisis makes governments, and especially those with low fiscal space, unable to continue to support traditional welfare programs. This makes those affected by the market shocks shift to demand of simple exclusionary commitments.

There is an immense literature on the economic and cultural causes of populism. The survey article by Guriev and Papaioannou (Reference Guriev and Papaioannou2022) contains almost all of them. Gidron and Bonikowski (Reference Gidron and Bonikowski2013), Mudde and Kaltwasser (Reference Mudde and Kaltwasser2017), and especially Norris and Inglehart (Reference Norris and Inglehart2019) highlight the relevance of cultural backlashes, which clearly occur simultaneously with an increase in economic insecurity. Both economic and cultural factors matter, but here we want to zoom in particular on the strands of this literature related to the financial crisis and to the role of the fiscal space. Foster and Frieden (Reference Foster and Frieden2017) present the correlation between distrust measures and debt using the Eurobarometer survey. Algan et al. (Reference Algan, Guriev, Papaioannou and Passari2017) show that in elections after 2008, the regions where unemployment rose saw the sharpest decline in the trust placed in institutions and traditional politics,Footnote ¹⁰ while Dustmann et al. (Reference Dustmann, Eichengreen, Otten, Sapir, Tabellini and Zoega2017) show that in the aftermath of the crisis, the distrust of European institutions was correlated with the populist vote. Using the age-earnings profiles typical of different occupations, we are able to directly identify the channel through which the specific features of the financial crisis (primarily the inability to borrow) impacted each cohort of citizens in Europe. We are therefore able to provide evidence of causal effects by differentiating such effects across cohorts with different occupational distributions.

There is an interesting literature on financial crises as determinants of extremism (see, for example, Funke et al. Reference Funke, Schularick and Trebesch2016 and Galofré-Vilà et al. Reference Galofré-Vilà, Meissner, McKee and Stuckler2021).Footnote ¹¹ Our analysis of the 2008 financial crisis emphasizes their impact on the political orientation and feasibility of policy platforms on either side of the political spectrum, rather than focusing on extremism specifically.Footnote ¹² Voth et al. (Reference Voth, Doerr, Gissler and Peydró2020) were the first to present causal evidence that a financial crisis can fan extreme right populism, based on variation in failing banks in Germany prior to the 1932 elections. In contrast to their findings, we show that the 2008 financial crisis shifted politics in the direction of populism on a broader scale, rather than only on the right. Indeed, the transformation after 2008 occurred primarily on the left.Footnote ¹³ Furthermore, our method of identification makes it possible to zoom in on the heterogeneous impact of the financial crisis by occupation and to identify the mechanism that creates economic insecurity, regardless of preexisting anti-Semitism or some similar types of ideology.

On the relevance of the fiscal space, Arias and Stasavage (Reference Arias and Stasavage2019) and Fetzer (Reference Fetzer2019) look at the political costs of austerity politics, although they ignore the dynamic transformation we focus on. The role of fiscal constraints in shaping political trust and populist support has also been documented in the Latin American context by Baccini et al. (Reference Baccini, Hicks and Rettl2025). Indeed, we find that it is precisely in the countries with the smallest fiscal space that the financial crisis had a greater impact on politics (and especially on the left).Footnote ¹⁴

Rodrik (Reference Rodrik2018) traces the increase in populism to the globalization shock. While this may be true when considered in isolation and for specific events,Footnote ¹⁵ Guiso et al. (Reference Guiso, Herrera, Morelli and Sonno2019) show that globalization shocks alone cannot account for the cross-country pattern of populism in Europe. They show that the interaction of globalization with a euro-dummy accounts for the large majority of the explanatory power, and, in the presence of such an interaction variable, globalization shocks alone have little effect. In Section 7, we show that whereas the supply of populism displays a discontinuous jump in 2008, there was no similar increase in 2004, the year of the globalization shock due to the expansion of the EU. One may argue that, in a deep sense, the globalization shock is the root cause of populism, because a macro literature traces the burst of the crisis to the imbalances that originated from globalization. What we show is that a globalization that does not involve a financial crisis is unlikely per se to generate the spread of populism that we have witnessed in Western countries, particularly in countries that are heavily dependent on credit and financial markets.

Moreover, while globalization primarily reduced trust in free markets, the conjunction of the financial crisis and shrinking fiscal space reduces trust in all other institutions as well. It is the collapse of confidence in representative democracy that increases the demand and supply of simple protection commitment policies such as walls, protectionism, and Brexit.

The rest of the Element is organized as follows: Section 2 draws the conceptual framework that generated all our ideas on the role of all crises, and in particular the financial crisis, on voters’ policy preferences and on asymmetric turnout. Parties enter the picture simply in terms of whether they are likely to adopt a populist protectionist commitment or not and why. Section 3 describes our data and measurements. Section 4 describes the important role of voters’ participation incentives and displays descriptive results that are in line with the asymmetric turnout effects of crises that we explain in the theory. Section 5 describes our empirical methodology and identification strategy, and Section 6 displays causal estimates of the effects of the financial crisis on all aspects of voters’ preferences and behavior. section 7 uses manifestos of parties to show the different effects of globalization and the financial crisis on the party structure and policy platforms. Section 8 analyzes strategic entry and strategic positioning by parties in the aftermath of the crisis and looks at the convergence of platforms across parties on each side of the political spectrum using expert surveys. Section 9 concludes with some remarks on the persistence of some of the effects of the financial crisis.

2.1 Voters’ Preferences on Closed vs Open Markets

We start from a voter’s evaluation of the tradeoffs involved when considering her expected utility from an open market status quo versus a protectionist reform that would cut out competitors to the voter’s employing firm or potential employing firms, reducing also the instability and market dependence on prices and wages.Footnote ¹⁶ The perception of threat to jobs and wages created by the technological changes that were permitting greater and greater open market integration was the central cause of populist voting already in the United States in the last quarter of the nineteenth century.Footnote ¹⁷ The globalization literature on the role of the China shock in the determination of populist voting in many Western democracies in the twenty-first century gave rise to an analogous tradeoff. The same tradeoff must have been a central one in the decision by American voters to vote or not to vote for Donald Trump, who made protectionism a central pillar of his campaign(s). Since this key tradeoff has been at the center of so many studies on the fears and insecurities that led to populist voting, let us zoom on this tradeoff.

When evaluating her prospects in an open market status quo, we can simplify the voter’s thought process as follows: she has a belief $p\in [0,1]$ to be a winner from the open market and a probability $(1-p)$ to be a loser from it. Let us normalize the utility for a winner to 1 and the loser’s utility to 0.Footnote ¹⁸ However, in the open market status quo, with a representative democracy determining the functioning of government institutions, the voter expects that in case she turns out to be a loser from the market process, she will be compensated by such institutions with probability $q\in (0,1)$. For example, if a voter expects that her government has both “the objective” to protect the losers from market competition with redistribution or public jobs and “the capacity” to do so, then $q$ is a high number; whereas if one expects the government either to be attentive to other objectives (or interest groups’ influences) or not to have the capacity to compensate, then $q$ is a low number. Putting things together, the expected utility from an open market status quo ($s$) with a representative government that could, in principle, try to compensate losers is

$U(s)=p+(1-p)q.$(1)

We can think of $p$ as representing the voter’s self-confidence in her skill and the fit of such skills for the open competition. Hence, $p$ can be very different from voter to voter, because it depends on education, acquired skills, psychology, and so on.Footnote ¹⁹ The second parameter, $q$, captures in a nutshell “trust” in the government, where once again trust can be high or low depending on both the government or politicians’ objectives and incentives on one hand and capacity on the other. One might have low trust because she does not think that politicians will even care and try to compensate, derailed by other objectives, or may think that the objectives are aligned but there are not enough resources to effectively counter the negative market prospects of a market loser. We will come back to these components of trust when we introduce the mechanism through which the financial crisis and fiscal space interact.

A protectionism reform can be captured for simplicity by a fixed expected utility $r\in (0,1)$ that the voter expects if all external competitors are shut out of the market.Footnote ²⁰ The expected utility $r<1$ parsimoniously captures the perception of many workers and/or firms that, on the one hand, the market winner’s payoff is higher in an open market, but, on the other hand, with no competitors, her industry, job, and wage are sheltered, and hence the low loser’s payoff can be avoided.

Thus, a voter prefers a protectionist commitment iff

$r>p+(1-p)q.$(2)

Hence, a voter is the more likely to support a protectionist commitment the lower are her confidence $p$ and her trust $q$, and the higher is her expected utility from protectionism $r$. Even though in reality voters are heterogeneous in all these three parameters, let us simplify the analysis by assuming $q$ is fixed for all; and as far as $r$ is concerned, let’s assume for simplicity that it can take two values, $r^r$ and $r^l$, where $r^j$, $j=l,r$ represents the expected utility of the protectionist policy for voters of ideology $j$, where $j=l(r)$ denotes the standard left(right) ideology. Given that the populist exclusionary commitments typically include not only protectionism in trade policies but also closed borders and nationalist policies on multiple dimensions, we can safely assume that $r^r>r^l$: The typical left ideology dislikes exclusions, closed borders, closure to other cultures, and so on; hence, a policy bundle based on exclusions and closures gives lower overall utility to left-wing voters. Assuming a fixed share $\alpha \in (0,1)$ of left-wing voters in the population, we denote by $\hat{r}\equiv \alpha r^l+(1-\alpha )r^r$ the average utility from a bundle of closure policies.Footnote ²¹ Finally, we let $p$ be distributed on the interval [0,1] with density $f(\cdot )$. Assuming that a polity has a unit mass of voters, the standard economic belief that free trade is good in the aggregate can be captured by the following assumption:

Market Efficiency Assumption (MEA): ${\int }_0^{1}p+(1-p)qf(p)dp>\hat{r}$.

Even if all voters believe in the MEA, consistent with most economic studies, it is clear that those for whom (2) holds prefer the protectionist commitment.

2.2 The Effects of Different Types of Crises

A globalization shock, caused, for example, by a technological change making transportation of goods easier or by innovations and new competitors (humans or robots) who appear as external threats, typically determines a reduction in $p$ for some classes or occupational groups. Thus, the common opinion that globalization shocks like the entry of China into the WTO may have been a trigger of populism can be parsimoniously captured by this heterogeneous reduction in $p$ values in the model. In the binary simplification below, where $p$ can take only two values, $\underset{\_}{p}$ and $\overline{p}$, the globalization and automation shocks affect mostly $\underset{\_}{p}$, namely the probability of success for low-skill individuals (or individuals working in exposed industries).

A financial crisis such as the Great Recession affects all levels of $p$ and $q$: an economic agent who was counting on good access to reliable financial markets and banking for her economic prospects and future now sees the probability $p$ of the good outcome go down and, at the same time, perceives a lower $q$ that market and government institutions can “bail” her out in case of a negative realization of economic conditions. The financial crisis made most market operators tighten credit, and government institutions had a hard time helping the many losers. Thus, and naturally more so in countries with very low fiscal space, $q$ must have been shocked downward. Given that $q$ applies to all, this common component of the crisis effects is what relates in the model to the enlargement of the set of people for whom (2) starts holding after the crisis.

The importance of the trust parameter $q$ can also be appreciated through a number of historical comparisons. At the time of increased import exposure and price instability exposure caused by technological change in the US after the Civil War, there were no income taxes (only una tantum taxes to recover from the Civil War and reconstruction costs). Hence, when the Democratic Party saw the growing success of the People’s Party in 1890 and 1892 with a protectionist campaign due to the perceived effects of market integration on $p$, they managed to take back such voters by acknowledging such grievances and by making a pledge to support the demand for a graduated income tax, which had been one of the economic policy components of the People’s Party platform. A federal progressive income tax system came later, but the alignment of the Democratic Party with those stances of the People’s Party was enough, because in case of success of the introduction of redistributive taxes, the perception was that $q$ could be high. On the other hand, in the twenty-first century the vast majority of liberal democracies are plagued by high debts and low fiscal space, in spite of income taxes already on the downward-sloping portion of a hypothetical Laffer curve. Thus, in the twenty-first century the “capacity” part of the trust parameter $q$ is very low. Hence, it is more difficult to gain back the support of those who shifted their support to populist protectionist parties.

In all historical cases it is the case that the populist campaigns focused a lot on corruption, and on capture by the elites, and on difficult accountability of traditional politicians, because that “objectives” component of $q$ does not depend on the capacity component, and hence regardless of the latter, an anti-elite and corruption denunciation rhetoric play well as complementary to a protectionist commitment.Footnote ²² In other words, if a party or politician proposes itself as a credible protectionist, then it is in their campaign strategy interest to focus the campaign on further reducing $q$ with claims or pointers to the capturability of traditional discretionary politicians by interest groups. Anti-elite rhetoric is therefore a clear complement to a policy commitment strategy, especially when running against a candidate proposing themselves as able to handle the potential consequences for some people of a free market economy that is efficient.Footnote ²³ Given that anti-elite rhetoric is one of the key elements of what constitutes the measure of populism in parties and politicians in political science, the association of populists with parties and politicians championing protection policy commitments is an “equilibrium” one. Anti-elite rhetoric can also be linked directly to $p$ and $q$: if the financial crisis and its handling have determined an increase in the belief that elite capture of the institution is responsible, then a corresponding reduction in $p$ for non-elite members follows naturally, as well as a reduction of $q$.

A similar logic could be invoked to capture the tradeoff between open- and closed-border policies: on the one hand, in many European countries the demographic trends and the sustainability issues for social security systems make many economists think that free movement of labor is good in the aggregate for a country, and the pros of open borders on average dominate the cons. But, on the other hand, those who fear losing the job or jobs for their children or fear lower wages may support closed borders in spite of what economists claim on the aggregate effects. This is especially true when there is low trust in the willingness or capacity of institutions to protect potential losers under open competition. As the literature on social identification suggests (see, e.g., Bonomi et al. (Reference Bonomi, Gennaioli and Tabellini2021) and the survey by Nouri and Roland), voters who fear being on the losing side of markets have shifted from identifying with other poor people and demanding redistribution, to identifying with their social or ethnic group and demanding protection through exclusion of outsiders – external competitors, other groups, and minorities.

Thus, the logic behind closed border policies is very similar to that behind closed market protectionism. If the protectionist policy bundle on the voter’s table also contains closed borders and anti-immigration stances, the difference between $r^r$ and $r^l$ can be expected to be larger than in the case of economic protectionism alone. This is because left-wing ideology is socially more inclusive, and hence the threshold to switch to an exclusionary immigration policy is much higher for left-wing than for right-wing voters. The implication is that when $q$ is shocked downward by a crisis, a disproportionately large fraction of the voters who shift from opposing to supporting condition (2) are right-wing.

2.3 The Consequent Cultural Clashes

One of today’s most pronounced cultural cleavages pits the traditional left’s progressive, inclusive vision – centered on diversity and equal opportunity – against a conservative impulse to protect identity by excluding immigrants and minority cultures from the political agenda. In other words, an “open” versus “closed” society. While other external factors (see Section 1.6) may also have heightened this divide, our key argument is that polarization on cultural issues can emerge naturally from shifting preferences over economic openness versus protectionism, even if individual ideologies remain unchanged.

In fact, for a moderate voter with no animosity against other groups based on hate or conflict history or whatever, but simply concerned about jobs and wages and alike, the decision to endorse a protectionist policy in the economic domain goes hand in hand with a closed borders support, which naturally leads to support a politician who campaigns on a set of exclusionary commitments. Thus, even moderate people now split in supporters for exclusion versus inclusion, even though they are still the same ideologically moderate people.

While any type of economic crisis may shift individual preferences toward protectionist commitments, it is important to distinguish which occupational groups are primarily affected. In our simplified model with two types of agents – those with low and high baseline probability $p$ of being a market winner – globalization shocks (e.g., the China shock) mainly reduce $\underset{\_}{p}$, thereby increasing insecurity particularly among occupations already vulnerable to external competition. Financial crises, by contrast, typically affect both $q$ and $\overline{p}$. Thus, the above implications for cultural division coming from the policy preferences effects extend to the classes that were not affected by the globalization shock. The increase in survey responses highlighting divisions on inclusion versus exclusion preferences is, therefore, potentially a simple byproduct of the extension of protection commitment preferences to the middle class.

If the model is correct, we should observe that the China shock increased reported economic insecurity mostly among the “low-$p$” occupations, whereas the financial crisis triggered declines across the board but relatively more for the “high-$p$” group. This suggests that cultural polarization may be an indirect effect of differentiated economic shocks across occupational strata, and the empirical analysis will support this mechanism. Posturing on culturally divisive issues has always been an equilibrium behavior (see Ash et al., Reference Ash, Morelli and Van Weelden2017), but especially so when traditional economic policy positions on left and right lost support across classes due to the crises affecting most of them.

2.4 Turnout Effects

In the simple theoretical framework described so far, we used the language of “preferences” for one or the other policy; but having a preference for one policy over another does not necessarily mean that the voter will turn out to vote. Participation is, of course, a crucial component of elections.

Using the as-if-pivotal voting assumption (see, e.g., Alesina and Rosenthal, Reference Alesina and Rosenthal1996), combined with the standard assumption of an orthogonal distribution of stochastic voting costs, the simplest theory of turnout predicts that a citizen will vote whenever the difference in expected utility between her preferred policy and the alternative exceeds her realized cost of voting. In this framework, citizens who are nearly indifferent between alternatives (i.e., for whom $r$ is very close to $p+(1-p)q$) are the least likely to vote.

Thus, a downward shock in $q$ caused by a crisis or corruption scandal is likely to increase abstention among those who previously had $p+(1-p)q>r$. Given that the $q$ shock affects everyone in the same direction, those who were previously abstaining may now shift to voting for a populist protectionist platform. For citizens experiencing a very large drop in the expected utility of the status quo, there can be a direct switch from one voting choice to the other. More commonly, however, the two effects operate separately: abstention becomes more attractive for those who previously supported open markets, while the populist option becomes more appealing to those who were previously near-indifferent and abstaining.

Moreover, given $r^r>r^l$, the prediction on turnout for left- and right-wing voters when $q$ is shocked downward by a crisis is different: turnout of left-wing voters may go down because the distribution of $p$ values is correlated with income and class, such that the increased fraction of potential losers is more concentrated on the left, but since $r^l$ is lower such left-wing citizens are more likely to station in the abstention pool; on the other hand, among those right-wing voters who were close to indifferent before the crisis, it is more likely that they jump over to vote for a protectionist closed markets and closed border party, since $r^r$ is higher.

Let us illustrate the point with a numerical example.

Suppose $\alpha =0.5$, $r^r=0.7$, $r^l=0.4$; the initial $q$ was 0.5 and the new after crisis is 0.3. Finally, suppose that among left-wing voters $f(p)$ is simply that half have $\underset{\_}{p}=0.2$ and the other half have $\overline{p}=0.5$. Among right-wing voters, those with $\overline{p}$ are actually 70%. Thus, within this numerical example, condition (2) has to be evaluated differently not only for left and right but also for $\overline{p}$ and $\underset{\_}{p}$ types within each ideology; hence there are four different conditions.

For a high-ability right-wing citizen, the expected utility of open market is 0.75 before the crisis, thus greater than the 0.7 of protectionism; hence before the crisis all the high-ability right-wing prefer open markets. Low ability $\underset{\_}{p}$ types among right-wing citizens have 0.6 from open market, hence they prefer protectionism. On the left, all prefer open markets or borders over closed markets or borders before the crisis because of the much lower $r^l$. So, in terms of preferences between the two alternative policies, only low-ability right-wing citizens prefer to close everything, but for any cost of voting $c$ we need to zoom in on willingness to turnout as well: for high-ability right-wing voters, the difference in expected benefit between the two alternative policies is only 0.05; hence only those in that set with very low cost realization will turn out to vote. On the other hand, for low-ability right-wing people the difference in expected benefit is 0.1 (in favor of protectionism), and thus the expected turnout among right-wing voters is higher within the low-ability segment. On the left, on the other hand, for high-ability citizens the expected benefit gap is 0.35, and hence they likely vote, if the distribution of costs, for example, goes from zero to 1. Finally, among low-ability left-wing people the expected benefit gap is 0.2, and therefore their turnout rate is somewhere between that of high-ability left-wing and that of right-wing voters. Thus, in the status quo precrisis only low-ability right-wing people would be interested in a populist protectionist policy, and they would almost surely lose, given the higher turnout on the left and low expected turnout on the right. Hence, a populist party might not even enter the race.

After the crisis, which shocks downward $q$ to 0.3, the picture changes:

Now all right-wing citizens prefer the protectionist policy, but the benefit gap is 0.05 (the opposite direction wrt precrisis) for high ability and 0.26 for low ability. So the high-ability right-wing people have the same turnout rate as before crisis, but the turnout rate among low-ability right-wing citizens jumps significantly up. On the other hand, on the left we have that the high-ability types continue to prefer openness but with an expected benefit gap of 0.25, hence a reduction in turnout; the low-ability left-wing have a benefit gap of only 0.04, hence they almost all stay home rather than voting. Thus, this example clarifies that

Formally, assuming costs of voting distributed uniformly from zero to one and a discrete support for $p\in \{\underset{\_}{p},\overline{p}\}$, the turnout rate within group $\theta ,j$, $\theta \in \{\underset{\_}{p},\overline{p}\},j\in \{l,r\}$ is

$t_{\theta j}(q)=|\theta +(1-\theta )q-r^j|$(3)

and hence expected turnout for group $\theta ,j$ increases (decreases) when $q$ decreases if and only if (2) holds (doesn’t hold) before the crisis for such a group.

Whether the market efficiency assumption holds in the status quo before crisis obviously depends on the percentage ($\alpha$) of left-wing voters, because, given the correlations mentioned above, left-wing voters typically not only have $r^l<r^r$ but also have a lower percentage $\beta$ of members with $\theta =\overline{p}$, ${\beta }^l<{\beta }^r$.

Formally, MEA holds iff

$\begin{array}{ll}\widehat{r}\hfill & <[\alpha {\beta }^l+(1-\alpha ){\beta }^r][\overline{p}+(1-\overline{p})q]\hfill \\ \hfill & +[\alpha (1-{\beta }^l)+(1-\alpha )(1-{\beta }^r)][\underset{\_}{p}+(1-\underset{\_}{p})q].\hfill \end{array}$(4)

With $\alpha =0.5,{\beta }^l=0.5,{\beta }^r=0.7,q=0.5,r^r=0.7,r^l=0.4,\overline{p}=0.5,\underset{\_}{p}=0.2$ the LHS of (2.4) is $\hat{r}=0.55$, while the RHS is $0.60.75+0.40.6=0.7$, hence MEA holds. If the crisis is such that $q$ drops to 0.3, then the RHS drops to $0.60.65+0.40.44=0.566$, and hence MEA continues to hold in the numerical example even after the big drop in $q$.

Let us now compare the probability of winning for a party proposing the closure commitment before and after the crisis. With $q=0.5$ the expected turnout of voters supporting the closure commitment is $(1-\alpha )(1-{\beta }^r)t_{\underset{\_}{p},r}(0.5)=0.015$, whereas the total expected turnout in favor of the open market status quo (defended by traditional parties based on MEA rhetoric) is $(1-\alpha ){\beta }^r{t}_{\overline{p},r}(0.5)=0.0175$ plus $\alpha [{\beta }^l{t}_{\overline{p},l}(0.5)+(1-{\beta }^l)t_{\underset{\_}{p},l}(0.5)]=0.34$. Thus, with these expected turnouts, the probability of winning for the closure party is almost zero. Hence, as intuitively argued below, one might not even expect the closure party to materialize in the electoral race in the first place before the crisis.

After the crisis, expected turnouts change as follows: now the right-wing are all in favor, and their total expected turnout is roughly 0.21, where the left is all against but turnout drops to below that, and hence the probability of winning for the closure commitment proposer is significant and entry by populist protectionist party is guaranteed.

While the example computations above are made by assuming that a crisis shocks only $q$, it is easy to obtain a similar example even for the case in which a crisis also shrinks $\overline{p}$.

To illustrate the example in a compact way, Table 1 reports the expected utility from the open market status quo, the utility from protectionist policies, and the resulting preference and turnout gap for each type of voter. The table compares outcomes before and after a crisis that reduces $q$ from 0.5 to 0.3, using the same numerical assumptions discussed earlier. The turnout gap corresponds to the absolute difference in utility between the preferred and alternative options, which directly maps into the share of individuals within each type expected to vote under the uniform cost distribution assumption.

Table 1Voter preferences before and after the crisis

Notes: Each row in the table reports the expected utility from the open market status quo $U(s)=\theta +(1-\theta )q$ for a given voter type $(\theta ,j)$, where $\theta \in \{\underset{\_}{p},\overline{p}\}$ denotes ability (or perceived benefit from openness), and $j\in \{l,r\}$ denotes ideology. The table compares this to the utility from protectionist policies $r^j$ and indicates which policy is preferred by the voter (based on which utility is higher), along with the absolute utility difference between the two options (Turnout gap), which determines the likelihood of turning out under the assumption of uniformly distributed voting costs. A higher turnout gap implies a greater incentive to vote. The table highlights how a crisis-driven decline in $q$ can shift preferences toward protectionism among right-wing voters and simultaneously reduce turnout among left-wing voters, thereby increasing the relative electoral viability of populist protectionist parties even if the open market equilibrium remains efficient.

An important note about this example: It shows that even after a big negative shock in trust $q$ in the objectives and or capacity of institutions to make inclusive welfare policies work, the open market status quo can still remain the efficient solution (MEA holds even with $q=0.3$); but the probability of winning (and hence the incentive to enter) for a right-wing populist party with a closure bundle of policy commitments jumps up significantly. This is exactly what we observe in the data.

2.5 Entry of Populist Parties

Like for voters, obviously there is a cost of entry also for a political party (organizational cost and opportunity costs). Intuitively, the expected benefit from entering on the competitive supply of politics with a protection commitment campaign is higher, hence outweighing the costs, the larger the pool of citizens who could turn out to vote for them conditional on running. Thus, given the above analysis, it is natural that drops in $p$ or $q$ eventually enlarge enough the pool of disillusioned voters such that entry occurs. Thus, if the financial crisis caused a reduction of $p$ for a new class of voters and a reduction of $q$ for all, it is natural to expect the financial crisis as a moment of great endogenous transformation also on the supply side.

Moreover, in principle multiple protection policy commitments could come to the voter’s table, perhaps one championing almost exclusively economic protection commitments such as tariffs, quotas, and so on while another bundle of protection commitments by another (typically more right-wing) party includes anti-immigration commitments of a brutal type and other social exclusion policies on other dimensions. The intuitive reason why in a country both a left-wing populist party with weaker closures on the social dimension and a right-wing populist party championing a tougher set of exclusionary policies may simultaneously or sequentially decide to enter the electoral race after a big crisis relates to the turnout message conveyed by the above example: left-wing voters, if they find the status quo depressing, may simply abstain if the only protection alternative on the table has ugly components of anti-diversity, anti-pluralist, and anti-integration stances; hence, if there exists only a tough right-wing populist option the predicted asymmetric turnout effects described above are unavoidable.

A populist party with softer closures on the social dimension can instead attract those depressed and abstaining left-wing voters. Since the entry conditions depend on too many factors outside our model, we do not attempt to obtain a set of sufficient conditions for the existence in equilibrium of one or two or more populist parties on one or both sides of the ideological spectrum, but it is clear that such sufficient conditions always exist, and hence in the empirical analysis we expect to find evidence of greater appeal for populist parties coded to be on left or right for voters of the two sides. Thus, ideology simply partitions incentives. In the empirical analysis of the transformation of the political supply, we will show that indeed the six most important determinants of the populist transformation of manifestos have to do with closures of different intensities on the two ideological sides.

2.6 Transition to Empirics

We present empirical evidence supporting the predictions of our simple model. The survey data described below quantify economic insecurity across occupations before and after the crisis. Because populist parties did not gain the power to immediately change policy in its aftermath, we can map these measures of insecurity directly on our theoretical framework as follows: the economic security of an individual of type $\theta$ at a time with trust $q$ is given by $e_{\theta }(q)=\theta +(1-\theta )q$. Denoting by $\overline{e}=e_{\overline{p}}$ and $\underset{\_}{e}=e_{\underset{\_}{p}}$, our arguments above point to the fact that the globalization and automation shocks should be expected to affect downward $\underset{\_}{e}$, whereas the financial crisis affects $\overline{p}$ and $q$ probably more, and hence empirically we should see a significant change in $\overline{e}$ around the financial crisis. $\overline{e}$ could, for example, be constructed from the survey responses as the average level of economic security declared by individuals with an age-earning profile above the median.

In the empirical analysis, we will also need to consider voters’ preferences and participation decisions when multiple populist parties exist. In particular, when a populist party positions itself on the left but embraces protectionist economic policies – without adopting the exclusionary or deportation measures that left-wing voters oppose – we would expect left-wing turnout to rise. In this scenario, voters’ perceived utility from a “softer” protectionist platform exceeds their utility under the status quo.

Even if the radical right-wing party and the moderate populist-left party split the protectionist vote, the overall shift in demand toward closure policies makes it more likely that some form of protectionism becomes politically dominant. Stylized predictions on convergence toward protectionism follow naturally: when economic insecurity is widespread, even voters with inclusive social preferences may favor market closure – if such a policy is offered in an acceptable ideological form. The supply dynamics will display a consistent pattern.

5.1 Building the Instrument

Let ${\beta }_k$ denote the steepness parameter of the age-earnings profile of workers in occupation $k$ and let $s_{jkc}$ denote the share of workers that belong to cohort $j$ in country $c$ who are employed in occupation $k$, relative to all workers in the cohort at the beginning of our sample. Our instrument is then:

$z_{jct}=y_{ct}\left({\displaystyle \sum _{k=1}^K{\beta }_k}{s}_{jkc}\right)$(7)

where $y_{ct}$ is country $c$’s GDP in year $t$, which is set to 1 in 2008. Hence, it captures the dynamics of GDP in a country relative to 2008 and differences across countries in terms of the crisis’ impact on the country’s economy. The instrument allocates the shock to a country’s GDP among the various cohorts according to the relative importance of the various occupations in each cohort and the exposure of each occupation to financial shocks. Hence, when a country’s GDP falls following the onset of a financial crisis, cohorts that have a higher incidence of occupations with greater dependence on external financing should experience a larger increase in economic insecurity. In other words, the instrument $z_{jct}$ should correlate negatively with cohort economic insecurity $E{I}_{jct}$. In Section 6, we discuss in detail how the measures of ${\beta }_k$ were obtained and how we constructed the instrument.

To build the instrument, we first use the EU-SILC panel data for the period 2003–2012 and estimate for each of the two-digit ISCO occupation codes (comprising twenty-six occupations) the following model of labor income:

$y_{ict}^k={\alpha }_0^k+{\alpha }_1^k{Z}_{ict}+{\beta }^{k}log{(age)}_{ict}+f_i+f_t+{\epsilon }_{ict}^k$(8)

where $k$ is occupation, $i$ is the individual, $c$ is the country, and $t$ is time. The variable $y_{ict}$ is logged labor income, $ag{e}_{ict}$ is the individual’s age, and ${\epsilon }_{ict}$ is the residual shock to labor income. The model also includes an individual fixed effect $f_i$ that captures relevant time-invariant individual characteristics, a vector $Z_{ict}$ of controls that include time-varying individual demographics (marital status and education, which vary over time for some members of the sample), and year fixed effects ($f_t$) reflecting business cycle and aggregate productivity dynamics.

From this estimation, we retrieve the slope of the profile ${\hat{\beta }}^k$. The estimated values of ${\beta }^k$ range from 1.99 to 4.37, with a mean of 3.18 and a standard deviation of 0.53, suggesting that there is a remarkable degree of heterogeneity in ISCO occupation age-earnings profiles. At age 40, an extra year on the job is associated with an increase in labor income that ranges from 5% in the occupation with the flattest profile to 11% in the occupation with the steepest. Table A4 in Online Appendix E lists the occupations and their relative ${\hat{\beta }}^k$.

We next use the EU-SILC 2003, 2004, and 2005 waves, for each cohort $j$ and country $c$, to compute the weights ${\hat{s}}_{jkc}$, that is, the share of workers in occupation $k$ belonging to cohort $j$ in country $c$, prior to the financial crisis shock. We then compute the instrument:

${\widehat{z}}_{jct}=y_{ct}\left({\displaystyle \sum _{k=1}^K{\widehat{\beta }}^k}{\widehat{s}}_{jkc}\right).$(9)

This instrument captures exogenous variation in $E{I}_{jct}$ that corresponds to theory-based shocks to $p$ and $q$: occupations with steeper earnings profiles rely more on financial access in early stages, making them more vulnerable to a fall in perceived success probability $p$ and in institutional support $q$ after a financial crisis.

Figure 5 presents a bin scatter plot of the instrument plotted against the index of economic insecurity across the cohorts in our sample, highlighting a clear negative correlation between the instrument and economic insecurity. Thus, cohorts with a higher than average share of individuals in occupations characterized by steep age-earnings profiles experienced a greater increase in economic insecurity when the financial crisis reduced a country’s GDP, that is, when $y_{ct}$ dropped below 1 in the years following the financial crisis.

Figure 5

The instrument and economic insecurity

Figure 5 Long description

The x-axis ranges from 1 to 4 and the y-axis from 0.16 to 0.24. Around 20 binned dots are scattered mostly between instrument values of 1 and 3.5. The fitted line slopes downward from approximately 0.235 at the left to 0.185 at the right. Cohorts with steeper age-earnings profiles experienced greater increases in economic insecurity following the financial crisis.

Notes: The figure shows the bin scatter plot of the instrument (Eq. 9) against the index of economic insecurity across the cohorts in our sample.

Table 7 presents the results of regressing the instrument against economic insecurity, where the first column controls for cohort, country, and wave fixed effects, and the second adds the cohort-level time-varying controls described in Section 3, thus reproducing the first stage of our IV estimation. The instrument has the expected sign and is highly statistically significant, suggesting that we are likely to have enough power to use this instrument in regressions estimating the effect of economic insecurity on voting and the other outcome variables.

Table 7Economic insecurity and populist vote – First stage

Notes: OLS estimation. Dependent variable: Economic insecurity. Instrument is the variable computed as in equation 9. In both regressions we control for cohort, country, and wave fixed effects. In column 2, we add the cohort-level time-varying controls described in section 3. Table A12 in Appendix J includes the control variables’ coefficients. Errors are clustered at the cohort level.

A potential concern with our identification strategy is that the occupational composition of cohorts – which determines their borrowing sensitivity – may also correlate directly with political preferences, violating the exclusion restriction. In Appendix F, we show that the borrowing sensitivity component of our instrument does not predict political outcomes in the absence of the GDP shock interaction, supporting the validity of our identification strategy.

6.1 Economic Insecurity

We begin our empirical analysis by documenting a salient descriptive pattern: the evolution of economic insecurity across income groups in the wake of the 2008 financial crisis. This provides an initial, non-parametric validation of the theoretical mechanism discussed in Section 2, before turning to more rigorous regression-based identification.

Using repeated cross-sectional data from the ESS, we construct a principal component index of economic insecurity (described in Section 3) and track its evolution across two waves: the one closest to 2006 (precrisis) and the one closest to 2010 (postcrisis). To capture heterogeneous effects by baseline economic status, we split individuals within each country into two groups based on whether their income is above or below the national median.

Figure 6 presents the resulting comparison. The left panel shows that among individuals below the median income, economic insecurity remains largely stable across the two periods. The small postcrisis decline is not statistically significant. In contrast, the right panel reveals a sharp and statistically significant increase in perceived insecurity among individuals above the median. Their average PCA index of insecurity rises by approximately 30%, indicating that the financial crisis caused substantial subjective economic distress even among relatively advantaged individuals.

Figure 6

Change in economic insecurity, below and above median income

Notes: The figure shows the average value of economic insecurity before and after the 2008 financial crisis by populist vote. The left panel shows values for individuals below the median income of their country. The right panel shows values for individuals above the median income of their country.

This pattern is consistent with our theoretical framework, where the financial crisis is modeled as a common shock that reduces both the probability of market success (especially $\overline{p}$) and the trust in institutional compensation mechanisms ($q$). While globalization shocks mainly affect low-$p$ individuals (typically low-income), financial crises extend economic insecurity to previously secure groups, thus reducing the expected utility $U(s)=p+(1-p)q$ for a broader segment of the population.

To further investigate the specific role of credit market conditions, we exploit variation in the tightening of lending standards across European countries. Using the ECB Bank Lending Survey, we construct a country-year measure of the worsening of credit access conditions for households. For each ESS wave, we classify countries as experiencing above- or below-median credit tightening. Figure 7 presents the results. Focusing on individuals above the 25th percentile of the income distribution – to isolate voters for whom credit constraints are most plausibly binding while excluding those least likely to rely on credit markets – we find that populist vote shares are significantly higher in countries experiencing more severe credit tightening. The difference is statistically significant at the 5% level (p-value = 0.029). This finding provides direct evidence for our theoretical mechanism: the financial crisis affected political preferences not merely through general economic distress but specifically through the credit channel. When banks tightened lending standards, individuals who relied on borrowing to smooth consumption or finance investments – particularly those with steeper age-earnings profiles – experienced a sharper decline in economic security, leading to greater support for populist alternatives.

Figure 7

Populist vote by credit tightening intensity

Figure 7 Long description

Left panel, below-median income: pre-2008 insecurity is about 0.267 and post-2008 is about 0.266, with overlapping confidence intervals. Right panel, above-median income: pre-2008 insecurity is about 0.141 and post-2008 rises to about 0.181, a substantial increase with non-overlapping confidence intervals. The crisis disproportionately affected middle and upper-income groups.

Notes: The figure shows the average populist vote share among individuals above the 25th percentile of the income distribution (within each country-wave), in countries where at least one populist party is present. Countries are classified as above or below the median of credit tightening based on the ECB Bank Lending Survey measure of worsening household credit conditions. Error bars represent 95% confidence intervals. The difference between groups is statistically significant (t-test p-value = 0.0288).

In what follows, we estimate the causal effect of economic insecurity on turnout and populist voting, and we examine how political supply responded to the new distribution of insecurity.

6.2 Voting Populist

In light of equation (2) in the theoretical model, rising economic insecurity implies that more individuals satisfy $r>p+(1-p)q$ and thus prefer closure policies. We test this implication by estimating the effect of $E{I}_{jct}$ on populist vote shares.

Table 8 presents the estimation results for model (6) where the outcome variable is the share of votes going to populist parties. The first two columns present the OLS estimates, where the first column controls for cohort, wave, and country fixed effects while the second adds the time-varying cohort controls described in Section 3. In both cases, a cohort’s support for populist parties correlates positively with the variation over time in the level of economic insecurity experienced by the cohort in a particular country. The effect is highly statistically significant and of similar magnitude in both specifications. The other two columns show the 2SLS estimation results for the corresponding specification. The effect of economic insecurity on the populist vote is strongly positive and significant. The IV estimates show a larger effect than the OLS estimates, which is consistent with the idea that the index of economic insecurity imperfectly captures economic distress, thus resulting in attenuation bias in the OLS estimates. Based on the estimates in the last column, an increase of one standard deviation in economic insecurity leads to an increase in the populist vote of 7 percentage points, which is about 94% of the sample mean. The Kleibergen-Paap Wald F statistic reported at the bottom of the table indicates that the estimates do not suffer from a weak instrument problem.

Table 8Economic insecurity and the populist vote

Notes: OLS estimation in columns 1 and 2, 2SLS estimation in columns 3 and 4. Dependent variable: Populist vote. Economic insecurity is the survey-based measure of economic insecurity. In all regressions we control for cohort, country, and wave fixed effects. In columns 2 and 4 we add also cohort-level time-varying controls described in section 3. Table A13 in Appendix J includes control variables’ coefficients. Errors are clustered at the cohort level. Kleibergen-Paap F-statistic are reported for columns 3 and 4.

6.3 Back-of-the-Envelope Calculation

To quantify the contribution of the middle class to the rise of populism, we perform a back-of-the-envelope calculation. Using our 2SLS estimate of the effect of economic insecurity on populist voting ($\hat{\beta }=0.85$), the observed increase in economic insecurity among the middle class (approximately 10 percentage points, as shown in Figure 1, panel b), and the weight of the middle class in the population (50%), we can compute the share of the increase in populist voting attributable to rising middle-class insecurity. The calculation yields: $\frac{0.85\times 0.1\times 0.5}{0.06}\approx 0.71$. This implies that the extension of economic insecurity to the middle class – a distinctive feature of the financial crisis relative to earlier globalization shocks – can account for roughly 70% of the observed increase in populist vote shares. This finding underscores the political significance of the crisis’s reach beyond traditionally vulnerable low-income groups: The enlargement of the pool of economically insecure voters to include substantial portions of the middle class was a critical driver of populism’s electoral success in Europe.

6.4 Turnout

Table 9 shows the estimation results for voter turnout, based on OLS (column 1) and 2SLS (columns 2–4) estimates, including in all cases all the controls (cohort, wave, and country fixed effects and time-varying cohort controls). The results indicate that an increase in economic insecurity discourages people from voting. This is true for both the OLS estimates and the 2SLS estimates. As in the case of voting populist, the IV estimate of the effect of economic insecurity is much stronger than the OLS estimate, which is consistent with the presence of measurement error in the index of economic insecurity. From an economic perspective, the estimates in the second column imply that an increase of one standard deviation in economic insecurity reduces turnout by more than 8 percentage points or about 10% of the sample mean.

Table 9Economic insecurity and voter turnout

Notes: OLS estimation in column 1, 2SLS estimation in columns 2-4. Dependent variable: Populist vote. Economic insecurity is the survey-based measure of economic insecurity. New populist same orientation is a dummy equal to one if a new populist party exists in the country and whether it is of the same orientation as that of the voters in the cohort. New populist same orientation post-crisis is a dummy assuming value one when the new populist appears just after the burst of the financial crisis. In all regressions we control for cohort, country, and wave fixed effects, together with cohort-level time-varying controls described in section 3. Table A14 in Appendix J includes control variables’ coefficients. Errors are clustered at the cohort level. Kleibergen-Paap F-statistic are reported for columns 2-4.

The evidence is consistent with the idea that the economic insecurity resulting from the crisis led to disappointment with traditional political parties, and as a result voters either became more receptive to populist, anti-elite, and protectionist rhetoric and more likely to vote for populist parties, or they became more inclined to abstain.

We next investigate whether the appearance of a populist party constitutes a valid alternative for disappointed voters, thus mitigating their tendency not to vote. In column 3, we add an indicator for whether a new populist party with the same orientations as the voters in the cohort exists in that country, as well as its interaction with the economic insecurity variable.Footnote ³⁷ The results show that neither of the two variables is statistically significant. In column 4, we replace the indicator for a new populist party with one indicating whether or not a populist party appears just after the onset of the financial crisis. This is based on evidence, albeit weak, that the presence of a populist party of the same orientation at the onset of the crisis reduces the tendency to abstain from voting. This suggests that political orientation matters and that a populist party can more easily attract the disappointed voters who share its ideology. Among such voters, the effect of economic insecurity on turnout is 1/3 smaller (in absolute value) than among voters who do not share the orientation of the populist party (−0.976 as compared to −0.622, which is the difference between −0.976 and 0.354).

This is aligned with the model’s prediction that abstention is mitigated only when voters are offered a protectionist platform with a compatible $r^j$ value – i.e., left-oriented voters are more likely to turn out if a non-exclusionary protectionist party enters, thus raising their perceived utility from closure above $U(s)$.

6.5 Robustness and Trust

In Appendix G, we show how economic insecurity also triggered substantial distrust in politicians and political parties. Importantly, all these results survive several robustness checks (Appendix H), which include using alternative (data-driven and not self-reported) measures of economic insecurity (following Guvenen et al., Reference Guvenen, Ozkan and Song2014), potential nonrandom exposure to the shocks (Borusyak and Hull, Reference Borusyak and Hull2020) and different clustering of the standard errors and/or different fixed effects schemes.

The results, taken together, lend support to our thesis that deterioration in individual economic security leads to a loss of confidence in political parties, which may in turn lead to disappointment and voters’ abstention.

8.1 The Entry Choice Between Left and Righ

The choice of entering on the left or on the right should depend on the relative entry space. The latter, in turn, depends on the ideological orientation of the electorate and on the salient features of the main economic insecurity determinants. We verify this by estimating (12).

The party orientation is observed in the CHES survey and measured on a scale from 1 (far left) to 10 (far right), so our data are limited by the CHES coverage. The shares of left- and right-oriented voters, also on a 1-to-10 scale, are obtained from the waves of the ESS. As a measure of left-salient factors, we use the Gini coefficient of income inequality (from the World Bank World Income Inequality Database) and as a measure of the right-salient factor the share of immigrants from Muslim countries in the total population. This variable, obtained from the World Bank Bilateral Migration Matrix, is available for three years (1999, 2010, and 2013). We predict ${\delta }_1<0$ and ${\delta }_2>0$.

Relative entry space should be a critical determinant of the orientation choice whenever the individual characteristics of left- and right-leaning voters are similarly distributed in terms of the key variables of economic insecurity, trust, and ability to assess populist policies that drive the consensus for populist parties. Table 14 confirms that this is indeed the case. The distribution of proxies for the determinants of voting, summarized by mean and standard deviation, is extremely similar between left- and right-oriented voters.

Table 14Comparison left/right oriented

Notes: The table reports summary statistics of characteristics of left-oriented and right-oriented voters in our sample.

Figure 9 shows that in the CHES data, the distribution of the orientation of populist parties is sharply different from that of non-populist parties: populists have a much higher density on the right.

Figure 9

Left/right orientation

Figure 9 Long description

The dashed non-populist curve rises from near zero, peaks around 0.155 at position 3, dips to about 0.13 near position 5, rises again to about 0.155 near position 7, and then drops sharply toward zero by position 10. The solid populist curve stays below 0.05 until position 3, rises steadily, crosses above the non-populist curve around position 6, peaks at approximately 0.18 near position 8, and then declines. Populist parties are concentrated toward the right of the ideological spectrum.

Notes: The figure shows the kernel density of the ideological orientation on the left/right scale of populist and non-populist parties in Europe.

Table 15 (first column) shows that the heterogeneity in populist party orientation is consistent with our hypothesis. Income inequality weighted by the population share of left-oriented voters tends to pull the orientation of populist parties to the left, and the effect is statistically significant. A 1-standard-deviation increase in this factor shifts orientation to the left by more than one unit in the scale or 21% of the sample mean. The share of immigrants from Muslim countries weighted by the share of right-oriented voters has a positive and highly statistically significant effect, pulling populist parties’ orientation to the right. A 1-standard-deviation increase in this factor increases the score by 1.51 points or 29% of the sample mean orientation. Interestingly, it is not immigration per se that affects the populist orientation but its origin from Muslim countries (see, e.g., Laitin, Reference Laitin2018). If we replace immigration from Muslim countries with the population share of all immigrants or of immigrants from EU countries, the immigration variable (weighted by the share of right-oriented voters) is not statistically significant. This strengthens our interpretation of the results, as it strongly suggests that the orientation chosen is the one most susceptible to effective populist rhetoric (see Rodrik, Reference Rodrik2017). Results are unchanged if we account for selection due to endogenous populist party entry (second column) by adding as a control a Mill’s ratio computed from a first-stage Probit for the presence of a populist party using as instruments the controls in the third column of Table 13.

Table 15Populist parties’ orientation choice

Notes: The table reports regressions of the orientation of the populist parties in our sample on measures of relative entry space. The measure of party orientation is defined on a scale from 1 (extreme left) to 10 (extreme right). In the second column Mill’s ratio computed from a first-stage probit for the presence of a populist party using as instruments the controls in the third column of Table 12 is added. Standard errors clustered at the country level are shown in parenthesis. *** significant 1% or less; ** significant at 5%; * significant at 10% confidence level.

These results corroborate the theoretical intuition that the composition of $r^j$ – which includes preferences over both economic and social exclusion – determines the relative attractiveness of entry on the right versus the left. In environments where $q$ falls and social exclusion gains salience, right-wing entry becomes more likely. These findings also corroborate the tenet that populist platforms are a strategy and can adjust the content of the strategic exclusionary commitment to the ideology of their targeted voters.

Populist parties and platforms appear when the popular disappointment is sharp enough to raise realistic hopes of winning a share of the total vote - a scale effect - large enough to outweigh the entry cost. Conditional on entry, the party chooses its political orientation strategically, tilting toward voters, ideology and where the factors behind the crisis are more salient - a relative size effect.Footnote ⁴³ There may have been other ideological and cultural reasons for the orientation choice of a new party, but we have shown that even the most standard office-seeking motivation can explain the observed variation.

8.2 Populist Platforms’ Content

Having shown the determinants of entry and orientation choices, we can now illustrate the third supply choice, namely the policy platform choices.

Using the 2014 Chapel Hill Expert Survey (CHES), in which national experts rate European parties on a range of positions, policies, and salient issues, we construct continuous measures of three important dimensions of populism: (1) anti-elite rhetoric, (2) protectionism, and (3) concealment of long-term costs for all the parties in the CHES database. We use these measures to see if it is true that those parties identified as populist in the categorization of PopuList which was proposed by Rooduijn et al. (Reference Rooduijn, Van Kessel, Froio, Pirro, De Lange, Halikiopoulou, Lewis, Mudde and Taggart2019) (using dimension 1) are indeed significantly more likely to choose policy platforms that conform to the second and third dimensions. The rhetoric variable averages the scores assigned to a measure of the salience of anti-establishment and anti-elite rhetoric and of reducing political corruption, on a scale from 0 (not important at all) to 10 (very important). The protectionism measure is the average of the scores for the position on five policies that may offer economic protection in different domains: deregulation (10 strongly opposes deregulation of markets); immigration (10 strongly in favor of tough policy); tax policy: (10 strongly favors tax cuts vis-a-vis improving government services); economic intervention (10 fully in favor of state intervention); cosmopolitanism (0 strongly advocates cosmopolitanism, 10 strongly advocates nationalism); redistribution of wealth (10 fully in favor of redistribution). To capture the third dimension, we average parties’ positions on two long-term issues: the environment and international security or peacekeeping. Policies on these issues will pay off in the long run, the first by limiting global warming and the second by guaranteeing a stable international order. We interpret a high score on downplaying the importance of these issues as the gauge of a strategy of hiding the long-term costs of protectionism. Table 16 shows regressions of each of the three indexes on the populist party identifier, after controlling for the political orientation of the party (0, far left; 10, far-right).

Table 16Correlation of the three dimensions

Notes: The table shows OLS regressions of each of the three indexes (Anti-Elite Rhetoric, Protection, and Concealment of the long-term costs of short-term protection) as well as of the principal component on the populist identifying dummy. Each regression controls for the left/right orientation of the party. The last row shows the difference in the score of populist parties from the sample mean.

Independently of political orientation, populist parties all score higher in each of the three indexes. The difference between populist and non-populist parties is sharpest on the anti-elite/anti-corruption dimension (63% above the sample average), but it is substantial for the other two (36% and 26% above average).

8.3 Non-populist Parties’ Reaction to Populism

Recalling that our theoretical framework strongly suggests that populist commitments are a rational strategy to accommodate the drop in $p$ and $q$ (trust in markets and governments), a consistent testable hypothesis is that after the financial crisis even non-populist parties moved their platforms in the same general direction. To test this hypothesis, we use the five waves of the Chapel Hill Expert Survey (CHES). For each of a list of issues (see Appendix K for the full description), the CHES reports the position of the party on a scale of 0 to 10 (for some issues, the CHES scale is 1 to 7, but we rescale them to 0-10). To assess the party position, CHES questions a pool of experts in each country. For instance, on the issue of deregulation/regulation the position of the party is gauged by a number, running from 0 (strongly opposed to deregulation) to 10 (strongly in favor). We disregard issues present in only one or two surveys, considering only those that are assessed in at least three and preferably all five surveys. We group the positions into four families: overall European integration (P_EI); European policy (P_EU, obtained by summing the scores on three issues: powers of European institutions, European cohesion policy, and EU foreign and security policy); ideology (P_ID, obtained by summing the scores on three issues: general ideological stance (left/right), stance on intensity of government intervention in the economy, and libertarian versus traditional/authoritarian stance); and an index of the positions on a set of eleven policy issues (P_PD: government expenditure versus taxation, deregulation, redistribution of wealth, civil liberties versus law and order, social lifestyle, religious principles in politics, immigration policy, multiculturalism, urban versus rural interests, political decentralization to regions/cities, and position toward ethnic minorities). The first three indexes are available for all surveys, the fourth for the last three waves.

To compare platforms we proceed as follows. Let $y_{icjt}$ denote the position of party $i$ in country $c$ on issue $j$ (EI, EU, ID, PD) in year $t$. We distinguish between platforms of populist parties, P, and non-populist, NP, and let $D_{ijct}=(y_{ijct}^{NP}-y_{jct}^P{)}^2$ denote the distance between the platform of non-populist party $i$ and the main populist party in its country, if there is one. Let $s_{t-1}^P$ denote the share of the vote going to the populist party or parties in the last election before the survey. We test the electoral competition hypothesis by running the regression:

$D_{ijct}=f_t+f_{NP}+\gamma s_{t-1}^P+u_{icjt}$(13)

where $f_t$ are time fixed effects, $f_{NP}$ are non-populist party fixed effects, and $u_{icjt}$ an error term. Because parties are country-specific, the party fixed effects also capture systematic national differences across countries. Most electoral competition models would predict a negative value for $\gamma$, that is, the platforms of non-populist parties should move closer to that of the populist party as the latter becomes more successful.

Figure 10 plots the relation between the distance of the platforms of non-populist parties from those of the populist, and the populist share of the vote in the most recent election for each of the issues and for the overall index. To pick up possible non-linearities, we plot a local polynomial regression with the 95% confidence band. In all the issues the distance decreases as populist parties gain support, which jibes with the thesis that populist policies are more palatable to the electorate at times of systemic crisis. Table 17 shows the estimates of the linear regression specified above, confirming the visual inspection of Figure 10: as populist parties gain support, their non-populist adversaries appear to adapt their platforms to reduce the distance from the populists. The effects, which account for endogenous presence of populist parties,Footnote ⁴⁴ are substantial: increasing the share of votes to the populist party by 1 standard deviation (15 percentage points) shortens the distance between the non-populist and populist platforms by 34%, 18%, 27%, and 28% of the sample mean of our issues EI, EU, ID, and PD, respectively. Table 18 rules out the possibility that it is the populist party that moves closer to the traditional parties as it gains consensus. To show this, we regress the change in populist positions on the populist share of the vote. We find that populist parties do not revise their position as their share of votes increases. Overall, this evidence means that simply counting the number of populist parties, or tallying their share of votes/seats, understates the supply of populist policies in a country.

Figure 10

Distance from the populist platform and share of votes to populist parties

Figure 10 Long description

Each panel plots share of votes to populist parties on the x-axis from 0 to 60 with a fitted curve and 95 percent confidence band in gray. Top-left, Distance EI: starts near 38, declines to about minus 10 at vote share 50. Top-right, Distance EU: starts near 75, declines to about minus 25. Bottom-left, Distance IQ: starts near 40, declines to about 15. Bottom-right, Distance PD: starts near 100, declines to about minus 25. In all four dimensions, non-populist parties converge toward populist platforms as populist vote share increases.

Notes: The figures show the local polynomial smooth relation between measures of distance of non-populists from populist platforms and the share of the vote that went to populist parties in the last most recent election.