The Long Shadow of Jewish Persecution on Financial Decisions

Francesco D’Acunto Marcel Prokopczuk Michael Weber ∗

Abstract For centuries, Jews have been associated with financial services. We find that present-day households in German counties where Jewish persecution was higher in the Middle Ages and the Nazi period invest less in stocks, have lower savings in bank deposits, and are less likely to get a mortgage, but not to own a house. Current antisemitism, generalized trust, or supply-side forces do not appear to explain these correlations, which are consistent with a norm of distrust in finance, transmitted across generations. The forced migrations of the Ashkenazi communities across the German lands after the 11th century help assess the extent to which the effect of Jewish persecution on present-day financial decisions is causal.

JEL: D91, G11, J15, N90, Z10, Z12.

Keywords: Cultural Economics, Intergenerational Transmission of Norms, Religious Identity, Household Finance, History & Finance.

∗D’Acunto is at Haas School of Business, UC Berkeley (francesco [email protected]). Prokopczuk is at Leibniz University Hannover ([email protected]). Weber is at the Booth School of Business, University of Chicago ([email protected]). The views expressed in this paper do not reflect those of the DIW, the Bundesbank, the Banque de France, or the Friedrich Ebert Stiftung. We thank Joachim Voth and Nico Voigtländerfor making their data on anti-Jewish violence publicly available, Tarek Hassan and Konrad Burchardi for sharing their data on the postwar migrations of Eastern European Germans, Thomas Kick for his help with the Bundesbank data on county-level bank branches, Tobias Lauter for his assistance with the Hoppenstedt data, the ifo-Prussian Economic History project, Ralph Melzer, Elmar Braehler, and Oliver Decker for sharing the micro-data underlying the FES Mitte-Studies, and Martin Eisele for his great assistance in the access of the PHF Bundesbank data. For very helpful comments and discussions, we thank Ran Abramitzky, Nick Barberis, Zahi Ben-David, Kelley Bergsma, Michael Brennan, Johannes Buggle, Filipe Campante, Davide Cantoni, Jason Chen, James Choi, Pierluigi D’Acunto, Stefano DellaVigna, Barry Eichengreen, Joey Engelberg, Ruben Enikolopov, Paola Giuliano, Will Goetzmann, Rick Green, Luigi Guiso, Tarek Hassan, Danling Jiang, Samuli Knupfer, Ross Levine, Sonya Lim, Dmitry Livdan, Ulrike Malmendier, Gustavo Manso, Adair Morse, Petra Moser, Rabbi Adam Naftalin-Kelman, Terry Odean, Martha Olney, Christine Parlour, Chris Parsons, Hagit Perry, Caitlin Rosenthal, Paola Sapienza, Markus Schmid, Andrei Shleifer, Stephan Siegel, Andrei Simonov, Paul Smeets, Robert Vishny, Nico Voigtländer,Jason Wittenberg, Noam Yuchtman, and especially Joachim Voth, and participants at several seminars and conferences. All errors are our own. 1 Introduction

For centuries, Jews have been associated with financial services, largely because of bans on lending money at interest for Christians and Muslims.1 The ban for Christians was lifted in the 15th century. Yet, in 1882, 3% of German workers were Jewish, but 23% of the overall financial sector workforce, and more than 85% of the brokers in the Berlin stock exchange, were Jewish.2 The diffusion of government-owned banks and local credit unions largely reduced the Jewish representation in German banking (Mosse, 1987). But Jews were still accused of dominating and manipulating the stock market, especially after major market crashes as in 1873 and 1905 (Marr(1879) and Koehler(2005)). The alleged Jewish manipulation of the German economy through the control of the stock market was an important tenet of the antisemitic propaganda until the Second World War.3 At the same time, Germans had persecuted Jews since the Middle Ages, and localized anti-Jewish sentiment had persisted for centuries (Voigtlaender and Voth, 2012).

We document that present-day households in German counties where Jewish persecution was higher in the Middle Ages and the Nazi period invest less in stocks than other households. Figure 1 plots the negative correlation between stock market participation and historical Jewish persecution at the county level, conditional on a large set of historical and present-day observables. Households in counties where Jewish persecution was one standard deviation higher are 11% less likely to hold stocks. The size of this association is about one fourth of the eﬀect of holding a college degree on investing in stocks, and college education is one of the most relevant determinants of stock market participation.

Our analysis focuses mainly on stock market participation, because the defamation of Jews as stock-market manipulators was salient in the press and popular culture even

1For instance, Pope Leo IX banned Christians from lending money at interest as far back as 1049, and Gratianus formalized the ban in the Corpus Iuris Canonici in 1150: “Usura est, ubi amplius requiritur quam datur: [...] Ecce evidenter ostenditur, quod quicquid ultra sortem exigitur usura est.” (“Usury is when one requires more than what he gives: [...] It is therefore obvious that whatever is required in addition shall be usury.”) The human capital Jews had accumulated since the second century may have facilitated their sorting into trade and ﬁnance well before 1049 (Botticini and Eckstein(2005) and Botticini and Eckstein(2012)). 2Table 10 of the Online Appendix reports the share of Jewish workers across sectors in 1882 Germany. For the share of Jewish brokers, see Glagau(1876) and Fritsch(1892). Gross(1975) argues Jewish brokers founded the Berlin stock exchange, and the few non-Jewish brokers would not go on Saturdays, although the market was open, because of the lack of traders. 3The notion that Jews would manipulate the economy and culture through their stock ownership was already common in Wilhelmine Germany (e.g., Treitschke(1879), Sombart(1912), Poliakov(1977)). In the delirious words of Hitler, ”In times of distress a wave of public anger has usually arisen against the Jew; [...]the Jew gained an increasing inﬂuence in all economic undertakings by means of his predominance in the stock exchange” (Hitler, 1939).

1 Figure 1: Stock Market Participation and Historical Jewish Persecution .2 0 -.2 Ratio Households who Own Stocks (residuals)

-2 -1 0 1 2 Principal Component Measures of Persecution (residuals) Each point is a German county. The vertical axis plots the residuals from estimating the following equation,

0 RatioStockholdersk = α + Kk × δ + k, where Kk is the set of historical and present-day observables described in section 3. The horizontal axis plots the residuals of a regression of the principal component of the measures of Jewish persecution described in section 2 and the same set of covariates Kk. after the Jewish presence in banking had faded (Koehler, 2005). But consistent with the historical stereotype of Jews as moneylenders (Reuveni and Wobick-Segev, 2010), we also ﬁnd that present-day households in German counties where Jewish persecution was higher use banking services less than other households, even if they do not face a lower supply of banking services. Households in high-persecution counties are 10% less likely to have a mortgage, but as likely as other households to own a house. The ratio of retail deposits over total assets of the banks in high-persecution counties is 2% lower than in other counties. Instead, households in high-persecution counties are as likely to save as other households, and the concentration of bank branches they can access is the same as in other counties. We ﬁnd suggestive evidence that they are more likely to hold cash.

After documenting the associations between historical Jewish persecution and the ﬁnancial decisions of present-day German households, we assess their robustness, and then we investigate the channels that might have transmitted these associations over time.

Our results obtain similarly in two different cross sections of households: (i) the SOEP data set, which includes 13,569 German households interviewed in nine waves over the period 1984-2010, and (ii) the PHF data set of Deutsche Bundesbank, which only includes one wave of 3,500 households interviewed in 2010, but collects unique information on household heads’ risk aversion, generalized trust, financial literacy, and religiosity. Reverse causality is not a major concern in our setting, because of the timing of the outcomes we consider. But testing for a causal effect of Jewish persecution on present-day

2 households’ ﬁnancial decisions would require a quasi-random assignment of persecution across similar German counties in the distant past. We cannot construct such an ideal experiment, which is why we assess the concern that unobservables correlated with Jewish persecution and current ﬁnancial decisions drive the results in two ways.

First, we test for the robustness of our results when using alternative sources of spatial variation in Jewish persecution. All the results go through if we only look at counties that hosted documented Jewish communities in the Middle Ages, so that the presence of Jews, irrespective of violence against them, cannot fully explain the results.4 The results also hold when we only exploit the variation in Jewish persecution across geographically close counties, which delimit likely homogenous areas along several historical and present-day dimensions, as well as geography, the quality of institutions, and cultural heritage. We limit the variation in Jewish persecution to counties in the same states (Bundesl¨ander), in the same occupation zones after the Second World War, and in the same virtual states. Virtual states are arbitrary partitions of Germany that account for the endogeneity of state borders. Our results are also robust when accounting for the mass migrations from Eastern Europe to Western Germany after the Second World War.

Second, we exploit the forced migrations of the ﬁrst Jewish communities of Germany, the Ashkenazi communities, across the German lands in the Middle Ages because of the Crusades. Forced migrations give us some quasi-exogenous variation in the settlement of Jewish communities across Germany, and hence in the likelihood of medieval persecution against Jews at the county level. The forced migrations get us closer to the ideal experiment of randomly assigning Jewish persecution across similar counties, but they clearly raise a set of concerns. In section 4, we discuss the concerns and the advantages this source of variation implies. In the Online Appendix, we construct an alternative test using a source of variation of historical anti-Jewish sentiment based on the support for the Nazi party, and we also ﬁnd results in line with the baseline analysis.

Because of the long-run nature of the association between Jewish persecution and ﬁnancial decisions, we are especially interested in studying the supply- and demand-side channels that might have transmitted the eﬀect over the centuries.

We ﬁnd that supply-side forces are unlikely to explain our results, because counties with higher historical Jewish persecution did not have a lower supply of banking services in

4We thank Nico Voigtl¨anderand Ruben Enikolopov for suggesting this test.

3 the past, and do not have a lower supply of banking services today.5 Counties with higher persecution did not have a lower ratio of Jewish employees in the financial sector as of 1882 either. Persecution did not reduce the supply of finance or the quality of the human capital running financial services at the county level in the long run.

On the demand side, after controlling for the effects of historical and present-day household- and county-level characteristics such as income, education, age, risk aversion, financial literacy, and generalized trust, a cultural norm of distrust in finance transmitted through generations may explain our findings. In those counties where anti-Jewish sentiment was higher, such sentiment might have involved financial services, which, contrary to other sectors, Jews had run almost exclusively for centuries. Guiso et al. (2009a) discuss the link between distrusting a population and distrusting the attitudes and behaviors with which the population is associated. They exploit such a link in the context of international trade.

A norm of distrust in finance might explain our results if it has survived independently of anti-Jewish sentiment, and of the “backwardness” of households, which instead cannot explain our results for at least three reasons. First, current German banking is unrelated to Jews, whose presence in Germany was tragically reduced by the end of the Second World War. Second, the effect of historical Jewish persecution on stockholdings is similar for households with different education levels, whereas current anti-Jewish sentiment and backwardness are more prevalent among the least educated, a result of the sociology literature that we replicate with novel data from the Friedrich Ebert Stiftung. Third, the effect of historical Jewish persecution on financial decisions is similar across cohorts and over time, whereas Voigtlaender and Voth(2014) show the present-day antisemitic beliefs of those raised during the Nazi period are substantially higher than the antisemitic beliefs of younger cohorts.

The persistence of a cultural norm of distrust in ﬁnance across education levels and cohorts is plausible, because the atrocities of the Nazi period against the Jewish population are a compulsory topic in the curricula of primary and secondary schools in Germany,6 which allows for antisemitic beliefs to dissipate with schooling, consistent

5Note institutions supported by the Church to compete with the Jewish monopoly in banking did not diffuse in Germany. These institutions drive the supply-side effects of Jewish persecution on long-run financial development, which Pascali(forthcoming) studies for the case of Italy. 6The document Behandlung des Nationalsozialismus im Unterricht of April 1978 bound all German states to introduce the Holocaust and 19th-century antisemitism as compulsory topics in their primary and secondary education curricula. For a discussion, see der StändingenKonferenz der Kultusminister der Länderin der Bundesrepublik Deutschland(1997).

4 with our findings and previous findings. Instead, German curricula do not cover topics related to financial literacy, and do not discuss the benefits of investing in the stock market, so that schooling would hardly change a norm of distrust in finance transmitted across generations.

To test directly for the scope of a persistent cultural norm of distrust in ﬁnance, we ran our own survey on a representative sample of 1,000 present-day Germans to elicit their distrust toward the stock market and banks.7 The survey also elicited measures of risk tolerance and generalized trust at the individual level. We do ﬁnd that distrust in the stock market and local banks is indeed higher for respondents in counties with higher historical Jewish persecution, even after controlling for their risk tolerance and generalized trust.

We cannot test directly whether distrust in finance caused past anti-Jewish sentiment, or whether past anti-Jewish sentiment caused distrust in finance over time, because we cannot observe the attitudes toward financial services of German households in the Middle Ages. Several historians argue the motivations for Jewish persecutions in Europe were merely religious at first, similar to the persecution against other religious minorities, such as the Cathars, and the hatred against the Jews as economic exploiters only appeared at a later stage (e.g., Poliakov(1975) and Perry and Schweitzer(2002)). If the cross-sectional variation of distrust in finance existed prior to anti-Jewish sentiment, it should reflect the extent of religiosity across counties in the distant past.8 Although we cannot measure the extent of religiosity of the broader population in the Middle Ages, we always control for the historical and current mix of religions across German counties in our analyses. For a subset of household heads, we also observe their religiosity, independent of the religious denomination, and controlling for individual-level religiosity does not change our results either. That distrust in finance caused Jewish persecution seems implausible also because Muslims did not commit pogroms against Jewish communities during the Black Death period, which is the time when we measure medieval persecution, even if Islam condemned moneylending as Christianity did. Because stock markets did not exist in the Middle Ages, salient stock market crashes cannot have determined the historical variation of Jewish persecution either.

We conclude that supply-side channels hardly drive the centuries-long association between Jewish persecution and the use of ﬁnancial services by present-day German

7We thank Stefano Della Vigna and Noam Yuchtman for inspiring this test. 8We thank Paola Sapienza for emphasizing this point.

5 households. The effect is consistent with a cultural norm of distrust in finance, which has lingered until the present day, and which still affects the financial decisions of Germans even if the underlying source of such a norm, the Jewish population, was forced out of German financial services long ago.

This paper contributes to the literature on the economic effects of culture, initiated by Banfield(1958) and Putnam(1993). In particular, it belongs to the research on the effects of trust on economic outcomes. Guiso et al.(2004) show that social capital positively affects stock market participation, illiquid investments, and the use of checks. In Algan and Cahuc(2006), the inherited component of trust by immigrants affects economic growth in the United States. We contribute by studying the deep-rooted origins of a persistent cultural norm that has lingered and still affects economic decisions, even if the Jewish population has not run German financial services for decades.

The paper also relates to the body of research that uses historical natural experiments to understand present-day outcomes, surveyed by Spolaore and Wacziarg(2013) and Nunn(2014). For the case of financial outcomes, D’Acunto(2015a) labels this nascent approach “History & Finance.” Recent contributions include Pascali(forthcoming), who shows Italian provinces where Jewish-managed banking developed in the past, and where Christian competing institutions flourished, have a more developed banking system today; Pierce and Snyder(2014), who find present-day firms in African countries with higher historical extraction of slaves face lower access to formal and informal credit, and Brown, Cookson, and Heimer(2015), who exploit the differences in the adjudication of civil-contract disputes in Native American reservations to state courts or tribal courts determined by the Public Law 280 of 1953. D’Acunto(2015b) shows the variation in basic education across European regions has persisted for centuries, and it predicts the innovation and investment of traditional manufacturing firms.

We build on a recent literature on the persistence and the effects of Jewish persecution, and especially on Voigtlaender and Voth(2012). We use their historical data and compute their measures at the county level. Our work aims to test whether the persistence of localized anti-Jewish sentiment documented by Voigtlaender and Voth(2012) and Voigtlaender and Voth(2013a) may affect the economic behavior of households in the long run, and if yes which channels have transmitted such a long-run effect on economic decision-making. We also relate to Grosfeld, Rodnyansky, and Zhuravskaya(2013), who find an effect of the Pale of Settlement, an area within Poland and Russia where Jews were confined in the past, on the support for socialist/communist parties after the fall

6 of the Soviet Union, on generalized trust, and on the ratio of self-employed workers in the local population, but not on households’ income or economic decision-making. They interpret the results as evidence of anti-market beliefs generated by ethnic hatred against Jews, which increased the bonding trust within the non-Jewish population. In our paper, we find no effect of Jewish persecution on generalized trust, political preferences, or the likelihood of self-employment and entrepreneurship, and hence generic anti-market beliefs cannot explain our results.9 Instead, we document that Jewish persecution is related to the use of financial services by present-day German households, and to a their actual economic and financial decisions. We find these effects are consistent with a cultural norm of distrust in the financial sector, which has lingered over the centuries. Recent papers document supply-side effects of Jewish persecution (e.g., Acemoglu et al.(2011), Waldinger(2010), Pascali(forthcoming), Akbulut-Yuksel and Yuksel(Forthcoming)). We find supply-side channels do not drive our findings.

This paper also speaks to the research on the stock market participation puzzle. Explanations include background risk (Paxson(1990), Guiso et al.(1996)), social interactions (Hong et al.(2004)), awareness (Guiso and Jappelli(2005)), generalized trust (Guiso et al.(2009b)), insurance motives (Gormley et al.(2010)), ﬁnancial literacy (van Rooij et al.(2011)), macroeconomic experiences (Malmendier and Nagel(2011)), labor income risk (Betermier et al.(2012)), and corporate scandals (Giannetti and Wang (forthcoming)).

2 Measuring Jewish Persecution over Time

Building on Voigtlaender and Voth(2012), we propose four main measures of Jewish persecution at the level of German counties. Intensive and extensive margins of Nazi persecution. We capture an extensive margin and an intensive margin of Jewish persecution during the Nazi period. The extensive margin, namely, the overall extent of Jewish persecution, is the number of Jews deported from each German county to a concentration or extermination camp. This measure increases with the size of Jewish communities, and hence is higher for counties with large cities. The intensive margin, namely, the relative extent of Jewish persecution, is the ratio of deported Jews during the Nazi regime over the total Jewish population in a county as of 1939, which is before the Nazi regime implemented their criminal “Final Solution.”10 German Jews started to hide and live in captivity after the Kristallnacht

9Table 15 of the Online Appendix includes these non-results. 10Results are virtually unchanged if we use the county-level Jewish population as of 1933.

7 pogrom of 1938. On top of the deportations planned at the central level, to locate hiding Jews, the Nazis relied on delation by locals. Households in areas where anti-Jewish sentiment was higher were presumably more likely to delate. Black Death pogroms. The third measure, Pogrom 1349, is a dummy that equals 1 if a county experienced at least one anti-Jewish pogrom during the years of the Black Death, that is, around 1349. The Black Death was arguably the worst pandemic in human history, and up to one third of the Europeans may have died. Unsubstantiated theories on the origins of the pandemic diffused all over Europe. Accusations against the Jews were common and led to persecution, especially in the German lands. Voigtlaender and Voth (2012) find pogroms during the Black Death predict the extent of Jewish persecution during the Nazi period at the town level. In our county-level analysis dictated by the level of resolution of our financial data, we do not use the number of pogroms in a county, but the occurrence of any pogrom in the county. The number of pogroms is meaningful in a city-level analysis, because cities that experienced more pogroms developed stronger anti-Jewish sentiment in the past. But the number of pogroms in a county would be driven by the number of cities in the counties for which data on pogroms are available, as opposed to the concentration of actual pogroms in cities. Principal component of measures. We also compute the first principal component of the three measures described above. The principal component isolates the common variation across the other three alternative margins of persecution. For this reason, the principal component is our preferred measure of Jewish persecution, but we report results using all measures throughout the paper. Other measures of Jewish persecution. In Table 11 of the Online Appendix, we also report the results when using additional measures of historical anti-Jewish sentiment used by Voigtlaender and Voth(2012). Data sources. We collect data from 10 sources. The characteristics of German households are from the Socio-Economic Panel (SOEP) run by the Deutches Institut für Wirtschaftsforschung Berlin (DIW). Research in Economics has already used the SOEP data.11 The SOEP has conducted interviews on a yearly basis since 1984. For each wave, the SOEP includes households that have been interviewed in previous waves as well as new households. Because we are interested in the cross-sectional association of anti-Jewish sentiment and stockholdings, we only include non-repeated observations when running the main analysis. A drawback of the SOEP data is that they do not provide the complete financial portfolios of households; hence, we cannot document how Jewish persecution

11For instance, see Fuchs-Schuendeln and Schuendeln(2005), Dohmen et al.(2012), and Burchardi and Hassan(2013).

8 affects every component of households’ financial portfolios, the fraction of wealth invested in stocks, or whether lower stockholdings translate into higher deposits and other safe assets. Moreover, the SOEP data set does not include measures of the household head’s risk aversion, financial literacy, or generalized trust. To address these shortcomings, we replicate our results using the balance sheets of the German households in the 2011 wave of the Panel of Household Finances (PHF) run by the Deutsche Bundesbank, which is the only wave with data available to researchers so far. The PHF sample we can match with the historical data consists of 1,256 households across 99 counties, and hence is too small to make it our main working sample. But the PHF data allow averaging out a direct estimate of households’ wealth, as well as the elicited risk aversion, financial literacy, and generalized trust of household heads, in our cross-sectional analysis. We use the historical data in Voigtlaender and Voth(2012) on Jewish persecution in the Middle Ages and during the Nazi period. We aggregate all city-level data at the level of present-day counties, which is the finest geographical level at which the SOEP data allow observation of households’ location. We obtain county-level historical characteristics from the Ifo Prussian Economic History Database, described in detail by Becker et al.(2014). We collect current county-level controls: socio-demographics from DeStatis, the index of land quality of Ramankutty et al.(2002), and the coordinates of the centroid from Eurostat. In the analysis of the channels that transmitted the long-run association between Jewish persecution and financial decisions, we use three novel sources of data: (i) the micro- data underlying the Friedrich Ebert Stiftung’s yearly survey on antisemitism and racial discrimination for a representative cross-section of 4,000 Germans. This survey allows observation of the incidence of anti-Jewish sentiment and ”backwardness” of present-day households across German counties; (ii) the data on the foundation date of German’s Volksbanken and Raiffeisenbanken from the Hoppenstedt database, which allows us to construct the spatial pattern of diffusion of credit unions across the German lands in the second half of the 19th century; and (iii) our own survey aiming at eliciting the distrust towards the stock market and financial services of present-day German households. We ran the survey through the company Clickworker, because we are not aware of any data on a representative set of German households regarding their trust towards financial services. Statistics and baseline correlations. Table 1 reports summary statistics for the variables in the main analysis at the household level.12 In the left panel, we use all available observations. In the right panel, we only look at households in counties with

12For county-level variables, we attribute the same values to all households living in a given county.

9 non-missing historical information. The far-right column reports p-values for testing the null hypothesis that the means across the samples are equal. The full sample of non-repeated households in the SOEP county-level data set includes 29,680 observations. The county of residence is not available for 2,655 households. Moreover, we are missing the county-level historical information for 9,207 households.13 The remaining missing observations are due to denial to answer demographic questions, such as the income or age of the household head. Measures of Jewish persecution do not differ between the whole sample and the subsample without missing information. On average, 57% of households live in counties where at least one pogrom happened during the Black Death. The average ratio of Jews deported during the Nazi period is 34%. This ratio may be higher than 100%, because Jews may have migrated across counties between 1939 and 1945.14 To compute this measure, we exclude counties where the Jewish population in 1939 was reported to be zero. The ratio of stock-owning households across counties and over time is about 16% in both samples. The average age of the person who makes financial decisions is 49 years. The two samples do not differ in terms of the geographic characteristics of the counties. They are similar with respect to the average population size and the average number of Jews in 1939. However, households in the running sample are more likely to be homeowners (42% vs. 39%), and they have lower self-reported income (30,500 euros vs. 31,500 euros). The SOEP survey does not ask households for an estimate of their overall wealth. We use income and homeownership to proxy for wealth. Households in the running sample are less likely to hold only a high school degree.15 Households in the running sample live in counties with lower unemployment and fewer blue-collar workers in 1933, and a higher proportion of Catholics in 1925, which controls for the historical religious composition in the local population. Figure 2 depicts the properties of Jewish persecution and present-day stock market participation at the county level. Panels (a) and (b) of Figure 2 show the spatial distribution of the ratio of Jews deported during the Nazi Period and of the average ratio of households who own stocks from 1984 to 2010. In both maps, the darker the county is, the higher the value of the variable. The data are not available for blank counties. Jewish persecution was higher in Western counties. Stock market participation

13All results are similar if we attribute values of historical variables from neighboring counties to those for which we have missing information. 14The ratio exceeds 1 only in one county (Viersen, in Nordrhein-Westfalen). Excluding Viersen from the analysis keeps all results unchanged. 15The average ratio of household heads who claim to hold a college degree across counties from 1984 to 2010 is extremely low. The results are virtually unchanged if we use a dummy for holding a high school degree or higher levels of education.

10 is higher in the south and in the north. As expected, participation is systematically lower in Eastern Germany. Panels (c) and (d) of Figure 2 plot the densities of the ratio of Jews deported and present-day stock market participation.16 Panel (e) plots the correlation between the ratio of deported Jews and the average ratio of households who own stocks from 1984 to 2010, which is negative (-0.13, p-value=0.03). Panel (f) of Figure 2 shows the average participation across counties with and without pogroms during the Black Death. Participation is higher in counties with no pogroms, but a t-test for the diﬀerence between the two means does not reject the null that the means are equal. The Online Appendix describes additional characteristics of the raw data. In Figure 7, we show persecution was higher in states closer to the Rhine Valley than those in Southern, Northern, and Eastern Germany. In Figure 8, we plot the correlations between stockholdings and the other measures of Jewish persecution in the raw data, all of which are negative.

3 Jewish Persecution and Stock Market Participation

In the baseline analysis, we estimate the association between historical Jewish persecution and stock market participation by German households in the period from 1984 until 2010. The following is our most general speciﬁcation:

0 0 P r(HoldsStocksikt|Xikt,Kik) = Φ(α+β ×P ersecutionk +Xikt ×γ +Kik ×δ+Income deciles+ηt +ikt), (1) where HoldsStocksikt is a dummy that equals 1 if household i in county k and surveyed in wave t holds any stocks and P ersecutionk is one of our four measures of Jewish

Persecution. Xikt includes the following individual-level controls: gender, single/marital status, age (2nd degree polynomial), education level, homeownership, and investment in life insurance products. Kk includes the following county-level current and historical controls: latitude, income per capita, share of college-educated population, index of quality of cultivable land, log of population in 1933, log of Jewish population in 1933, share of population employed in the retail sector in 1933, share of population employed in manufacturing in 1933, share of Catholic population in 1925. Income deciles are dummies indicating the decile of the income distribution to which the household belongs, and Φ is 17 the standard normal cdf. ηt are a set of survey-wave group ﬁxed eﬀects, each capturing a group of four adjacent years.18 We allow for correlation of unknown form across residuals

16We have excluded Viersen, where the ratio of deported Jews exceeds 100%. 17All the results are virtually identical if we include second- or third-degree income polynomials instead of deciles. Because we do not observe the overall wealth of households, income and homeownership serve as proxies for the overall wealth. 18Results do not change if we make the survey-wave fixed effects coarser or finer. Because we exclude repeated observations across survey waves, no time variation exists at the household level.

11 at the county level, because attributing county-level measures to each household induces a mechanical correlation of residuals across households in a same county. Table 2 reports the average marginal eﬀects for our baseline speciﬁcation. All the variables are standardized, with the exception of dummy variables. In columns (1)-(2),

P ersecutionk is the first principal component of the three measures of Jewish persecution. In column (1), we include only historical socio-demographic controls at the county level, and hence variables that are measured contemporaneously to the Jewish persecution. In column (2), we add present-day controls, a dummy that equals 1 for households in Eastern Germany, and survey-wave group fixed effects. A one-standard-deviation increase in persecution (1.016) is associated with a 1.84-percentage-point lower stock market participation. The effect is statistically significant and economically large. It amounts to 11% of the average participation in our sample. The effect is about one fourth the size of the effect of holding a college degree on the likelihood of investing in stocks. Higher education is one of the strongest determinants of stock market participation in the literature. In columns (3)-(4) of Table 2, we look separately at the extensive and intensive margins of persecution. A one-standard-deviation increase in the logarithm of the Jews deported (2.31) is associated with a 1.41-percentage-point lower likelihood that a household invests in stocks (column (3)). In column (4), a one-standard-deviation higher ratio of deported Jews over the total Jewish population in 1939 (0.192) is associated with a 1.22-percentage- point drop in the likelihood that a household invests in stocks. In column (5) of Table 2, we measure the Jewish persecution as the occurrence of one or more pogroms during the Black Death of 1349. Households in counties with at least one pogrom are about two percentage points less likely to hold stocks. The size and statistical significance of the effect of individual- and county-level observables on current stockholdings are comparable across specifications. All the (untabulated) results are similar if we estimate the marginal effects using a linear probability model. In columns (6)-(9) of Table 2, we repeat the estimation of equation1 only within the set of counties that hosted a documented Jewish community before the Black Death of 1349. The intuition is that in those counties where Jews did not exist in the past, the process of persistence of anti-Jewish sentiment might be different than in other counties. Moreover, unobservables that favored the settlement of Jewish communities in the past could drive both historical persecution and present-day stockholdings. The restriction reduces the sample size by 26%, but unobservables that explain the settlement of Jewish communities in the past cannot be driving the variation in persecution across this subsample of counties.

12 The specification in columns (6)-(9) provide results that are qualitatively similar to the baseline analysis, but the size of the estimated effects are smaller than the corresponding sizes for the same measures of persecution in the baseline analysis, and we fail to detect statistical significance for the effect of pogroms in 1349 on current stockholdings (column (9)). We exploit the fact that the margin of variation of the existence of historical Jewish communities is relevant to our effect when exploiting a historical natural experiment in the next section. Other Measures and Outliers. In Table 11 of the Online Appendix, we replicate the analysis using other measures of historical anti-Jewish sentiment computed by Voigtlaender and Voth(2012). We use the logarithm of the number of anti-Jewish letters sent to the Nazi tabloid Die Stürmer from each county, a dummy equal to 1 if a local synagogue or prayer room was destroyed or heavily damaged, and the first principal component of these measures plus the votes for the Nazi Party in 1928 and the occurrence of a pogrom in the county in the 1920s. We also augment our principal component with these additional variables. In all cases, we find results in line with the baseline analysis. The effect for the attacks against synagogues is statistically insignificant. In Table 12 of the Online Appendix, we also replicate our main analysis excluding four outlier counties in which the ratio of households holding stocks exceeds 40%. All the results are similar to the baseline analysis.

4 Alternative Sources of Variation of Anti-Jewish Sentiment

German counties are likely to differ along several dimensions, such as geography, history, and the quality of current and historical institutions. Geographically close counties. In columns (1)-(5) of Table 3, we augment Equation 1 with geographic fixed effects, which only exploit the variation in Jewish persecution across counties close in space. Panel A uses all counties, whereas in Panel B, the sample is limited to the counties where a Jewish community was documented as of 1349. In column (1), we look at counties in the same state (Bundesland). Counties in the same state are exposed to the same current institutions, but are not necessarily exposed to the same historical institutions, because the borders of states do not always coincide with the borders of historical administrative regions. Moreover, some states consist of a few counties, and hence have minor variation in Jewish persecution. To address these issues, in columns (2)-(3) of Table 3, we divide Germany into 9 and 16 arbitrary squares of similar size by longitude and latitude, which we label ”virtual states.” We only exploit the variation

13 in Jewish persecution across counties that fall in the same squares.19 This method also overcomes the endogeneity of state borders. In column (4) of Table 3, we only exploit the variation in Jewish persecution within occupation zones after the Second World War. These zones do not perfectly overlap with the present-day state boundaries, and occupants implemented different denazification policies across zones. All specifications replicate the baseline results. In column (5), we estimate the baseline specification at the county level. This test addresses the concern that individual observations in a county may be spatially correlated in a way not properly accounted for by the clustering of the standard errors at the county level. Across all specifications, the results are similar to the baseline analysis. Refugees from Eastern Europe after WWII. Millions of Germans from Eastern Europe moved forcibly or voluntarily to Germany after the Second World War, and were not granted a right to go back (e.g., see Burchardi and Hassan(2013)). The refugees were distributed across the occupation zones of Western Germany.20 We test if the effects we document survive when controlling for the number of refugees per inhabitant across counties. In column (6) of Table 3, we add the share of the population coming from Eastern Europe in German counties in 1961 as a covariate to our baseline probit specification.21 The estimated effects of persecution on stockholdings are in line with the baseline analysis for the full set of counties, as well as the counties where a Jewish community was documented as of 1349. Subsamples. In columns (7)-(10) of Table 3, we conduct a set of subsample analyses. In columns (7)-(8), we exclude households in counties in the bottom third of the distribution of present-day income per capita and of college-educated inhabitants. In columns (9)- (10), we exclude households in counties that were wealthy in the distant past, that is, counties hosting any Hanseatic League cities, or any free imperial cities in the Middle Ages. The baseline results are replicated, although we detect no statistical significance for the coefficients in columns (7)-(8) when limiting the sample to the counties with a documented Jewish community as of 1349. PHF sample: Financial literacy, risk aversion, wealth. The SOEP sample does not allow us to keep constant dimensions such as household heads’ financial literacy, risk aversion, or even the estimated total wealth of households. We therefore run our cross- sectional analysis on the German households in the 2011 wave of the Panel of Household

19Figure 9 of the Online Appendix plots the composition of the virtual states. 20France did not accept any refugees in their occupation zone, so the within-occupation-zone analysis of column (4) of Table 3 indirectly informs us on the robustness of the baseline eﬀect to controlling for refugees. 21We are indebted to Konrad Burchardi and Tarek Hassan for sharing their hand-collected data on the amount of refugees in Western Germany at the county level from the 1961 census.

14 Finances (PHF), which is the only wave available to researchers to date. The size of this sample is more than one order of magnitude lower than the SOEP sample, and we can only exploit the variation in Jewish persecution across 99 German counties for which we have both historical data on persecution and PHF observations. The PHF questionnaire asks households to provide an estimate of their overall wealth. It also elicits households’ financial literacy and risk aversion using qualitative scales. The PHF questionnaire also collects information on households’ religiosity (frequency of church attendance), which helps us to assess whether religiosity may have persisted over the centuries and caused Jewish persecution to begin with. In column (1) of Table 4, we estimate equation1 on the PHF sample. In column (2) of Table 4, we add the individual-level measures of financial literacy, risk tolerance, and religiosity, and in column (3) we also add deciles for the overall household wealth. As expected, the measures of risk tolerance and financial literacy are positively associated with the likelihood of holding stocks, on top of the effect of being male and holding a college degree. The association between Jewish persecution and the likelihood of holding stocks is negative, statistically significant, and stable across these specifications. One- standard-deviation higher persecution in the county decreases the likelihood that the household owns stocks by 7 percentage points, which is about 24% of the average likelihood of holding stocks in the sample.22 In columns (4)-(6) of Table 4, the outcome variable is a dummy equal to 1 if the household directly holds stocks and trades them. Only 4% of the sample has direct ownership of stocks, and as expected being financially literate does not increase direct trading of individual stocks, which is known to guarantee lower excess returns than index investing (Barber and Odean, 2001). Jewish persecution is also negatively associated with the likelihood that households trade stocks directly. Forced migrations of Ashkenazi Jews and medieval persecution. Unobservable characteristics of German counties may have jointly determined Jewish persecution and stockholdings. Comparing geographically close counties helps alleviate this concern, but this test does not establish a causal effect of Jewish persecution on present-day financial decisions. Ideally, we would have randomly assigned anti-Jewish sentiment across similar German counties before the Black Death of 1349, because the variation of Jewish persecution across counties has persisted since the Middle Ages. To get close to such an experiment, we look at the forced migrations of Ashkenazi Jews out of the Rhine Valley after the 11th century. In the top map of Figure 3, the darker a county is, the older the first Jewish community documented in the county. Blank counties

22The likelihood of holding stocks in the PHF sample is 29%, which is similar to the likelihood of holding stocks for the SOEP households in the 2010 wave (28%).

15 are those with missing data. The earliest Jewish presence in the German lands was found in the cities of Trier, along the Mosel, and Cologne, along the Rhine. Archaeologists date this presence to the ninth century. Research has found evidence of Jewish communities in the 10th century along the entire Rhine Valley.23 The Jewish population in other areas of current Germany was sparse before the 11th century (Engelman(1944)). At the onset of the Crusades, Christian knights traveling from England and France to the Holy Land persecuted Jewish communities. Several towns on the Rhine expelled Jews, causing a massive Jewish migration toward Eastern, Northern, and Southern Germany. Evidence of sizable Jewish communities dates back to the late 13th and 14th centuries in Munich (south) and Berlin (east) (Toch, 2012).24 The bottom maps of Figure 3 show the location of the cities of Trier, on the Mosel, and Emmerich, on the northern end of the German Rhine. The age of the first documented Jewish community in a county increases as one moves toward each of these cities. We argue the distance of counties from the Rhine Valley determined the existence of Jewish communities at the time of medieval persecutions. In a first step, we use the distance of a county from the Rhine Valley to predict the probability that a Jewish community existed in the county before the Black Death. In a second step, we use the existence to predict the extent of Jewish persecution. The rationale is as follows: in counties with no Jewish communities before the Black Death, violence against Jews cannot have emerged, because no targets for such violence existed. In counties where early Jewish communities existed, the probability of a historical pogrom against the local Jews is strictly positive ex ante, because of the mere presence of Jews.25 In a third step, we use the persecution to predict present-day stock investments. Note this source of variation does not capture the different attitudes toward Jewish persecution across the counties that hosted Jewish communities, but only the variation in the likelihood of persecution between the counties that hosted a community and those that hosted no communities. Both margins of variation in persecution are relevant to our effect, as the lower estimated effect confirms when we restrict the analysis to the variation in persecution within the group of counties that hosted Jewish communities in the past (see columns (6)-(9) of Table 2).26

23We refer to Toch(2012) as a comprehensive economic history of European Jews in the Middle Ages. 24Only in the 15th century did Ashkenazi Jews merge with the communities of Khazar origin who had moved from the Black Sea to current Poland. See van Straten(2004) for archaeological evidence and Elhaik(2012) for genetic-based evidence. 25Of course, we will not necessarily observe a positive realization, that is, a pogrom in all of these counties ex post. 26Consistently, in Table 17 of the Online Appendix, we estimate the eﬀect within groups of counties at a similar distance from the Rhine Valley, which also gives us lower point estimates than the baseline speciﬁcations in Table 2.

16 We consider three measures of the distance of a county from the Rhine Valley. They are

the residuals ( ˆk) from regressions of the following type:

0 DistanceRhinek = Kk × δ + k, (2)

where Kk includes dummy variables for the historical characteristics of counties, that is, whether the county hosted a monthly market, a bishop see, free imperial cities, Hanseatic League cities, a city incorporated before 1349, and geographic characteristics (the land quality index of Ramankutty et al.(2002), whether a navigable river existed in the county, and whether the county lies in eastern Germany). We compute the minimal Euclidean distance of a county from the cities of Trier, on the Mosel, and Emmerich, on the northern end of the German Rhine (Figure 3). The lowest of these two distances is our third measure. The shortest distance is about 2 km, whereas the greatest distance is 1100 km. The alternative measures aim to capture two gradients of the distance from the Rhine Valley, that is, the southwest/northeast gradient and the northwest/southeast gradient. As Figure 3 shows, counties lying at the same distance from each of the two cities, that is, on the same isodistance curves, are diﬀerent; hence, our measures indeed capture two alternative gradients of the distance from the Rhine Valley. Across both gradients, the likelihood that a Jewish community existed in the Middle Ages increases toward the Rhine Valley.27

If we wanted to interpret this strategy as a causal test for the effect of Jewish persecution on present-day stockholdings, we should assume a demanding exclusion restriction. The residual distance of a county from the Rhine Valley should not affect current stock market participation through channels different from the county-level historical persecution against Jewish communities. Moreover, Jewish communities escaping from the Rhine Valley should have been equally likely to settle in any counties at the same distance from the Rhine Valley. Note if the latter condition did not hold, we would expect, if anything, that Jewish communities were more likely to settle in counties with higher demand for financial services; hence, the selection would bias our reduced-form coefficients downward. We propose four tests to assess the extent to which this exclusion restriction may be economically plausible. First, in Panel A of Table 5, we estimate the reduced-form effect of the distances from the Rhine Valley on the ratio of households that own stocks when the distances enter as regressors instead of persecution, and when both the distances

27To make our speciﬁcations directly comparable to the baseline results, we include the latitude of counties when computing the residual distances, but all the results are virtually identical if we exclude the latitude from the vector Kk in equation2.

17 and one of the persecution measures enter separately. All the coefficients refer to OLS regressions. In columns (1), (4), and (7), all three distances are positively associated with the likelihood that households hold stocks. Once the measures of Jewish persecution enter the reduced-form specifications, the estimated autonomous associations of the distances with stockholdings drop in magnitude, whereas the estimated standard errors for each coefficient stay virtually identical. Hence, the insignificant effects are not due to the imprecision of our estimates. Second, in Panel B of Table 5, we look at the effect of the distance on the likelihood that French households own stocks. Data for French households are from the Enquete Patrimoine run by the Banque of France in 2004. If the distance from the Rhine captures anything peculiar to the spatial diffusion of development or wealth, we should observe an effect of the distance on the stockholdings by French households who live to the west of the Rhine. Across all our measures, we find no economically or statistically significant association between the distance from the Rhine and the stockholdings of French households. Third, in Table 13 of the Online Appendix, we regress several county- and individual-level observables that correlate with stock market participation on our distance measures. Most distance residuals are not significantly associated with current county- and individual-level observables. In most cases, the size of the estimated coefficients is small, which suggests the imprecision of the estimates does not drive the non-results. An exception is the ratio of college graduates at the county level, which is positively associated with the residuals on the distance from Emmerich, but the estimated coefficient is negative when we use the distance from Trier. Fourth, in the Online Appendix, we derive the conditions for identifying our three-stage OLS system presented below, and detect no correlation between the instrument and the residuals from the structural equations.

We estimate the stages described above in a three-stage OLS framework (see Becker and Woessmann(2009)) at the county level: 28

0 Community1906k = α + β × LogDistanceRhinek + Kk × δ + k 0 P ersecutionk = α + β × Communityˆ 1096 + Kk × δ + k 0 RatioStockholdk = α + β × P ersecutionˆ k + Kk × δ + k,

where Communityˆ 1096k and P ersecutionˆ k are the predicted values for county k when

28Results are similar if we estimate the system at the individual level. Because the ﬁrst and second stages involve county-level outcomes, we ﬁnd it more appropriate to report the test at the county level.

18 estimating the system of three simultaneous equations. Columns (1)-(3) of Table 6 report the coefficient on LogDistanceRhinek when estimating the system using the three alternative measures of distance from the Rhine Valley. In column (1), a one-standard-deviation increase in the residual distance from Trier (0.69) reduces by nine percentage points the likelihood that an early Jewish community existed in a county. This figure is 27% of the average likelihood that a county hosted a Jewish community in 1096 (33%), the date before the Black Death, for which we have data on most counties. Estimated magnitudes when using the distance from Emmerich and the minimal distance are -13 and -11 percentage points.29 Hence, the farther a German county is from the Rhine, the less likely a Jewish community is to have existed there in 1096. In columns (4)-(6) of Table 6, we report results for the second stage. A one-standard-deviation increase in the likelihood of an early Jewish community increases the principal component of the persecution measures by 0.53, which is about one half of a standard deviation. The magnitudes are similar when we predict the likelihood of a Jewish community with the other measures of distance. Columns (7)-(9) of Table 6 document the third stage. Consistent with the reduced-form results, an increase in the instrumented persecution of Jews significantly reduces stock market participation at the county level.

5 Jewish Persecution and Banking: Mortgages and Deposits

So far, we have focused on the likelihood that German households hold stocks. Focusing on stock market participation is meaningful, because the defamation of Jews as stock-market manipulators survived in the press and popular culture even after the Jewish presence in other financial institutions, such as banking, had faded. The role of Jews in banking services started to decrease substantially with the foundation of the first Raiffeisenbank in 1843, and the subsequent diffusion of Volksbanken across German counties. Several generations of Germans have accessed banking services run by the non-Jewish population. But if the historical association between Jews and financial services affects current financial decisions through channels different from current antisemitism, we would expect to also find an effect of historical Jewish persecution on the attitudes of present-day Germans toward banking. We first look at the decision to get a mortgage to finance homeownership. This decision allows observation of whether households increase their debt through bank financing, or if they use their own savings, keeping constant the likelihood that they are homeowners. For the case of Germany, looking at this margin is quite relevant: in 2001, 43% of German

29The standard deviations of the distance from Emmerich and the minimal distance are 0.62 and 0.77.

19 households owned their home, and 20% of households had ever hold mortgages; that is, only 47% of homeowners had financed their homeownership via a mortgage (Georgarakos et al.(2010)). In Table 7, we find historical Jewish persecution is unrelated to households’ decision to buy their home, but it decreases significantly the likelihood that households hold a mortgage. In columns (1)-(2), we report the coefficients for estimating two probit specifications whose outcome variable is a dummy equal to 1 if the household owns any real estate properties. The effect of Jewish persecution on the likelihood of homeownership is economically and statistically insignificant. In columns (3)-(4), we report the coefficients for the same specifications, but where the outcome variable is a dummy equal to 1 if the household has ever held a mortgage. A one-standard-deviation increase in persecution reduces the likelihood of holding a mortgage by 0.7 percentage points, which is 10% of the average likelihood of holding mortgages in our sample (6.9%).30 The size and the statistical significance of this association are in line with the effect of Jewish persecution on present-day stockholdings, which we documented above. The second decision that relates households to banking services is their likelihood of saving through bank deposits. In columns (5)-(6) of Table 7, we find Jewish persecution is unrelated to the likelihood that the households in our sample declare they regularly save part of their monthly income. Ideally, we would like to observe how much of such savings are deposited in a bank. Unfortunately, we do not observe this dimension in the SOEP sample. We therefore compute the ratio of deposits to total assets for the banks that operate in a county, and we regress this ratio on the persecution measures. This test aims to check the amount of money households deposit in local banks, keeping constant the size of the local banks’ activities. In columns (7)-(8) of Table 7, we find one-standard-deviation higher Jewish persecution reduces the county-level ratio of deposits over the sum of local bank assets by 1.5 percentage points, which is 2% of the average ratio of deposits over assets across counties (76%). Mortgages and deposits in the PHF data. In Table 15 of the Online Appendix, we obtain all the results above in the PHF sample of German households surveyed in 2010 as well. Historical Jewish persecution is unrelated to the likelihood that households save a part of their monthly income regularly, and it is unrelated to the likelihood that the household is a homeowner. Instead, higher Jewish persecution is associated with a lower likelihood of holding a mortgage, even after controlling for wealth and for the elicited risk tolerance, financial literacy, and religiosity of the household head.

30The average in our cross section of households observed from 1984 to 2010 is lower than the average for the cross section of households studied by Georgarakos et al.(2010), which are all observed in 2001.

20 6 Channels Mediating the Eﬀect of Jewish Persecution on Financial Decisions

In this section, we study the channels through which Jewish persecution may affect the financial decisions of present-day German households. We consider supply- and demand- side channels. Supply-side channels may mediate the negative effect through a historical and/or current lower supply of financial investment opportunities in counties with higher Jewish persecution. Demand-side channels may mediate the effect of Jewish persecution on financial decisions through the present-day demand for financial services. Supply-side channels. Starting in 1843, credit unions (Volks- and Raiffeisenbanken) have diffused across Germany. They were and still are specialized in financing local businesses and collecting households’ savings. If credit unions diffused early into more antisemitic areas, current households in those areas might be less aware of stock investment. We collect novel data on the foundation dates of credit unions across German counties from the proprietary registry of the Hoppenstedt Firmendatenbank.31 Panel A of Figure 10 of the Online Appendix shows the diffusion path of credit unions across Germany. In Panel B of Figure 10 of the Online Appendix, we plot the year when the first credit union is documented in a county against the ratio of deported Jews over the 1933 Jewish population at the county level. The two dimensions are not negatively correlated, so credit unions did not diffuse earlier in the more antisemitic counties. Another supply-side channel concerns the skill structure of counties. Persecution may have reduced local financial services because a large share of finance workers were Jewish. If the depletion of human capital needed to run financial institutions drove our baseline effect, the effect should be larger in counties with a higher ratio of Jewish workers in finance. In Panel A of Figure 11 of the Online Appendix, we find no association of the ratio of Jews in finance as of 1882 on present-day stockholdings. In Panel B, we proxy the ratio of Jews in finance as of 1933 with the ratio of Jews in the total population in 1933, and we find no association between the main effect and the ratio of Jews in a county either. We also look at the present-day supply of financial services. In Table 16 of the Online Appendix, we regress the present-day number of retail bank branches per inhabitant at the county level from the Bundesbank data-base on the measures of historical Jewish persecution, and we find these two dimensions are also unrelated. Moreover, different from the case of Italy, institutions supported by the Church to compete

31The registry reconstructs the chains of mergers and acquisitions over the decades for currently existing German banks. They collect the foundation date, the type, and other characteristics of any entity involved in these chains as far back as any information is retraceable.

21 with the Jewish monopoly in banking did not diffuse in Germany. These institutions are the backbones of the more developed local banking systems after the expulsion of Jews from the the Italian south that Pascali(forthcoming) studies for the case of Italy. These results suggest that supply-side channels are unlikely to explain the effect of historical Jewish persecution on the financial decisions of present-day German households. Jewish persecution and distrust in finance. Moving on to the demand-side channels, we first test whether historical Jewish persecution predicts present-day distrust in financial institutions, including the stock market, commercial banks, and local banks. We do not observe distrust in financial institutions for the households in the SOEP sample, and hence, we run our own survey on a sample of 1,000 German households, asking them the extent to which they trust the stock market, commercial banks (Privatbanken), and local banks (Sparkassen and Genossenschaftsbanken). The survey is administered by the company ClickWorker on a stratified sample of the German population that sign up to the platform to perform tasks and surveys for pay. The respondents only know they are part of a survey, and they ignore the identity or scopes of the researchers. This protocol is crucial to avoid demand effects invalidating the procedure. We also ask for demographics and the zip code, which we map into counties. We adapted the questions developed for the United States in the Kellogg-Booth Index of Financial Trust (Sapienza and Zingales(2012)). In the Online Appendix, we report the original survey. After providing their demographics, the respondents faced a set of four questions asking the extent to which they trust others (generalized trust), the stock market, commercial banks, and local banks, on a scale from 1 (do not trust at all) to 7 (trust completely). We also elicited households’ willingness to take risks in general and in financial decisions, on a similar scale from 1 to 7. Finally, the respondents reported whether they owned stocks and whether they had a primary banking relationship with a commercial bank or with a local bank. The final sample on which we can run the analysis is about one half of the original 1,000 households. This drop in the sample size is due to the loss of respondents in counties for which we have no historical persecution data. Also, our final sample is distributed across 57 counties, because we don’t have respondents for all German counties in the survey, and we only consider counties for which we have at least five respondents, to guarantee the representativeness of the answers.32

To make the results directly comparable to the analysis of the eﬀect of Jewish persecution on ﬁnancial decisions, we transform the trust measures into dummy variables, which equal 1 if the measure is larger or equal to 5, and zero otherwise, and we estimate probit

32The results are similar if we change this threshold.

22 specifications whose outcomes are the trust in each financial institution.33 Table 19 of the Online Appendix reports the summary statistics and the correlational structure across these trust measures. The Table also provides t-tests for whether the average of the trust measures is the same in the full sample of households, and in the subsample of households that enter the analysis. Interestingly, the correlations across the financial trust measures, and with generalized trust, are positive but not high. A ”pecking-order” path of trust exists in financial institutions: on average, 40% of respondents trust local banks, 17% trust commercial banks, and only 13% trust the stock market. The average generalized trust is 39%, and its correlation with the measures of financial trust varies from 9% to 11% across measures. Risk tolerance is correlated with the trust in the stock market (38%), but not with the trust in local banks (2%), which suggests controlling for the respondent’s risk tolerance in the analysis is important.34 In Panel A of Table 9, we regress the trust toward the stock market on the principal component of the measures of Jewish persecution (columns (1)-(2)), after controlling for generalized trust, risk tolerance, and demographics at the individual level. A one-standard-deviation increase in persecution is associated with a 4-percentage-point drop in the respondents’ trust in the stock market. Columns (3)-(4) of Table 9 report a similar specification whose outcome is the dummy for whether respondents trust commercial banks, and columns (5)-(6) of Table 9 for whether respondents trust local banks. We do not detect an association of Jewish persecution with the trust in commercial banks, but we find that a one-standard-deviation increase in persecution is associated with an 8-percentage-point drop in the trust in local banks. Panel B of Table 9 shows the trust measures we elicit are positively associated with the trusted outcomes, even when controlling for generalized trust, risk tolerance, and demographics. Disentangling the demand-side channels. Although distrust in financial services seems a plausible channel that mediates the effect of Jewish persecution on present-day financial decisions, we could think of at least two other demand-side channels. First, Jewish persecution may capture households’ ”backwardness,” that is, a set of cultural cues and beliefs that promote distrust toward the unfamiliar. Second, current households that are antisemitic may still associate the stock market with Jews, and thus invest less in stocks. Third, distrust in the financial sector may be a by-product of historical anti-Jewish sentiment. It may have transmitted across generations even if Jews are not

33All the results are similar if we instead use the original categorical variables and estimate multinomial logit speciﬁcations. 34In a previous survey, we only elicited the trust toward the stock market, and we did not elicit generalized trust or risk aversion at the individual level. We thank Luigi Guiso for encouraging us to run this novel version of the survey.

23 associated with the financial sector anymore in Germany because they were tragically eliminated after the Second World War. We cannot disentangle these three channels based on the main effect. We exploit the heterogeneity of households by education and cohorts, two dimensions for which the channels have alternative predictions. More educated households are less prone to backwardness, and are less antisemitic today: In Panels B and C of Figure 4, we show the backwardness and antisemitism of current Germans decline substantially and monotonically with their education level, using novel data from the Friedrich Ebert Stiftung, described in Decker et al.(2012). These unique data allow for the first time observation of whether present-day households think Jews have too much influence in financial markets. The association of Jews with the stock market decreases with education similarly to other measures of antisemitism and backwardness. Distrust in the stock market, in commercial banks, and in local banks, instead, do not decline monotonically with education, as we show in Panel A of Figure 4.35

Backwardness or present-day antisemitism could explain the effect only if the size of the effect decreased with the education level of households. But the effect of Jewish persecution on the likelihood of holding stocks does not vary across education levels. Panel A of Table 8 shows that the baseline effect of Jewish persecution on stockholdings does not reverse for college-educated household heads. The same holds true if we allow for differential effects of college and high school education (columns (2), (4), (6), and (8)). The effect of Jewish persecution on current stock market participation is invariant to the education level of household heads. This result seems inconsistent with backwardness and antisemitism, and consistent with the non-sensitivity of distrust in finance to education levels in our survey. Morever, Voigtlaender and Voth(2014) find current antisemitism is significantly higher for the cohort of those raised during the Nazi period, compared to the younger cohorts. Hence, if a conscious association of Jews with finance drove our results, the effect should be higher for the cohort raised during the Nazi period. In Panel B of Table 8, we interact our measures of Jewish persecution with dummies for two cohorts of Germans: those born before 1945 and those born after 1965, so that the omitted category is the set of household heads born after the Second World War and before 1965. Across all the measures, on average, the oldest cohort invests more than the two younger cohorts. When using the principal component measure of persecution (columns (1)-(2)), the size of the effect of persecution on the oldest cohort is lower than the size of the effect for the younger cohorts,

35To make patterns easily comparable, we have deﬁned the measures of distrust as ((7-answer)*5)/7, where ”answer” is the response to the questions on trust in ﬁnancial institutions in the survey, which range from 1 to 7.

24 and is indifferent from the effect on other cohorts when using the other measures. These results are inconsistent with a conscious association of present-day antisemitic households driving our findings, because Germans raised during the Nazi period display significantly higher antisemitism than all other cohorts of Germans (Voigtlaender and Voth, 2014).36 An additional result consistent with a persistent norm of distrust in finance, but hardly consistent with backwardness or the conscious association of Jews with finance by present-day antisemitic households, is that the size of the effect does not vary from 1984 to 2010 (see Table 14 in the Online Appendix).

The persistence of a cultural norm of distrust in finance irrespective of the presence of Jews in the financial sector is plausible, because the Holocaust is a compulsory topic in the curricula of primary and secondary German schools,37 which may allow for antisemitic beliefs to dissipate with schooling, consistent with our findings and previous findings. German curricula do not cover topics related to financial literacy, or the benefits of investing in the stock market, so that schooling would hardly change a norm of distrust in finance transmitted across generations.

7 Conclusions

Households in German counties where Jewish persecution was higher in the Middle Ages and the Nazi period access financial services less than other households. They are less likely to hold stocks, they have fewer mortgages, but not lower homeownership, and they put less savings into bank accounts, but are as likely as other households to save. Households in high-persecution counties appear more likely to keep their money in cash form. Among several possible channels, we find evidence consistent with a cultural norm of distrust in finance, which might have transmitted across generations.

Future research should study the transmission mechanism of norms that affect the investment and saving decisions of households, and how such norms could be modified. Can financial education dissipate a norm of distrust in finance, by making individuals aware of the costs of not investing in remunerative instruments? Or do individuals decide not to invest despite knowing the negative effects of their actions on the long-run accumulation of financial wealth? The results contribute to the interdisciplinary debate on hatred beliefs and their long-term

36We cannot look at our survey across cohorts, because we have only three respondents born before 1945. 37German states have included the Holocaust and the 19th-century antisemitism as compulsory topics in their primary and secondary school curricula since 1978 (see der St¨andingenKonferenz der Kultusminister der L¨anderin der Bundesrepublik Deutschland(1997)).

25 consequences on societies. Because of the historically high equity premium, and of the reliance of ﬁrms on equity capital, the persecution of the Jewish minority in the past reduced not only the long-term wealth of the persecuted, but of the persecutors as well.

References D. Acemoglu, T. Hassan, and J. Robinson. Social structure and development: A legacy of the holocaust in russia. Quarterly Journal of Economics, 126(2):895–946, 2011. M. Adena, R. Enikolopov, M. Petrova, V. Santarosa, and E. Zhuravskaya. Radio and the rise of the nazis in prewar germany. Working Paper, 2015. M. Akbulut-Yuksel and M. Yuksel. The longer-term direct and external effects of jewish expulsions in nazi germany. American Economic Journal: Economic Policy, Forthcoming. Y. Algan and P. Cahuc. Inherited Trust and Growth. Journal of Political Economy, 114(2):280–316, 2006. E. Banfield. The Moral Basis of a Backward Society. The Free Press, 1958. B. Barber and T. Odean. Gender, overconfidence, and common stock trading. Quarterly Journal of Economics, 1(116):261–292, 2001. S. Becker and L. Woessmann. Was Weber Wrong? A Human Capital Theory of Protestant Economic History. Quarterly Journal of Economics, 124(2):531–596, 2009. S. Becker, F. Cinnirella, E. Hornung, and L. Woessmann. ipehd - the prussian ifo economic history database. Historical Methods, 47(2):57–66, 2014. S. Betermier, T. Jansson, C. Parlour, and J. Walden. Hedging labor income risk. Journal of Financial Economics, 3(105):622–639, 2012. M. Botticini and Z. Eckstein. Jewish Occupational Selection: Education, Restriction or Minorities? Journal of Economic History, 65(4):922–948, 2005. M. Botticini and Z. Eckstein. The Chosen Few. Princeton University Press, 2012. J. Brown, A. Cookson, and R. Heimer. Law and finance matter: Lessons from externally imposed courts. Working Paper, 2015. K. Burchardi and T. Hassan. The economic impact of social ties: Evidence from german reunification. Quarterly Journal of Economics, 128(3):1219–1271, 2013. F. D’Acunto. History & finance: A nascent field. Working Paper, 2015a. F. D’Acunto. Innovating to invest: The role of basic education. Working Paper, 2015b. A. de Bromhead, B. Eichengreen, and K. O’Rourke. Political extremism in the 1920s and 1930s: Do the german lessons generalize? Journal of Economic History, 73(2):371–406, 2013. O. Decker, J. Kiess, and E. Brahler. Die Mitte im Umbruch: Rechtextreme Einstellungen in Deutschland 2012. Bonn: Verlag J.H.W. Dietz Nachf. GmbH, 2012. Das Sekretariat der Ständingen Konferenz der Kultusminister der Länder in der Bundesrepublik Deutschland. Zur auseinandersetzung mit dem holocaust in der schule. 1997. T. Dohmen, A. Falk, D. Huffman, and U. Sunde. The intergenerational transmission of risk and trust attitudes. Review of Economic Studies, 2(79):645–677, 2012. E. Elhaik. The Missing Link of Jewish European Ancestry: Contrasting the Rhineland and the Khazarian Hypotheses. Genome Biology and Evolution, 5(1):61–74, 2012. U. Z. Engelman. The Rise of the Jew in the Western World. Behrman’s Jewish Book House, NY, 1944. T. Fritsch. Statistik des judenthums. Verlag von Beyer, 1892. N. Fuchs-Schuendeln and M. Schuendeln. Precautionary savings and self-selection: Evidence from the german reunification experiment. Quarterly Journal of Economics, 120(3):1085–1120, 2005. D. Georgarakos, A. Lojschova, and M. Ward-Warmedinger. Mortgage indebtedness and household financial distress. ECB Working Paper Series No. 1156, 2010. M. Giannetti and T. Wang. Corporate scandals and stock market participation. Journal of Finance, forthcoming.

26 O. Glagau. Der boersen- und gruendungs-schwindel in berlin. Leipzig: Verlag v. Paul Frohberg, 1876. T. Gormley, H. Liu, and Z. Guofu. Limited participation and consumption-saving puzzles: A simple explanation and the role of insurance. Journal of Financial Economics, (96): 331–344, 2010. I. Grosfeld, A. Rodnyansky, and E. Zhuravskaya. Persistent Anti-Market Culture: a Legacy of the Pale of Settlement after the Holocaust. American Economic Journal: Economic Policy, 5(3):189–226, 2013. A. Gross. Economic history of the jews. New York: Schocken Books, 1975. L. Guiso and T. Jappelli. Awareness and Stock Market Participation. Review of Finance, 9(4):537–567, 2005. L. Guiso, T. Jappelli, and D. Terlizzese. Income Risk, Borrowing Constraints and Portfolio Choice. American Economic Review, 86:158–172, 1996. L. Guiso, P. Sapienza, and L. Zingales. The Role of Social Capital in Financial Development. American Economic Review, 94(3), 2004. L. Guiso, P. Sapienza, and L. Zingales. Cultural Biases in Economic Exchange. Quarterly Journal of Economics, 124(3):1095–1131, 2009a. L. Guiso, P. Sapienza, and L. Zingales. Trusting the Stock Market. Journal of Finance, 63(6):2557–2600, 2009b. A. Hitler. Mein Kampf - Murphy Translation, Vol. I, chapter XI: Race and People. Hurst and Blackett, 1939. H. Hong, JD. Kubic, and J. Stein. Social Interaction and Stock Market Participation. Journal of Finance, 2004. G. King, O. Osen, M. Tanner, and A. Wagner. Ordinary economic voting behavior in the extraordinary election of adolf hitler. Journal of Economic History, 68(4):951–996, 2008. I. Koehler. Die ”ariesirung” der privatbanken im dritten reich. Verlag C.H. Beck, 2005. U. Malmendier and S. Nagel. Depression Babies: Do Macroeconomic Experiences Affect Risk-Taking? Quarterly Journal of Economics, 126(1):373–416, 2011. W. Marr. Der Sieg der Judentums ueber das Germanentum. Bern, 1879. W. Mosse. Jews in the German Economy - The German-Jewish Economic Elite 1820- 1935. Clarendon Press, 1987. N. Nunn. Historical development. Handbook of Economic Growth. Vol. 2., pages 347–402, 2014. L. Pascali. Banks and development: Jewish communities in the italian renaissance and current economic performance. Review of Economics and Statistics, forthcoming. C. Paxson. Borrowing constraints and portfolio choice. Quarterly Journal of Economics, 105(2):535–543, 1990. M. Perry and T. Schweitzer. Antisemitism - myth and hate from antiquity to the present. Palgrave Macmillan, 2002. L. Pierce and J. Snyder. The development roots of firm access to finance: Evidence from the african slave trade. Working Paper, 2014. L. Poliakov. The history of anti-semitism - vol. 3: From voltaire to wagner. The Vanguard Press, Inc., 1975. L. Poliakov. The history of anti-semitism - vol. 4: Suicidal europe. The Vanguard Press, Inc., 1977. R. Putnam. Making Democracy Work. Princeton University Press, 1993. N. Ramankutty, J. Foley, J. Norman, and K. McSweeney. The global distribution of cultivable lands. Global Ecology and Biogeography, 11:377–392, 2002. G. Reuveni and S. Wobick-Segev. The Economy in Jewish History: New Perspectives on the Interrelationship between Ethnicity and Economic Life. Berghahn Books, 2010. P. Sapienza and L. Zingales. A trust crisis. International Review of Finance, 12(2): 123–131, 2012. W. Sombart. Die zukunft der juden. Leipzig:, 1912. E. Spolaore and R. Wacziarg. How deep are the roots of economic development? Journal of Economic Literature, 51(2):325–369, 2013. M. Toch. The Economic History of European Jews: Late Antiquity and Early Middle Ages. Brill Academic Publishers, Leiden, 2012. H. Treitschke. Unserer Aussichten. Preussische Jahrbuecher 44/45, 1879.

27 M. van Rooij, A. Lusardi, and R. Alessie. Financial Literacy and Stock Market Participation. Journal of Financial Economics, 101(2):449–472, 2011. J. van Straten. Jewish migrations from germany to poland: the rheinland hypothesis revisited. The Mankind Quarterly, 44(3-4):367–384, 2004. N. Voigtlaender and H.J. Voth. Persecution perpetuated: The medieval origins of antisemitic violence in nazi germany. Quarterly Journal of Economics, 127(3):1339–1392, 2012. N. Voigtlaender and H.J. Voth. Married to intolerance: Attitudes towards intermarriage in germany, 1900-2006. American Economic Review Papers and Proceedings, 103(3): 79–85, 2013a. N. Voigtlaender and H.J. Voth. Taught to hate: Nazi indoctrination and anti-semitic beliefs in germany. Working Paper, 2014. F. Waldinger. Quality matters: The expulsion of professors and the consequences for phd students’ outcomes in nazi germany. Journal of Political Economy, 118(4):787–831, 2010.

28 Figure 2: Data Properties: Jewish Persecution and Stock Market Participation

(a) Jewish Persecution Nazi Period (b) Present-day Stock Market Participation

(c) Jewish Persecution Density (d) Stock Market Participation Density 8 .04 6 .03 4 .02 Density Density 2 .01 0 0 0 20 40 60 80 100 0 .1 .2 .3 .4 .5 Ratio Deported Jewish Population (base: 1933) Ratio of Households which Own Stocks (1984-2010) (e) Persecution and Participation (f) Pogroms and Participation .18 .6 .17 .4 .16 .2 .15 0 .14 0 20 40 60 80 No Pogroms in 1349 At Least one Pogrom in 1349 Ratio of Households who Own Stocks (average 1984-2010)

Ratio Deported Jewish Population Ratio of Households who Own Stocks (average 1984-2010)

In Panels (a) and (b), the darker a county is, the higher the value of the depicted variable. Blank counties are those for which the data are not available. Panel (a) plots the ratio of Jews deported during the Nazi period over the total Jewish population in 1939. Panel (b) plots the average yearly ratio of households who have invested in stocks from 1984 to 2010. Panels (c) and (d) plot the sample distributions of the same measures as above. Panel (e) depicts the unconditional correlation between stock market participation and the ratio of Jews deported across German counties. Panel (f) shows the mean stock market participation in counties that experienced and did not experience a pogrom in 1349.

29 Figure 3: Settlement of Jewish Communities and Distance from the Rhine Valley

(a) Year when the ﬁrst Jewish community was documented

(b) Counties at same distance from Trier (c) Counties at similar distance from Emmerich

In the maps of Figure 3, the darker a county is, the earlier a Jewish community was documented in that county. Blank counties are those for which the data are not available. The bottom maps show the location of the cities of Trier, on the Mosel, and Emmerich, on the northern end of the German Rhine. The isodistance curves out of the two cities emphasize which counties are at the same distance from Trier or from Emmerich.

30 Figure 4: Distrust in Finance, Antisemitism, and Backwardness by Education Levels (a) Distrust in Financial Institutions 5 5 5 4 4 4 3 3 3 Distrust in Local Banks 2 Distrust in the Stock Market Stock in the Distrust 2 2 Distrust in Commercial Banks 1 1 1

Below High School High School College Below High School High School College Below High School High School College

(b) Present-day Antisemitism 5 5 5 4 4 4 31 3 3 3 2 2 2 There is Something Dodgy About Jews Jews use Tricks to Obtain what They want what Obtain Jews use Tricks to Jews have too much Influence in Financial Markets 1 1 1

Below High School High School College Below High School High School College Below High School High School College

(c) Present-day Backwardness 5 5 5 4 4 4 3 3 3 2 2 2 Women Should Stay at Home Kids should be Forced to Obey their Parents 1 1 1 Women Should Sacrifice their Career to Help Husbands Below High School High School College Below High School High School College Below High School High School College

Each graph in Figure 4 reports the sample mean and the 95% conﬁdence interval around the mean of the outcomes described on the vertical axes split in three groups sorted by the respondent’s education level: Below High School includes respondents who have a Realschule degree or lower level of education; High School includes respondents who have an Abitur or correspondent level of education in the German system. College includes respondents with a Vordiplom or higher level of education.

Table 1: Summary Statistics

All available observations Non-missing information p-value

Obs. Mean St. dev. Min. Max. Obs. Mean St. dev. Min. Max. Persecution of Jews Pogrom 1349 17818 0.576 0.494 0 1 13870 0.570 0.495 0 1 0.358 Log deported Jews 17818 4.956 2.311 0 10.93 13870 4.950 2.295 0 10.93 0.881 Ratio deported Jews 17477 34.06 19.61 0 163.5 13599 34.34 19.81 0 163.5 0.158 Principal component persecution 17477 -0.223 1.106 -3.082 3.571 13599 -0.222 1.112 -3.082 3.571 0.980

Household characteristics Holds Stocks 26761 0.163 0.369 0 1 13870 0.158 0.364 0 1 0.092 Homeowner 27064 0.391 0.488 0 1 13870 0.417 0.493 0 1 0.009 Has life insurance 26761 0.469 0.499 0 1 13870 0.462 0.499 0 1 0.198 Income 26761 31522 26614 -36 986400 13870 30514 27061 -36 986400 0.000 Age 21981 48.62 17.64 17 97 13870 48.49 17.58 17 96 0.371 Female 21982 0.490 0.500 0 1 13870 0.473 0.499 0 1 0.000 Single 27064 0.178 0.382 0 1 13870 0.218 0.413 0 1 0.000 High School or higher 27079 0.766 0.423 0 1 13870 0.718 0.450 0 1 0.000 32

County characteristics Log Population 1933 17818 11.57 1.371 7.94 14.29 13870 11.56 1.368 7.94 14.29 0.638 Log Jews 1933 17818 6.271 2.047 0 11.99 13870 6.261 2.041 0 11.99 0.708 Percentage unemployed 1933 17818 18.21 8.768 2.618 40.52 13870 18.01 8.838 2.618 40.52 0.028 Percentage blue collars 1933 17818 44.85 10.56 16.49 72.40 13870 44.67 10.68 16.49 72.40 0.023 Percentage self employed 1933 17818 20.01 4.211 9.096 32.74 13870 20.08 4.265 9.096 32.74 0.064 Percentage Catholics 1925 17818 40.11 32.71 0.502 98.77 13870 41.27 32.53 0.517 98.77 0.006 Eastern Germany 27079 0.129 0.335 0 1 13870 0.071 0.258 0 1 0.000 Latitude 24959 50.69 1.727 47.95 54.03 13870 50.65 1.642 47.95 54.03 0.643 Land quality index 24959 0.562 0.149 0.306 0.870 13870 0.566 0.149 0.306 0.870 0.628 Income p.c. 2005 22748 18033 2301 12846 27253 13870 18099 2220 13115 25027 0.452 College graduates 2005 24766 24.40 4.929 17.60 34.60 13870 23.55 4.463 17.60 34.60 0.010

This Table reports summary statistics for measures of the historical persecution of Jews in the German lands, and for the characteristics of households and counties where households live. Each observation is a German household interviewed by the SOEP any time between 1984 and 2011. For each variable, the table reports the number of observations for which the variable is observed, its mean, standard deviation, minimal, and maximal values. The left panel reports statistics for households interviewed by the SOEP (repeated observations are excluded) and whose county of residence has been collected. The right panel reports statistics for the sample of households for which we do not miss any household or county-level characteristics. The latter is the sample we will use throughout the analysis. The column on the far right reports p-values for tests of the diﬀerence in the means across the two samples. Table 2: Jewish Persecution and Stock Market Participation

(1) (2) (3) (4) (5) (6) (7) (8) (9) All Counties Only if Jewish Community in the Middle Ages

P.C. Jewish Extensive Intensive Pogrom P.C. Jewish Extensive Intensive Pogrom Persecution Margin Margin 1349 Persecution Margin Margin 1349

Persecution of Jews -0.015 -0.018 -0.015 -0.011 -0.020 -0.012 -0.013 -0.009 -0.011 (column heading) 0.007** 0.007*** 0.006** 0.002*** 0.009** 0.005** 0.006** 0.002*** 0.011 Log Jews 1933 -0.003 -0.001 0.001 -0.003 -0.002 -0.002 -0.013 -0.020 -0.018 0.005 0.001 0.005 0.004 0.004 0.009 0.010 0.009** 0.010*

% Catholics 1925 0.028 0.000 -0.001 -0.001 0.001 0.000 0.001 0.001 0.000 0.012** 0.011 0.010 0.010 0.010 0.001 0.001 0.001 0.001

Age -0.002 -0.002 -0.002 -0.002 -0.002 -0.002 -0.002 -0.002 0.001 0.001 0.001 0.001** 0.001 0.001 0.001 0.001

Age^2 /100 0.004 0.004 0.004 0.004 0.004 0.004 0.004 0.004 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** Female -0.006 -0.006 -0.006 -0.006 0.004 0.004 0.004 0.004 0.006 0.006 0.006 0.006 0.007 0.007 0.007 0.006 Single 0.054 0.054 0.054 0.055 0.059 0.058 0.059 0.060 33 0.010*** 0.010*** 0.010*** 0.010*** 0.010*** 0.010*** 0.011*** 0.010*** College 0.011 0.011 0.011 0.011 0.015 0.015 0.016 0.016 0.004*** 0.004*** 0.004** 0.003** 0.004*** 0.005*** 0.005*** 0.005*** Eastern Germany -0.033 -0.040 -0.041 -0.031 -0.002 -0.001 0.005 -0.002 0.019* 0.021* 0.020** 0.020 0.011 0.011 0.020 0.012 Income p.c. 2005 0.034 0.0041 0.038 0.034 0.060 0.058 0.063 0.053 0.016** 0.016** 0.017** 0.016** 0.020*** 0.020*** 0.025** 0.020*** % College graduates 2005 0.001 0.002 0.002 0.001 -0.001 -0.001 -0.001 -0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.002 0.001

Income deciles X X X X X X X X X Other historical controls X X X X X X X X X Wave groups f.e. X X X X X X X X Regional controls X X X X X X X X Observations 17,701 13,870 13,870 13,870 13,870 9884 9884 9884 9884 N. of clusters 296 270 270 270 270 190 190 190 190 (Pseudo-) R2 0.01 0.10 0.10 0.10 0.10 0.10 0.10 0.11 0.10 This Table reports average marginal eﬀects computed after estimating the following probit speﬁcation:

0 0 P r(HoldsStocksik|Xik,Kik) = Φ(α + β × P ersecutionk + Xik × γ + Kik × δ + Income deciles + ηt + ik), Each observation is a German household interviewed by the SOEP any time between 1984 and 2011. In all columns, the dependent variable is a dummy which equals one if the household holds stocks. The main covariate of interest, P ersecution, is the measure of persecution of Jews in the past described at the top of each nd nd column. Xik includes the following individual-level controls: gender, single/marital status, income (2 degree polynomial); age (2 degree polynomial); college education; homeownership; and life and social insurance. Kk includes the following county-level current and historical controls: latitude; income per capita; share of college-educated population; index of quality of cultivable land; log of population in 1933; log of Jewish population in 1933; ratio of population employed in the retail sector in 1933; ratio of population employed in manufacturing in 1933; and ratio of Catholic population in 1925. Income deciles are dummies indicating the decile of the income distribution to which the household belongs, and Φ is the standard normal cdf. S.e. are clustered at the county level. Statistical signiﬁcance is reported as follows: *10%, **5%, ***1%. Table 3: Geographically Close Counties and Robustness

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Geographically close counties Robustness

ities

s s W WW W poorest poorest level States Zone No Free Free No

Within League counties counties County County Within 9 Within 16 Refugees No Occupation Occupation Post No Hanseatic Imperial C Virtual States Virtual States Within Controlling for No least educated educated No least

Panel A. All Counties

Principal Component -0.012 -0.012 -0.010 -0.018 -0.021 -0.020 -0.028 -0.026 -0.016 -0.019 Persecution of Jews 0.006** 0.005** 0.006* 0.006*** 0.007*** 0.006*** 0.009*** 0.009*** 0.007** 0.007** Observations 13,870 13,870 13,870 13,870 12,686 9,371 8,348 11,560 11,087 N. of clusters 270 270 270 270 270 227 156 168 244 225 (Pseudo-) R2. 0.12 0.12 0.11 0.10 0.25 0.09 0.08 0.08 0.08 0.08

Panel B. Only if Jewish Community in the Middle Ages

Principal Component

34 -0.009 -0.010 -0.010 -0.013 -0.022 -0.014 -0.023 -0.010 -0.012 -0.015 Persecution of Jews 0.004** 0.005** 0.005** 0.006** 0.007*** 0.008* 0.010** 0.006 0.008 0.008* Observations 9,884 9,884 9,884 9,884 9,031 6,990 6,661 8,219 7,173 N. of clusters 190 190 190 190 190 159 116 132 169 147 (Pseudo-) R2. 0.11 0.12 0.11 0.10 0.26 0.09 0.07 0.08 0.07 0.07

Individual contr. X X X X X X X X X Historical contr. X X X X X X X X X X Regional controls X X X X X X X X X X Wave group f.e X X X X X X X X X This Table reports average marginal eﬀects computed after estimating the following probit speﬁcation:

0 0 P r(HoldsStocksik|Xik,Kik) = Φ(α + β × P ersecutionk + Xik × γ + Kik × δ + Income deciles + ηt + ik), adding the geographic fixed effects described on the column headings (columns (1)-(4)), or across subsamples defined as described by the column headings (columns (5)-(10)). Each observation is a German household interviewed by the SOEP for the first time between 1984 and 2010. In all columns, the dependent variable is a dummy that equals 1 if the household holds stocks. The main covariate of interest, P ersecution, is the principal component of the three measures of Jewish persecution described in the main text. Xik includes the following individual-level controls: gender, single/marital status, nd nd income (2 degree polynomial); age (2 degree polynomial); college education; homeownership; and life and social insurance. Kk includes the following county-level current and historical controls: latitude; income per capita; share of college-educated population; index of quality of cultivable land; log of population in 1933; log of Jewish population in 1933; ratio of population employed in the retail sector in 1933; ratio of population employed in manufacturing in 1933; and ratio of Catholic population in 1925. Income deciles are dummies indicating the decile of the income distribution to which the household belongs, and Φ is the standard normal cdf. S.e. are clustered at the county level. Statistical significance is reported as follows: *10%, **5%, ***1%. Table 4: Wealth, Financial Literacy, and Risk Tolerance: The PHF Sample

(1) (2) (3) (4) (5) (6) Holds Stocks Holds Stocks (trades directly)

Probit Probit Wealth Probit Probit Wealth deciles deciles -0.073 -0.070 -0.071 -0.017 -0.018 -0.017 Persecution Jews 0.032** 0.031** 0.031** 0.007** 0.007** 0.008**

Log Jews 1933 -0.007 -0.010 -0.005 -0.001 -0.001 -0.001 0.010 0.010 0.010 0.003 0.002 0.002

% Catholics 1925 0.053 0.044 0.033 0.012 0.013 0.011 0.034 0.035 0.036 0.010 0.010 0.010

Age 0.008 0.006 0.004 0.001 0.001 0.002 0.006 0.006 0.006 0.002 0.002 0.002

Age^2 /100 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 0.005 0.001 0.001 0.001 0.001 0.001 Male 0.071 0.065 0.069 0.002 0.002 0.003 0.018*** 0.019*** 0.017*** 0.005 0.005 0.005 Single -0.013 -0.015 -0.031 -0.006 -0.005 -0.007 0.047 0.047 0.047 0.017 0.017 0.018 College 0.063 0.057 0.055 -0.006 -0.006 -0.001 0.031** 0.031* 0.029* 0.066 0.068 0.007 Eastern Germany -0.021 -0.024 0.004 0.009 0.009 0.007 0.051 0.052 0.050 0.008 0.009 0.010 Financially Literate 0.063 0.062 -0.000 -0.000 0.026** 0.025** 0.008 0.008 Wants to take Risk 0.242 0.228 0.020 0.023 0.022*** 0.021*** 0.009** 0.010** Goes to church often -0.002 -0.002 -0.003 -0.006 0.027 0.026 0.009 0.009

Other historical controls X X X X X X Regional controls X X X X X X Observations 1,256 1,256 1,256 1,256 1,256 1,256 N. of clusters 99 99 99 99 99 99 Pseudo-R2 0.21 0.26 0.27 0.22 0.22 0.28 This Table reports average marginal eﬀects computed after estimating the following probit speﬁcation:

0 0 P r(HoldsStocksik|Xik,Kik) = Φ(α + β × P ersecutionk + Xik × γ + Kik × δ + W ealth deciles + ηt + ik), where the main covariate of interest, P ersecution, is the first principal component of the measures of Jewish persecution in the past described in Section 2 of the paper. Xik includes the following individual-level controls: gender, single/marital status, income (2nd degree polynomial); age (2nd degree polynomial); college education; homeownership; and life and social insurance. Kk includes the following county-level current and historical controls: latitude; income per capita; share of college-educated population; index of quality of cultivable land; log of population in 1933; log of Jewish population in 1933; ratio of population employed in the retail sector in 1933; ratio of population employed in manufacturing in 1933; and ratio of Catholic population in 1925. W ealth deciles are dummies indicating the decile of the wealth distribution to which the household belongs, and Φ is the standard normal cdf. The outcome variables and the household-level demographics are from the first wave of the Panel of Household Finances of the German Bundesbank. S.e. are clustered at the county level. Statistical significance is reported as follows: *10%, **5%, ***1%.

35 Table 5: Exclusion Restriction 1: Distance and Persecution in Reduced Form

Panel A. (1) (2) (3) (4) (5) (6) (7) (8) (9) Reduced Form Distance Trier Distance Emmerich Minimal distance

Log Distance 0.014 0.005 0.002 0.017 0.009 0.009 0.017 0.009 0.008 0.006** 0.006 0.006 0.007** 0.007 0.007 0.005*** 0.06 0.006

P.C. Persecution -0.011 -0.011 -0.011 0.004** 0.004** 0.004**

Pogrom 1349 -0.019 -0.018 -0.019 0.009** 0.009** 0.009**

Historical contr. X X X X X X X X X County controls X X X X X X X X X Individual controls X X X X X X X X X

Observations 13870 13426 13679 13870 13426 13679 13870 13426 13679 Adjusted R2 0.08 0.09 0.09 0.08 0.09 0.09 0.08 0.09 0.09 36 Panel B. (1) (2) (3) (4) (5) (6) (7) (8) (9) Placebo Distance French Households: West of the Rhine Valley, no Jewish migrations

Distance Trier Distance Emmerich Minimal distance

Log Distance -0.003 -0.005 -0.004 0.005 -0.003 -0.003 -0.003 -0.005 -0.004 0.013 0.011 0.011 0.028 0.024 0.024 0.013 0.011 0.011

Regional controls X X X X X X X X X Individual controls X X X X X X Income quintiles X X X X X X Education quintiles X X X X X X Size of town X X X

Observations 9692 9383 9383 9692 9383 9383 9692 9383 9383 Adjusted R2 0.01 0.13 0.13 0.01 0.13 0.13 0.01 0.13 0.13 Panel A of Table 5 reports average marginal eﬀects for estimating the following probit speciﬁcation:

0 0 P r(HoldsStocksi|Xi,Kk) = Φ(α + β1 × DistanceRhinek + β2 × P ersecutionJewsk + Xi × δ + Kk × δ + ηt + ik) (3)

where DistanceRhine is the measure of distance from the Rhine Valley indicated above each column, and P ersecutionJews is the principal component of the margins of persecution against Jews in columns (2), (5), and (8), or an indicator for counties where at least a pogrom against the local Jewish community was documented during the Black Death period. Panel A of Table 5 reports the coeﬃcients for regressing a dummy that equals 1 if the household holds stocks on the measure of distance from the Rhine Valley indicated above each column, on the subsample of French households from the Enquete Patrimoine, which live to the west of the Rhine Valley. In both Panels, standard errors are clustered at the level of the county (NUTS 3). Statistical signiﬁcance is reported as follows: *10%, **5%, ***1%. Table 6: Three-stage OLS: Distance from Rhine, Persecution, Stockholdings

(1) (2) (3) (4) (5) (6) (7) (8) (9) First stage: Second stage: Third stage: Existence Jewish Persecution of Jews Holds stocks community in 1049

Dist. Dist. Minimal Dist. Dist. Minimal Dist. Dist. Minimal Trier Emmerich distance Trier Emmerich dist. Trier Emmerich distance Residual Distance -0.136 -0.213 -0.148 Rhine 0.044*** 0.043*** 0.037***

Existence Jewish 1.063 1.508 1.274 community in 1049 0.504** 0.430*** 0.490***

Persecution of Jews -0.063 -0.050 -0.057 0.027** 0.021** 0.024**

Historical controls X X X X X X X X X Regional controls X X X X X X X X X

Observations 270 270 270 270 270 270 270 270 270 Adjusted R2 0.09 0.08 0.06 0.33 0.35 0.25 0.16 0.16 0.16 This Table reports OLS coefficients for the three-stage instrumental variable procedure described in Section 4. In the first stage (columns (1)-(3)), 37 the probability that a Jewish community existed in 1049 in each German county is predicted by the residuals of the distance of the county from the Rhine Valley regressed on all available observables for medieval economic conditions of counties. In the second stage (columns (4)-(6)), the extent of Jewish persecution during the Nazi period is predicted with the predicted probability that a Jewish community existed in a county in 1096. In the third stage (columns (7)-(9)), the ratio of households who own stocks in each county is predicted with the predicted extent of Jewish 3SLS: usingpersecution. purged In alllog stages distance observations with medieval are German controls counties and residuals coefficients and are estimatedthen three with outcomes. OLS. Hubert-White Predicted s.e. are measure reported is below principal each component,coefficient. similar Statistical results significance to be tables is reported for Online as follows: Appendix *10%, **5%, for ***1%. other 3 measures individually (only weird coeff for pogrom 1349) Table 7: Jewish Persecution, Mortgages, and Deposits

(1) (2) (3) (4) (5) (6) (7) (8) Home Owner Has a Mortgage Regularly Saves part Deposits/Assets of Income Local Banks Probit Probit Probit Probit Probit Probit OLS OLS

-0.009 -0.008 -0.007 -0.007 0.006 0.005 -0.015 -0.015 P. C. Persecution Jews 0.010 0.011 0.003** 0.003** 0.007 0.020 0.007** 0.006**

Log Jews 1933 -0.007 -0.011 0.004 0.004 -0.001 -0.001 -0.025 -0.021 0.011 0.012 0.003 0.003 0.009 0.007 0.011** 0.010**

% Catholics 1925 0.002 0.000 -0.000 -0.001 -0.000 -0.000 0.000 -0.000 0.002 0.001 0.001 0.001 0.001 0.001 0.001 0.001

Age 0.000 -0.001 0.011 0.010 -0.001 -0.005 -0.002 0.002 0.002 0.001*** 0.001*** 0.002 0.002*** 0.002

Age^2 /100 -0.000 -0.001 -0.001 -0.001 0.001 0.001 0.001 0.001 0.000*** 0.000*** 0.000*** 0.000*** Female 0.019 0.019 -0.005 -0.004 -0.034 -0.032 0.005 0.010* 0.010* 0.004 0.004 0.009*** 0.008*** 0.044 Single 0.153 0.147 -0.001 0.006 0.077 0.092 -0.035 0.013*** 0.013*** 0.006 0.007 0.017*** 0.018*** 0.072 College 0.036 0.033 -0.001 0.003 -0.016 -0.012 0.230 0.006*** 0.005*** 0.002 0.002 0.005*** 0.005** 0.177 Eastern Germany 0.068 0.065 -0.019 -0.019 0.005 0.005 0.079 0.049 0.048 0.011* 0.010* 0.030 0.028 0.020*** Income p.c. 2005 -0.004 -0.003 0.001 0.002 0.003 0.001 -0.001 0.004 0.004 0.001 0.001 0.003 0.003 0.007 % College graduates 2005 0.001 -0.003 0.001 0.001 0.001 0.001 0.03 0.003 0.001 0.001 0.002 0.002

Wealth deciles X X X X X X Other historical controls X X X X Wave groups f.e. X X X Regional controls X X X X Observations 11,484 11,484 11,484 11,484 10,900 10,900 236 236 N. of clusters 236 236 236 236 236 236 (Pseudo-) R2 0.05 0.06 0.16 0.18 0.09 0.11 0.07 0.19 Columns (1)-(6) of this Table repors average marginal eﬀects computed after estimating the following probit speﬁcation:

0 0 P r(Depvarik|Xik,Kik) = Φ(α + β × P ersecutionk + Xik × γ + Kik × δ + W ealth deciles + ηt + ik), where the main covariate of interest, P ersecution, is the first principal component of the measures of Jewish persecution in the past described in Section 2 of the paper. Xik includes the following individual-level controls: gender, single/marital status, income nd nd (2 degree polynomial); age (2 degree polynomial); college education; and life and social insurance. Kk includes the following county-level current and historical controls: latitude; income per capita; share of college-educated population; index of quality of cultivable land; log of population in 1933; log of Jewish population in 1933; ratio of population employed in the retail sector in 1933; ratio of population employed in manufacturing in 1933; and ratio of Catholic population in 1925. W ealth deciles are dummies indicating the decile of the wealth distribution to which the household belongs, and Φ is the standard normal cdf. W ealth deciles are dummies indicating the decile of the wealth distribution to which the household belongs, and Φ is the standard normal cdf. The outcome variable, Depvarik is a dummy that equals 1 if the household owns any real estate property in columns (1)-(2), a dummy that equals 1 if the household has a mortgage outstanding, or has had a mortgage in the past, in columns (3)-(4), and a dummy that equals 1 if the household declares they save part of their monthly income regularly in columns (5)-(6). The outcome variables and the household-level demographics are from the first wave of the Panel of Household Finances of the German Bundesbank. Columns (7)-(8) report the results for a county-level OLS regression of the ratio of total deposits in the branches of banks in each county on the sum of the assets of all branches of banks in the county. We aggregate the branch level deposits and assets from Bankscope at the county level. S.e. are clustered at the county level. Statistical significance is reported as follows: *10%, **5%, ***1%.

38 Table 8: Non-sensitivity of the Eﬀect to Education Levels and Cohorts

(1) (2) (3) (4) (5) (6) (7) (8) P.C. Measures of Log deported Jews Ratio deported Jews Pogrom Persecution (extensive margin) (intensive margin) 1349

Panel A. Effect by level of education of investors Persecution -0.017 -0.012 -0.015 -0.008 -0.011 -0.011 -0.017 -0.019 0.007** 0.007 0.005*** 0.006 0.002*** 0.006 0.010* 0.011 Persecution*College 0.001 0.003 -0.003 0.007 0.000 0.000 0.004 0.003 0.004 0.008 0.010 0.016 0.002 0.003 0.005 0.009 Persecution*High School -0.005 -0.015 0.000 0.001 0.011 0.015 0.008 0.011 College 0.011 0.015 0.014 0.009 0.011 0.015 0.009 0.013 0.004*** 0.006** 0.009 0.013 0.004** 0.007** 0.005* 0.008 High School -0.012 0.012 -0.012 -0.013 0.012 0.024 0.014 0.017 Panel B. Effect by cohorts of investors Persecution -0.022 -0.026 -0.019 -0.022 -0.011 -0.006 -0.025 -0.030 0.007*** 0.008*** 0.006*** 0.007*** 0.003*** 0.003* 0.010** 0.011***

Persecution*Born pre 1945 0.010 0.014 0.004 0.005 -0.000 -0.000 0.015 0.018 0.006* 0.007* 0.003 0.003 0.001 0.001 0.013 0.014

Persecution*Born post 1965 0.008 0.003 -0.001 0.014 0.009 0.004 0.001 0.015

Born pre 1945 0.040 0.043 0.017 0.013 0.037 0.044 0.028 0.029 0.011*** 0.011*** 0.018 0.020 0.012*** 0.013*** 0.014** 0.013**

Born post 1965 -0.031 -0.047 -0.017 -0.047 0.012*** 0.020** 0.013 0.013***

Individual, historical controls X X X X X X X X Wave groups f.e. X X X X X X X X Current regional controls X X X X X X X X reports average marginal eﬀects computed after estimating the following probit speﬁcation:

0 0 P r(HoldsStocksik|Xik,Kik) = Φ(α + β × P ersecutionk + Xik × γ + Kik × δ + W ealth deciles + ηt + ik),

Each observation is a German household interviewed by the SOEP for the ﬁrst time between 1984 and 2010. In all columns, the dependent variable is a dummy equal to 1 if the household holds stocks. The main covariate of interest, P ersecution, is the measure of Jewish persecution described at the top of each column. In Panel A, P ersecution is interacted with dummies that equal 1 if the household head holds a college degree or a high school degree. In Panel B, P ersecution is interacted with dummy variables for two groups of households: those in the top third and those in the bottom third of the distribution based on the county-level share of college graduates. S.e. are clustered at the county level (Kreis). Statistical signiﬁcance is reported as follows: *10%, **5%, ***1%.

39 Table 9: Persecution, Distrust in Finance, and Financial Decisions

(1) (2) (3) (4) (5) (6) Panel A Trust the Stock Trust Commercial Trust Local Market Banks Banks Probit Probit Probit Probit Probit Probit

P.C. Persecution of Jews -0.047 -0.041 -0.001 0.001 -0.083 -0.082 0.020** 0.018** 0.021 0.021 0.025*** 0.026*** Generalized Trust 0.029 0.041 0.066 0.078 0.100 0.102 0.035 0.032 0.039* 0.039* 0.047** 0.047**

Risk Tolerance 0.089 0.078 0.024 0.018 0.004 0.002 0.014*** 0.011*** 0.015* 0.014 0.020 0.020

Eastern Germany -0.059 -0.061 -0.070 -0.049 0.047 0.058 0.069 0.062 0.040* 0.039 0.075 0.081

Age group f.e. X X X Gender f.e. X X X Education group f.e. X X X

Observations 495 495 495 495 495 495 N. of clusters 57 57 57 57 57 57 Pseudo-R2 0.21 0.26 0.03 0.06 0.03 0.04 (1) (2) (3) (4) (5) (6) Panel B Holds Stocks Relationship with Relationship with Commercial Bank Local Bank Probit Probit Probit Probit Probit Probit

Trust the Stock Market 0.224 0.230 0.032 -0.018 -0.093 -0.076 0.065*** 0.062*** 0.070 0.068 0.055* 0.057 Trust Commercial Banks 0.070 0.060 0.260 0.279 -0.140 -0.145 0.060 0.061 0.068*** 0.065*** 0.051*** 0.049***

Trust Local Banks 0.014 0.011 -0.039 -0.049 0.203 0.211 0.049 0.048 0.055 0.051 0.048*** 0.045***

Generalized Trust -0.011 -0.005 -0.089 -0.090 0.096 0.097 0.044 0.042 0.046* 0.043** 0.040*** 0.038**

Risk Tolerance 0.099 0.099 0.041 0.034 0.018 0.014 0.013*** 0.012*** 0.019** 0.015** 0.013 0.012

Eastern Germany -0.124 -0.133 0.207 0.172 0.020 0.014 0.056** 0.057 0.073*** 0.065*** 0.051 0.047

Age group f.e. X X X Gender f.e. X X X Education group f.e. X X X

Observations 488 488 490 490 490 490 N. of clusters 57 57 57 57 57 57 Pseudo-R2 0.22 0.24 0.06 0.10 0.08 0.12 This Table repors average marginal eﬀects computed after estimating the following probit speﬁcation:

0 P r(Depvarik|Xik,Kik) = Φ(α + β × P ersecutionk + Xik × γ + ik), where the main covariate of interest, P ersecution, is the first principal component of the measures of Jewish persecution in the past described in Section 2 of the paper. Xik includes the following individual-level controls: gender, age-group fixed effects (15-28, 29-35, 36-45, 46-60, 61+); education-group fixed effects (Hauptschule, Realschule, Abitur, Hochschulabschluss); respondent’s elicited generalized trust; respondent’s elicited financial risk tolerance; and a dummy that equals 1 if the respondent resides in Eastern Germany.Φ is the standard normal cdf. The outcome variables, Depvarik are dummy variables indicated on top of each column. The outcome variables and the household-level demographics are from our own survey of present-day Germans run by the company Clickworker, as described in Section 6 of the paper. S.e. are clustered at the county level. Statistical significance is reported as follows: *10%, **5%, ***1%.

40 Online Appendix to The Long Shadow of Jewish Persecution on Financial Decisions

41 Anti-Jewish Ideology, Nazi Votes, and Stockholdings. Incentives unrelated to anti-Jewish sentiment may have driven Jewish persecution. For instance, individuals and political leaders may have hoped to seize Jewish property if they took part or promoted the attacks against Jews. We therefore study the association between historical anti-Jewish sentiment and current stockholdings using a source of variation of ideological anti-Jewish sentiment different from persecution, that is, voting. Unlike persecution, voting is unobservable and not verifiable. Voting choices were unlikely to raise the expectation of rewards by the Nazis before the start of the Third Reich. The Nazis rose to power during a long and deep economic crisis: hyperinflation was a major concern, and unemployment plagued several counties. Many voters have supported the Nazi party (which had not been in power before 1933) in the hope of improving their economic conditions. de Bromhead et al.(2013) show that persistently depressed economic conditions are a strong predictor of the electoral support of right-wing, anti-system parties in Europe in the 1930s. Thus, voting for the Nazi party in 1933 should be a valid proxy for anti-Jewish sentiment in areas where unemployment in 1933 was low. The economic crisis was less severe in those areas, and votes for the Nazis were more likely to capture the local support for their ideological platform. By contrast, votes for the Nazis should be a noisier proxy for anti-Jewish sentiment in counties where unemployment was high, that is, where voters were likely driven by economic motives when voting for the Nazis. This argument does not imply the Nazis had higher support, on average, in counties with higher unemployment,38 but it exploits the different motives within the group of Nazi voters in 1933. If indeed past anti-Jewish sentiment reduces stock market participation, we would expect a negative association between Nazi vote shares in 1933 and present-day stockholdings in counties where unemployment was low in 1933, and a less negative effect in other counties. We find exactly this pattern. In Figure 5, the left vertical axis reports the average marginal effect of the Nazi vote share in 1933 elections on households’ stockholdings, and is associated with the histograms. The horizontal axis indicates the percentile of the distribution of counties by the unemployment rate in 1933. We sort counties in cumulative percentiles of this distribution. For instance, the histogram labeled “20”

38In fact, King et al.(2008) ﬁnd that voters hit by the economic crisis but without a high risk of unemployment supported the Nazi party in 1933.

42 reports the average marginal effect for households in counties in which the unemployment rate in 1933 was below the 20th percentile. The right vertical axis reports the standard errors attached to the marginal effects, which are clustered at the county level. They are associated with the black line. The average marginal effect of Nazi vote shares in 1933 is negative in counties with low unemployment in 1933, up to the 45th percentile of the distribution. The magnitude of the effect and its statistical significance decrease up to the 80th percentile. The effect becomes economically and statistically insignificant once we add households in counties above the 85th percentile of the distribution of unemployment in 1933. In Figure 6 of the Online Appendix, we find a virtually identical pattern as in Figure 5 for the September 1930 elections. These elections were held during the economic crisis but in a time when the Nazis had no control of the German mass media, such as the radio (Adena et al., 2015). We also propose placebo analyses to corroborate our interpretation of the evidence. Figure 6 of the Online Appendix also shows that the pattern for Nazi votes is not replicated for the Social-Democrat votes or the Communist votes. Whereas the economic motives to vote for these parties might be similar as for the Nazis, these parties should have not attracted antisemitists more than the Nazis. In Panels (d), (e), and (f), we show that the Nazi votes are uncorrelated with the likelihood that households invest in life insurance products, that the household head is a woman, or with the households’ income.39

39In unreported results, we also ﬁnd no signiﬁcant associations if using the age, the education level, or the homeownership status of the household head as alternative outcomes.

43 Figure 5: Nazi Votes, Economic Crisis, and Stock Market Participation

Percentile county-level unemployment 1933 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 0.00 0.00

‐0.05 0.05

‐0.10 0.10

‐0.15 0.15

‐0.20 0.20 Standard Errors

44 ‐0.25 0.25

‐0.30 0.30

‐0.35 0.35 Effect of 1933 Nazi Votes on Stockholdings Votes Effect of 1933 Nazi

‐0.40 0.40

Figure 5 plots the average marginal eﬀects of the variable V oteShareNazi1933 computed after after estimating the following probit speciﬁcation in subsamples of households sorted by the unemployemnt rate in their county of residence as of 1933:

0 0 P r(HoldsStocksik|Xi,Kik) = Φ(α + β × V oteShareNazi1933k + Xi × γ + Kik × δ + ηt + ik) Each observation is a German household interviewed by the SOEP any time between 1984 and 2011. The left vertical axis reports the average marginal effect of V oteShareNazi1933, and it is associated with the histograms. The horizontal axis indicates the percentile of the distribution of counties by the unemployment rate in 1933 below which the estimation is performed. For instance, the histogram labeled ”20” reports the average marginal effect for estimating the probit model only for households who live in counties were the unemployment rate in 1933 was below the 20th percentile; the histogram labeled ”30” reports the average marginal effect for estimating the probit model only for households who live in counties were the unemployment rate in 1933 was below the 30th percentile. The right vertical axis reports standard errors attached to each marginal effect and clustered at the county level. They are associated with the black line. Dark brown histograms are marginal effects that are significant at the 1% level or lower; orange histograms at the 5% level; white histograms are not significant at any conventional level. Figure 6: Nazi Votes and Placebo Outcomes

Nazi Votes in 1930 SPD Votes in 1933 KPD Votes in 1933

0.05 -0.05 0.4 0.4 0 0 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 -0.05 0.05 0.3 0.3 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 -0.1 0.1 0 0

-0.15 0.15 0.2 0.2 -0.2 0.2 -0.2 0.2

-0.25 0.25 -0.4 0.4 Stamdard errors 0.1 0.1 -0.3 0.3 -0.6 0.6 -0.35 0.35 0 0 -0.8 0.8 -0.4 0.4 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 M.E. Perc. Votes Nazi 30 on current M.E. Perc. Votes stockholdings 30 Nazi

-0.45 0.45 -1 1 Percentile unemployment Kreis 1930 -0.1 -0.1

Female Household Head Household Income Invest in Life Insurance

6000 ‐6000 45 0 0 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 4000 ‐4000 0.25 0.25

2000 ‐2000 ‐0.05 0.05 0.2 0.2

0 0 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 0.15 0.15 ‐0.1 0.1 ‐2000 2000 0.1 0.1

‐0.15 0.15 ‐4000 4000

0.05 0.05 ‐6000 6000 ‐0.2 0.2 0 0 ‐8000 8000 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100

‐0.25 0.25 ‐10000 10000 ‐0.05 ‐0.05

Figure 6 plots the coeﬃcent on the variable V oteShareNazi1933 computed after after estimating the following OLS speciﬁcations in subsamples of households sorted by the unemployemnt rate in their county of residence as of 1933:

0 0 DepV arik = α + β × V oteShareNazi1933k + Xi × γ + Kik × δ + ηt + ik Each observation is a German household interviewed by the SOEP any time between 1984 and 2011. DepV ar is indicated on top of each graph. In each graph, the left vertical axis reports the OLS coefficient on V oteShareNazi1933, and it is associated with the histograms. The horizontal axis indicates the percentile of the distribution of counties by the unemployment rate in 1933 below which the estimation is performed. For instance, the histogram labeled ”20” reports the OLS coefficient for estimating the OLS specification only for households who live in counties were the unemployment rate in 1933 was below the 20th percentile; the histogram labeled ”30” reports the average marginal effect for estimating the OLS specification only for households who live in counties were the unemployment rate in 1933 was below the 30th percentile. The right vertical axis reports standard errors attached to each coefficient and clustered at the county level. They are associated with the black line. Dark brown histograms are marginal effects that are significant at the 1% level or lower; orange histograms at the 5% level; white histograms are not significant at any conventional level. Figure 7: Densities of County-level Persecution across German States

(a) Out of Rhine Valley towards North .5 .4 .3 .2 .1 Density P.C. Anti-Jewish Violence 0

-4 -2 0 2 4

Nordrhein-Westfalen Niedersachsen Schleswig-Holstein (b) Out of Rhine Valley towards South .4 .3 .2 .1 Density P.C. Anti-Jewish Violence 0

-3 -1 1 3

Nordrhein-Westfalen Hessen Baden-Wuerttemberg Bayern (c) Farthest Distance from Rhine Valley in Both Directions .5 .4 .3 .2 .1 Density P.C. Anti-Jewish Violence 0

-3 -1 1 3

Nordrhein-Westfalen Bayern Brandenburg

46 Figure 8: Jewish Pesecution and Stockholdings - Raw Data

(a) Principal Component Jewish Persecution (b) Logarithm Jews Deported .6 .6 .4 .4 .2 .2 0 0

-4 -2 0 2 4 Ratio Households who Own Stocks (average 1984-2010) 0 2 4 6 8 10 Ratio of Households who Own Stocks (average 1984-2010) Principal Component Measures of Jewish Persecution Log Total Jewish Deportations (c) Ratio Jews Deported (d) Pogrom in 1349 .18 .6 .17 .4 .16 .2 .15 0 .14

Ratio Households who Own Stocks (average 1984-2010) 0 20 40 60 80 100 No Pogroms in 1349 At Least one Pogrom in 1349

Ratio Deported Jewish Population (base: 1933) Ratio of Households who Own Stocks (average 1984-2010)

47 Figure 9: Virtual States

56 (a) 9 Virtual States 54 52 LAT 50 48

6 8 10 12 14 LON

56 (c) 16 Virtual States 54 52 LAT 50 48

6 8 10 12 14 LON

48 Figure 10: Supply-side Channel 1: Jewish Persecution and the Diﬀusion of Credit Unions

(a) Diﬀusion Path of Credit Unions across Space and Over Time Until 1860 Until 1875 54 54 52 52 Latitude Latitude 50 50 48 48

6 8 10 12 14 16 6 8 10 12 14 16 Longitude Longitude Until 1900 Up to the present 54 54 52 52 Latitude Latitude 50 50 48 48

6 8 10 12 14 16 6 8 10 12 14 16 Longitude Longitude (b) Jewish Persecution and the Foundation of Credit Unions 1950 1900 1850 Year First VB-Raiffeisenbank founded in Kreis

0 20 40 60 80 100 Fraction Deported Jewish Population (base: 1933)

Panel (a) of Figure 10 plots the Germans counties where credit unions were first introduced at different points in time. Panel (b) of Figure 10 plots the correlation between the year when the first credit union of a county was founded and the ratio of Jews deported during the Nazi period over the total Jewish population in 1933. Each point is a German county. We exclude the eastern German counties where credit unions where founded after the fall of the Berlin Wall, because their year of foundation is affected by the ban on private banks within the German Democratic Republic.

49 Figure 11: Supply-side Channel 2: Jews in Finance and Present-day Stockholdings

(a) Ratio of Jews in Finance in 1882 and Present-day Stockholdings .06 .04 .02 0 -.02 Likelihood Household Holds Stocks (residuals) -.04 10 20 30 40 50 60 70 80 90 100 Ratio Jews in Finance 1882 (county)

(b) Ratio of Jews in 1933 and Present-day Stockholdings .08 .04 0 -.04 Likelihood Household Holds Stocks (residual) -.08 10 20 30 40 50 60 70 80 90 100 Ratio Jews over Population 1933 (county)

Panel (a) and Panel (b) of Figure 11 plot the average residuals from regressing a dummy that equals 1 if the household holds stocks on the controls of Equation 1, across the deciles of the distribution of the share of ﬁnancial employees of the Jewish religion in a county as of 1882 and of the ratio of Jews over the total German population in a county as of 1933. Intervals represent 95% conﬁdence intervals for the estimated averages.

Table 10: Jewish Population and Geman Sectors in 1882

Jews Total Jews/Total Total sector/Overall sector sector (%) Workers (%)

Total working population 357546 11037320 3.239 100 Unemployed 2.598 6.116

Agriculture, Manifacturing, Trade Agriculture 1641 4625893 0.035 41.911 Forestry and Fishing 65 66455 0.098 0.602 Mining and metal transf. 1255 866794 0.145 7.853 Chemical 266 28908 0.920 0.262 Textile 1724 385565 0.447 3.493 Food and Beverage 9239 363837 2.539 3.296 Building and Construction 1312 533925 0.246 4.837 Printing 555 35352 1.567 0.320 Retail trade 70175 466249 15.051 4.224 Bookshops, art dealers 485 9580 5.063 0.087

Services Engineering Services 591 146650 0.403 1.329 Health Care 1108 40883 2.710 0.370 Transportation 421 128136 0.329 1.161 Hotels and restaurants 3654 147061 2.485 1.332 Household services 692 278927 0.248 2.527 Military 918 25860 0.355 2.343 Administration 1093 119140 0.917 1.079 Finance 3042 13324 22.986 0.120 Insurance 223 6655 3.351 0.060

51 Table 11: Jewish Persecution and Stock Market Participation - Other Measures of Persecution

(1) (2) (3) (4) (5) (6) (7) (8) (9) All Counties Only if Jewish Community in the Middle Ages

Log Letters Attacks P.C. P.C. Log Letters Attacks P.C. P.C. Stuermer Synagogue VV augment Stuermer Synagogue VV augment

Persecution of Jews -0.009 -0.006 -0.018 -0.009 -0.021 -0.006 -0.007 -0.012 -0.015 (column heading) 0.003*** 0.003** 0.013 0.004** 0.006*** 0.003** 0.021 0.005** 0.007** Log Jews 1933 -0.001 0.001 -0.002 -0.000 0.004 -0.011 -0.017 -0.014 -0.011 0.001 0.004 0.005 0.003 0.005 0.010 0.010* 0.010 0.011

% Catholics 1925 0.029 0.002 -0.005 0.002 0.000 0.001 0.001 0.002 0.002 0.012** 0.010 0.011 0.011 0.010 0.013 0.002 0.001 0.001

Age -0.002 -0.003 -0.002 -0.002 -0.003 -0.003 -0.003 -0.002 0.001** 0.001** 0.001** 0.001** 0.001* 0.001* 0.001* 0.001*

Age^2 /100 0.004 0.004 0.004 0.004 0.004 0.004 0.004 0.004 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** Female -0.004 -0.004 -0.004 -0.002 0.006 0.006 0.006 0.007 0.006 0.006 0.006 0.006 0.007 0.007 0.007 0.008 Single 0.052 0.048 0.052 0.049 0.057 0.057 0.056 0.058 0.010*** 0.009*** 0.009*** 0.009*** 0.010*** 0.010*** 0.010*** 0.010*** 52 College 0.008 0.009 0.008 0.008 0.012 0.012 0.012 0.010 0.003** 0.003*** 0.003** 0.004** 0.004*** 0.004*** 0.004*** 0.004** Eastern Germany -0.039 -0.032 -0.038 -0.025 0.005 0.001 0.001 0.006 0.021* 0.019* 0.021* 0.021 0.030 0.029 0.029 0.033 Income p.c. 2005 0.039 0.0046 0.043 0.042 0.061 0.059 0.064 0.061 0.018** 0.018*** 0.017** 0.020** 0.026** 0.026** 0.026** 0.033* % College grad. 2005 0.001 0.009 0.001 0.000 -0.002 -0.002 -0.002 -0.002 0.001 0.013 0.001 0.001 0.002 0.002 0.002 0.002

Income deciles X X X X X X X X X Other historical controls X X X X X X X X X Wave groups f.e. X X X X X X X X Regional controls X X X X X X X X Observations 17,701 13,870 13,870 13,870 13,870 9884 9884 9884 9884 N. of clusters 296 270 270 270 270 190 190 190 190 Pseudo-R2 0.01 0.09 0.08 0.09 0.09 0.09 0.09 0.09 0.09 This Table reports average marginal eﬀects computed after estimating the following probit speﬁcation:

0 0 P r(HoldsStocksik|Xik,Kik) = Φ(α + β × P ersecutionk + Xik × γ + Kik × δ + Income deciles + ηt + ik), Each observation is a German household interviewed by the SOEP any time between 1984 and 2011. In all columns, the dependent variable is a dummy which equals one if the household holds stocks. The main covariate of interest, P ersecution, is the measure of persecution of Jews in the past described at the top of each column. nd nd Xik includes the following individual-level controls: gender, single/marital status, income (2 degree polynomial); age (2 degree polynomial); college education; homeownership; and life and social insurance. Kk includes the following county-level current and historical controls: latitude; income per capita; share of college-educated population; index of quality of cultivable land; log of population in 1933; log of Jewish population in 1933; ratio of population employed in the retail sector in 1933; ratio of population employed in manufacturing in 1933; and ratio of Catholic population in 1925. Income deciles are dummies indicating the decile of the income distribution to which the household belongs, and Φ is the standard normal cdf. S.e. are clustered at the county level. Statistical signiﬁcance is reported as follows: *10%, **5%, ***1%. Table 12: Jewish Persecution and Stock Market Participation- Excluding Outlier Counties

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Principal Component Log Deported Jews Ratio Deported Jews Pogrom 1349 Measures of Persecution (extensive margin) (intensive margin)

Probit Probit OLS Probit OLS Probit OLS Probit OLS

Persecution of Jews -0.013 -0.014 -0.014 -0.015 -0.016 -0.054 -0.053 -0.009 -0.010 (column heading) 0.005*** 0.005*** 0.005*** 0.006** 0.006*** 0.032* 0.031* 0.005** 0.005**

Log Jews 1933 0.032 0.001 0.036 0.008 0.007 -0.022 -0.011 -0.021 -0.024 0.046 0.057 0.056 0.042 0.041 0.061 0.060 0.040 0.041

% Catholics 1925 0.029 0.007 0.005 0.001 -0.008 0.004 0.002 0.002 -0.005 0.013** 0.012 0.017 0.011 0.012 0.012 0.013 0.011 0.011

Age -0.002 -0.002 -0.002 -0.003 -0.002 -0.002 -0.002 -0.003 0.001* 0.000** 0.001** 0.001** 0.001* 0.001** 0.001** 0.001**

Age^2 /100 0.004 0.004 0.004 0.004 0.004 0.004 0.005 0.004 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** Female -0.020 -0.022 -0.031 -0.032 -0.019 -0.021 -0.032 -0.032 0.060 0.062 0.061 0.061 0.062 0.061 0.062 0.061 Single 0.054 0.051 0.053 0.050 0.054 0.051 0.054 0.051

53 0.010*** 0.009*** 0.010*** 0.009*** 0.010*** 0.009*** 0.010*** 0.009*** College 0.056 0.061 0.055 0.057 0.061 0.083 0.056 0.087 0.019*** 0.028** 0.020*** 0.029*** 0.020*** 0.031*** 0.021*** 0.030*** Eastern Germany -0.034 -0.028 -0.042 -0.036 -0.035 -0.029 -0.033 -0.025 0.022 0.020 0.022* 0.019* 0.025 0.020 0.021 0.018 Income p.c. 2005 0.038 0.043 0.040 0.045 0.041 0.046 0.033 0.038 0.016** 0.017*** 0.017** 0.017** 0.017** 0.018** 0.017* 0.017** % College graduates 2005 0.116 0.100 0.164 0.149 0.117 0.121 0.093 0.068 0.143 0.143 0.141 0.141 0.146 0.151 0.128 0.126

Income deciles X X X X X X X X X Other historical controls X X X X X X X X X Wave groups f.e. X X X X X X X X Regional controls X X X X X X X X Observations 17,350 13,591 13,591 13,862 13,862 13,591 13,591 13,862 13,862 N. of clusters 285 259 259 268 268 259 259 268 268 (Pseudo-) R2 0.01 0.10 0.08 0.10 0.08 0.10 0.08 0.10 0.09 This Table reports average marginal eﬀects computed after estimating the following probit speﬁcation:

0 0 P r(HoldsStocksik|Xik,Kik) = Φ(α + β × P ersecutionk + Xik × γ + Kik × δ + Income deciles + ηt + ik), Each observation is a German household interviewed by the SOEP any time between 1984 and 2011. In all columns, the dependent variable is a dummy which equals one if the household holds stocks. The main covariate of interest, P ersecution, is the measure of persecution of Jews in the past described at the top of each column. nd nd Xik includes the following individual-level controls: gender, single/marital status, income (2 degree polynomial); age (2 degree polynomial); college education; homeownership; and life and social insurance. Kk includes the following county-level current and historical controls: latitude; income per capita; share of college-educated population; index of quality of cultivable land; log of population in 1933; log of Jewish population in 1933; and ratio of Catholic population in 1925. Income deciles are dummies indicating the decile of the income distribution to which the household belongs, and Φ is the standard normal cdf. S.e. are clustered at the county level. Statistical signiﬁcance is reported as follows: *10%, **5%, ***1%.

Table 13: Exclusion Restriction 2: Distance Residuals and Observables

Dependent Variable (1) (2) (3) (4) (5) (6) Distance Trier Distance Emmerich Minimal distance

Log County Income p.c. 0.009 0.011 -0.007 -0.008 0.006 0.009 0.008 0.009 0.011 0.013 0.007 0.008

% College graduates -0.096 -0.224 0.718 0.795 0.230 0.147 0.293 0.303 0.341** 0.439* 0.250 0.276

Land Quality Index -0.003 -0.003 -0.005 -0.005 -0.003 -0.003 0.017 0.020 0.014 0.019 0.014 0.016

Female 0.001 -0.006 0.006 -0.003 0.004 -0.002 0.012 0.012 0.011 0.014 0.010 0.010

Single 0.009 0.000 0.015 0.008 0.009 0.002 0.012 0.014 0.010 0.014 0.010 0.011

Home owner -0.006 -0.012 0.016 -0.000 -0.002 -0.013 0.020 0.023 0.020 0.025 0.017 0.019

Log Income 0.006 0.019 -0.000 0.010 0.007 0.019 0.023 0.025 0.019 0.024 0.018 0.021

Age -0.401 -0.124 -0.630 -0.228 -0.434 -0.179 0.634 0.787 0.463 0.717 0.501 0.661

High School or higher -0.008 -0.005 -0.004 0.000 -0.007 -0.005 0.009 0.009 0.009 0.011 0.007 0.008

Eastern Germany f.e. X X X X X X Res. medieval X X X Res. medieval and 1930s X X X

Table 13 documents the association between the measures of distance from the Rhine Valley and current county- and individual-level observables. The measures of distance are the estimated residuals from the following OLS speciﬁcation:

0 DistanceRhinek = α + Kk × δ + k (4)

where DistanceRhine is one of three measures indicated above each column, and K is a set of observables capturing the economic characteristics of counties in the past. In odd columns, K includes all observable proxies for economic growth in the Middle Ages (whether the county hosted a monthly market, a bishop siege, any free imperial city, whether it was part of the Hanseatic League, whether it was incorporated before 1349) and geographic characteristics (latitude, the land quality index of Ramankutty et al.(2002), whether a navigable river existed in the county, whether the county is in eastern Germany). In even columns, K additionally includes the log of Jewish inhabitants as of 1933 and a set of socio-demographics of counties in 1933 (unemployment rate, ratio of workers in blue-collar jobs, in self-employment, ratio of Roman Catholics). Each line reports the estimated β in the following OLS speciﬁcation:

0 Depvarik = α + β × ˆk + Cik × δ + ik

where Depvar is a county- or individual-level observable enlisted in the left column for each line, ˆk are the residuals from Equation 4 and C is a set current individual and county-level observables. S.e. are clustered at the county level (hence equivalent to Huber-White heteroskedasticity-robust s.e. in speciﬁcations at the county level). Statistical signiﬁcance is reported as follows: *10%, **5%, ***1%.

54 Table 14: Eﬀect of Jewish Persecution on Stockholdings Over Time

(1) (2) (3) (4) (5) (6) (7) (8) Panel A. P.C. Measures Log deported Jews Ratio deported Jews Pogrom Effect over time of Persecution (extensive margin) (intensive margin) 1349

Probit OLS Probit OLS Probit OLS Probit OLS

Persecution -0.016 -0.015 -0.005 -0.006 -0.047 -0.044 -0.025 -0.024 0.006*** 0.006*** 0.003* 0.003** 0.025* 0.022** 0.013* 0.013* Persecution*1984-1987 0.017 0.014 0.005 0.004 0.102 0.101 0.003 0.001 0.013 0.011 0.007 0.006 0.057* 0.053* 0.025 0.021 Persecution*1988-1991 0.001 -0.001 -0.003 -0.004 0.028 0.024 0.002 0.001 0.015 0.017 0.008 0.008 0.082 0.089 0.034 0.038 Persecution*1992-1995 0.004 0.003 -0.002 -0.002 0.040 0.042 0.006 0.004 0.013 0.014 0.006 0.007 0.059 0.064 0.029 0.033 Persecution*1996-1999 -0.004 -0.008 0.001 0.001 -0.034 -0.059 0.010 0.019 0.014 0.017 0.008 0.009 0.088 0.098 0.029 0.033

Persecution*2000-2003 0.006 0.004 -0.003 -0.004 0.018 0.001 0.023 0.029 0.007 0.009 0.003 0.004 0.042 0.049 0.019 0.022

Persecution*2004-2007 0.008 0.007 0.001 0.001 0.009 0.010 0.013 0.009 0.008 0.007 0.004 0.003 0.047 0.040 0.018 0.017

Individual, historical controls X X X X X X X X Wave groups f.e. X X X X X X X X Current regional controls X X X X X X X X

Observations 13,870 13,870 13,870 13,870 13,870 13,870 13,870 13,870 N. of clusters 270 270 270 270 270 270 270 270 (Pseudo-) R2 0.10 0.08 0.10 0.08 0.10 0.08 0.10 0.08

55 Table 15: All Outcomes for Households in the PHF Sample

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Saves Has Homeo Entrepre Has Has Comm HH is a Has Credit Trusts Cash/ Mortgage wner neur Gov.bonds odities creditor Card Others Wealth Probit OLS Probit Probit Probit Probit Probit Probit Probit OLS 0.022 -0.041 -0.029 -0.017 -0.007 -0.013 0.038 -0.031 -0.026 0.035 Persecution Jews 0.032 0.021** 0.027 0.025 0.005 0.029 0.023* 0.040 0.042 0.021*

Log Jews 1933 -0.017 0.013 -0.028 0.015 0.001 -0.009 -0.011 0.040 0.011 0.004 0.011 0.007* 0.009*** 0.008* 0.002 0.009 0.006* 0.013*** 0.012 0.008

% Catholics 1925 0.015 0.028 -0.021 0.031 0.006 0.003 -0.007 -0.071 -0.034 -0.024 0.049 0.024 0.036 0.035 0.006 0.036 0.031 0.056 0.049 0.029

Age 0.013 0.027 0.020 0.015 0.001 -0.004 0.011 0.021 -0.015 -0.009 0.007* 0.006*** 0.005*** 0.006*** 0.001 0.004 0.005** 0.006*** 0.007** 0.004**

Age^2 /100 -0.001 -0.001 -0.001 -0.001 -0.000 0.000 -0.000 -0.001 0.001 0.001 0.000** 0.001*** 0.000*** 0.000** 0.001 0.001 0.000** 0.000*** 0.000** 0.000** Male -0.006 -0.016 -0.005 0.006 0.011 0.014 0.023 0.040 -0.038 0.018 0.028 0.015 0.022 0.021 0.006* 0.019 0.018 0.028 0.030 0.018

56 Single -0.042 -0.079 -0.038 0.040 0.015 -0.005 0.067 0.005 0.066 0.024 0.049 0.047* 0.038 0.043 0.010 0.041 0.034* 0.044 0.066 0.050 College 0.076 0.008 -0.058 0.043 0.010 0.031 -0.010 0.174 0.040 0.008 0.037** 0.021 0.030* 0.029 0.005* 0.024 0.022 0.031*** 0.041 0.021 Eastern Germany -0.107 -0.061 -0.062 -0.016 0.003 0.030 0.044 0.001 -0.042 0.096 0.047** 0.033* 0.036* 0.059 0.011 0.042 0.036 0.054 0.053 0.035** Financially Literate 0.003 0.017 0.078 0.032 -0.003 -0.036 0.009 0.036 -0.057 -0.073 0.038 0.023 0.028*** 0.024 0.004 0.022* 0.019 0.028 0.038 0.029** Wants to take Risk 0.032 0.034 -0.012 0.080 0.003 0.017 0.028 0.138 0.100 -0.066 0.032 0.019* 0.020 0.023*** 0.006 0.021 0.019 0.025*** 0.029*** 0.026** Goes to church often -0.026 -0.040 -0.039 -0.016 0.005 0.019 -0.023 -0.092 0.064 0.032 0.037 0.029 0.029 0.033 0.006 0.028 0.029 0.043** 0.044 0.030

Historical controls X X X X X X X X X X Regional controls X X X X X X X X X X Wealth deciles X X X X X X X X X X Observations 1,256 1,256 1,256 1,256 1,126 1,256 1,256 1,256 1,254 1,256 N. of clusters 99 99 99 99 98 99 99 99 99 99 Pseudo-R2 0.06 0.16 0.15 0.11 0.17 0.02 0.04 0.21 0.05 0.08 Table 16: Placebo Outcomes

(1) (2) (3) (4) (5) (6) (7) (8) P.C. Measures Pogrom Log distance IV of Persecution 1349 from Emmerich Panel A. Bank branches p.c. Persecution 0.002 0.001 0.002 -0.001 -0.000 0.001 -0.000 -0.004 0.001** 0.001 0.001** 0.001 0.001 0.001 0.001 0.003 Observations 268 267 277 276 277 276 268 267 Panel B. Home ownership Persecution -0.002 -0.006 -0.011 -0.011 -0.014 0.004 0.036 -0.011 0.010 0.008 0.016 0.015 0.013 0.011 0.040 0.033 Observations 17467 14169 17808 14440 17808 14440 17467 14169 Panel C. Life Insurance Persecution 0.006 0.009 -0.007 -0.008 0.008 0.005 -0.021 -0.014 0.007 0.007 0.014 0.014 0.010 0.011 0.034 0.033 Observations 17360 14169 17701 14440 17701 14440 17360 14169 Panel D. Log Income Persecution -0.035 -0.034 0.014 0.046 0.061 0.041 0.198 -0.137 0.037 0.033 0.063 0.063 0.048 0.050 0.154 0.143 Observations 17310 14123 17651 14394 17651 14394 17310 14123 Panel E. Female HH head Persecution -0.007 -0.004 -6*10-5 -0.001 0.026 0.014 -0.066 -0.046 0.005 0.005 0.010 0.010 0.008*** 0.010 0.036* 0.031 Observations 14258 13599 14529 13870 14529 13870 14258 13599 Panel F. High School or higher Persecution 0.001 0.004 0.004 0.009 0.002 -0.001 -0.012 -0.016 0.005 0.006 0.010 0.012 0.007 0.009 0.022 0.091

Observations 17477 12726 17818 12966 17818 12966 17477 12726

Individual, historical contr. X X X X Wave groups f.e. X X X X

Table 17: Keeping the Distance from the Rhine Valley Constant: Within-distance-group analysis

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Principal Component Log deported Jews Ratio deported Jews Pogroms 1349 Measures of Persecution (extensive margin) (intensive margin)

Persecution of Jews -0.011 -0.010 -0.010 -0.006 -0.006 -0.005 -0.027 -0.025 -0.021 -0.018 -0.015 -0.018 (see columns heading) 0.004*** 0.004** 0.004** 0.002*** 0.002*** 0.002*** 0.017 0.017 0.016 0.009** 0.009* 0.009**

Distance group f.e. X X X X X X X X X X X X Number distance groups 5 8 10 5 8 10 5 8 10 5 8 10 Avg. st.dev.distance within groups (Km) 135 127 120 135 127 120 137 127 120 137 127 120 Average n. counties across groups 52 41 31 54 42 33 52 41 31 54 42 33

Individual, historical controls X X X X X X X X X X X X Wave groups f.e. X X X X X X X X X X X X Observations 13,599 13,599 13,599 13,870 13,870 13,870 13,599 13,599 13,599 13,870 13,870 13,870 N. of clusters 261 261 261 270 270 270 261 261 261 270 270 270 (Pseudo-) R2 0.09 0.09 0.10 0.10 0.09 0.09 0.09 0.09 0.10 0.09 0.09 0.10 This Table reports average marginal eﬀects computed after estimating the following probit speﬁcation:

0 0 P r(HoldsStocksik|Xik,Kik) = Φ(α + β × P ersecutionk + Xik × γ + Kik × δ + Income deciles + ηt + ηs + ik), Each observation is a German household interviewed by the SOEP any time between 1984 and 2011. In all columns, the dependent variable is a dummy which equals one if the household holds stocks. The main covariate of interest, P ersecution, is the measure of persecution of Jews in the past described at the top of each column. nd nd Xik includes the following individual-level controls: gender, single/marital status, income (2 degree polynomial); age (2 degree polynomial); college education; homeownership; and life and social insurance. Kk includes the following county-level current and historical controls: latitude; income per capita; share of college-educated

58 population; index of quality of cultivable land; log of population in 1933; log of Jewish population in 1933; and ratio of Catholic population in 1925. Income deciles are dummies indicating the decile of the income distribution to which the household belongs, and Φ is the standard normal cdf. ηs are a set of fixed effects for counties at similar distances from the Rhine Valley when the counties are split into 5, 8, or 10 equal-sized bins based on the distance. S.e. are clustered at the county level. Statistical significance is shown as follows: *10%, **5%, ***1%. We aim to estimate the following system of three structural equations:

0 Existence1096k = α1 + β1Distance purgedk + Kkδ1 + 1k

0 P ersecutionJewsk = α2 + β2Existence1096k + Kkδ2 + 2k

0 RatioHoldStocksk = α3 + β3P ersecutionJewsk + Kkδ3 + 3k

where Existence1096, P ersecutionJews and RatioHoldStocks are endogenous variables, and Distance purged are the estimated residuals from a regression of distance from the Rhine on economic and socio-demographic observables for counties in the past. We obtain conditions for identiﬁcation of the system assuming that Distance purged is an exogenous variable. We solve the system for the implied reduced-form equation:

0 RatioHoldStocksk = α + βDistance purged + Kkδ + k where

α = α3 + β3α2 + β3β2α1

β = β3β2β1

δ = β3β2δ1 + β3δ2 + δ3

k = β3β21 + β32 + 3

The implied reduced-form equation suggests the following identifying condition:

Cov(Distance purged, ) = 0

⇔β3β2Cov(Distance purged, 1) + β3Cov(Distance purged, 2) + Cov(Distance purged, 3) = 0

Since Cov(Distance purged, 1) = 0 by construction, this is satisﬁed by any of the two suﬃcient conditions below:

A) Cov(Distance purged, 2) = Cov(Distance purged, 3) = 0

B) Cov(Distance purged, 3) = −β3Cov(Distance purged, 2).

59 Table 18: Residuals from Structural Equations and Purged Distance

(1) (2) (3) (4) (5) (6) Second stage residuals: Third stage residuals: Persecution of Jews on % Hold stocks on Existence 1096 Persecution of Jews

Dist. Dist. Minimal Dist. Dist. Minimal Trier Emmerich dist. Trier Emmerich dist. P.C. Persecution Jews -0.053 -0.179 -0.092 -0.001 0.006 0.001 0.115 0.103* 0.099 0.009 0.007 0.007

Extensive Margin -0.007 -0.067 -0.046 -0.002 0.009 0.001 0.115 0.106 0.093 0.009 0.007 0.007

Intensive Margin -1.946 -3.620 -2.463 -0.002 0.006 0.001 2.204 2.759 2.140 0.009 0.007 0.007

Pogrom 1349 0.011 -0.060 -0.005 -0.001 0.010 0.002 0.047 0.045 0.041 0.009 0.007 0.007

Historical contr. X X X X X X Regional controls X X X X X X

Table 18 documents the association between the measures of distance from the Rhine Valley and the residuals from the second- and third-stage structural equations of the 3SLS system. The measures of distance are the estimated residuals from the following OLS speciﬁcation: 0 DistanceRhinek = α + Kk × δ + k where DistanceRhine is one of three measures indicated above each column, and K is a set of observables capturing the economic characteristics of counties in the past. Columns (1)-(3) regress the residuals from the following structural equation on the distance measures: 0 P ersecutionJewsk = α2 + β2 × Exist1096k + Kk × δ2 + 2k where P ersecutionJews is the measure of persecution enlisted in the left column for each line, and Exist1096 is an indicator for whether a Jewish community existed in a county as of 1096. Columns (4)-(6) regress the residuals from the following structural equation on the distance measures:

0 RatioHoldStocksk = α3 + β3 × P ersecutionJewsk + Kk × δ3 + 3k

Hubert-White s.e. are reported below each coeﬃcient. Signiﬁcance is reported as follows: *10%, **5%, ***1%.

60 Table 19: Summary Statistics and Correlations - Trust Survey

(a) Summary Statistics

All available observations Non-missing information p-value

Obs. Mean St. dev. Min. Max. Obs. Mean St. dev. Min. Max.

Trust in the Stock Market 981 0.131 0.338 0 1 578 0.149 0.356 0 1 0.060 Trust in Commercial Banks 981 0.170 0.376 0 1 578 0.164 0.371 0 1 0.592 Trust in Local Banks 981 0.406 0.491 0 1 578 0.386 0.487 0 1 0.133 Generalized Trust 1000 0.387 0.487 0 1 586 0.398 0.499 0 1 0.422 Risk Tolerance (1 to 7) 989 3.037 1.439 1 7 580 3.028 1.451 1 7 0.814

(b) Correlational Structure Trust the Stock Trust Commercial Trust Local Generalized Market Banks Banks Trust Trust the Stock Market 1

Trust Commercial Banks 0.2635 1 [0.0000] 576 Trust Local Banks 0.1316 0.2523 1 [0.0016] [0.0000] 576 577 Generalized Trust 0.0926 0.0946 0.1055 1 [0.0261] [0.0230] [0.0133] 577 577 577 Risk Tolerance 0.3829 0.1271 0.0222 0.1609 [0.0000] [0.0022] [0.5944] [0.0001] 578 578 578 579