Financialization, wealth and the changing political aftermaths of banking crises

Abstract

Households face two politically salient risks associated with financial instability. The first risk, which has existed for perhaps centuries, is associated with the indirect effect of systemic banking crises on employment and income flows. The second risk arises from the direct effects of crises on asset prices and thus household wealth stocks. Historically, the second risk mainly affected only a narrow wealthy elite. We argue that the rapid expansion and financialization of middle-class wealth since the mid-20th century mean that many voters now have ‘great expectations’ regarding government responsibility to protect their wealth. The political risks of financial instability for incumbent governments have thus increased sharply, especially when institutional constraints hamper their ability to respond to voters’ new expectations. We show that the probability of incumbent governments facing significant institutional constraints retaining office after systemic banking crises has indeed fallen sharply in recent decades compared to the pre-1945 period.

The severe banking crises that struck many advanced countries from 2007 have had powerful, ongoing political consequences. Many governments in crisis-hit countries lost office in their aftermath. Meanwhile, political instability, polarization and populism appear to be rising, policy contestation is increasing and support for globalization and even democracy is declining (Hernández and Kriesi, 2016). Are these political effects typical or unusual? Thus far, with rare exception (Funke et al., 2016), there has been little long-run comparative work that addresses this question.

We argue that in one important respect—the probability of incumbent governments retaining office—the political impact of banking crises has become much greater in recent decades compared to the past. This is because many voters now face an additional, direct risk from crises due to their impact on household wealth stocks. This comparatively recent risk to middle-class wealth has added to a much longer-standing risk that crises pose to households’ employment and income flows. We argue that this growing risk exposure has induced ‘great expectations’ among many middle-class voters regarding government responsibility to protect their wealth. For some of this group, this may now be more important than income protection, which became the dominant expectation by the mid-20th century (Ruggie, 1982; Blyth, 2002). This growing overlap with elite pro-bailout preferences has considerably enlarged the constituency supportive of financial bailouts.

Great expectations have also modified the way in which political institutions shape voter assessments of government performance after crises, making it more difficult for incumbents to escape political punishment. The intensity of these expectations and informational asymmetries mean that many voters will be unforgiving if they suffer significant absolute or relative wealth losses. This leads us to different conclusions than the ‘clarity of responsibility’ hypothesis (Powell and Whitten, 1993; Duch and Stevenson, 2008; Hellwig, 2008). Governments facing more polarized veto players should find it increasingly difficult to survive banking crises because they cannot meet the higher expectational bar set by many voters. Severe banking crises typically require extraordinary policy interventions and often emergency legislation. Government delays in undertaking these responses to fast-moving financial crises tend to exacerbate market panic and thus wealth shocks. These direct effects of crises are more easily discernible for voters compared to their indirect effects on income flows, which are less immediate and more complex. For example, crisis policy responses often have longer-term employment effects and fiscal costs that are difficult for voters to predict with confidence (Gandrud and Hallerberg, 2015). Given these severe informational disadvantages, voters will be drawn to more immediately observable signals associated with policy delay and gridlock, and less forgiving when it puts their wealth at risk. Thus, by inhibiting timely interventions, polarized veto players in the presence of great expectations should be more politically salient than in earlier crises and as compared to normal recessions, and worsen prospects for government survival.

Utilizing a new data set of 59 democratic countries stretching back to the early 19th century, we find strong support for our main argument. The survival prospects for incumbent parties in democracies experiencing banking crises decline sharply after 1970, especially in political systems characterized by polarized veto players. We find that neither varying crisis severity nor rising democratization can by themselves account for the changing political consequences of crises. Instead of seeing democracy as a constraint on cronyist bailouts during crises (Rosas, 2006, 2009), we take a more pessimistic view of democracy as a potential generator of bailouts, further financialization and political instability.

Our argument relates to a number of important themes and debates in social science. It builds on recent work by economists and economic historians that emphasizes how the accelerating ‘democratization of finance’ since the 1970s has tied middle-class wealth increasingly closely to asset and credit markets (Piketty and Saez, 2014; Piketty and Zucman, 2014; Turner, 2015; Jordà et al., 2015, 2016). Our macro-level perspective on how evolutionary changes in wealth distributions and financial systems shape political outcomes has implications for understanding long cycles in the global economy (Blyth and Matthijs, 2017). In emphasizing the centrality of evolving mass expectations for policy and politics, it contributes to arguments about the importance of ideas for political outcomes (Blyth, 2002; Berman, 2006) and the political economy of ‘everyday’ financial life (Seabrooke, 2007; Finlayson, 2009; Langley, 2009; Schwartz and Seabrooke, 2009). In showing how these changing expectations modify the effect of political institutions on electoral outcomes, it also supports historical institutionalist accounts of the importance of time and context (Pierson, 2004; Capoccia, 2016).

The next section of this article situates our argument in the existing literature. We then outline our methods and present our data and results. In the conclusion, we discuss their further implications.

1. The changing political aftermaths of banking crises

While scholars have devoted insufficient attention to the changing politics of banking crises, the ‘economic voting’ literature, which links voter evaluations of economic performance to election outcomes, provides insights on which we build (Anderson, 2007, pp. 278–281). This literature departed from an older ‘sociological’ tradition emphasizing the role of class and religious attachments in party attachment, arguing that party loyalty was eroding and that issues and voter assessments of competence had become more important (Lewis-Beck, 1988; Clarke et al., 2004). From this perspective, because crises often accelerate and deepen recessions and thus income flows (Reinhart and Rogoff, 2009; Jordà et al., 2013), they should tarnish incumbents’ reputation for economic competence and prompt voter punishment. An important claim in this literature, the ‘clarity of responsibility’ hypothesis, asserts that ‘the greater the perceived unified control of policymaking by the incumbent government, the more likely is the citizen to assign responsibility for economic and political outcomes to the incumbents.’ (Powell and Whitten, 1993, p. 398).

This idea has prompted work on the political effects of recent banking crises (Pepinsky, 2012; Crespo-Tenorio et al., 2014). One recent study of post-2007 electoral outcomes in Europe confirms the expectation that clarity of responsibility matters and that ‘crisis, defined as negative economic growth going into the election, negatively impacts incumbent vote share in a general way’. (Lewis-Beck and Lobo, 2017, p. 625). In our much longer sample, whereas post-crisis output and income losses since 1970 have generally been less severe compared to the pre-1945 period (Reinhart and Rogoff, 2009; Bordo, 2001), the survival rate of incumbent parties in the 5 years following a systemic banking crisis is significantly lower since 1970 than before 1945. This result is potentially consistent with the argument that class and religion-based party affiliation has declined since the war while economic performance has become more important.

But it also leaves open an alternative possibility on which we build: that what is salient for most voters when they assess governments’ economic performance has changed over time. In particular, we argue that the performance of voters’ wealth stocks has increased in salience compared to their income flows. We suggest that this has altered how institutional factors condition voter assessments of government performance in and after banking crises.

The ‘embedded liberalism’ literature emphasizes the societal consensus that emerged by the mid-1940s regarding the responsibility of governments to ensure full employment and income stability. The political imperative to protect society from deep financial crises of the kind experienced in the 1930s was an important element of this consensus, which justified financial repression within and between countries (Allen et al., 1938; Helleiner, 1994, pp. 25–50; Teichova et al., 1994; Barth et al., 2006; Busch, 2009). Wealth protection was not a policy priority in the early post-war decades. Indeed, wealth was viewed with suspicion and its redistribution was prioritized until the 1970s in many countries, with high top tax rates on incomes and wealth compressing economic inequality (Scheve and Stasavage, 2016).

This changed as the middle class accumulated wealth in the form of marketable assets. House ownership and pension provision became far more widespread in many democracies, driving a rapid increase in the real net wealth of the middle class despite rising wealth inequality in some from the 1980s.¹Figure 1a and b, which plots real net private per capita wealth in a number of advanced and emerging market democracies, shows how dramatic this wealth accumulation has been.² Distributionally, it is historically unprecedented. Available long-run estimates for Europe and North America suggest that wealth was highly concentrated before the 1940s. Middle-class households then held at best about 5% of total wealth, with an extremely narrow elite controlling most (Piketty and Saez, 2014, p. 839). From the 1980s, mortgage lending accelerated in many countries, promoting rapid increases in real house prices and housing equity, further benefitting middle-class homeowners (Mian and Sufi, 2014; Turner, 2015; Jordà et al., 2016). Pension asset values also grew dramatically, with a strong trend towards defined contribution and away from defined benefit schemes (Brooks, 2005; Piketty and Zucman, 2014). Many households’ interest in wealth protection accordingly also rose sharply.

The political salience of this trend was strengthened because many of these households also became more exposed to financial instability as governments embraced financial liberalization (Helleiner, 1994; Frieden, 2006, pt. IV). Banking crises reappeared after more than three decades of financial tranquility during the era of financial repression. Changes in ‘everyday saving’ habits also increased middle-class vulnerability to financial instability (Crouch, 2009; Finlayson, 2009; Langley, 2009). A shift toward ‘asset-based welfare’ and what Crouch describes as ‘privatized Keynesianism’ in many advanced democracies meant that households became ‘financial subjects’ out of necessity to provide for their retirement, to educate their children, to pay for health care, to supplement income and to access housing (Crouch, 2009). Rising economic inequality, by increasing competition for positional goods such as education and housing, fed the demand for credit, further boosting leverage and housing prices (Rajan, 2010; Jordà et al., 2015, 2016; Ahlquist and Ansell, 2017). Housing leverage grew rapidly in many democracies since the 1980s (Figure 2),³ increasing the risk of outsized losses during banking crises.⁴

We suggest that rising middle-class exposure to financial instability has fostered the recent historical emergence of mass “great expectations” concerning government responsibility for wealth protection. Unfortunately, extensive longitudinal data directly measuring these changing voter expectations are unavailable. Nonetheless, we identified two national household surveys on voter attitudes toward wealth protection, one in the Netherlands in 2010 and multiple waves in UK from 2003 to 2012, that provide evidence consistent with great expectations.⁵ About half of Dutch respondents believed that the banking supervisor should ‘never … let a bank fail.’ Three-quarters also incorrectly assumed that supervisors will refund any deposits when a bank goes bankrupt (van der Cruijsen et al., 2013). From 2003 to 2007, over 55% of British survey respondents with an opinion believed that in a crisis, the authorities would bail out some or all failing financial firms. When asked in 2007 to explain this view, the main reason (18%) given was ‘too many consumers would be affected’; others responded that ‘people would lose confidence in the financial system’ (6%) and ‘government would never allow consumers to lose money’ (5%) (Financial Services Authority (UK), 2009, p. 26). These expectations were not irrational, as government bailouts and intra-crisis extensions of deposit protection became increasingly common in democracies since the 1980s despite neoliberal policy rhetoric (Goodhart, 1999; Jeffrey and Andrew, 2019). Furthermore, many voters do not see financialization and their own rising vulnerability as matters of personal choice (Langley, 2009).

One straightforward observable implication of our argument is that governments that fail to protect middle-class wealth will suffer rising rates of voter punishment. Yet this prompts a further question. If we are correct that the political coalition favoring government financial stabilization has grown over time, why would any government risk disappointing it? Our answer is that institutional constraints can prevent elected governments from meeting voters’ great expectations. Polarized veto players in different branches of government, including those across and within parties in government coalitions, can hamper policy responses to economic shocks by raising transaction costs (Haggard and Kaufman, 1995; Haggard, 2000; Oatley, 2004). They can produce gridlock, delay and/or limited interventions. In contrast to slower-moving ‘normal’ recessions, fast-moving financial crises can greatly amplify the impact of such institutional constraints. If financial markets lose confidence in governments’ ability to respond quickly or adequately, market contagion can compound wealth losses in highly financialized economies (Alesina and Drazen, 1991; Tsebelis, 2002).⁶

This leads us to a different prediction than the clarity of responsibility hypothesis. By frustrating modern voters’ great expectations, such institutional constraints will prompt more rather than less voter punishment of incumbent governments. Modern voters with substantial wealth at risk are unlikely to be forgiving of institutionally-constrained governments since they will often be unable to discern whether bad policy outcomes in crises are the result of institutional constraints or incompetence. Delayed cost realization—in which politicians choose deliberately to hide them from cost-conscious voters (Gandrud and Hallerberg, 2015)—compound these informational asymmetries, drawing voter attention to more observable signals associated with policy delay and gridlock. Thus, the post-crisis electoral success of governments should be dependent not just on the response itself, but on its perceived timeliness and quality. Visible policy disarray and delay that generates wealth-destroying market contagion should reinforce the incompetence signal that the onset of the crisis has already sent.

Hence, our first empirical expectation is that severe banking crises will elicit greater electoral punishment in the modern era than the pre-1945 period when institutional obstacles produce delayed policy responses.

Second, in the modern era, banking crises should have more powerful political consequences than severe non-financial economic downturns because the former is a much greater threat to middle-class wealth stocks (Funke et al., 2016).

Third, modern voters should also react negatively to policy responses that are substantively redistributive and perceived as unfair. People are attached to norms of fairness and concerned with relative as well as absolute losses; they may also want contradictory things (Thaler, 2015, chap. 15). Governments facing greater institutional constraints may favor selective intervention benefitting concentrated interests rather than diffuse unorganized groups, compounding perceptions of unfairness (Alessandri and Haldane, 2009; Johnson and Kwak, 2010; Mian et al., 2014). In the US crisis from 2007, for example, most banks, senior creditors and depositors (including those uninsured ex-ante) were protected, whereas equity investors and house-owners often suffered large losses. These losses were almost certainly lower than they would have been without the interventions, but government appeals to such counterfactuals failed to convince many voters. Crises often produce zero-sum conflicts between creditors and debtors (Mian et al., 2014), widen economic inequality between rich and poor in the modern era due to higher leverage (Atkinson and Morelli, 2011; Bordo and Meissner, 2012), and sharply raise government spending and debt because financialization has made bailouts more costly (Alessandri and Haldane, 2009; Johnson and Kwak, 2010). These forces intensify distributional conflict and can increase ideological fragmentation in society (Mian et al., 2014; Funke et al., 2016). In addition, more extensive interventions in the modern era can disproportionately benefit the very wealthy, including protecting financial sector employees (Bank of England, 2012; Mian et al., 2014; Wolff, 2017). This compounds widespread perceptions of political and policy ‘capture’ by large financial institutions (Hacker and Pierson, 2010; Johnson and Kwak, 2010; Culpepper and Reinke, 2014).

There is considerable survey evidence consistent with the claim that modern voters will react negatively to perceived distributional unfairness in crisis aftermaths (Gallup, 2009; McCarty et al., 2013, pp. 234–237). On average, two-thirds of respondents in a 2013 Pew Global Attitudes survey of 39 countries believed the gap between rich and poor had increased in the 5 years since the 2007–2008 financial crises (Pew Research Center, 2013). This concern was significantly higher in crisis-afflicted countries (80.7%) than in crisis-free countries (61.6%) (t = 3.06, P<0.01). For six of these crisis-afflicted countries, we matched 2009 data from a BBC World Service Poll about distributional fairness to their views about 2007–2012 inequality trends (BBC World Service, 2009). In these countries, on average, two-thirds of respondents believed that recent economic outcomes were distributionally unfair. This correlated strongly with the percentage who thought inequality had risen between 2007 and 2012 (r=0.50).⁷

For these reasons, it can be exceptionally difficult for modern governments to convince middle-class voters that bank bailouts are the necessary cost of protecting their own wealth, incomes and jobs. Such governments face an acute dilemma. They can expect severe punishment if they fail to meet voters’ great expectations regarding wealth protection. When they attempt to meet them, success is far from guaranteed. Voter punishment is still likely if asset prices fall, if governments respond in a delayed and limited manner, or in ways perceived as highly costly and redistributive. Avoiding punishment depends in large part on delivering prompt, effective interventions that protect most middle-class wealth and spread the burden reasonably fairly between different groups (debtors, creditors, taxpayers and welfare beneficiaries). Later, we briefly discuss a case—Sweden in 2008—that approximates this outcome, but it is far from typical.

2. Data and method

We use a new database stretching from 1822 to 2013 to test our argument and its three empirical implications. Tables 1 and 2 provide the sample and summary statistics. Our primary interest is to assess how institutional constraints associated with the veto player environment condition the impact of crises on the incumbent political party’s hold on office. We first make use of survival data, following others by constructing an annual ‘Partisan spells’ indicator, which measures when the incumbent chief executive’s political party loses office (Crespo-Tenorio et al., 2014). This indicator allows us to abstract from institutional differences among democracies that shape electoral outcomes, including parliamentary and presidential regimes, term limits for individual leaders and the length of electoral cycles. We summarize our coding rules and sources in Section 2 of the Online Appendix.

We then extend the analysis by considering a ‘thinner’ version of political accountability, the vote share of incumbent political parties (Samuels and Hellwig, 2010). We use the last election in a given year when there are two or more elections in one year. We consider only elections for the chief executive and analyze the incumbent vote share of the president’s party in presidential systems and the prime minister’s party in parliamentary systems.

We identify democracies using data from Boix, Miller, and Rosato (Boix et al., 2013). Importantly, in addition to free and fair contestation, this dichotomous measure requires that democracies meet a minimal suffrage requirement (a majority of the male adult population). By using this measure, we aim to rule out possible objections that the changing political effects of banking crises may be due to suffrage expansion and democratization.

We focus on systemic banking crises rather than isolated banking failures, relying on the Reinhart and Rogoff (R&R) data set (Reinhart and Rogoff, 2009; Reinhart, 2010). In our sample, this yields 114 systemic crises in 59 democracies over 1822–2013.⁸Figure 3 plots the distribution of crises across our sample. We employ data on 59 crises in 21 democracies in the pre-1939 era, and 55 crises in 59 democracies in the post-war era. Our analysis below takes into account that the period 1945–1970 was one of unusual tranquility with no observed crises.

For the survival data, following Funke, Schularick, and Trebesch (Funke et al., 2016) and Mian, Sufi, and Trebbi (Mian et al., 2014), we code the Banking crisis measure as ‘1’ during the full five years after the start of a systemic crisis, and ‘0’ otherwise.⁹ If a new systemic crisis occurs in this 5-year period, we restart the 5-year horizon from the most recent crisis. In the R&R sample, we analyze 592 partisan spells, of which 173 overlapped with a post-crisis window (29.2%). We observe quantitatively similar results when we use the vote share data.

Our unit of analysis is the partisan spell. Figure 4 shows the basic structure of our survival data for the case of Britain from 1831–2011. The horizontal bars capture the duration of each partisan spell.

Our argument emphasizes the importance of ideological differences among veto players as a core element shaping government policy responses. The lack of available data on veto player preferences prevents us from directly capturing polarization for our broad sample. Instead, we use Witold Henisz’s Political Constraints Index Data set, which considers the number of independent branches of government, the extent of alignment across these branches and the heterogeneity of preferences within these branches (Henisz, 2017).¹⁰ Given the absence of data on veto player preferences, Henisz utilizes partisan fractionalization in the legislature to calculate preference heterogeneity for a broad sample of countries over two centuries. In addition to its international comparability and availability since 1800, we view the strength of this measure as based on the connection between the theory underlying it and our own argument. The Henisz measure is based on a simple single dimensional spatial model of political interaction that permits the status quo and the preferences of all actors to vary across the entire space. We follow Rosas in suggesting that government responses to banking crises, and political preferences regarding these responses, range on a continuum from ‘Bagehot’ to ‘bailout’ (Rosas, 2009). The single policy dimension underlying Henisz’s model thus provides a simple means to assess our argument about the influence of domestic institutional arrangements and political preferences. Finally, Henisz’s measure fits with our argument that additional political constraints on policy change in the form of homogeneity (heterogeneity) of party preferences within the opposition (governing) coalition have a positive but diminishing constraining effect on policy change (Henisz, 2002, p. 363). The Henisz measure is a continuous variable ranging between 0 and 1. Higher values indicate more veto players with different preferences.¹¹ We create an interaction term that combines this measure with the banking crisis variable, which enables an assessment of whether the effect of a banking crisis on incumbency survival varies across different veto player environments.

Our control variables include the age and level of democracy, growth, gross domestic product per capita, and prior history of financial instability using the cumulative number of systemic crises. All variables enter the model as annual data lagged by one year.

We control for the age and level of democracy (using Polity IV’s cumulative democracy score) because incumbents in more consolidated systems might have higher post-crisis survival rates (Marshall et al., 2017). We also seek to address the alternative hypothesis that voter punishment has increased because of the declining salience of class and religion in party attachments and the rising importance of performance-related indicators, such as growth and employment. Growth is a crucial control variable since we seek to account for its confounding effects on banking crisis onset and on incumbent party survival.¹² However, since severe downturns often follow banking crises, we later take additional steps to distinguish the direct wealth effects of the crisis from the indirect effects on income flows and employment in the recession that often follows. In the spirit of Barro and Ursua (Barro and Ursua, 2008), we also address this issue by comparing banking crises to severe recessions not involving a crisis but in which output declines exceed those that follow systemic banking crises. As noted above, a straightforward empirical implication of our argument is that political survival will be less likely following banking crises than after severe recessions, because of the rising salience of wealth effects. We follow Funke, Schularick, and Trebesch by identifying ‘non-financial macro-disasters’ where the annual output decline is higher than the average output decline that follows banking crises and occurs outside a five-year crisis window (Funke et al., 2016). We apply this cut-off separately for the pre-1945 sample (with a threshold of 4.5%) and for the post-1970 sample (with a threshold of 3.8%). On average, these disasters see an annual gross domestic product contraction of 9.8% in the pre-1945 sample and 6.9% in the post-1970 sample.

We control for prior history of financial instability because voter expectations of governments may be highest—and voter punishment most severe—in countries where crises have been less frequent, irrespective of the time period or institutional environment.

Using the survival data, we estimate a series of Cox proportional hazard models that model the expected length of a partisan spell for an incumbent party. Since we emphasize how changes over calendar time (years elapsed since 1800) shape the occurrence of partisan spell termination within a country—an event that occurs more than once—we use a conditional elapsed time model with stratification (Box-Steffensmeier and Jones, 2004). This allows us to frame our results as the likelihood of partisan spell termination in a specific year. This model can address the possibilities that partisan spells develop sequentially and that their timing differs across occurrences (or strata).¹³ For the vote share data, we follow the same specifications found in the survival analysis but use ordinary least squares.¹⁴ We code the Banking crisis measure as ‘1’ when a banking crisis began during the incumbent party’s term preceding the election, and ‘0’ otherwise.

3. Results

Our first step is to estimate survival models using the full sample from 1822 to 2010, excluding the 1946–1969 period because the absence of systemic banking crises in these years risks biasing the result (see Figure 3).¹⁵Table 3 provides the results. We plot the simulated marginal effect of a banking crisis as the veto player variable varies from its observed minimum to its maximum values and include a histogram of the distribution of the veto player variable.

Figure 5, utilizing results from Model 1, reveals that while the effect of banking crises is positive in the full sample, it fails to attain statistical significance for any values of veto players. Since Figure 5 may obscure the possibility of a time-dependent conditional relationship between crises and the institutional environment, we divide our data into two time periods that capture the long-term shift in voter expectations: 1822–1938 and 1970–2010. We then reestimate Model 1 across these different subsets and then plot the marginal effect by using the results from Models 2 and 3.

Figures 6 and 7 are consistent with our argument that changing societal expectations and institutional contexts have together reshaped political outcomes after banking crises. Figure 6 shows that in the period before 1939, banking crises have an insignificant and trivial effect on partisan spell termination for all observed values of veto players. Separately, we removed the interaction and found no unconditional relationship between crises and partisan spell termination, consistent with voters’ prevailing lower expectations regarding government policy responsibility. Since we limit our sample of democracies to countries where the electoral franchise was extended to a majority of the adult male population, these results are not likely due to suffrage restrictions. An illustration of this general result may be found in the aftermath of deep banking crises in 1907 in the USA and Canada, where long-entrenched incumbents undertook very limited policy responses and sharp recessions ensured (Conant, 1915, pp. 469–71; Bruner and Carr, 2007). Both governments retained office in elections in late 1908.

Figure 7 reveals that the conditional relationship between crises and veto players after 1970 is positive and significant. As the histogram shows, nearly 50% of observations fall in the range of statistical significance. Modern voters deliver a much harsher verdict to incumbent governments in higher veto player environments following a crisis, especially when compared to the pre-war era. These electoral judgments were particularly prominent in Latin America following crises in the 1980s, in Scandinavia in the early 1990s, in East Asia and Latin America in the late 1990s and early 2000s, and in Western democracies after 2008.

The magnitude of these effects can be indicated by comparing risk ratios for the post-1970 period. Comparing two governments in a ‘higher’ veto player environment, one experiencing a banking crisis is 2.21 [1.38, 3.41] times more likely to lose office than a counterpart in a tranquil financial setting.¹⁶ Comparing two crisis-hit governments, one in a higher veto player environment is 1.66 [1.03, 2.82] times more likely to lose office than one in a lower veto player environment. Notably, crisis-hit governments in lower veto player settings do not experience a significantly higher termination risk than counterparts in tranquil settings. These results support our argument about the very different expectations of post-1970s voters. The results from a Chow test provide additional supportive evidence, indicating that our findings are not stable across the two time periods in our sample.¹⁷

Results for incumbent party vote share, reported in Table 4, confirm the conditional relationship uncovered using the survival data. Although the marginal effects shown in Figure 8 reveal that banking crises lead to a significant decline in incumbent party vote share in higher veto player environments in the full sample, Figures 9 and 10 show that observations in the modern period once again underpin this result, with many observations falling in the range of statistical significance in Figure 10. The post-1970 era results suggest that following a crisis the vote share of governments in higher veto player environments drops by 14.44 [−21.67, −7.21] percentage points. This effect compares to a mean value of 36% for the incumbent vote share. We find no significant effect when post-1970 crises occur in lower veto player environments. The Chow test statistic is once again significant.¹⁸

We find conflicting and, at best, weak support for the conjecture that voter punishment has increased due to the rising importance of performance-related indicators such as growth. Growth has a significant effect on incumbent party performance in both eras in models reported in Table 4, but only before World War II in the models presented in Table 3. Moreover, the results from a Chow test fails to uncover a significant difference in the coefficients reported in Table 4 across the two eras, which casts further doubt on the idea that growth was more politically salient in the post-1970 era. On the contrary, these results provide highly suggestive evidence that the indirect risks associated with banking crises were present even in the pre-World War II era. It is in the modern era when our results suggest the direct risks to wealth stocks acquire the greatest political salience.

Indeed, we also find that the political impact of modern banking crises differs substantially from that of severe non-financial macro-disasters. In contrast to modern crises, in macro-disasters, voter punishment tends, if anything, to weaken in higher veto player environments (see Table 5). This benchmarking exercise thus also permits a clearer identification of the direct wealth impact of banking crises as distinct from the indirect impact via recession. It is consistent with our argument that institutional constraints will generally worsen the survival prospects for incumbent governments after modern crises because of their unusual impact on household wealth as well as on inequality and polarization compared to non-financial macro-disasters.¹⁹ We also show in Section 3 of Online Appendix that housing and stock market wealth tend to contract sharply after modern banking crises but not after macro-disasters. The former clashes with the great expectations of modern voters, but the latter may seem less threatening.

We perform a number of robustness checks, reported in the OA. Whether we use Laeven and Valencia’s (L&V) data set of post-1970 systemic crises (Online Appendix Section 4), a different sample of democracies using the Polity IV measure (Online Appendix Section 5), consider the role of austerity policies in the wake of banking crises (Online Appendix Section 6), include International Monetary Fund conditionality and Eurozone membership as additional institutional constraint control variables (Online Appendix Section 7), or address collinearity between veto players and democracy using residualization (Online Appendix Section 8), our results still hold. We also consider alternative indicators of the domestic structure of political competition (Online Appendix Section 9), including a measure of whether the executive’s party enjoys majority status in the lawmaking house(s) of the legislature, and another that captures the effective number of parties in the legislature. That we fail to uncover significant results for either suggests that our results are not merely a function of the presence of divided government or a consequence of turnover being more likely in more fractionalized party systems. Our results are also robust when we use entropy balancing to ameliorate concerns about endogenous selection of governments into banking crises as well as covariate balance and covariate overlap (Online Appendix Section 10).

4. Assessing causal mechanisms

Ideally, we would directly test the changing impact of the size, share and composition of middle-class wealth on post-crisis political outcomes. Unfortunately, such data—or plausible proxies for them²⁰—are unavailable, even for most advanced countries. Our empirical models offer a second-best alternative in this data-constrained environment.

We therefore turn to other supplemental tests of the causal mechanisms at work and report these results in Section 12 of Online Appendix. Using granular data on the buildup of financial market stress and on the timing of policy interventions, we find veto players to be a leading cause of policy delays during the 2008–2009 crisis. We also find in additional tests that veto players and longer policy delays heighten the buildup of financial stress over the electoral cycle, which in turn is linked to lower vote shares for the incumbent party following a crisis (Online Appendix Section 12; Online Appendix Table A.18).²¹

To further assess our claim that polarized veto player environments are prone to post-crisis policy gridlock, we searched for cross-national measures of legislative performance.

Excepting Binder’s data on legislative deadlock in the post-war USA (Binder, 2003, 2015), we were unable to identify data related to post-crisis stabilization during our time frame. USA is recorded with an intermediate value in the Henisz index, and many claim that its policy makers had a relatively ‘good’ crisis in 2007–2009 (Culpepper and Reinke, 2014; Drezner, 2014, p. 177; Geithner, 2015). If we find substantive policy delay with negative consequences for wealth and voter perceptions of unfairness in this case, then cases with higher values in the Henisz index are likely prone to larger effects. Importantly, this case also permits us to exploit the DW-NOMINATE (Dynamic Weighted Nominal Three-step Estimation) ideological scores for members of Congress to assess the effect of actual veto player preferences on post-crisis gridlock (Carroll et al., 2015).

The results we report in Online Appendix Section 12, Table A.19 are supportive of our hypothesized mechanism. We find polarization-induced greater policy gridlock in years following crises. Consider the pattern of delay during the 2007–2008 crisis (Hetherington and Rudolph, 2015). Policy makers were aware of the early solvency problems at the government-sponsored enterprises (GSEs)—Fannie Mae and Freddie Mac—and the stress building in the financial system since house prices started to decline in August 2006, and that public money would be needed to stabilize the banks (Acharya et al., 2011; Bernanke, 2015, p. 229; Frame et al., 2015; Geithner, 2015, pp. 169–170). Yet serious legislation action, including an April 2008 Treasury proposal to recapitalize the banks, was delayed amidst growing opposition to bailouts after the rescue of Bear Stearns in March 2007, and conflict within Congress over reforms to the GSEs and mortgage debtor relief (Wessel, 2009, p. 179; McCarty et al., 2013, p. 184). Anticipation of resistance from veto players thus delayed both the rescue of the GSEs and proposed legislation to recapitalize banks until financial stress had become so severe that their opposition could be overcome. As David Wessel observes, the highly polarized veto player environment had led Bernanke and Paulson to conclude ‘that there wasn’t any point in asking Congress – unless the crisis intensified to the point where there were no other options. Either way, it boiled down to the same result: waiting until it was too late’ (Wessel, 2009, pp. 179–80; emphasis added). Not until the failure of Lehman Brothers and American International Group in September, when complete financial collapse loomed, did the government propose its Troubled Asset Relief Program to Congress. Even then, only accelerating financial market collapse and further political negotiation ensured its ultimate passage, by which time massive and widespread wealth destruction had occurred.

Gridlock also contributed to a bailout that appeared highly selective and redistributive to many voters. Wall Street and senior bank creditors received a bailout but proposals for effective mortgage foreclosure relief foundered. Only limited help was provided to struggling homeowners, mainly due to Republican resistance (Barofsky, 2012, pp. 127 and 196–200; Nocera, 2011). These characteristics contributed to the heavy Republican defeat in the November 2008 election, as predicted by our argument.

Governments facing less polarized veto players can perform better in systemic crises. In Sweden, for example, ‘elite consensus facilitated a swift and effective crisis response’ (Bermeo and Pontusson, 2012, p. 15; see also McCarty, 2012). The center-right alliance government introduced a guarantee only 6 months after the initial buildup of financial stress, considerably faster than the median response of 11 months in our sample analyzed in Section 12 of Online Appendix. Its 2008 and 2009 stimulus packages were relatively uncontested (Lindvall, 2012). Unlike America’s Republicans, Sweden’s center-right government retained office in the 2010 elections for the first time in a century, even increasing its vote share.

5. Implications

Our findings indicate that the direct risks that systemic banking crises pose to household wealth have become much more politically salient in recent decades. This has substantially eroded prospects for political survival in democracies experiencing financial instability. This effect is most powerful when institutional constraints associated with polarized veto players prevent governments from satisfying voters’ great expectations. Voters appear to regard such institutional constraints on government responses to such crises differently than in ‘normal’ recessions and even in exceptionally severe macroeconomic downturns.

This modern political predicament is very different than in an earlier era when the vulnerability of most of society to financial shocks was mainly indirect. We are not suggesting that these indirect vulnerabilities are no longer important; the economic voting literature has shown that they remain so (Lewis-Beck and Lobo, 2017). However, because of the financialization of middle-class wealth, employment and income shocks associated with severe crises may be important in new ways that compound wealth anxiety, notably by jeopardizing leveraged households’ control and stake in key assets such as housing.

Our panoramic approach reveals the crucial importance of long cycles and time-dependent processes, including the changing stakes of middle-class households in financial stability. It also situates institutional political economy arguments in a much wider context, suggesting that the clarity of responsibility effect may be contingent on voter understandings of government policy responsibility, on the type of economic crisis and the size of the constituency vulnerable to wealth losses.

There are a number of related issues worthy of further investigation. Since women voters may demand greater trade protection (de Bromhead, 2018), it would be relevant to investigate gender differences regarding wealth protection. Future research could also consider how the “new interdependence” generates policy conflict and cooperation among jurisdictions during crises (Farrell and Newman, 2016). There is also potential to extend the ‘comparative disaster’ literature by investigating whether great expectations have emerged over the long run in other policy areas—such as pandemics or terrorism—and to what effect (Rodriguez et al., 2018).

The developments we identify also have broader consequences. Since modern governments facing great expectations find it difficult not to provide extensive wealth protection during crises, this may have generated ever-greater moral hazard and leverage. This fosters the very financial fragility that arouses deep voter and government concern and suggests an overlooked political origin for ‘Minskian’ cycles of credit boom and bust (Minsky, 1992; Haldane, 2013; Turner, 2015). Whether the ‘ideological strain’ generated by crisis interventions, and the fiscal austerity that often follows, is greater for left- or right-wing political parties is worthy of further research. Given the rising risk that most governments in democracies will not meet middle-class voters’ great expectations of wealth protection, this increases the fragility of political tenure and exacerbates distributional conflict. Severe banking crises can be a political gift to opposition parties, but the prolonged social conflict they often unleash can be politically unmanageable. This means that great expectations may have contributed to the erosion of voters’ faith in mainstream parties and democratic institutions.

Supplementary material

Supplementary material is available at Socio-Economic Review online.

Footnotes

Acknowledgements

An earlier version of this article was presented at the annual meeting of the International Political Economy Society, Madison, WI, November 11 and 12, 2011. The support of the Economic and Social Research Council (ESRC) in funding the Systemic Risk Centre is gratefully acknowledged [grant number ES/K002309/1]. This research also received financial support from J.M.C.’s Mid-Career Fellowship from the British Academy for the Humanities and the Social Sciences [MD130026], from Chwieroth’s AXA Award from the AXA Research Fund, from J.M.C. and A.W.’s Discovery Project award from the Australian Research Council [DP140101877], and from a seed fund grant from the Melbourne School of Government. The authors are grateful for helpful comments on earlier drafts provided by Christopher J. Anderson, David Andrews, Leonardo Baccini, Ken Benoit, Sarah Binder, Frederick Boehmke, Janet Box-Steffensmeier, Mark Buntaine, Richard Carney, Keith Dowding, Zachary Elkins, Eric Helleiner, Christopher Gandrud, Julia Gray, Dominik Hangartner, Philip Keefer, Luke Keele, Jouni Kuha, David Leblang, Amanda Licht, Christopher Meissner, James Melton, Paul Preston and Dennis Quinn. The authors would like to acknowledge the research assistance of Hortense Badarani, Wendy Chen, Noemie Chomet, Alexia Leckie, Alexander Parsons, Richard Reid, Bill Roosman and Sam Wilkins.

References