From Finance to Fascism: The Real Effects of Germany’s 1931 Banking Crisis

Sebastian Doerr Stefan Gissler José-Luis Peydró Hans-Joachim Voth

Abstract

Do financial crises radicalize voters? We analyze a canonical case – Germany in the early 1930s. Following a large bank failure in 1931 caused by fraud, foreign shocks and political inaction, the economic downturn went from bad to worse. We collect new data on bank-firm connections. These suggest a direct link from the banking crisis to both economic distress and extreme radical voting. The failure of the Jewish-led Danatbank strongly reduced the wage bill of connected firms, leading to substantially lower city-level incomes. Moreover, Danat exposure strongly increased Nazi Party support between 1930 and 1933, but not between 1928 and 1930. Effects work through both economic and non-economic channels. Greater economic distress caused by Danat led to more Nazi party votes. In addition, the failure of a Jewish bank was skillfully exploited by propaganda, increasing support above and beyond economic effects. This was particularly true in cities with a history of anti-Semitism. In towns with no such history, Nazi votes were only driven by the economic channel. Greater local exposure to the banking crisis also led to fewer marriages between Jews and gentiles just after the Summer of 1931, and to more attacks on synagogues after 1933, as well as more deportations of Jews.

Keywords: financial crises, banking, real effects, Great Depression, democracy. JEL codes: E44, G01, G21, N20, P16.

This version is from September 2018. Sebastian Doerr is at University of Zurich ([email protected]); Stefan Gissler is at the Federal Reserve Board ([email protected]); José-Luis Peydró is at ICREA- Universitat Pompeu Fabra, CREI, Barcelona GSE, Imperial College London, CEPR ([email protected]); Hans-Joachim Voth is at University of Zurich and CEPR ([email protected]). We would like to thank participants at Goethe University House of Finance-SAFE-Institute for Banking and Financial History “The Real Effects of Financial Crises: Past, Present, Future”, the 2018 European Summer Symposium in Financial Markets, the 1st Endless Summer Conference on Financial Intermediation and Corporate Finance, and The Halle Institute for Economic Research “Challenges to Financial Stability” conferences, seminar participants at CREI, Federal Reserve Board, MIT, and University of Zurich, Daron Acemoglu, Fabio Braggion, Fernando Broner, Paola Bustos, Stijn Claessens, Carola Frydman, Luc Laeven, Ralf Meisenzahl, Atif Mian, Isabel Schnabel, David Schoenherr, Moritz Schularick, and Peter Temin. We also thank Stéphanie Collet and the Research Center Sustainable Architecture for Finance in Europe (SAFE) for providing us with interwar data on German companies. Peydró acknowledges financial support from both the Spanish Ministry of Economics and Competitiveness Feder EU (project ECO2015-68182-P) and the European Research Council Grant (project 648398). The views in this paper are solely the authors’ and should not be interpreted as reflecting the views of the Board of Governors of the Federal Reserve. I. Introduction

Do financial crises fan the flames of fanaticism? From the election of Donald Trump to the recent elections in Italy, France, Austria, Germany, Sweden and Greece, the period since 2008 has witnessed dramatic gains for anti-establishment parties and politicians. On the 10th anniversary of the Lehman collapse in the summer of 2018, both the Financial Times and the New York Times featured op-eds arguing that "Populism is the true legacy of the financial crisis". At the same time, the driving forces behind the recent populist surge remain unclear: In addition to the aftermath of the financial crisis, rising concerns over immigration, growing income inequality, fiscal austerity, and the adverse effects of foreign trade have all been argued to play a role (Dippel et al 2016; Autor et al 2017; Becker and Fetzer 2017). While recent evidence (Mian et al. 2014; Algan et al 2017) and historical studies (De Bromhead et al 2013; Funke et al 2016) using aggregate cross-country data suggest that financial crises are linked to radical voting, there is a striking shortage of evidence based on micro data documenting a direct effect of financial shocks on political radicalization, and the economic and non-economic channels through which they operate. We examine the canonical case of a radical government coming to power amidst economic and financial disaster: The rise of the Nazi Party, which established a genocidal dictatorship responsible for starting World War II, causing millions of casualties. As aggregate GDP in Germany fell by 40% and unemployment surged towards 6 million, the Nazi party went from 2.6% of the popular vote in 1928 to 43.9% in March 1933. Economic analyses of the rise of the Nazis have mainly focused on the link between unemployment and radical voting. They find relatively little evidence that the German slump caused the Nazi Party to gain in the polls (Falter 1991; Childers 1983; Stachura 1978). Instead, voting studies find that the turn towards the Nazis was broad-based across social classes (Falter and Zintl 1988; Kolb 1997). The mass army of unemployed overwhelmingly supported the Communists, not the Nazis (Falter 1991; King et al 2008), suggesting only a weak direct link between economic distress and Nazi voting.1 In this paper, we show that the German banking crisis in the summer of 1931 was crucial in boosting the Nazi movement’s electoral fortunes. It not only aggravated the German economy’s downturn, leading to more radical voting because of declining incomes, but also increased the Nazis’ popularity directly: Their central, long-standing claim that “the Jews are

1 King et al. (2008) have argued that groups hurt economically by the depression did not all have the same interests. Rather, in their view, Nazi support was a case of ‘ordinary economic voting’ – those at low risk of unemployment turning to the Nazis and those at high risk turning to the Communists. 1 our [Germany’s] misfortune”2 was seemingly borne out by indisputable fact. The German banking crisis of mid-1931 was triggered by the collapse of Danatbank, one of Germany’s four universal banks. Danatbank had been led by the prominent Jewish banker Jakob Goldschmidt. The bank faced unsustainable losses because one of its borrowers, a large textile firm, defaulted. The shock to Danat resulted in a full-blown financial crisis because of both internal and external factors: The international financial system was under severe strain because of the concurrent failure of Creditanstalt, an Austrian bank. This led to foreign depositor bank runs in Austria and Germany. Political inactivity because of conflict between Germany and France over a proposed customs union with Austria undermined international cooperation. Because its own gold stock was low, and because foreign help was not forthcoming, the German central bank could not bail out Danat. The ensuing run on one bank led to a suspension of most bank deposits, a bank holiday, and Germany’s de facto exit from the gold standard.3 The banking collapse was quickly exploited by Nazi propaganda. We argue that the electoral surge of the Nazi Party occurred in part because one of its central messages – the hatred of Jews – went from extreme to mainstream after 1931 on the back of the banking crisis’ impact. We first present new evidence on the real effects of the German banking crisis, and then document the crisis’ effects on Nazi support and anti-Semitic attitudes. Instead of focusing on unemployment, we collect new data on the wage bill of firms. Shorter working hours and lower wages caused income losses, as did disruptions to the flow of credit and a collapse of demand from under- and unemployed workers. To identify the effect of bank failures on the real economy and voting, we use newly- collected data on firm-bank pair lending relationships from a contemporary directory of listed firms. Because German banks lent nationwide in the 1930s (in contrast to the US), we can use cross-sectional variation in firm- and city-level exposure to identify effects, despite the fact that the German banking crisis erupted after several years of recession. Information on bank connections was recorded prior to the banking crisis. There is no evidence that firms borrowing from the Danatbank were ex ante any more risky than client firms of other (great) banks – nor were they different in size, age, or leverage. Finally, our statistical results on the effect of Danat exposures do not change when we control for a plethora of additional factors and fixed effects,

2 This was the motto of Der Stürmer, a highly anti-Semitic Nazi weekly magazine that published the slogan on its front page in every issue. 3 For the banking crisis, see Born 1967, Kindleberger 1978, Ferguson and Temin 2003, Schnabel 2004. Kindleberger (1978) argues that the Austrian banking crisis was crucial for the German one. Ferguson and Temin (2003) highlight the inaction of German politicians, including the lack of bail-outs, while Schnabel (2004) emphasizes crisis of the Mark and the fact all the universal banks took substantial risks before the crisis. 2 suggesting that connection to the Danat is orthogonal to unobservable and observed firm fundamentals, following Altonji et al. (2005). Figure 1 illustrates our main results. The left-most panel shows the strong decline in lending by the two banks that collapsed in the summer of 1931, Danat and Dresdner. While loan volume declined by 10% at Commerzbank and Deutsche Bank, aggregate lending at Danat and Dresdner fell twice as much. Accordingly, at Danat-connected companies, liabilities also declined twice as fast as at unconnected firms (a reduction of 30% instead of 15%). Finally, the wage bill at Danat-exposed firms collapsed by 40%, while the average firm only experienced a reduction of 15% in wage and salaries.4 Cities with more firms exposed to the Danatbank prior to the crisis experienced larger income declines, and unemployment rose more after the banking crisis. These adverse effects were greatest in cities where Danat-connected firms were non- exporting – local demand spillovers there were larger. The right-most panel shows the distribution of changes in votes for the Nazis, for cities with high and low exposure to Danatbank (measured by the assets of connected firms). There is a clear shift to the right for cities with higher exposure, where cities that suffered more from the Danat collapse also saw the largest increases in support for the Hitler movement. Why did cities with higher exposure to Danat vote more for the Nazis, a singularly extreme party? Economic effects of the banking collapse could be responsible. These indeed mattered, but there is also evidence of direct effects, over and above the economic channel. With the financial crisis, the Nazis had seemingly incontrovertible proof for their misguided theories of Jewish domination and destruction. It was an easy story to tell and to sell: As in many other countries, Jews were vastly overrepresented in 1930s German high finance (Mosse 1987). Nazi propaganda consistently blamed the Jewish population for Germany’s economic ills. Figure B1 in the appendix reproduces a cartoon from Der Stürmer, a Nazi weekly, depicting a gigantic, all-powerful, obese Jewish banker hanging a small German businessman. Goebbels, later Minister for Propaganda, instructed the party in its quarterly journal for propaganda to emphasize that banking crisis was fully in line with the party’s anti-Semitic line. Diaries from the time, such as Hans Schäffer’s (a leading German-Jewish civil servant), point to a general surge in anti-Semitic sentiment after the banking collapse, even at the highest levels of society.5 To demonstrate the importance of both direct and indirect channels empirically, we first show that higher Danat exposure at the city level significantly increased Nazi Party support

4 Because they collapsed at the same time and were merged by the government, we combine Danat and Dresdner in our empirical analysis. In our paper, we refer to firms borrowing from Danat or Dresdner as ‘Danat connected’. Our results hold if we only use Danat instead of Danat and Dresdner. 5 Schäffer documents meetings of the cabinet in which members of the Catholic centre party openly discussed alleged plans by Jews to ‘take over’ the German economy. 3 between 1930 and 1933 elections (but not between 1928 and 1930). Danat exposure is based on firms’ ex-ante connection to Danat.6 An important mechanism by which Danat exposure increased Nazi votes is economic in nature.7 While unemployment changes have little predictive power, income decline predicted by exposure to Danat strongly increases support for NSDAP -- a one std. dev. decline in income is associated with a 9.6. percentage points rise in Nazi support (while the average change in NSAP vote share from 1930 to 1933 was 22 p.p.). In contrast, average income changes in general only increased Nazi votes by 1.7 p.p. Non-economic mechanisms from Danat exposure to the rise of the Nazi party are also crucial. After controlling for income changes,8 Danat exposure by local firms increased Nazi voting by the same order of magnitude as Danat-induced changes in income. We interpret this additional effect as evidence for the differential success of Nazi propaganda in affected cities, maliciously spreading a narrative of how Jews destroyed the German economy. In line with this, effects of Danat exposure on Nazi voting are greater in towns with an earlier history of anti-Semitism (as proxied by voting for anti-Semitic parties, 1890-1914). There, not only did income variation induced by the Danat collapse lead to an even stronger increase in NSDAP support, but the banking crisis’ direct effect – over and above economic fundamentals – on Nazi voting was markedly greater. In contrast, in localities without a prior history of anti-Semitism, the non-economic mechanism played no role, and income changes in general or predicted by Danat failure similarly explain Nazi voting. In other words, where hatred of Jews had no deep historical roots, there is no effect of the banking collapse on Nazi voting over and above to income effects. In combination, these results suggest that the banking crisis and Nazi propaganda directly created negative attitudes toward the Jewish population, mapping into Nazi votes – and all the more so in areas where the banking crisis caused the greatest suffering, and where it reactivated deep-rooted anti-Semitism. The Nazi Party stood for many things, and the importance of its anti-Semitic message before 1933 is not universally seen as key (Brustein 1998; Falter 1991). What independent evidence is there that relations between Jews and gentiles actually worsened more in towns and cities affected by the Danat collapse? We collect monthly data on Jewish mixed marriages for

6 Danat exposure is similarly significant for the change in elections between 1930 and 1932. Also, the level of Danat exposure is orthogonal to NSDAP voting in 1930. 7 Danat exposure affects both city-level change in income and unemployment. City-level Danat exposure or change in income affect Nazi votes, but not change in unemployment. This finding in particular provides new evidence for the hypothesis in King et al. (2008) that income losses amongst groups not threatened by unemployment generally benefitted the Nazis. Note also that trade shocks (global trade collapse) affected city-income but not Nazi votes. 8 City-level income and unemployment, in levels or also including polynomials. 4 a subsample of cities. Marriages are rare events and indicative of deep involvement between Jews and gentiles. And yet, mixed marriages also act as a ‘canary in the coalmine’, reflecting not only romantic attachment but also the social acceptability of intermarriage. Cities with more Danat exposure saw a sharp decline by 15-17% of mixed marriages just after the banking crisis; unaffected places saw no change (and neither did intra-faith marriages). The financial crisis had serious repercussions – and not only in economic terms and through voting for the Nazi Party. Anti-Semitism heightened by the banking crisis also led to more vociferous forms of hate even during the years after 1933. We find evidence that cities more exposed to the Danat collapse witnessed higher deportation rates of Jewish citizens to concentration camps, and more attacks on synagogues, Jews, and their property during the 1938 pogroms (“Reichskristallnacht”). Our findings contribute to three literatures – the real effects of banking crises, the effects of economic shocks on conflict and instability, and the history of the Nazi Party’s rise to power in Germany. Since Bernanke’s (1983) classic paper, a growing literature has documented the effects of financial crises.9 After the 2008-09 financial crisis, there is clear evidence that firms suffered from a credit crunch – a bank credit crunch as in Ivashina and Sharfstein (2010), or in Jiménez et al (2012). More recent evidence shows the real effects of the 2008 financial crisis (Chodorow-Reich 2014; Paravisini et al., 2014; Jiménez et al. 2017; Huber 2018; Bernanke 2018).10 Our main contribution is to show the impact of a banking crisis on political extremism, and not only the real effects of a credit crunch.11 Our evidence suggests that income reductions arising from a banking crisis can be special. In our particular case, income reductions due to the Danat crisis implied five times more Nazi voting than a similarly-sized income change not due to Danat. This is true despite of the fact that Danat- and non-Danat induced income changes have the same average and the left tail is similarly large. The conflict literature has investigated the effects of a variety of adverse economic shocks, such as rainfall variation or commodity price shocks. Results typically show that exogenously induced economic distress makes civil war and other forms of conflict more likely

9 See, for example, Reinhart and Rogoff (2009) 10 This evidence as well as recent theoretical work on the transmission of financial shocks to the real economy during the 2008-09 financial crisis is summarized in Gertler and Gilchrist (2018). For an analysis of the US Great Depression and its banking crisis, see Calomiris (1993) and Calomiris and Mason (2003). Benmelech et al (2017) use an empirical strategy similar to Almeida et al (2009), and demonstrate a link from the 1930s US banking crisis to firm employment 11 Financial crisis can bring medium- or long-term costs by leading to the wrong (economic) policies thereafter (Mian et al 2014). In the case of Germany in the 1930s, there is no doubt that the rise of the Nazi party to power ultimately triggered one of the worst catastrophes of history, the Second World War, with over 60 million casualties. 5 (Collier and Hoeffler 1998; Miguel et al 2004).12 Autor et al. (2016) use exposure to trade with China as a source of identification and demonstrate that US electoral districts were more likely to support extreme candidates the more adverse the trade shock was. Similarly, Dippel et al. (2016) argue that negative trade shocks increased support for radical right-wing parties in Germany in recent years. However, Funke et al (2016), analyzing financial crises over the past 140 years covering 20 advanced economies and more than 800 general elections, find that political extremism does not increase during normal recessions or after severe macroeconomic shocks that are not financial in nature. Instead of aggregate cross-country data, we use new micro-data (firm-bank connections and firm- and city-level data) in conjunction with a bank shock to show how a financial crisis radicalized voters and facilitated the rise of the Nazi power in 1933 via both economic and non-economic mechanisms.13 The rise of the Nazi Party has attracted extensive scholarly attention over the last 80 years. Initial analyses emphasized either class-based theories (Lipset 1960; Hamilton 1983) or theories of the masses (Ortega-y-Gasset 1932; Arendt 1973). The findings based on voting records have largely superseded this earlier literature, demonstrating that, far from being a party dominated and supported principally by members of the lower middle class, it was a “catch-all” party that drew support from all walks of German society instead (Falter 1981; Childers 1983). Nonetheless, some differences in the cross-section emerge: Research using district-level voting results shows that Protestants were much more likely to offer support than Catholics, that the better to-do increasingly turned towards the Nazis after 1930, and that the unemployed overwhelmingly supported the Communists instead.14 King et al (2008) use ecological inference to show that while a broad-based shift underpinned the Nazi’s rise to electoral success, some groups were more susceptible. This is especially true of the self-employed from high-unemployment areas, and domestic employees from low- to medium- unemployment areas. While few doubt that the rise of the Nazis was facilitated by the Great Depression (see e.g. Evans 2004; Kersaw 2016), in the literature there is as of now no compelling evidence that areas of Germany more affected by economic distress turned to the Hitler movement at the

12 At the same time, democratic transitions also appear to become more likely during periods of exogenously low income (Lipset 1960; Brückner and Ciccone 2009). 13 Two recent, innovative papers also study financial shocks and political outcomes using micro data and bank shocks. Braggion et al (2018) and Gyongyosi and Verner (2018) respectively analyze bank lending and social unrest in China in 1933 and bank FX lending in the 2008 crisis and political extremism voting in Hungary. 14 We do not find significant robust evidence that the banking failure led to more votes for the Commnists. 6 polls.15 By collecting new data on city incomes and bank-firm connections, we provide a direct link from economic distress to electoral success. We proceed as follows. We first provide historical context and background, then describe our data and empirical strategy. Next, we present our main empirical results, before discussing the robustness of our findings. Finally, we offer some concluding remarks.

II. Historical background

In this section, we briefly describe three aspects of the historical context – the Great Depression in Germany, the banking crisis of 1931, and the rise of the Nazi Party to power.

A. The Great Depression in Germany

Three features distinguish Germany’s Great Depression – its early onset, origin, and severity. While the US downturn began in 1929, Germany’s industrial output had already begun to contract in 1927. In contrast to the US experience, German investment began to fall first. Consumption only began to fall later, whereas in the US ‘autonomous’ declines in consumption kick-started the depression (Temin 1976, Romer 1990, Olney 1999). After 1929, declines in German output accelerated. Peak-to-trough, German industrial output fell by 40%, while the corresponding figure is around 20% in Britain and 10% in Japan. The only other major industrialized country with a similarly severe decline in economic activity were the US. In 1933 Germany counted 6 million unemployed, equivalent to one third of the workforce. Unemployment spelled misery, as elsewhere. While the unemployment insurance system looked after those losing their jobs, benefits became progressively smaller. After 20-27 weeks, the unemployed only qualified for emergency aid, which offered only minimal support.16 However, unemployment was only the most visible manifestation of economic misery. Many small business owners and ´´entrepreneurs suffered severe income declines. In desperate bid to balance the budget wages of civil servants were repeatedly cut, as were pension payments. Fiscal austerity was a distinguishing feature of the German slump. The federal government, states, and municipalities had borrowed heavily before 1929. A good share of the money raised came from abroad. Once international debt markets froze, authorities had to try

15 One notable exception is Galofré-Vilà et al (2018), who argue that austerity in the form of higher taxes was a key reason for pro-Nazi voting. When we control for austerity among other variables, our results remain unchanged in statistically significance and in the size of the estimated coefficients. 16 At the height of the depression, unemployment benefits became means-tested. 7 and balance their books by raising taxes and cutting expenditure. Germany’s export industries were not helped by the surge of protectionism after 1929. While already saddled with relatively high labor costs, new tariffs and difficulties in encountering export financing translated into rapidly falling sales of German products abroad. By 1933, German exports had declined by 70 % relative to their 1929 value.

B. The banking crisis of 1931

Germany’s 1931 banking crisis began in Austria. In May of the year, the Austrian Credit- Anstalt revealed large losses. When it collapsed, foreign deposit withdrawals accelerated in other countries, including Germany (Kindleberger 1978). While the Austrian banking crisis unfolded, huge losses at a German textile firm, Nordwolle, came to the attention of its main creditor, the Darmstädter Nationalbank (Danatbank or simply Danat). Nordwolle management had dabbled in ill-timed speculation, and also outright fraud. Losses on the loans to the textile firm were large, equivalent to 80% of Danatbank’s equity, and accordingly threatened the bank’s survival. Once Danatbank realized the scale of its losses, it turned to the German central bank, the Reichsbank. A bank holiday was declared for two days in July, transfers and other transactions remained barred for over a month (Ferguson and Temin 2003). While the Reichsbank attempted to offer support, its ability to do so was severely circumscribed by its commitment to the gold standard (James 1985, Schnabel 2004). International politics did not help. International support for the Reichsbank – for example by the Bank of England and the Banque de France – could have helped its attempts to shore up the banking system, and to stay on the gold standard. However, conflict between Germany and France had been brewing since the German government announced its plans to pursue a customs union with Austria. Lingering international tensions, one of the Versailles Treaty’s repercussions, undermined any bid for multilateral support. In the end, the Reichsbank failed to rescue the banks and it had to suspend convertibility of the Mark into gold. In addition, animosities between Deutsche Bank, Germany’s largest bank at the time, and Danatbank prevented a private solution to the banking panic (Born 1967).

C. The rise of the Nazi Party

8 From relatively obscure beginnings in post-war Munich, the Nazi Party grew in influence and membership during the hyperinflation. In 1923, it made a violent, but unsuccessful, bid for power, in the so-called Beerhall Putsch. After its bloody collapse, Nazi leaders were tried and sent to prison, Hitler most prominently among them; the party was declared illegal. Using his time in prison, Hitler wrote Mein Kampf (“My Struggle”) about his political vision and experiences so far. A growing number of prominent right-wing politicians beat a path to the door of his prison cell. After an early release from internment, Hitler returned to politics at the head of the newly-legalized party. While membership continued to grow, it initially had paltry success at the polls. In the Reichstag election in 1928, the Nazi Party received a mere 2.6% of the vote. During Weimar’s “Golden Years”, this right-wing fringe party was languishing in obscurity. All of this changed after 1929. As the Great Depression took an ever-greater hold in the German economy, politics turned acrimonious. The last democratically elected Chancellor Müller resigned in 1930, after a row over the rapidly rising cost of unemployment insurance. Thereafter, Chancellor Brüning governed without a parliamentary majority, supported by the emergency powers of President von Hindenburg. After its poor showing at the polls in 1928, the Nazi Party had changed its tune. It no longer advocated a violent overthrow of the established democratic order. Instead Hitler emphasized that only legal means would be used to come to power. As the party seemingly moved towards the political middle, it made itself acceptable to middle and upper classes. Hitler increasingly gave talks in front of businessmen. In the September 1930 election, the Nazi Party scored its first big success, winning 18.3% of the vote. The party gained 4.6 million voters between 1928 and 1930. Its success was in large part due to agitation against the Young Plan, a rescheduling of Germany’s reparations obligations: it lowered the annual payments in exchange for loans from abroad, but also lengthened debt maturity. While the plebiscite against ratification of the Young Plan was ultimately defeated, it provided a platform for the Nazis to argue that Germany was being enslaved for generations to come. After the 1930 election, the ranks of the Nazi Party continued to grow, as did its street- fighting group of paramilitaries, the Sturmabteilung (SA). As the depression worsened, Weimar Germany’s politics became more and more violent. Street fighting between Communists, Nazis, and Republican paramilitaries became commonplace, especially in large cities. Police frequently intervened, often with deadly results (Kurz 1988). The Nazi Party’s big electoral breakthrough came in March and July 1932, some nine to twelve months after the banking crisis. In two rounds, von Hindenburg narrowly defeated

9 Hitler for President of the German Reich; nonetheless, the Nazi Party’s candidate polled eleven million votes in the first round and 13 million in the second. In the July 1932 election, the Nazi Party received 13.7 million votes, its highest share in a fully democratic election. The Nazi Party had become the largest party in parliament, receiving more votes than social democrats and communists combined. Fully confident of his claim to the chancellorship, Hitler negotiated his entry into the government – and failed to convince an aging President von Hindenburg. By November 1932, in another round of elections, electoral support began to slip. Their vote count fell by 2 million. By late 1932 many political commentators confidently predicted that the Nazis were on their way out. Barely a month later, after lobbying from arch-conservative advisors around him, President von Hindenburg finally appointed Hitler as Chancellor, in a cabinet where leading Nazi politicians were in a minority. Nonetheless, within two months, the Nazis had staged another set of (partially free) elections, and taken over effective power in all of the country by March 1933.

III. Data and main variables

A. Data

We use various data sources for interwar Germany, several of them hand-collected and digitalized for the first time. The main challenge is to establish the connection between firms and banks, since historical data on individual loans are unavailable. Instead, we proxy bank- firm connections by collecting information on which banks paid out firms’ dividends. Each year, investors would receive dividends at the branches of their main bank for each listed company (so-called “Zahlstellen”). This information is provided in the Handbook for German Stock Companies (“Handbuch der deutschen Aktien-Gesellschaften”), a yearly 4,000-pages compendium of balance sheet information for each listed German company. We use yearbooks for 1929 and 1934 to collect firm-level data on bank connections and balance sheet items.17 We begin our analysis on the firm level. In a first step, we collect data on total assets, total capital, as well bank connections in 1929 for 5,610 firms. For each firm, we record whether “Zahlstelle” lists Danatbank or Dresdner bank, or one of the other Great Berlin Banks (Deutsche Bank, Commerzbank, BHG). Since the focus of the first part of our analysis is how the banking crisis

17 Each handbook generally reports balance sheet data either for July of the previous year or February of the current year. For example, the 1934 Handbook reports July 1933 or February 1934 data items, which is why the 1934 Handbook reflects firm balance sheets during the trough of the recession in 1933. 10 affected employment and wages, in a second step we identify all firms reporting their wage bill in 1929 and 1934. For this sample we collect pre-crisis balance sheet items in 1929 on founding date, total assets and capital, return on assets, dividends, industry, and city, as well as total wage bill in 1929 and 1934.18 This results in a sample of 386 firms in 239 cities and 20 industries. Of these, 59 firms list either Danatbank or Dresdner Bank as “Zahlstelle”, 63 firms list any of the other Great Banks, and 17 borrowed from Danat/Dresdner and at least one other Grossbank.19.For the sample of firms reporting their wage bill, we additionally collect information on liabilities in 1929 and 1934, which are available for a total of 258 out of our 386 firms. We use the term “Danat” for firms connected with either Danatbank and Dresdner Bank, as both were merged in 1932 and suffered similar declines in lending; we also use the terms connected and borrowing interchangeably. We then analyze our data at the city level. We use the universe of listed firms in the 1929 Handbook to collect data on city level exposure to Danat (explained in detail below). For a total of 187 cities, we assemble information on city population and unemployment from 1930 to 1934 from the Statistical Yearbooks of German Cities (“Statistisches Jahrbuch deutscher Städte”), as well as total city labor force from the 1933 census. We gather data on major German federal elections in May 1928, September 1930, July 1932, November 1932, and March 1933. For each election, we record the number of votes for the different electable parties at the city level from German federal statistics (“Statistik des Deutschen Reichs” (ICPSR 42)). In addition, we collect information on destroyed and damaged synagogues (Alicke 2008) and deportations (Bundesarchiv). The 1925 census provides information on the share of blue-collar workers, Protestants, and Jews for each city. We also collect new data on city income in 1928 and 1934 from Germany’s statistical handbooks (“Statistik des Deutschen Reichs, Neue Folge 1884- 1944”, bulletins 378 and 492).20 To shed light on the underlying mechanism and control for pre-existing trends, we also collect data on votes cast for anti-Semitic parties in elections during the late Imperial period (1890-1914). Our final data set contains 187 cities for which we have data on elections, income,

18 While today data sources like Compustat provide easy access to comparable information across firms, historical handbook data do not. While for some firms we have information on the dividend-paying bank, assets, and wages, other firms provide none of this information. There are no filing requirements or any consistent form of balance sheet across firms. Firms that report a wage bill in 1929 can be missing in 1934 for several reasons: they do not report the wage bill anymore; they exited the market; they delisted; or they merged. Unfortunately, we can only verify the first of these points and thus analyze the intensive margin only. 19 Grossbank is the German word for Great Banks and is the common term used to refer to the large Berlin banks. 20 The German government collected data on city incomes every two years, but due to tight budgets not in 1930. Hence, 1928 and 1934 are the closest available data before and after the crisis. 11 and exposure to Danat. Table A1 in the Appendix gives a comprehensive overview of variables used in our analysis and the sources of the data.

B. Measures of bank-firm connections and city exposure

In the first part of our analysis, we look at firms’ responses to the collapse of Danat. To measure a firm’s connection with a bank, we create two dummies. Danat equals 1 if in 1929 a firm’s bank affiliation lists Danatbank or Dresdner Bank, and 0 otherwise; Grossbank equals 1 if the firm is connected to any of the other Great Banks. As dependent variable we use the growth in total wage bill from 1929 to 1934, denoted by D wages. The change in wage bill reflects changes in employment, as well as average wage, i.e. firms’ total payroll. As controls we use pre-crisis firm age and firm size (log of total assets), leverage, and return on assets. In the second part, we analyze the aggregate effects of a credit crunch and move to the city level. To study the effects of Danat’s failure and subsequent credit reduction on cities in Germany, we establish a measure of city exposure to Danat or Dresdner. In each city, we sum across firms connected to Danatbank or Dresdner Bank. Each firm is weighted by its reliance on external (bank) financing and its relative size in a city. In our main analysis, we proxy firms external credit needs by their leverage ratio (defined as liabilities over capital). City c’s exposure to Danatbank is calculated as

/0120/030!%- 1%%!3%- !"#$%&'!( = + ,-( . 5 ∗ 7 8 ∗ 91:13- (1) 41#031/- 1%%!3%( -

is an indicator for whether firm f is located in city c. Each observation is then weighted by firm f’s share of assets out of total firm assets in city c. To construct exposure, we use data on the universe of listed firms in 1929 that report total assets and total capital (N=5,610). Our main outcome variables are the change in city income from 1928 to 1934 (dire economic conditions forced the German government to omit data collection in 1930) and the change in NSDAP votes from 1930 to 1933. The change in income is defined as the growth rate from 1928 to 1934. The change in NSDAP votes is the change in the share of votes from September 1930 to March 1933. To further measure radicalization, we define the dummy NSDAP majority that takes on value 1 if the NSDAP won the majority of votes in 1933, but not in 1930 for each city. Additionally, we define the dummy synagogues that takes value 1 if a synagogue was damaged or destroyed in a city after 1933; as well as total deportations from

12 1933-1945 over total Jewish population in 1933, as in Voigtländer and Voth (2012). For robustness, we construct the change in the unemployment rate from 1930 to 1933. The unemployment rate is defined as the yearly unemployment over total labor force in 1933. We also use city’s exposure to exports as an additional exogenous shock to income and unemployment. From 1929 to 1934, Germany’s total exports declined by almost 70% from 13,486 billion Reichsmark to 4,178 billion Reichsmark.21 While most of the decline happened prior to 1932, the fall in export volume likely contributed to the recession. To measure a city’s exposure to the decline in exports, we first aggregate firm assets to the city-industry level. We then match total industry exports in 1929, provided by the German Statistical Yearbook (“Statistisches Jahrbuch des deutschen. Reiches 1930”), to each town-industry cell. Finally, we aggregate to the city level, where we weight each industry i by its share of assets in the respective city. Thus, exposure of city c to industry exports is given by

1%%!3%> !"#$'3%( = + 7 8 ∗ 3$31/ !"#$'3%> (2) 1%%!3%( >

For our analysis, we define exports as log(1+exports). The assumption is that cities’ industrial structure was predetermined at the time of the Depression. To further shed light on the mechanism connecting the banking crisis to changes in income and unemployment, we define tradable and non-tradable industries as industries in the top and bottom tercile in terms of export shares. Similarly, we define cities as high exports and low exports cities if they are in the top and bottom tercile of variable exports.

C. Descriptive statistics

Table 1 shows summary statistics for main variables. Panel (a) displays firm-level variables. The overall economic decline is clearly visible by an average decline in total wage bill by 19.5 percent. The average firm is 30 years old and relatively large, reflecting the fact that our sample covers listed companies. Since only 386 out of 5,610 firms report their wage bill, the issue of sample selection arises. For example, larger firms could be more likely to report their wage bill. To show that our subsample of firms that report their wage bill is representative of the universe of listed firms, Figure 3 panel (a) compares the distribution of log(assets) for the sample of firms that report their wage bill in 1929 (386 observations) and the universe of listed firms in

21 Sozialgeschichtliches Arbeitsbuch III, 1978. 13 1929 (5,610 observations). While the full sample shows slightly more dispersion, both distributions are similar and the difference in means is insignificant. This suggests that our sub- sample of firms reporting a wage bill is representative of the average listed firm in terms of size. Next, we ensure that firms borrowing from Danat are comparable in terms of pre-crisis riskiness to other borrowers. If Danat-connected firms were significantly riskier prior to the crisis, the fall in wage bill reflects weak demand, instead of changes in loan supply. We define firm leverage (or debt-to-equity-ratio) as liabilities over capital and compare leverage for firms borrowing from Danat or Dresdner, any other Grossbank, and other banks for the full 1929 firm sample. Figure 3 panel (b) shows that Danat (blue solid line) and Grossbank (red dashed line) borrowers were almost identical in terms of pre-crisis leverage. Firms that borrowed neither from Danat nor Dresdner, nor any other Grossbank (black dashed line), had higher leverage instead. Thus, in the full sample of listed firms, firms borrowing from Danat or Dresdner were not riskier before the crisis. Finally, Figure 4 plots the geographical distribution of Danat- connected firms and Danat-exposed cities in panels (a) and (b). Danat-connected firms are spread out over Germany, reflecting the fact that Germany’s large banks served borrowers all across the country. Hence, there is no geographical clustering of Danat-connected firms. Table 2 panel (a) compares the main firm-level variables for the treatment (Danat) and control group (Other) used in our baseline specifications for the sample of 386 firms reporting their wage bill. The 59 firms connected to Danat are significantly older and larger, but have lower leverage. They are similar in terms of return on assets and capital-to-labor ratio (wage bill over assets). In panel (b) we restrict the control group to the set of 63 firms that are connected to other large banks. Except for return on assets, the sample between both groups is mostly balanced, which alleviates concerns about self-selection of firms and banks. If anything, we would see Danat borrowers as more likely to weather a crisis: lower leverage, larger size, and higher profits should make them better able to withstand a recession than smaller and riskier firms. Thus, we expect that the selection bias arising from bank-firm pairings works against us.22

22 One could still be concerned that Danat extended credit to inferior firms during its rapid expansion in the 1920s. In 1922 Darmstädter Bank für Handel und Industrie merged with Nationalbank für Deutschland, creating Darmstädter und Nationalbank. Over the following years Danat expanded by opening branches all over Germany, and aggressively poached clients from other banks. By 1929 Danat had become the second largest German bank. Tougher competition forced banks to fight for deposits and increase their risk taking (for a detailed discussion of the German banking sector before the crisis, see Schnabel 2004). In particular, anecdotal evidence suggests that

14 Figure 1 panel (b) shows that negative wage growth is also reflected in our aggregate city-level data. On average, cities lost 14.3 percent of their income. The NSDAP won on average 22.3 % of votes from 1930 to 1933 and won the majority in 71 % of the cities where it did not win in 1930. Panels (a) and (b) of Table 3 compares cities with positive Danat exposure (exposed) to all remaining cities (not exposed) and cities with positive Grossbank exposure. In panel (a) Cities more exposed to Danat were on average larger and had a lower share of blue-collar workers than cities with zero exposure. Danat cities also had a higher share of Jews, but the difference is less than 0.5% in absolute terms. In terms of 1930 unemployment rate, share of votes cast for NSDAP, as well as the share of Protestants, there are no significant differences. These differences and similarities persist when we limit the sample to cities with positive exposure to Grossbanken in panel (b), although differences narrow. To shed further light on how firm and city characteristics differ across groups, Table 4 runs multivariate regressions on the firm (columns 1 to 4) and city level (columns 5 to 7). Columns (1)-(4) use the dummy Danat as dependent variable. In the multivariate analysis, Danat-connected firms are significantly larger and have lower leverage, similar to univariate statistics (column (1)). Column (2) adds industry fixed effects, results remain similar. Once we repeat the analysis for the sample of firms

Danat’s CEO Jakob Goldschmidt was thought of engaging in reckless lending by providing extensive credit to financially unsound firms. If Danat selectively picked borrowers with weak fundamentals during its period of rapid expansion, our estimates are biased: Danat borrowers would be less able to weather the general economic downturn. The decline in wage bill growth would then due to weak firm fundamentals, and not because of a contraction in credit supply to Danat borrowers. To ensure that our coefficient on Danat exposure reflects the negative effects of a loan supply shock on firms’ wage bill, we hand-collect data on borrowers from Darmstädter Bank für Handel und Industrie and Nationalbank für Deutschland in 1922, before the merger. If firms that already borrowed from either bank in 1922 were similarly or less risky compared to borrowers in 1929, we can rule out that Danat acted irresponsible when acquiring new borrowers. In the Online Appendix, we compare balance sheet information in 1929 for firms that borrowed from either bank in 1922 (1922 borrower) with all other firms in the wage-bill sample (all other firms). In a further test presented in the Online Appendix, we compare firms that borrowed from Danat in 1922 and 1929 to firms that did not already borrow from Danat in 1922, but were clients by 1929 (all Danat borrowers). Except for size and age, differences are not statistically significant. The significant difference in age and size is to be expected, as firms that already appear in 1922 are older almost by definition. 1922 borrowers have insignificantly higher dividend payments, return on assets, and profits, as well as lower leverage, which suggests that 1922 borrowers were slightly sounder than firms that connected with Danat later.

15 connected to other large banks, differences disappear (columns (3)-(4)). With industry fixed effects, coefficients on all explanatory control variables are insignificant. At the city level in columns (5) to (7), positive exposure to Danat is significantly correlated with city size, and negatively correlated with the share of blue-collar workers. Once the sample is reduced to cities with positive exposure to large banks (column (6)) or cities for which we have information on historical anti-Semitism in column (7), most differences are no longer statically significant. In column (6) cities with exposure to Danat only had a slightly higher share of Protestants.

IV. Empirical strategy

A. Firm level At the firm level, we model the effect of Danat’s collapse on firms’ wage bill as:

ΔA1B!%- = C + E 91:13- + 4$:3'$/%- + F- (3)

is the change in firm f’s total wage bill between 1929 and 1933 and reflects the change in total firm expenditures on employees.23 The main independent variable is , which is a dummy that equals 1 if firm f was connected to Danat or Dresdner in 1929 and 0 otherwise. denotes pre-crisis firm controls log total assets, age, return on assets, and leverage. Additionally, in some regressions we include dummy that takes on value 1 if a firm was connected to any other Great Bank in 1929. We expect that a contraction in loan supply by Danat leads to a decline in firms’ wage bill if they borrowed from Danat, so . To account for the fact that shocks to firms within the same city might be correlated, we cluster standard errors at the city level.24 Our research design relies on two assumptions. First, that the German banking crisis was exogenous to borrower characteristics and loan demand. Second, firm connections to banks were sticky and their options to obtain credit from other sources were limited. We discuss each assumption in turn. Banking crises and recessions often go hand in hand and it is difficult to establish causality (Reinhard and Rogoff 2009; Schularick and Taylor 2012). If some banks lend to

23 As described in Section III, data in the 1934 handbook usually records 1933 balance sheet items. 24 In the online appendix, we show that results are robust to clustering on the main bank-city or main bank- industry level, as well as the city-industry level. 16 riskier firms, deteriorating firm performance and bank failures coincide (Kiyotaki and Moore 1997; Khwaja and Mian 2008; Jimenez et al 2012). Disentangling changes in banks’ loan supply from changes in firms’ loan demand thus poses a major identification challenge. The German Banking crisis offers a unique setting to address these issues. First, the reduction in credit by Danatbank and Dresdner Bank was exogenous to all but one single firm in Germany. Nordwolle, a large textile company, had been expanding at a rapid rate during the latter part of the 1920’s. Its stock was rising, and Danatbank and Dresdner Bank were not hesitant to lend ever larger sums. In 1931 Danat had a sizeable loan to Nordwolle outstanding, equal to 80 percent of Danat’s equity. In May 1931, Danatbank discovered that Nordwolle had forged their books and a large bet against wool prices caused large losses. When news broke in June of the same year, investors realized that the failure of Nordwolle was imminent. Already thinly capitalized, Danat’s equity was in peril and the situation was equally dire at Dresdner Bank; depositors started to run. Danatbank and Dresdner Bank saw massive outflows from domestic and foreign depositors. On July 12, Danatbank’s liquidity was depleted and it could no longer open its branches. The national bank holidays imposed by the government did not ease the problems and Danatbank’s and Dresdner Bank’s businesses experienced a major disruption. As Figure 1 shows, loan supply to their borrowers decreased substantially. Crucially for identification, this shift in loan supply was unanticipated: similar to Enron in the 2000s, the sudden discovery of accounting fraud had unanticipated negative consequences. Since Germany’s large banks served clients all over Germany, the unexpected default of Nordwolle affected firms all over Germany through banking linkages. In the US during the Great Depression, local banks served clients in local markets. This makes it difficult to disentangle demand and supply effects during a local shock. In contrast, since the largest German banks operated on a national scale, deposits were taken and loans provided in towns across the country. Crucially, by including city fixed effects, we compare firms that depended on Danatbank or Dresdner Bank with other firms in the same city, while controlling for heterogeneity and common shocks at the local level. Moreover, by comparing coefficients and across specifications, we can gauge how sensitive firm connections to Danat are to unobservable factors (Altonji et al 2005, Oster 2017). Even if a shock that leads to a banking crisis is exogenous to firms’ loan demand, swift government intervention often disrupts demand and supply simultaneously in the immediate aftermath of a banking failure. Yet neither the German government nor the central bank interfered in a substantial way. The government’s fiscal position was too weak to save a bank. The central bank’s gold reserves were at historic lows and liquidity provision was not an option.

17 International opposition prevented a public solution, and resistance by Deutsche Bank a private solution. This resulted in a prolonged crisis; it took over one year to merge Danatbank and Dresdner as a minimal response to the crisis. With government help unavailable and bank lending distressed, firms often turn to alternative sources of funding. However, the German economy was based on close ties between banks and firms. Bank directors often sat on firms’ supervisory boards and banks held a substantial equity stake in connected companies. As a result, switching costs were often prohibitive, making firms’ bank relationships sticky. On top, other banks in the German economy also experienced deposit withdrawals, although to a lesser extent. In the face off the overall economic conditions in 1931, other banks would have had difficulties to meet a sudden increase in credit demand from firms banking with Danat or Dresdner.

B. City level Did bank failures radicalize voters? To address this question, we aggregate firm connections to the city level. Our main specification at the city level is:

H( = C + E !"#$%&'!( + 4$:3'$/%( + F( (4)

is an outcome variable such as the change in income or votes for the NSDAP in city c. is our main explanatory variable of interest; it is city c’s exposure to borrowers from Danat and Dresdner, calculated from firm-level data (given in equation (1)). To further control for confounding factors we include city latitude and longitude, log population in 1930, as well as its share of Protestants, Jews, and blue-collar workers in 1925 out of its total population. In some regressions we also control for city exposure to other Grossbanken, as well as exposure to the decline in exports. We use robust standard errors. We expect that cities with a higher exposure to Danat-borrowing firms see a stronger contraction in bank lending and thus a decline in income ( ) and an increase in votes for the NSDAP ( ). Regression equation (4) relies on two main assumptions. First, cities with more firms connected to Danatbank or Dresdner Bank were more affected by the failure of those banks than other cities. Second, bank failures had a larger impact on a city if connected firms were larger or had a higher need for external financing. When firms lay off workers or cut their wages as a result of a banking crisis, local economies suffer. As the number of affected firms grows, the situation worsens. Our sample of firms covers the universe of listed firms in 1929 and therefore represent the largest firms of the German economy. While the number of firms may

18 be low in some cities, it is a sensible assumption that our firm sample represents a significant part of each city’s labor market.25 To measure a city’s exposure, we focus on banks’ assets side—the impact of changes in lending on firms in a given city. Yet it is not only the sheer number of firms that had adverse effects. What also matters is how important connected firms were for the local economy and how they could cope with a loss in credit. We take this into account by weighting firms by their size and their external financing needs (see equation (1)). While regression equation (4) controls for several city characteristics, unobservable shocks could still bias our estimates. To address this issue, in alternative specifications we run panel regressions of the following form

H(J = C + E!"#$%&'!( ∗ log(/$1:%)J + 403H NO( + 30P! NOJ + F(J (5)

Outcome variable y is the unemployment rate or log number of listed firms in city c in year t.26 We interact city exposure with log loan volume of Danat and Dresdner over time. As controls, we include yearly log city population. The panel specification allows us to control for common shocks through time fixed effects, as well as unobservable city characteristics through city fixed effects. A decline in loan volume should lead to a stronger increase in unemployment and fall in the number of firms if a city had higher exposure to Danat. We cluster standard errors at the city level.

V. Main results

A. Firm-level results

This section presents results for firm-level regression equation (3). Table 5 shows that firms that borrowed from Danat saw a significantly stronger decline in their total wage bill and liabilities. Without controlling for covariates, Danat borrowers’ wage bill fell by 25 percentage points than those of firms that did not have connections with Danatbank or Dresdner Bank (column 1). Column (2) adds firm characteristics to control for the fact that Danat borrowers were on average larger and older. After controlling for firm size and age, return on assets, and leverage, as well as connections to other large banks, Danat borrowers still had significantly

25 For example, the 1925 German business register reports that companies with five or more employees employ 42 % of the workforce, but represent only 9.7 % of all companies (Rahlf (2015) : Deutschland in Daten. Zeitreihen zur Historischen Statistik, BpB, Chapter 18, Table 1, p252) 26 Unfortunately, yearly income data is unavailable. 19 lower wage bill growth (-19.8%). Column (3) allows for heterogeneity by industry. Industry fixed effects for 20 distinct industries absorb any unobservable characteristics that affected all firms within an industry, such as changes in exports.27 The coefficient on Danat remains significantly negative, but falls somewhat in magnitude compared to column (2). A possible concern is that local shocks bias our estimation. Since Danat served clients all across Germany, we can include city fixed effects to absorb unobservable differences across cities. Since several cities only have one firm, our sample size falls once we include city fixed effects. To make results comparable column (4) first restricts the sample to firms located in cities with more than one firm. Borrowing from Danat leads to an almost identical effect on wage bill growth as in our baseline specification with the full sample in column (2) (-19.6% vs -19.8%). In column (5) we control for common citywide shocks. The coefficient increases in size and remains significant and negative. Column (6) additionally adds industry fixed effects. After absorbing common shocks to cities and industries, the coefficient on danat remains significant and close in magnitude to the baseline specification (column (4)). Comparing two firms within the same city and industry, borrowing from Danat reduced the wage bill by 25.4 percentage points. Importantly, the R2 increases by a factor of 10 from column (4) to column (6), while our coefficient of interest remains similar. This suggests that Danat connections are orthogonal to the (considerable) variation captured by firm observables and unobservables (Altonji et al., 2005, Oster 2017). Column (7) addresses the concern that Danat acquired a selection of risky borrowers during its rapid expansion since 1923. We use the dummy Danat1922 (equal to one if a firm already borrowed from Danat already in 1922, before Danat’s rapid expansion). The coefficient is negative and significant, and similar in magnitude to our baseline results in column (2). Hence, we conclude that our main coefficient is not biased by Danat’s selection of firms after its merger. Overall, results in columns (1)-(7) show a strong negative effect of connections to Danat on firms’ wage bill growth: pre-crisis borrowing from Danat or Dresdner reduces firms’ total expenditure on employee wages and salaries by 15 to 25 percentage points during the crisis. Column (8)-(10) use the subsample of wage bill-reporting firms that also report their liabilities in 1929 and 1934. Similar to results for the wage bill, firms with connections to Danat reduced their total liabilities (column (8)). Comparing specifications without and with city and industry fixed effects shows that the coefficient on danat is similar across specifications, although R2 increases significantly. Hence, danat likely reflects changes in loan supply and is orthogonal to several potentially confounding demand factors on the city and industry level.

27 We lose two observations in column (2) since some industries only have one firm. 20

B. City-level results

In this section, we analyze wage changes and voting behaviour in cities hit by Danatbank’s failure. The turmoil at Danat and the subsequent credit crunch for related firms led to a significant reduction in city-wide economic activity, as measured by income. Table 6 provides the results of a regression of the change in income between 1928 and 1934 on city exposure to Danatbank. Column (1) shows that without controls or fixed effects, higher exposure significantly decreases city incomes. Taking underlying socio-economic differences across cities such as the share of Protestants into account, the coefficient increases significantly (column (2)). Moving a city from the 50th to the 90th percentile in terms of exposure to Danatbank reduces city incomes by an additional (0.24 x 0.295=) 6.9% (48% of the average decline or 0.38 standard deviations). In column (3) we further control for common shocks through province fixed effects for Germany’s 15 electoral areas. The magnitude decreases only slightly when we allow for heterogeneity across regions and the coefficient remains significant. Importantly, since Germany underwent a period of harsh austerity (Galofré-Vilà et al. 2018), which was decided at the state level, use of state (and province) fixed effects allows us to control for changes in public spending that could have radicalized voters. Columns (4)-(6) repeat the previous specifications, yet restrict the sample to cities with strictly positive exposure to Danat or Dresdner. In all cases, this slightly strengthens the effect of Danat's credit crunch on cities' local economies. Hence, even within the sample of cities that have positive exposure to Danat or Dresdner, higher exposure significantly reduces incomes. Danatbank’s failure reduced incomes. Did it also lead to radicalization among the electorate, and if so, through which channels? Table 7 suggests that voters radicalized. Regressing a city's change in the vote share for the Nazi party between 1930 and 1933, higher exposure to Danat significantly increased voting for the Nazis. (column (1)). The effect is even stronger once we take regional differences into account in column (2). Moving a city from the 50th to the 90th percentile in terms of exposure increases the NSDAP vote share by 2.2% (9.8% of the mean or 0.41 standard deviations). How important was the increase in votes due to the banking crisis? If the NSDAP won cities with high exposure by a large margin, the additional effect of the banking crisis may be statistically significant but not decisive to lift the Nazis to power. Columns (3) and (4) use an indicator variable equal to 1 if the Nazis won the majority in 1933, but not in 1930, and 0 otherwise. Results in both columns suggest that the banking crisis was important for the Nazis'

21 seizing power. Cities at the 90th percentile in terms of exposure have a 10.9% higher probability of an NSDAP majority than cities at the 50th percentile. So far, we have only looked at voter movement to the extreme right in response to the banking crisis. Columns (5) and (6) study voter behaviour toward the opposite side of the spectrum. The dependent variable is a city's change in vote share for the communist party, the KPD. Yet the evidence for a voter shift to the extreme left is rather weak. While significant in the baseline specification, the coefficient is significantly smaller and becomes insignificant once we allow for heterogeneity across regions. The radicalization we observe on the right part of the political spectrum in response to the banking crisis is not visible on the left. While we control for city characteristics and regional heterogeneity, we may still pick up voters' reaction to the general economic downturn the Great Depression brought about. Alternatively, cities with higher exposure could generally favour conservative or right-wing parties. To exclude this possibility, we run a placebo test of our main specification. We use the change in the Nazi Party's vote share between 1928 and 1930 as the outcome variable. Over the course of these two years, the Great Depression severely affected the German economy. As voters went to the polls, unemployment had increased. Yet voters in cities with higher exposure to Danat did not flock to the NSDAP more than in other cities, as column (7) shows. There is an insignificant negative effect of higher exposure on Nazi vote shares. This provides support for the hypothesis that our results reflect the impact of the banking crisis in 1931, and not the impact of the Great Depression that had already been underway once Danat went bust. Two years after the banking crisis, voters were comfortable with voting for a party that promoted anti-Semitism. Did these voters just turn a blind eye toward the Nazis' hateful racial agenda, or did the banking crisis have an effect on people's tolerance for the myriad of crimes that followed? After the Nazis came to power, the fate of Germany’s Jews took a turn for the worse. Columns (8) and (9) use the destruction of synagogues between 1933 and 1945 and the share of deported Jews as dependent variables. In towns and cities hit by the banking crisis, the persecution of Jews was markedly worse than in unaffected areas. The probability of a damaged or destroyed synagogue during the 1938 pogrom was significantly greater if a city had more exposure to Danatbank. In addition, these cities deported a markedly higher share of Jews during the Holocaust. While the use of only these two variables does not do justice to the cruelty of the Nazi regime and its victims, these findings suggest that anti-Semitic sentiment triggered by the banking crisis had dire consequences even years after Danat's failure. Voters not only radicalized at the ballots, they also radicalized in their actions.

22 C. Mechanism

Why did the banking crisis increase the support for the Nazi party? There are two plausible channels through which banking distress may increase support for extremist parties. First, Danat’s default led to economic misery by reducing incomes, and voters with a negative income shock might have increasingly supported the NSDAP. Additionally, Nazi propaganda, centered on blaming the Jewish population for the depression, may have increased anti-Semitic sentiments. This section shows that both channels were at work. While falling incomes lead to radicalization, exposure to the banking crisis had effects on radicalization above and beyond pure economic effects. We argue that Hitler’s propaganda machine exploited the fact that Danat had been led by a prominent Jewish banker, Jakob Goldschmidt. Danat’s failure allowed the Nazi party to scapegoat the Jewish population for Germany’s economic woes, despite the fact that the crisis originated at Nordwolle. Anecdotal evidence suggests that during the crisis, anti- Semitic newspapers featured several articles that paint the distorted picture of Jewish bankers ruining the ‘German business man’ (see Figure XX). Nazi propaganda also invented wild conspiracy theories, alleging that inter alia that in 1932 Chancellor von Papen was carrying out orders ``of his Jewish superior, Jakob Goldschmidt’’.. Table 8 disentangles the overall effect of an income increase and the effects of the crisis above and beyond economic factors on voting for the extreme right. Columns (1) and (2) show that the decrease in income had a direct effect on voter polarization. Cities where income fell more turned significantly more toward the Nazi Party. A one standard deviation drop in income is associated with (0.19 x 0.057) 1 % percent more votes for the NSDAP (0.19 standard deviations). The decline in income used in columns (1) and (2) stemmed from various sources, including the banking crisis. Columns (3) and (4) concentrate on the income loss due to the Danat collapse. We first regress the overall change in income in a city on city exposure to Danat to separate the part of income loss explained by the banking crisis and the income loss orthogonal to Danat's failure. Using only the income loss inflicted by the banking crisis to predict voter polarization, the effect increases dramatically compared to the effect of overall income loss. A one standard deviation percent income loss generated by the banking crisis increased Nazi electoral gains by 10 percentage points in column (4). While income and predicted income could be collinear, the correlation between both is 0.11. In columns (5) and (6) we include both variables in regressions and coefficients remain similar in sign, size, and magnitude.

23 Finally, in columns (7) and (8), we return to our measure of a city's exposure to Danat. Yet similar to the exercise in the preceding columns, we isolate the effect of exposure on radicalization that is not due to economic factors. In a first step, we regress the change in NSDAP votes on the change in unemployment and income. We then use only the residual as explanatory variable and regress it on exposure. The strong positive and significant coefficient suggests that the banking crisis radicalized voters even after taking into account any direct economic effects through falling incomes and rising unemployment. Its effect on voters' movement to the right goes above and beyond pure economic effects. Why did some voters increasingly flock to the Nazi party, even above and beyond any shift in behaviour that could be attributed to economic hardship? Table 9 explores this question in more detail and shows that Nazi propaganda needed fertile ground to flourish: pre-existing hostility toward Jews. Anti-Semitism historically differed across regions, with attitudes toward Jews being persistent over long periods of time (Voigtlaender and Voth 2012). We use these differences across cities to split our sample into two groups: cities with above-median historical anti-Semitism and cities with below-median anti-Semitism. We measure historical anti- Semitism using vote shares for anti-Semitic parties in elections around the late 19th century. Columns (1) and (2) show that the effect of Danat exposure on voter radicalization found in the full sample is driven by cities where anti-Semitism has deep historical roots. There, moving a city from the 50th to the 90th percentile in terms of exposure led to an increase in NSDAP votes by 2.5%.28 Moreover, this effect is independent of economic circumstances. In contrast, exposure to Danat had a positive, but only insignificant effect on voters moving to the right in cities with low anti-Semitism. In columns (3) and (4) we use the change in NSDAP votes orthogonal to changes in income and unemployment as the dependent variable. The effect above and beyond economic conditions is only present in areas where people were there was pre- existing anti-Semitism. The fact that extreme voting was less driven by economic factors in cities with high historical anti-Semitism than in cities with low anti-Semitism is confirmed in columns (5) and (6). While a fall in income has a significant effect on NSDAP voting in both areas, the magnitude of the economic effect is smaller in anti-Semitic cities. In other words, economic factors are more important for radicalization in areas with no history of anti-Semitic violence. This picture changes when we use income changes explained by the banking crisis. Columns (7) and (8) provide a stark contrast: In cities with low pre-existing anti-Semitism, income losses due to the banking crisis did not push voters into the arms of the Nazis. Yet in cities where

28 The difference in effect size across columns (1) and (2) is significant at the 17% level. 24 residents were more receptive to the Nazis' hateful anti-Jewish propaganda, the opposite is true (column 8). This effect emerges despite the stronger effect of income losses on radicalization in areas with low anti-Semitic attitudes. This result still holds when using both variables of the disaggregated changes in income (columns 9 to 11); if anything, differences in coefficients across specifications provide stronger support for the role of non-economic factors in anti- Semitic areas, since the effect of income (increase) decreases, but the effect of predicted income (decreases) increases in (low) high anti-Semitic areas when including both factors. As a final piece of evidence, we show that the banking crisis also affected relations between Jews and gentiles more generally, and not only voting. We collect monthly data on Jewish mixed marriages for a subsample of cities. Cities with more Danat exposure saw a sharper decline of mixed marriages, with a sudden change shortly after the outbreak of the banking crisis. Table 10 shows the results for a difference-differences specification, where we regress monthly data on the log number of mixed marriages in a city on an interaction term whether the city was exposed to Danat’s failure and whether the month was after July 1931. Each specification includes city and/or year fixed effects to control for unobservable city- specific characteristics, as well as common shocks. Columns (1) to (3) support our explanation that the banking crisis increased anti-Semitism. Cities more exposed to the failing bank experienced a significant decline in mixed marriages—marriages between Jews and non-Jews declined by 17.3% compared to pre-crisis times after controlling for common trends and unobservable city characteristics. This effect is not driven by an overall decrease in the frequency of Jewish marriages. Columns (3) to (5) use the log number of Jewish marriages as a placebo test. Cities exposed to the crisis did not see a significant change in Jewish marriages. While a rare event and indicative of deep involvement between Jews and gentiles, we consider it as a ‘canary in the coalmine’, reflecting not only romantic attachment but also the social acceptability of marrying across religious and ethnic lines. In combination, our results strongly suggest that Nazi propaganda successfully exploited the banking crisis, and that it was particularly successful in areas where it reactivated deep-rooted anti-Semitism – and that it affected relations between Jews and gentiles far from the voting booth as well.

VI. Robustness

This section addresses various concerns about the robustness of our results. In a separate Online Appendix, we show that our results are robust to different specifications such as quantile

25 regressions, alternative clustering of the standard errors, and the exclusion of individual firms, regions, or industries from the sample. So far, our analysis is based on a sample of firms connected to two banks, Danat and Dresdner. Consequently, we disaggregate our main explanatory variable to gain a deeper understanding on the relative importance of each bank. While both banks were creditor to Nordwolle, their experience during the crisis differed, with Danat being hit harder. If the banking crisis was the leading factor of our results, we expect these differences between Danat and Dresdner to have trickled down to their connected firms. Table 11 shows that our main result is comprised of a combination of the effects on firms connected to the individual banks. First, firms connected to either of the two banks suffered more than other firms. Second, Danat- connected firms experienced a larger effect on their wage bill than Dresdner-connected firms. These results are therefore in line with prior expectations and further strengthen our claim that the banking crisis (as opposed to the Great Depression in general) led to a decline in firms’ wage expenditures. The decline in wage bill for firms only connected to Danat is almost double as large compared to firms only connected to Dresdner. Yet, for both types of connected firms, the decline in wage bill is stronger (and significant) when compared to exposure to other Great Banks. A related concern is that given the sample size, outliers may have a significant effect on the results. However, Figure 5 Panel (a) shows that the coefficient from our main specification does not depend on outliers. We re-estimate the coefficient and change the sample every time by excluding each firm one-by-one. The coefficient is very stable across these different samples. While individual firms may not be a concern, firms located in a specific regions are. The region of Bremen did experience the fall of Nordwolle, which may have had significant effects on the local economy. If Danat had a strong foothold in Bremen and was lending to several firms, our effects may rather reflect Bremen’s economic distress and not the banking crisis. A similar argument can be made by firms located at the border with Austria, which saw a dramatic crisis in May 1931 as the Oestrreichische Credit-Anstalt barely escaped collapse. The third region of concern is the Ruhr region, where a large part of German economic activity was concentrated. An overrepresentation of firms in that region may limit the economic significance and representativeness of our findings for Germany as a whole. To address this concern, the final columns in Table 11 shows that our firm-level results are robust to excluding each of those regions.

26 On the city-level, we argue that the banking crisis did not only lead to a decrease in income but also to radicalization above and beyond economic reasons. The Nazis preyed upon the overrepresentation of Jews in the banking sector. To further strengthen the claim that propaganda had a significant effect, we use the difference between the ease of the population to connect Danat and Dresdner to the Jewish population. While Danat was run by one of the most prominent and visible figures of the German banking system, the link between Dresdner and Jewish bankers was less obvious. Table 12 therefore disaggregates cities’ exposure into exposure to Danat and exposure to Dresdner. The table shows that Danat-exposed cities were the driving force behind income decline as well as radicalization. For all outcome variables, the effects on Danat-exposed cities are larger than the effects on Dresdner-exposed cities. Second, qualitatively the results support a negative effect of the banking crisis stemming from both banks, although for Dresdner the effects are now less precisely estimated. And third, in a regression of NSDAP vote changes that are orthogonal to economic factors on both bank expose measures, Danat-exposure played a larger role than Dresdner-exposure. Cities that suffered from the crisis mainly because of Danat’s failure could be more easily convinced that Jewish bankers were to blame for citizens’ misery. Still, the concern persists that the Great Depression (even in only Danat-exposed cities) played a larger role than the banking crisis. Our main results are based on cross-sectional regressions and we cannot pinpoint when between 1930 and 1933 the main changes occurred. To alleviate that concern, we interact our city-specific measure of exposure with the yearly (log) loan volume given out by Danat and Dresdner. This allows us to construct a panel specification of our main regressions. Given the lack of a consistent time series of income data, we use the unemployment rate as well as the number of firms per capita as outcome variables. Table 13 shows that even when we are able to control for city-specific heterogeneity as well as year- specific differences and several fixed effects, the banking crisis had significant effects on local economies. Given a certain level of exposure to Danat or Dresdner, a city’s unemployment rate rose and number of firms decreased in response to a decrease in the banks’ loan supply. In addition to these economic effects, we showed that the banking crisis had significant effects on the elections in 1933, yet not in 1930. However, in 1933 a considerable amount of time had passed since the banking crisis and other factors in addition to the banking crisis may be at work. Table 13 addresses this concerns and shows the consistence of our results when using elections closer to the banking crisis. By showing that Danat-exposure had a significant effect on voting in both elections in 1932, we are able to narrow the window where Danat- exposure starts to matter to 1930 to 1932. The banking crisis falls squarely within this window.

27 While this window is now fairly narrow and the banking crisis seems the most plausible explanation, other changes in the German economy still occurred within this window. The large decline in Germany’s international trade being one major factor. The decline had been going on since the start of the Great Depression. If Danat-exposed cities were also large exporters, we may measure a response to a trade shock instead to a shock to the financial system. Table 14 therefore uses a city’s (log) exports as explanatory variables. And while exports do explain some part of the income decline at the city-level, they do not explain voter radicalization. If at all, a city’s openness and trade activity seems to dampen some of the effects of the banking crisis. When we sort cities into low-export and high-export cities, low-export cities exposed to Danat experienced a larger decline in incomes.

VII. Conclusion

This paper documents three main empirical facts. First, we show that firms that were connected to Danatbank – Germany’s second-largest bank in 1929 – suffered more when it collapsed. In particular, we show that employment and wages declined more sharply in firms that had previously paid their dividend through offices of Danat, which we argue is a reliable indicator of firm-bank linkages. Second, we show that distress originating in the financial sector led to increases in aggregate unemployment, and the more so in locations with stronger links to banks’ in distress. We create a measure of city-level exposure to Danat and demonstrate that unemployment surged more where larger and more numerous firms were exposed to its collapse. Third, economic distress induced by the banking collapse provided a major boost to the electoral fortunes of the Nazi Party. Unemployment changes after 1930 in Germany as a whole have no predictive power for radical voting – but in towns and cities affected by Danat’s collapse, this is radically different. We also offer evidence that these effects were sharper where the Nazis could tap into pre-existing anti-Semitic sentiment. Finally, we demonstrate that the severity of local persecution of Jews after 1933 was greater where Danat’s bankruptcy had caused the greatest economic harm.

28

References

Abelshauser, W., Faust, A., & Petzina, D. (1978). Sozialgeschichtliches Arbeitsbuch III. C.H. Beck.

Algan, Y., Guriev, S., Papaioannou, E., & Passari, E. (2017). The European Trust Crisis and the Rise of Populism. Working Paper.

Almeida, H., Campello, M., Laranjeira, B., & Weisbenner, S. (2009). Corporate Debt Maturity and the Real Effects of the 2007 Credit Crisis. NBER Working Paper, 14490, 1–36.

Altonji, J. G., Elder, T. E., & Taber, C. R. (2005). Selection on Observed and Unobserved Variables: Assessing the Effectiveness of Catholic Schools. Journal of Political Economy, 113(1), 151–184.

Arendt, H. (1973). The Origins of Totalitarianism. New York: Harcourt Brace Jovanovich.

Autor, D., Dorn, D., Hanson, G., & Majlesi, K. (2017). Importing Political Polarization? The Electoral Consequences of Rising Trade Exposure.

Benmelech, E., Frydman, C., & Papanikolaou, D. (2017). Financial Frictions and Employment during the Great Depression. NBER Working Paper, 23216.

Bernanke, B. S. (1983). Nonmonetary Effects of the Financial Crisis in the Great Depression. American Economic Review, 51(3), 257–276.

Bernanke, B.S. (2018): The real effects of disrupted credit, Brookings Papers on Economic Activity.

Bowlby, C. (1986): Blutmai 1929: Police, parties and proletarians in a Berlin confrontation. The Historical Journal, 29(1), 137-158.

Born, K. E. (1967). Die deutsche Bankenkrise 1931. Finanzen und Politik. Munich: R. Piper & Co. Verlag.

Brückner, M., & Ciccone, A. (2011). Rain and the Democratic Window of Opportunity. Econometrica, 79(3), 923–947.

Brustein, W. (1998): The logic of evil: The social origins of the Nazi Party, 1925-1933. Yale University Press.

Childers, T. (1983). The Nazi Voter. Chapel Hill: North Carolina Press.

Chodorow-Reich, G. (2014). The Employment Effects of Credit Market Disruptions: Firm- level Evidence from the 2008-09 Financial Crisis. Quarterly Journal of Economics, 129(1), 1–59.

29 Collier, P., & Hoeffler, A. (1998). On Economic Causes of Civil War. Oxford Economic Papers, 50(4), 563–573.

De Bromhead, A., Eichengreen, B., & O’Rourke, K. H. (2013). Political extremism in the 1920s and 1930s: Do german lessons generalize? Journal of Economic History, 73(2), 371–406.

Dippel, C., Gold, R., & Heblich, S. (2016). Globalization and Its (Dis-)Content: Trade Shocks and Voting Behavior. NBER Working Paper, 21812.

Duchin, R., Ozbas, O., & Sensoy, B. A. (2010). Costly external finance, corporate investment, and the subprime mortgage credit crisis. Journal of Financial Economics, 97(3), 418– 435.

Evans, R. J. (2004). The Coming of the Third Reich. Penguin.

Falter, J. W. (1981). Radicalization of the middle classes or mobilization of the unpolitical? The theories of Seymour M. Lipset and Reinhard Bendix on the electoral support of the NSDAP in the light of recent research. Historical Social Research/Historische Sozialforschung, 4(26), 389–430.

Falter, J. W. (1991). Hitlers Wähler. C.H. Beck.

Falter, J.W., & Zintl, R. (1988): The economic crisis of the 1930s and the Nazi vote, The Journal of Interdisciplinary History, 19(1), 55-85

Ferguson, T., & Temin, P. (2003). Made in Germany: The German currency crisis of July 1931. Research in Economic History, 21, 1–53.

Funke, M., Schularick, M., & Trebesch, C. (2016). Going to extremes: Politics after financial crises, 1870–2014. European Economic Review, 88(2011), 227–260.

Galofré-Vilà, G., Meissner, C. M., Mckee, M., & Stuckler, D. (2017). Austerity and the Rise of the Nazi Party. NBER Working Paper, 24106.

Gertler, M., & Gilchrist, S. (2018): What happened: Financial Factors in the Great Recession. Journal of Economic Perspectives, 32(3), 3-30

Goodman, S. (2015). The Orpheus Clock: The Search for My Family’s Art Treasures Stolen by the Nazis. Simon and Schuster.

Hamilton, R. (1983). Who Voted for Hitler. New Jersey: Princeton University Press.

Huber, K. (2018). Disentangling the Effects of a Banking Crisis: Evidence from German Firms and Counties. American Economic Review, 108(3), 868–898.

Ivashina, V., & Scharfstein, D. (2010). Bank lending during the financial crisis of 2008. Journal of Financial Economics, 97(3), 319–338.

James, H. (1985). The Reichsbank and Public Finance in Germany, 1924–1933. A study of the politics of economics during the great depression. Frankuft a. M.: Knapp Verlag.

30 Jiménez, G., Ongena, S., Peydró, J.-L., & Saurina, J. (2014). Hazardous Times for Monetary Policy: What do Twenty-Three Million Bank Loans Say About the Effects of Monetary Policy on Credit Risk Taking? Econometrica, 82(2), 463–505.

Jiménez, G., Ongena, S., Peydró, J.-L., & Saurina, J. (2012). Credit supply and monetary policy: Identifying the bank balance-sheet channel with loan applications. American Economic Review, 102(5), 2301–2326.

Kershaw, I. (2016). To Hell and Back: Europe, 1914-1949. Penguin.

Khwaja, A. I., & Mian, A. (2008). Tracing the Impact of Bank Liquidity Shocks: Evidence from an Emerging Market. American Economic Review, 98(4), 1413–1442.

Kindleberger, C. (1978). Manias, Panics, and Crashes: A History of Financial Crises. New York: Basic Books.

King, G., Rosen, O., Tanner, M., & Wagner, A. F. (2008). Ordinary economic voting behavior in the extraordinary election of Adolf Hitler. Journal of Economic History, 68(4), 951–996.

Kiyotaki, N., & Moore, J. (1997). Credit Cycles. Journal of Political Economy, 105(2).

Kolb, E. (1997): Was Hitler’s seizure of power on January 30, 1933 inevitable? German Historical Institute, 18

Koudijs, P., & Voth, H.-J. (2016). Leverage and beliefs: Personal experience and risk-taking in margin lending. American Economic Review, 106(11), 3367–3400.

Kurz, Thomas. " Blutmai": Sozialdemokraten und Kommunisten im Brennpunkt der Berliner Ereignisse von 1929. Dietz, 1988.

Lipset, S. M. (1960). Political man: The social bases of politics. Baltimore: John Hopkins University Press

Mian, A., Sufi, A., & Trebbi, F. (2014). Resolving debt overhang: Political constraints in the aftermath of financial crises. American Economic Journal: Macroeconomics, 6(2), 1–28.

Miguel, E., Satyanath, S., & Sergenti, E. (2004). Economic Shocks and Civil Conflict: An Instrumental Variables Approach. Journal of Political Economy, 112(4).

Mosse, W.E. (1987): Jews in the German economy: The German-Jewish Economic elite, 1829-1935, Oxford: Clarendon Press

Olney, M. L. (1999). Avoiding Default: The Role of Credit in the Consumption Collapse of 1930. Quarterly Journal of Economics, 114(1), 319–335.

Ortega y Gasset, J. (1932). The Revolt of the Masses. W. W. Norton and Company, Inc.

Oster, E. (2017). Unobservable Selection and Coefficient Stability: Theory and Evidence. Journal of Business and Economic Statistics, 1–18.

31 Reinhart, C. M., & Rogoff, K. (2009). This Time Is Differnt: Eight Centuries of Financial Folly. Princeton: Princeton University Press.

Romer, C. D. (1990). The Great Crash and the Onset of the Great Depression. Quarterly Journal of Economics, 105(3), 597–624.

Schäffer, H. (n.d.). Tagebuchaufzeichnungen und andere Papiere. Retrieved from http://digifindingaids.cjh.org/?pID=122626

Schnabel, I. (2004). The German Twin Crisis of 1931. Journal of Economic History, 64(3), 822–871.

Schularick, M., & Taylor, A. M. (2012). Credit Booms Gone Bust: Monetary Policy, Leverage Cycles, and Financial Crises, 1870-2008. American Economic Review, 102(2), 1029–1061.

Stachura, P. (1978). Der kritische Wendepunkt? Die NSDAP und die Reichstagswahlen vom 20. Mai 1928, Vierteljahreshefte fuer Zeitgeschichte, 26, 66-96

Temin, P. (2008). The German crisis of 1931: Evidence and tradition. Cliometrica, 2(1), 5– 17.

Voigtlander, N., & Voth, H.-J. (2012). Persecuation Perpetuated: The Medieveal Origins of Anti-Semitic Violence in Nazi Germany. Quarterly Journal of Economics, 127(3), 1339– 1392.

32 A Tables and Figures

A.1 Figures and Descriptives

Figure 1: Bank credit and firm wage bill during the crisis

(a) Bank loan growth (b) Firm wage bill

0 0

1933) −.1 − 1933

− −.2 −.1

−.3 wage bill 1929 ∆ −.4

change in loan volume % (1931 −.2

−.5 Commerz + Deutsche Danat + Dresdner No Danat borrowers Danat borrowers

Panel (a) shows the change in total bank loans from 1931 to 1933 for Germany’s four largest banks, Commerzbank and Deutsche Bank, as well as Danatbank and Dresdner Bank (Source: Handbuch deutscher Aktiengesellschaften 1934). Panel (b) shows average change in firm wage bill from 1929 to 1933 for firms not connected (No Danat borrowers) and connected (Danat borrowers) to Danatbank. Black bars denote standard deviation. For difference in means, N =386,t=2.34. Danatbank and Dresdner Bank cut their loan volume by more and Danat- or Dresdner-connected firms reduce their wage bill by significantly more.

Figure 2: City income and NSDAP votes

(a) City income (b) NSDAP votes

5 8

4

6

3

4 density density 2

2 1

High exposure High exposure Low exposure Low exposure 0 0 −.4 −.3 −.2 −.1 0 .1 .2 .3 ∆ income 1928−1934 ∆ NSDAP votes 1930−1933

Panel (a) shows density of change in city incomes from 1928 to 1934 for cities with zero (low) vs positive (high) exposure to Danat borrowers. Panel (b) shows density of change in NSDAP vote shares from March 1930 to September 1933 elections for cities with zero (low) vs positive (high) exposure to Danat borrowers. Cities with high exposure see a stronger fall in income and higher support for the far-right Nazi party during the crisis.

33 Figure 3: Danat borrowers before the crisis: size and leverage

(a) Assets by sample (b) Pre-crisis leverage

.3 .6 wage sample (n = 386) other borrowers full sample (n = 5,610) Danat borrowers Grossbank borrowers

.2 .4 density density

.1 .2

0 0 8 10 12 14 16 18 0 2 4 6 8 log(firm assets 1929) firm leverage 1929

Panel (a) shows distribution of log assets for wage bill sample of firms (blue line), as well as for full sample of firms in 1929 (black line). Panel (b) shows firm leverage (defined as total liabilities over capital) for the full 1929 sample of firms, split into Danat, Grossbank, and other borrowers. The wage bill sample has a similar distribution in terms of total assets, and Danat-connected firms (blue line) show similar pre-crisis leverage as other Grossbank-borrowers (red line), and lower leverage than borrowers from smaller banks (black line).

Figure 4: Geographical distribution of Danat borrowers

(a) Firm sample (b) City sample

no Danat borrower no Danat exposure Danat borrower Danat exposure

Panel (a) shows geographical distribution of firms with and without Danat connection, Panel (b) distribution of cities with positive and zero Danat exposure. Blue dots indicate firms/cities with Danat exposure, grey diamonds those with no Danat connection. Firms/Cities with Danat exposure are evenly spread across Germany, reflecting the fact that Germany’s large banks served clients all across the country.

34 Figure 5: Stability of coefficient and t-value

(a) Firm wage bill (b) City income (c) City NSDAP votes

0 0 0 4 .1 −.5 −.05 −.1 3 −1 .075 −.1 −1.5 −.2 2 value value value

− − .05 − t −2 t t −.15

−2.5 −.3 .025 1 −.2 coefficient on danat exposure coefficient on danat exposure coefficient on zahlstelle danat −3 coefficient (left axis) coefficient (left axis) coefficient (left axis) t−value (right axis) t−value (right axis) t−value (right axis) −.25 −3.5 −.4 0 0 0 100 200 300 400 0 50 100 150 200 0 50 100 150 200 firms ranked by impact cities ranked by impact cities ranked by impact

Each panel excludes one observations when estimating the underlying specification and then ranks observations by the effect that observation has on the estimated coefficient. Panel (a) plots coefficient and t-value of coefficient on danat in regression wagesf = Danatf + controlsf + ✏f on the y-axis. Dependent variable is growth in firm wage bill from 1929 to 1933. Each regression drops one individual firm. The x-axis ranks firms according to their impact on the coefficient, from highest to lowest. The blue dashed line denotes coefficient estimates, the black solid line the corresponding t-value. Panels (b) and (c) do the same for city level and run baseline city regression equation yc = exposurec + controlsc + ✏c with change in income or NSDAP votes as dependent variable. Across specifications, excluding firms or cities one-for-one does not change coefficients of interest in terms of sign, size, or significance.

Figure 6: “Der Stürmer” caricature

Note: This figure shows a caricature out of pro-Nazi newspaper “Der Stürmer” in May 1931 around the beginning of the German banking crisis. The caption says “The Jew banker and the German business man”, suggesting that Jewish-led banks are to blame for the dire economic situation of the German people.

35 Table 1: Summary statistics for main variables

Panel (a): firm level

Variable Obs Mean Std.Dev. P10 P25 P50 P75 P90 wages 386 -.195 .761 -.816 -.645 -.391 -.062 .497 DanatZahlstelle 386 .153 .36 0 0 0 0 1 GrossbankZahlstelle 386 .207 .406 0 0 0 0 1 age 386 29.813 28.298 11 11 18 43 61 logassets 386 13.844 1.396 12.016 12.987 13.824 14.77 15.582 leverage 386 3.298 4.654 1.39 1.679 2.182 2.997 4.712 returnonassets 386 .041 .129 -.032 0 .031 .062 .107 wagebill/assets 386 .344 .504 .056 .108 .237 .412 .714

Panel (b): city level

Variable Obs Mean Std.Dev. P25 P50 P75 NSDAPvotes 183 .223 .054 .187 .22 .263 NSDAPmajority1930-33 183 .71 .455 0 1 1 NSDAPvotes(residual) 181 .222 .052 .19 .219 .26 income 187 -.143 .181 -.229 -.141 -.073 income (predicted) 187 -.143 .077 -.197 -.143 -.088 exposure 187 .081 .117 0 .02 .136 exposure(residual) 185 .08 .111 .008 .045 .127 exposure(grossbank) 187 .13 .146 0 .089 .219 unemploymentrate1930 187 .064 .018 .05 .065 .077 NSDAPvotes09-1930 187 .194 .072 .139 .193 .242 log population 1930 187 4.009 .887 3.339 3.706 4.427 sharebluecollar 187 .417 .093 .349 .411 .48 shareJewish 187 .009 .008 .004 .006 .012 shareprotestant 187 .663 .29 .484 .795 .895 synagoguedam.ordest. 187 .754 .432 1 1 1 anti-Semiticvotes1900 81 2.912 7.273 0 .483 2.167 Panel (a) shows summary statistics for main firm-level variables, Panel (b) for main city-level variables.

36 Table 2: Summary statistics by group - firm level

Panel (a): All borrowers as control

Danat Other difference mean sd mean sd t

age 40.90 (36.67) 27.81 (26.08) 3.31 logassets 14.78 (1.34) 13.68 (1.34) 5.80 leverage 2.26 (0.88) 3.48 (5.02) -1.86 returnonassets 0.06 (0.09) 0.04 (0.13) 1.51 wagebill/assets 0.31 (0.26) 0.35 (0.54) -0.62

Observations 59 327 386

Panel (b): Grossbank borrowers as control

Danat Grossbank difference mean sd mean sd t

age 40.90 (36.67) 45.06 (36.27) -0.63 logassets 14.78 (1.34) 14.51 (1.36) 1.10 leverage 2.26 (0.88) 2.74 (2.91) -1.20 returnonassets 0.06 (0.09) 0.02 (0.08) 2.73 wagebill/assets 0.31 (0.26) 0.32 (0.24) -0.30

Observations 59 63 122

Summary statistics for main firm-level control variables as of 1929. Danat denotes all Danat-connected firms, Other all firms not connected to Danat, and Grossbank all firms connected to any other large German bank, but not Danat. diff shows t-value for difference in means.

37 Table 3: Summary statistics by group - city level

Panel (a): All cities as control

exposed not exposed difference mean sd mean sd t

exposure 0.15 (0.12) 0.00 (0.00) 10.85 exposure(grossbank) 0.17 (0.14) 0.08 (0.14) 4.49 unemploymentrate1930 0.06 (0.02) 0.06 (0.02) 0.53 NSDAPvotes09-1930 0.19 (0.06) 0.19 (0.08) -0.15 log population 1930 4.38 (0.97) 3.56 (0.46) 7.11 sharebluecollar 0.39 (0.08) 0.45 (0.10) -4.05 shareJewish 0.01 (0.01) 0.01 (0.00) 4.48 shareprotestant 0.65 (0.29) 0.68 (0.29) -0.81 anti-Semiticvotes1900 4.06 (9.15) 1.24 (2.13) 1.73

Observations 103 84 187

Panel (b): Grossbank-exposed cities as control

exposed GB exposed difference mean sd mean sd t

exposure 0.15 (0.12) 0.00 (0.00) 6.79 exposure(grossbank) 0.17 (0.14) 0.20 (0.16) -1.08 unemploymentrate1930 0.06 (0.02) 0.06 (0.02) 0.80 NSDAPvotes09-1930 0.19 (0.06) 0.19 (0.07) 0.23 log population 1930 4.38 (0.97) 3.80 (0.55) 3.24 sharebluecollar 0.39 (0.08) 0.43 (0.10) -1.98 shareJewish 0.01 (0.01) 0.01 (0.01) 2.73 shareprotestant 0.65 (0.29) 0.71 (0.27) -1.05 anti-Semiticvotes1900 4.06 (9.15) 1.88 (2.85) 0.88

Observations 103 33 136

Summary statistics for main city-level control variables as of 1929. exposed denotes all Danat-exposed cities, not exposed all cities not exposed to Danat, and GB exposed all cities exposed to any other large German bank, but not Danat. diff shows t-value for difference in means.

38 Table 4: Summary statistics - multivariate

Panel (a): All cities as control

(1) (2) (3) (4) (5) (6) (7) GB>0 Industry FE GB>0 ind FE GB exposure>0ASsample firm firm firm firm city city city VARIABLES danat danat danat danat exposure exposure exposure

age 0.001 0.000 -0.001 -0.002 (0.001) (0.001) (0.001) (0.001) logassets 0.069*** 0.078*** 0.039 0.049 (0.014) (0.014) (0.035) (0.036) returnonassets 0.140 0.029 1.399*** 0.837 (0.138) (0.142) (0.534) (0.596) leverage -0.008** -0.007* -0.030 -0.033 (0.004) (0.004) (0.021) (0.021) wagebill/assets 0.006 0.009 -0.132 -0.056 (0.036) (0.037) (0.184) (0.205) log population 1930 0.020** -0.001 -0.002 (0.010) (0.012) (0.017) sharebluecollar -0.174* -0.089 -0.169 (0.095) (0.131) (0.150) share jewish 0.894 0.761 2.542 (1.175) (1.364) (2.290) shareprotestant -0.037 -0.090** -0.006 (0.034) (0.045) (0.063) Latitudeofthecity -0.001 0.004 -0.001 (0.006) (0.008) (0.012) Longitudeofthecity -0.002 -0.000 -0.000 (0.003) (0.003) (0.007) anti-Semitic (dummy) 0.025 (0.033)

Observations 386 386 122 122 193 142 83 R-squared 0.100 0.164 0.089 0.263 0.091 0.051 0.064 Summary statistics for main firm and city-level control variables. firm denotes firm level and uses Zahlstelle Danat as dependent variable, city denotes city level and uses city exposure to Danat as dependent variable. ind FE includes industry FE. AS sample denotes sample of cities with information on anti-Semitic votes.

39 A.2 Results

Table 5: Danat-connected firms reduce their wage bill

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) FE sample FE sample FE sample FE sample FE sample VARIABLES wages wages wages wages wages wages wages liab liab liab

danat -0.250*** -0.198*** -0.154** -0.196*** -0.296** -0.254* -0.123** -0.192** -0.209* (0.062) (0.063) (0.061) (0.073) (0.115) (0.134) (0.049) (0.079) (0.113) danat 1922 -0.185** (0.086)

Observations 386 386 384 194 194 194 386 258 102 102 R-squared 0.014 0.024 0.077 0.043 0.362 0.421 0.020 0.075 0.091 0.523 Firm Controls - XX X X X XXX X Industry FE - - X --X -- - X City FE - - - - XX-- - X Cluster City City City City City City City City City City Dependent variable is growth in firm wage bill from 1929 to 1933. danat (grossbank) denotes dummy variable with value 1 if firm is connected to Danatbank or Dresdner Bank (any other Grossbank). danat 1922 is a dummy with value 1 if firm was connected to Danat in 1922 and 1929. Firm controls as of 1929 include age, log(assets), leverage, return on assets, and capital-to-labor ratio. Standard errors are clustered on the city level. Industry fixed effects cover 20 industries. Column (5) replicates baseline regression in column (2) on the smaller sample of cities and industries with at least two firms, which is subsequently used in columns (6) and (7). Key: *** p<0.01, ** p<0.05, * p<0.1.

40 Table 6: Danat-exposed cities see a fall in incomes

(1) (2) (3) (4) (5) (6) exposure>0 exposure>0 exposure>0 VARIABLES income income income income income income

exposure -0.177* -0.295*** -0.248** -0.239* -0.313** -0.263* (0.103) (0.110) (0.104) (0.142) (0.127) (0.132)

Observations 187 187 187 103 103 102 R-squared 0.013 0.182 0.237 0.035 0.119 0.210 City Controls - XX - XX Province FE - - X --X Dependent variable is change in city income from 1928 to 1934. exposure denotes city exposure to firms connected to Danatbank or Dresdner Bank. City controls include log population, share of blue collar, Jewish, and protestants, as well as city latitude and longitude. Standard errors are robust. Province fixed effects are dummies for Germany’s 15 electoral provinces. Columns (1)-(3) use the full sample, columns (4)-(6) only cities with exposure greater than zero. Key: *** p<0.01, ** p<0.05, * p<0.1.

41 Table 7: Danat-exposed cities see an increase in radicalization

(1) (2) (3) (4) (5) (6) (7) (8) (9) placebo VARIABLES NSDAP NSDAP NSDAP maj. NSDAP maj. KPD KPD NS 5/28-9/30 synagogue dep/pop

exposure 0.077*** 0.093*** 0.414* 0.452* 0.038** 0.022 -0.038 0.584*** 0.147* (0.029) (0.029) (0.248) (0.235) (0.016) (0.014) (0.027) (0.180) (0.077)

Observations 189 189 189 189 185 185 189 189 188 R-squared 0.284 0.399 0.119 0.222 0.134 0.366 0.309 0.270 0.214 Province FE - X - X - X --- Dependent variable is change NSDAP vote share from 1930 to 1933 in columns (1)-(2); a dummy with value 1 if NSDAP won majority in 1933 but not in 1930 in columns (3)-(4); change KPD vote share from 1930 to 1933 in columns (5)-(6); change NSDAP vote share from 1928 to 1930 in column (7), which constitutes a placebo regression of NSDAP support prior to the banking crisis; a dummy with value 1 if synagogues were damaged or destroyed after 1933 in column (8); and the share of deported Jewish population out of total population in column (9). exposure denotes city exposure to firms connected to Danatbank or Dresdner Bank. City controls include log population, share of blue collar, Jewish, and protestants, as well as city latitude and longitude. Standard errors are robust. Province fixed effects are dummies for Germany’s 15 electoral provinces. Key: *** p<0.01, ** p<0.05, * p<0.1.

42 Table 8: Economic and non-economic effects of exposure on voting

(1) (2) (3) (4) (5) (6) (7) (8) VARIABLES NSDAP NSDAP NSDAP NSDAP NSDAP NSDAP NSDAP NSDAP

income -0.051*** -0.057*** -0.042** -0.048** (0.019) (0.019) (0.019) (0.019) income(predicted) -0.450*** -0.558*** -0.382** -0.495*** (0.168) (0.167) (0.170) (0.165) exposure (residual) 0.067** 0.085*** (0.030) (0.029)

Observations 183 183 183 183 183 183 181 181 R-squared 0.281 0.385 0.283 0.395 0.300 0.414 0.306 0.433 Province FE - X - X - X - X Dependent variable is change NSDAP vote share from 1930 to 1933. income is change in income from 1928 to 1934; income (predicted) is predicted change in income from 1928 to 1934 by regressing change in income on exposure to Danat; exposure (residual) denotes city exposure to firms connected to Danatbank or Dresdner Bank, but the dependent variable is the residual of a regression of NSDAP on the change in income and change in unemployment. City controls include log population, share of blue collar, Jewish, and protestants, as well as city latitude and longitude. Standard errors are robust. Province fixed effects are dummies for Germany’s 15 electoral provinces. Key: *** p<0.01, ** p<0.05, * p<0.1.

43 Table 9: Voting: Heterogeneity

(1) (2) (3) (4) (5) (6) (7) (8) low AS high AS low AS high AS low AS high AS low AS high AS VARIABLES NSDAP NSDAP NSDAP NSDAP NSDAP NSDAP NSDAP NSDAP

income -0.117* -0.070** -0.116* -0.039 (0.058) (0.033) (0.060) (0.041) income (predicted) -0.064 -0.410*** -0.011 -0.342** (0.277) (0.135) (0.271) (0.163) exposure (residual) 0.006 0.140** (0.089) (0.064)

Observations 40 41 40 41 40 41 40 40 R-squared 0.388 0.304 0.338 0.353 0.388 0.369 0.388 0.333 Dependent variable is change NSDAP vote share from 1930 to 1933. income is change in income from 1928 to 1934; income (predicted) is predicted change in income from 1928 to 1934 by regressing change in income on exposure to Danat; exposure (residual) denotes city exposure to firms connected to Danatbank or Dresdner Bank, but the dependent variable is the residual of a regression of NSDAP on the change in income and change in unemployment. City controls include log population, share of blue collar, Jewish, and protestants, as well as city latitude and longitude. Standard errors are robust. low AS and high AS denote areas with low and high levels of historical anti-Semitism, measured by the share of votes cast for anti-Semitic fringe parties in elections around 1900. Key: *** p<0.01, ** p<0.05, * p<0.1.

44 Table 10: Mixed marriages decline after the crisis

(1) (2) (3) (4) (5) (6) VARIABLES log(JN) log(JN) log(JN) log(JJ) log(JJ) log(JJ)

high exposure post default -0.092*** -0.147** -0.173*** -0.002 -0.002 -0.002 ⇥ (0.032) (0.062) (0.065) (0.007) (0.007) (0.008)

Observations 1,342 1,342 1,342 1,342 1,342 1,342 R-squared 0.608 0.631 0.633 0.076 0.102 0.105 City controls - - X --X City FE XXXXXX Year FE - XX- XX Dependent variable in columns (1)-(3) is the log number of marriages in German cities between Jews and non-Jews. The dependent variable in columns (3)-(5) is the log number of marriages between two people of Jewish faith. Standard errors are clustered on the city level. high exposure is a dummy for cities with high exposure to Danat or Dresdner bank, post default adummywithvalue 1 for all months after July 1931. City controls include yearly population and unemployment rate, as well as an interaction term of post default and Grossbank exposure. Observations are on the city-month level from January 1930 to March 1933. Key: *** p<0.01, ** p<0.05, * p<0.1.

45 A.3 Robustness

Table 11: Firm level: Danat vs Dresdner and regions

(1) (2) (3) (4) (5) (6) (7) no austria no bremen no ruhr VARIABLES wages wages wages wages wages wages wages

danat+dresdner -0.198*** -0.185*** -0.181*** -0.204*** (0.063) (0.062) (0.064) (0.068) grossbank -0.081 -0.081 -0.075 -0.080 -0.062 -0.078 -0.019 (0.099) (0.101) (0.100) (0.100) (0.101) (0.101) (0.110) danat -0.269*** -0.264*** (0.082) (0.081) dresdner -0.143** -0.136** (0.070) (0.069)

Observations 386 386 386 386 323 375 332 R-squared 0.024 0.024 0.019 0.026 0.024 0.021 0.023 Firm controls XXXXX X X Cluster City City City City City City City Dependent variable is growth in firm wage bill from 1929 to 1933. Columns (1)-(4) contrast Danatbank and Dresdner Bank. danat+dresdner denotes dummy variable with value 1 if firm is connected to Danatbank or Dresdner bank, danat if firm is connected to Danatbank, and dresdner if firm is connected to Dresdner bank. grossbank is a dummy with value 1 if firm is connected to any other of the large German banks. Wages decline stronger for danat-connected firms than dresdner connected firms, and decline insignificantly for firms connected to other grossbanken. Columns (5)-(7) exclude different regions from the sample. Column (4) excludes the Austrian border region, column (6) cities in an 80km circle around Bremen, and column (7) Germany’s industrial power house, the Ruhr area. All regressions include firm controls as of 1929: age, log(assets), leverage, return on assets, and capital-to-labor ratio. Standard errors are clustered on the city level. Key: *** p<0.01, ** p<0.05, * p<0.1.

46 Table 12: City level: danat vs dresdner

(1) (2) (3) (4) (5) (6) (7) (8) (9) residual residual residual VARIABLES income income income NSDAP NSDAP NSDAP NSDAP NSDAP NSDAP

exposure(danat) -0.026** -0.025* 0.009*** 0.009** 0.008** 0.007* (0.012) (0.013) (0.003) (0.004) (0.004) (0.004) exposure(dresdner) -0.014 -0.003 0.005 0.001 0.006 0.003 (0.011) (0.013) (0.004) (0.004) (0.004) (0.004)

Observations 187 187 187 189 189 189 181 181 181 R-squared 0.168 0.154 0.169 0.286 0.265 0.286 0.307 0.298 0.309 City Controls XXX X X X X X X Dependent variable is change in income (columns (1)-(3)) or NSDAP votes (columns (4)-(9)).exposure (danat) denotes city exposure to Danat only, exposure (dresdner) denotes city exposure to Dresdner only. All explanatory variables are standardized to mean zero and standard deviation one. Columns (6)-(9) use the residual of a regression of change in Nazi votes on change in income and change in unemployment. Incomes fall stronger and voters radicalize by more in Danat-exposed cities. City controls include log population, share of blue collar, Jewish, and protestants, as well as city latitude and longitude. Standard errors are robust. Key: *** p<0.01, ** p<0.05, * p<0.1.

47 Table 13: City level: panel

(1) (2) (3) (4) (5) (6) VARIABLES u-rate u-rate u-rate firms firms firms

exposure loans -0.196*** -0.157*** -0.144** 0.514*** 0.128** 0.129** ⇥ (0.066) (0.058) (0.071) (0.059) (0.056) (0.056)

Observations 800 800 800 785 785 785 R-squared 0.761 0.911 0.911 0.960 0.970 0.973 City Controls - - X --X City FE XXXXXX Year FE - XX- XX Dependent variable is unemployment rate (columns (1)-(3)) or firms per capita (columns (4)-(6)).exposure denotes city exposure to Danat or Dresdner, loans denotes yearly log loan volume of Danat and Dresdner. GB denotes the same variables, but for other Grossbanken. city exposure to Dresdner only. City controls include log population and Grossbank exposure interacted with log loan volume by other Grossbanken. Standard errors are clustered on the city level. Observations are on the city-year level from 1930 to 1934. A decline in loan volume by Danat and Dresdner increases unemployment and decreases the number of firms in cities with higher exposure. Key: *** p<0.01, ** p<0.05, * p<0.1.

48 Table 14: City level: other elections, exports, and trade-intensive cities

(1) (2) (3) (4) (5) (6) (7) low exports high exports VARIABLES NS 30-7/32 NS 30-11/32 income NSDAP NSDAP income income

exposure 0.085*** 0.071** -0.564*** -0.246 (0.028) (0.033) (0.208) (0.211) log(exports) -0.011* 0.005 (0.007) (0.003) income (predicted) -0.643 (0.723)

Observations 181 181 187 189 183 63 62 R-squared 0.507 0.329 0.374 0.273 0.284 0.259 0.227 Dependent variable is change in NSDAP votes from March 1930 to July or November 1932 in columns (1) and (2). income denotes the change in income from 1928 to 1934, and NSDAP the change in NSDAP votes from March 1930 to September 1933. exposure denotes city exposure to Danat or Dresdner; log exports log city exports, and income (predicted) predicted income due to log exports. Columns (6) and (7) split cities into those with low and high exposure to tradable industries (bottom and top tercile). City controls include log population, share of blue collar, Jewish, and protestants, as well as city latitude and longitude. Standard errors are robust. Key: *** p<0.01, ** p<0.05, * p<0.1.

49

Appendix B: Data

Table A1: Variables and data sources

Variable Source Type

Firm-level variables

Assets Handbook of German Stock Companies in Reichsmark Wage bill Handbook of German Stock Companies in Reichsmark Bank-affiliation Handbook of German Stock Companies dummy Firm age, leverage, ROA Handbook of German Stock Companies in years / %

City-level variables

Exposure Handbook of German Stock Companies [0,1] Exports Statistisches Jahrbuch des Deutschen Reichs in Reichsmark Income Statistik des Deutschen Reichs, Neue Folge 1884-1944 In thousand Unemployment Statistisches Jahrbuch Deutscher Städte in thousand Population Statistisches Jahrbuch Deutscher Städte in thousand Labor force 1933 census in thousand Share of protestants 1925 census % Share of Jews 1925 census % Share of blue collar workers 1925 census % Election outcomes Statistik des Deutschen Reiches (ICPSR 42) % Debtor/creditor Enquetekommission zur Untersuchung der Bankenkrise dummy Anti-Semitic votes Statistische Jahrbücher des dt. Reichsamts für Statistik % Pogrom in 1349 Germania Judaica dummy Synagogues damaged or destroyed Alicke (2008) dummy Total deportations German federal archive (Bundesarchiv) in thousand

Note: This table lists main variables, data sources, and units. For details, see text.

50