Inflation and Individual Investors’ Behavior: Evidence from the German Hyperinflation

Abstract

We analyze how individual investors respond to inflation. We introduce a unique data set containing information on local inflation and security portfolios of more than 2,000 clients of a German bank between 1920 and 1924, covering the German hyperinflation. We find that individual investors buy fewer (sell more) stocks when facing higher local inflation. This effect is more pronounced for less sophisticated investors. Moreover, we document a positive relation between local inflation and forgone returns following stock sales. Our findings are consistent with individual investors suffering from money illusion. Alternative explanations, such as consumption needs, are unlikely to drive our results.

Authors have furnished an Internet Appendix, which is available on the Oxford University Press Web site next to the link to the final published paper online

Inflation is among the most important economic risks faced by individual investors. Following the outbreak of the COVID-19 pandemic, inflation resurfaced in many countries.¹ Even though individual investors play an increasingly important role in capital markets, little is known about how they respond to the prospect of higher inflation, and theory provides conflicting hypotheses on this question.² On the one hand, the hedging hypothesis predicts that investors are more likely to buy and less likely to sell stocks when expected inflation increases. This is because investors understand that stocks entitle them to a fraction of the income generated by the underlying real assets, allowing them to preserve the real value of their investments (e.g., Fama and Schwert 1977; Fama 1981; Boudoukh and Richardson 1993; Bekaert and Wang 2010). On the other hand, the money illusion hypothesis suggests that investors are less likely to buy and more likely to sell stocks in periods of higher expected inflation. This is because they adjust nominal interest rates but ignore that firms’ cash flows also grow with inflation, leading them to require higher dividend yields to hold stocks (e.g., Modigliani and Cohn 1979; Ritter and Warr 2002; Cohen, Polk, and Vuolteenaho 2005). Given these two competing hypotheses, understanding how investors react to expected inflation is an empirical question.

A test of individual investors’ response to inflation is subject to three main empirical challenges. First, one needs granular data on investors’ security transactions. This allows for a direct analysis of investment decision-making in inflationary periods. Second, one needs a time period in which inflation, if overlooked, produces sizable financial losses and thus attracts the attention of investors.³ Third, one needs a reliable measure of expected inflation that varies both over time and across investors. This is a necessary condition for a within-person analysis and enables one to control for the overall time trend.

This setup is not available in the most common investor-level data sets. In this paper, we therefore introduce a unique data set containing security portfolios of over 2,000 private clients of a German bank between 1920 and 1924, covering the German hyperinflation. While our sample is smaller compared to contemporaneous data sets, we show that the bank can be considered representative of a large German bank at the time, the clients representative of individual investors in Germany during the early 1920s, and the clients’ investment behavior comparable to the investment behavior of individual investors today. The data and the time period are well-suited to address each of the empirical challenges outlined above. First, we have detailed information on every trade executed by the bank’s clients, allowing for a direct analysis of individuals’ investment behavior. Second, between January 1920 and September 1923, inflation was high, potentially yielding large financial losses if overlooked, and arguably grabbing investors’ attention. Third, we have inflation data at the monthly level for hundreds of towns in Germany, resulting in an inflation measure that captures inflation experienced locally over time, which should be a reliable proxy for expected inflation.⁴

Figure 1 visualizes our main finding. Each month, we sort towns in Germany into deciles based on their local inflation and compute, for each inflation decile, the average buy-sell imbalance for stock trades of clients living in those towns. We then plot average buy-sell imbalances against inflation deciles. The figure shows a strong negative relationship between inflation deciles and investors’ buy-sell imbalances. This suggests that investors buy fewer (sell more) stocks when facing higher local inflation. Moving from the decile with the lowest inflation to the decile with the highest inflation reduces buy-sell imbalances by 17 percentage points.⁵ This result is consistent with the money illusion hypothesis, but inconsistent with the hedging hypothesis.

Fig. 1

Local inflation and stock trades

This figure shows the average monthly buy-sell imbalance for stocks of clients experiencing different local inflation. We focus on the time period from January 1920 to September 1923. Each month, we sort towns into deciles based on their local inflation. We assign clients to the closest town for which we have inflation data within a 25-km radius based on the place of residence. Internet Appendix A describes all variables used throughout the study in detail. The figure shows point estimates together with 99% confidence intervals.

In a more formal analysis, we regress investors’ buy-sell imbalances in stock trades on local inflation, including town-level controls, time fixed effects, and client fixed effects. We find that a 1% increase in local inflation is associated with a significant decline in the buy-sell imbalance for stocks of 3.5%. This regression analysis therefore confirms the negative slope across the bars observed in Figure 1.

We also analyze individual investors’ behavior around a reverse inflation shock. In October 1923, Germany successfully reformed its currency. As inflation declined close to zero within weeks, nominal and real discount rates converged. Hence, we expect that investors subject to money illusion no longer make a valuation error and increase their demand for stocks after the reform. The effect should be greater for clients who experienced higher inflation right before the reform as they made greater errors. We find evidence consistent with this prediction.

We then analyze the heterogeneity in the relation between local inflation and stock trades. We begin by examining whether our results vary across investor types. Existing research shows that sophisticated investors are less prone to behavioral biases (e.g., Feng and Seasholes 2005; Locke and Mann 2005; Grinblatt et al. 2016). Thus, we test whether sophistication reduces the effect. We find the negative relationship between local inflation and buy-sell imbalances for stocks to be attenuated for individual investors considered more sophisticated. Moreover, we document a positive and weakly significant association between local inflation and buy-sell imbalances for stocks when replicating our analysis for institutional investors, suggesting that institutional investors do not suffer from money illusion. These findings support the notion that sophistication reduces money illusion.

We also investigate whether our results vary across stocks. Since investors subject to money illusion do not properly adjust firms’ future cash flows to inflation, the documented effect should be stronger for stocks where future growth prospects are more important than current cash flows. We find the relation between local inflation and buy-sell imbalances to be significantly more negative for stocks with low dividend yield and for high-tech stocks. Thus, the effect is more pronounced for stocks where money illusion is more likely.

We then examine the relation between local inflation and the performance of stock sales. Investors subject to money illusion are more likely to sell stocks in inflationary periods because they perceive them to be overvalued. If these stocks were truly overvalued, we should observe negative real returns following inflation-induced stock sales. However, we find a positive relation between local inflation and foregone real returns following stock sales. This evidence is again consistent with investors committing an inflation-induced valuation error.

While our findings provide support for the money illusion hypothesis, we also explore several competing explanations. First, as argued by Fama (1981), inflation could proxy for economic prospects. Hence, if local inflation increases, investors might lower their cash flow expectations, which reduces their demand for stocks. Second, inflation could make individuals more risk averse, thus explaining the negative relationship between local inflation and buy-sell imbalances for stocks (e.g., Brandt and Wang 2003; Cohen, Polk, and Vuolteenaho 2005). Third, our results could be driven by investors liquidating stocks to buy consumption goods, which become more expensive as local prices rise. Fourth, the negative relation between local inflation and buy-sell imbalances for stocks could be due to investors shifting funds from stocks to other asset classes that potentially offer a hedge against inflation. We perform a battery of tests to shed light on these alternative explanations. However, we do not find much support for them. Thus, our evidence points toward money illusion driving our findings.

According to Modigliani and Cohn (1979), investors subject to money illusion make a second valuation error in inflationary periods. Investors do not understand that decreasing accounting profits of firms caused by higher nominal interest payments are offset by the depreciation of the real value of nominal liabilities. Thus, investors subject to money illusion reduce their demand for stocks of firms that issue new debt. In line with this second form of money illusion, we show that buy-sell imbalances for stock trades are on average lower for firms with greater increases in nominal liabilities when inflation rises at firms’ headquarters.

Our paper makes three contributions. First, we contribute to the empirical literature on investors’ response to inflation. Existing work focuses mostly on the relation between inflation and stock price changes. Lending support to the hedging hypothesis, some papers find inflation to be positively correlated with stock returns (e.g., Branch 1974; Firth 1979; Boudoukh and Richardson 1993). However, numerous studies also document a negative association between inflation and stock returns (e.g., Fama and Schwert 1977; Fama 1981; Bekaert and Wang 2010). Some articles rely on money illusion to explain this negative relation (e.g., Ritter and Warr 2002; Cohen, Polk, and Vuolteenaho 2005), while others have identified rational explanations. For instance, according to Fama (1981), the negative relationship between inflation and stock returns is due to higher expected inflation proxying for lower expected economic growth. Our approach is different from the approach of the existing literature. Rather than analyzing stock returns, which only provide indirect evidence of investors’ behavior, we study investors’ security transactions. This enables us to provide the first direct evidence of investors’ response to inflation. Our findings lend support to the money illusion hypothesis.

Second, we contribute to the literature on individual investors’ behavior. Extant studies show that individual investors are subject to various behavioral biases.⁶ To the best of our knowledge, we are the first to investigate individual investors’ response to inflation. We provide evidence that individual investors reduce their demand for stocks during inflationary periods, consistent with money illusion.

Third, we contribute to the literature on hyperinflations. Existing research mainly studies hyperinflations to understand how individuals form inflation expectations and how these expectations affect their demand for money (e.g., Cagan 1956; Frenkel 1977; Evans 1978; Salemi and Sargent 1979). However, little is known about individual investors’ decisions during hyperinflations. Our study fills this gap.

Our results stress the importance of the ongoing debate on the financial literacy of individuals. Recently, the European Commission pointed toward the limited financial literacy of households and advocated for making financial education a priority for Europe. Similar calls were made in the United States.⁷ Our results underscore concerns that the financial literacy of individuals may not be sufficient to respond appropriately to the currently resurfacing inflation.

1 Historical Background

1.1 The German hyperinflation

The origins of the German hyperinflation lie in the economic and political situation that characterize the First World War and its aftermath. At the onset of the war in 1914, the German government suspended the convertibility between the Mark and gold and switched to a fiat money system. The war effort was predominantly financed by domestic debt and newly printed money. As a result, when Germany surrendered in November 1918, the national consumer price index (CPI) had increased by more than 100% compared to the beginning of the war (e.g., Bresciani-Turroni 1937, pp. 23–28; Dalio 2018, pp. 7–11).

After the First World War, the newborn German republic needed to finance postwar reconstruction, current expenditures, and war reparations. However, tax revenues were low, and Germany lacked the political and administrative strength to cut spending or to impose new taxes. Uncertainty about tax collection also impaired the possibility to issue new debt to German citizens. The international debt market remained inaccessible as international investors had no confidence in the Mark and questioned Germany’s creditworthiness. Therefore, printing money increasingly became the way to meet financial obligations. Between the end of the war and the beginning of 1920, the price level had increased by a factor of four (e.g., Moulton and McGuire 1923, pp. 201–7; Bresciani-Turroni 1937, p. 30).

In 1920, both the internal price level and the exchange rates of the Mark to foreign currencies stabilized. The expansionary monetary policy of the German Central Bank (Reichsbank) made German exports more attractive and increased the demand for Mark as foreign consumers looked for the German currency to purchase German goods (e.g., Dalio 2018, pp. 16–19). Figure 2 shows the CPI for Germany from February 1920 onward. The average monthly national inflation rate between March 1920 and April 1921 was about 2.2%.

Fig. 2

Nominal prices

This figure shows the German consumer price index (CPI), the German stock market index, the dollar/Mark exchange rate, the price of one of the most liquid German government bonds, and German real estate prices in nominal terms between February 1920 and September 1923. All time series are normalized to one as of February 1920.

The London Ultimatum in May 1921 again worsened Germany’s financial situation and the trust in the Mark. The Allies now demanded reparations of 132 billion Mark, which represented an increase in government debt of around 330% of GDP (e.g., Dalio 2018, pp. 20–22). From May to December 1921, the average monthly national inflation rate was about 7.1% (see Figure 2).

At the beginning of 1922, optimism spread as the Allies acknowledged that reparation demands were unsustainable. However, when renegotiations failed in June, the Mark fell (e.g., Dalio 2018, pp. 26–32). The average monthly inflation rate was 13.7% in the first half of 1922 and 61.2% in the second half (see Figure 2).

In January 1923, France and Belgium invaded Germany’s industrial heartland, the Ruhr region, after the Allies found that Germany had defaulted on reparation payments. The consequences were a government-financed general strike in the Ruhr region and the need to import coal for the rest of Germany, which further burdened the state’s budget. By March 1923, inflation had spun out of control. In October 1923, the Mark stood at six-billion-to-one relative to its prewar value (e.g., Bresciani-Turroni 1937, p. 36; Dalio 2018, pp. 33–34).

In mid-October 1923, the government introduced a stabilization policy that stopped the hyperinflation. The main element of the policy was the introduction of a new currency, the Rentenmark, which was backed by gold as well as German land and was pegged to the dollar. Strict limits were placed on the amount of Rentenmark that could be printed. Stabilization also came with fiscal consolidation and renewed renegotiations with the Allies over reparation demands, which led to substantially reduced reparation claims and culminated in the Dawes Plan (e.g., Bresciani-Turroni 1937, p. 98; Dornbusch 1985; Dalio 2018, pp. 35–42).

1.2 Financial investments in Germany in the early 1920s

In the early 1920s, stocks and bonds were traded on about 20 different exchanges in Germany. Berlin was the country’s largest exchange and one of the largest in the world (e.g., Ferguson and Voth 2008; Moore 2012; Lehmann-Hasemeyer and Streb 2016). The investment universe in Berlin comprised over 4,000 securities issued by about 2,000 entities. Around 60% were fixed-income securities, the remaining were equity securities.⁸ Most issuers were companies, in particular manufacturing firms, iron and steel works, as well as railroads. To conduct a trade, investors commissioned a broker, often by phone, who traded on their behalf and was awarded a fee for the service. Trading was possible 6 days a week (all days but Sundays). For most securities, supply and demand were matched by dedicated market makers in one auction per day, resulting in one market price at which all trades were executed (e.g., Buchwald 1924, pp. 233–6).

Another potential investment was foreign exchange. However, to finance the war effort, between 1914 and 1918 German citizens had to surrender most of their foreign exchange to the government. After the war, Germans’ foreign assets in the Allied countries were expropriated. Moreover, during the 1920s, German authorities introduced even more rigorous rules that prevented investors from owning and purchasing foreign exchange. As a result, there was little ownership in foreign exchange and purchasing foreign exchange was difficult during our sample period.

Another asset class to potentially invest in was real estate. However, since the outbreak of the war, rents were fixed to preserve social peace (so-called Friedensmiete, or “peace rent”). Fixed rents disincentivize individuals to become landlords, even more so in a high-inflation environment. As prices increased, rents covered an ever-shrinking fraction of the maintenance costs, forcing many landlords to sell (Bresciani-Turroni 1937, p. 299). In the course of the hyperinflation, foreigners bought many of the available homes.⁹

Figure 2 shows the evolution of the national CPI, the German stock market index, the dollar/Mark exchange rate, the price of the 4.5% German government bond (one of the most liquid debt securities), and real estate prices for Germany in nominal terms between February 1920 and September 1923.¹⁰ Investments in stocks and the dollar closely follow the CPI, implying that investments in these two asset classes offered the ability to hedge against inflation. In contrast, prices of government bonds and real estate remain almost flat, indicating that such assets did not offer inflation protection.

2 The Money Illusion Hypothesis

Modigliani and Cohn (1979) describe how inflation influences investors’ valuation of stocks. They argue that investors mistakenly use nominal rates to discount real future cash flows of firms. Following Cohen, Polk, and Vuolteenaho (2005), we formalize this idea using the Gordon Growth Model and express the dividend-price ratio at time t as

$$\frac{D_{t+1}}{P_t}=R-G,$$

(1)

where $D_{t+1}$ is the nominal dividend per share paid at time t + 1, P_t is the price per share at time t, R is the nominal discount rate, and G is the nominal growth rate of future cash flows. A rise in inflation increases both R and G equally, leaving the dividend-price ratio unaffected. However, investors subject to money illusion adjust the discount rate R, but do not sufficiently update the growth rate G. As a result, when inflation increases, investors subject to money illusion require a higher dividend yield in order to hold stocks, which makes them less likely to buy and more likely to sell stocks. Money illusion, however, only leads to trading if investors disagree about the valuation of stocks. If all investors agree on the (mis)valuation of stocks, we expect to find no significant trading response to inflation.

Notice that investors subject to money illusion do not make the same mistake when they value bonds. As bonds offer constant cash flows, the growth rate G is irrelevant. Under increasing inflation, investors only have to adjust R, which they do correctly, even if they suffer from money illusion. Therefore, the valuation of bonds should be unbiased.

According to Modigliani and Cohn (1979), investors subject to money illusion commit a second valuation error. Such investors do not understand that decreasing accounting profits of firms caused by higher nominal interest payments are offset by an increase in the market value of equity resulting from the deprecation in the real value of nominal liabilities. As a result, investors suffering from money illusion reduce their demand for stocks of firms that issue new debt. We discuss this second form of money illusion in detail in Internet Appendix E.

3 Empirical Approach

We test for Modigliani and Cohn (1979) money illusion hypothesis using the following equations:

$$\mathit{Buy}-\mathit{sell}\hspace{0.17em}\mathit{imbalanc}{e}_{i,t}={\alpha }_t+{\alpha }_i+\beta \mathit{Local}\hspace{0.17em}\mathit{inflatio}{n}_{i,t}+\mathit{Control}{s}_{i,t}+{ϵ}_{i,t}.$$

(2)

The $\mathit{Buy}-\mathit{sell}\hspace{0.17em}\mathit{imbalanc}{e}_{i,t}$ of investor i in month t is defined as

$$\mathit{Buy}-\mathit{sell}\hspace{0.17em}\mathit{imbalanc}{e}_{i,t}=\frac{\#\mathit{buy}{s}_{i,t}-\#\mathit{sell}{s}_{i,t}}{\#\mathit{buy}{s}_{i,t}+\#\mathit{sell}{s}_{i,t}},$$

(3)

where $\#\mathit{buy}{s}_{i,t}$ (⁠$\#\mathit{sell}{s}_{i,t}$⁠) is the number of stock purchases (sales) by investor i in month t. The buy-sell imbalance captures investors’ net demand for stocks in a given month.¹¹α_t are year-month fixed effects that control for the overall time trend, thereby accounting for factors such as national inflation and overall economic conditions. α_i are client fixed effects, which control for time-invariant investor characteristics, such as gender. $\mathit{Local}\hspace{0.17em}\mathit{inflatio}{n}_{i,t}$ is the inflation in month t of the town where investor i lives. We assume that local inflation experienced by investors shapes their inflation expectations. This assumption is in line with Malmendier and Nagel (2016) and D’Acunto et al. (2021), who show that, when individuals form inflation expectations, they heavily rely on experienced price changes. $\mathit{Control}{s}_{i,t}$ represents a set of time-varying town-level characteristics. ${ϵ}_{i,t}$ is the error term. The money illusion hypothesis predicts a negative β for stock trades, that is, a negative relationship between local inflation experienced by investors and investors’ buy-sell imbalance for stocks. The hedging hypothesis predicts a positive β.

To test for Modigliani, and Cohn’s (1979) second form of money illusion, we analyze clients’ trading behavior in stocks of firms that experience changing inflation and changing net leverage. The empirical approach used to test this hypothesis is discussed in detail in Internet Appendix E.

4 Data

4.1 Investor data

We obtain security portfolio data from a German bank. The bank’s core business was in private and investment banking, serving private and institutional clients. The bank offered a broad range of wealth management services to its private clients, including securities accounts. While the bank was headquartered in Germany, and thus mainly targeted German clients, it also offered its services to clients living abroad.

In the predigital era, banks kept track of client-level security portfolios in so-called “deposit books” (Depotbücher). The Law of Deposits (Depotgesetz) required them to do so, which ensures that the information on transactions and holdings in these books is comprehensive (e.g., Buchwald 1924, pp. 427–8). Specifically, the deposit books record, for each client, every transaction, and after each transaction, the holdings in the respective security. The deposit books also provide several investor characteristics, such as clients’ place of residence and whether they hold accounts at other banks. The deposit books contain information on roughly 3,000 private clients with a security portfolio during our sample period. We drop around 700 clients for which we cannot clearly identify the account holder.¹² This leaves us with 2,262 clients who execute 49,415 transactions between January 1920 and December 1924. Figure IA1 in the Internet Appendix shows a sample page from the deposit books.

To shed light on the representativeness of our sample, we perform several comparisons. First, we compare our bank to other German banks at the time. Since only very limited information is available for certain types of banks (e.g., savings banks, cooperative banks, private banks), existing research counts the number of issuers for which a bank acts as paying agent to shed light on banks’ relative importance (e.g., Jeidels 1905; p. 127; Riesser 1911, p. 371; Wixforth and Ziegler 1997; Ziegler 2009).¹³ Based on this comparison, our bank ranks among the 30 largest banks in Germany, which is considerable given that around 20,000 banks operated in Germany at the time (Bundesbank 1976, pp. 67, 121). This suggests that our bank is representative of a large German bank.

Second, we compare our clients to the shareholder base in Germany during the early 1920s. We assign clients to social classes based on their profession and their title using the approach of Schüren (1989) and compare the obtained distribution to the one of Lehmann-Hasemeyer and Neumayer (2022), who assign attendees of annual general meetings of German firms during our sample period to social classes using the same approach. Figure IA2 in the Internet Appendix shows that the two distributions align closely, suggesting that our clients are representative of the German stockholding population.

We also compare the wealth of clients to the wealth of the German population. Data on the distribution of the population’s net wealth come from the wealth tax collected at year-end 1913.¹⁴ Note that only individuals with net wealth of more than 10,000 Mark were subject to this wealth tax, corresponding to about 2.8 million individuals (or 4.3% of the population). We use clients’ portfolio market value in January 1920 (the beginning of our sample period) and adjust it to obtain an estimate of clients’ net wealth in December 1913. Based on these estimates, around 76% (24%) of clients in our sample have a net wealth of more (less) than 10,000 Mark. Figure IA3 in the Internet Appendix shows that wealthy clients in our sample have a wealth distribution similar to individuals subject to the wealth tax, suggesting that these clients are representative of the wealthiest 5% of the population.

Finally, we compare the investment behavior of clients in our sample to the investment behavior of individual investors today. We find that our clients hold a similar number of securities and a similar fraction of stocks in their portfolio as in the data set of Barber and Odean (2000), that clients exhibit a strong preference for local investments (e.g., Seasholes and Zhu 2010), that they execute a similar number of trades per month as in the data set of Barber and Odean (2000), that men trade more than women (e.g., Barber and Odean 2001), and that clients are subject to the disposition effect (e.g., Shefrin and Statman 1985). This suggests that the investment behavior of our clients is comparable to the behavior of individual investors today.

4.2 Firm data

For each German firm whose shares the clients trade, we hand-collect annual balance sheet data from the Handbook of German Stock Corporations (Handbuch der Deutschen Aktiengesellschaften). We follow French, Ruback, and Schwert (1983) and Ritter and Warr (2002), who compute net leverage as the sum of nominal liabilities less the sum of nominal assets, all scaled by total assets. The change in net leverage is the difference between this year’s and last year’s net leverage. In our final sample, we have 623 German companies whose securities are traded by the clients and for which we have balance sheet data. Figure IA4 in the Internet Appendix shows a sample page from the handbooks.

We also hand-collect month-end market prices of stocks traded on the Berlin Stock Exchange from the Berlin Stock Exchange Newspaper (Berliner Börsen-Zeitung). We then use monthly stock prices to compute monthly stock returns. Clients in our final sample trade stocks of 553 firms for which we have return data. Figure IA5 in the Internet Appendix provides a sample page from the Berlin Stock Exchange Newspaper.

In addition, we hand-collect dividend data for each firm listed on the Berlin Stock Exchange from a book entitled The Coupon (Der Zinsschein). Our clients trade stocks of 485 firms for which we have dividend data. Figure IA6 in the Internet Appendix shows a sample page from this book.

4.3 Local inflation data

We additionally hand-collect information on monthly local consumer prices from the Quarterly Issue of the German Statistical Office (Vierteljahresheft zur Statistik des Deutschen Reichs). Starting in December 1919, the statistical office collected prices of a basket of goods considered representative for a family of five members in each German town with more than 10,000 inhabitants and constructed a local consumer price index.¹⁵ These data were originally compiled because the German Department of Labor (Reichsarbeitsministerium) needed information on local price changes as a basis for wage negotiations. We compute monthly local inflation as the percentage change in a town’s CPI between the current and the previous month. In total, we have monthly inflation data for 633 German towns between January 1920 and December 1924. We merge inflation data, investor data, and firm data by assigning clients and firms to the closest town for which we have inflation data within a 25-km radius.¹⁶ We end up with clients and firms being matched to 256 different towns with inflation data. Figure IA7 in the Internet Appendix provides a sample page showing the consumer price index data from the German Statistical Office.

4.4 Descriptive statistics

Table 1 reports descriptive statistics. Panel A presents sociodemographic variables of the clients in our sample. About 72% of the bank customers are male and 89% live in Germany. Moreover, 9% of clients hold an account with another bank. This suggests that clients typically do not have additional accounts with other banks and our bank appears to be the house bank of most clients. This allows for a comprehensive view of investors’ behavior.

Table 1

Descriptive statistics

	Mean	Min.	Median	Max.	SD	N
A. Client characteristics
Male (d)	0.72	0.00	1.00	1.00	0.45	2,262
Germany (d)	0.89	0.00	1.00	1.00	0.31	2,260
Europe (d)	0.97	0.00	1.00	1.00	0.18	2,260
Other bank account (d)	0.09	0.00	0.00	1.00	0.29	2,262
B. Portfolio characteristics
Avg. # securities	3.12	1.00	1.53	60.88	4.44	2,262
Avg. % stocks	48.70	0.00	50.00	100.00	42.56	2,262
Avg. % bonds	31.91	0.00	4.59	100.00	40.54	2,262
Avg. % foreign exchange	13.44	0.00	0.00	100.00	28.21	2,262
C. Trade characteristics
Avg. # trades per month	0.78	0.00	0.50	16.22	1.03	2,262
Avg. % buys	54.21	0.00	50.00	100.00	22.55	2,225
Avg. % stock trades	51.21	0.00	58.82	100.00	41.84	2,225
Avg. % bond trades	30.33	0.00	4.44	100.00	39.65	2,225
Avg. % foreign exchange trades	13.36	0.00	0.00	100.00	27.42	2,225
Avg. buy-sell imbalance for stocks	0.18	–1.00	0.11	1.00	0.40	1,508
Avg. buy-sell imbalance for bonds	0.07	–1.00	0.00	1.00	0.53	1,172
Avg. buy-sell imbalance for foreign	0.07	–1.00	0.00	1.00	0.52	817
exchange
D. Firm characteristics
Avg. net leverage (%)	14.31	–88.89	14.05	89.86	24.02	623
Avg. Δ net leverage (%)	1.10	–74.71	0.10	92.50	16.11	584
E. Local inflation
Raw local inflation (%)	537.92	–12.36	8.63	35,117.90	3,746.72	13,112

This table presents descriptive statistics on client characteristics (panel A), portfolio characteristics (panel B), trade characteristics (panel C), firm characteristics (panel D), and local inflation (panel E). We focus on the time period from January 1920 to December 1924. For time-varying variables, we report averages, except for panel E, where we report monthly observations. In panel D, the sample includes firms whose stocks the clients trade. In panel E, the sample includes towns in which at least one client lives or at least one firm is headquartered. We assign clients and firms to the closest town for which we have inflation data within a 25-km radius based on the place of residence and the location of headquarters, respectively. Internet Appendix A describes all variables used throughout the study in detail.

Panel B describes the composition of clients’ portfolios. On average, a portfolio consists of about three securities, of which 49% are stocks denominated in Mark, 32% bonds denominated in Mark, and 13% securities denominated in foreign currencies.¹⁷

Panel C provides information on clients’ trades. Clients on average execute 0.8 trades per month (that is, almost 10 trades per year). Around 54% of all trades are purchases, 51% involve stocks, 30% bonds, and 13% foreign securities. Our main variable capturing clients’ investment behavior is the monthly buy-sell imbalance. The average monthly buy-sell imbalance for stocks is 0.18. It is 0.07 for bonds and 0.07 for securities denominated in foreign currencies. The positive average buy-sell imbalance for stocks can be explained by many companies issuing new shares during our sample period. The equity issuance volume was primarily driven by substantial capital needs of firms in the early 1920s, when the German economy did relatively well (e.g., Aron 1927; Bresciani-Turroni 1937, p. 255).¹⁸

Panel D presents summary statistics on net leverage of firms in our sample. The average net leverage amounts to 14%. The average annual change in net leverage is 1%.

Panel E shows descriptive statistics on monthly local inflation of towns in which at least one client lives or at least one firm is headquartered. This is our main explanatory variable. The average (median) monthly local inflation rate amounts to 538% (9%) between 1920 and 1924. However, this number hides substantial cross-sectional and time-series variation. For instance, in October 1920, the town with the highest inflation rate was Beuel, which is a part of Bonn today, with 18%, while Aschaffenburg, near Frankfurt am Main, the town with the lowest inflation rate, experienced a reduction in prices of 3%. In 1920, the average monthly local inflation rate across all towns was 7%. It declined to 5% in 1921, rose to 38% in 1922, and reached 2,945% in 1923. In 1924, the year after the successful stabilization of the currency, the monthly local inflation rate averaged only 0.9%. To make the local inflation variable more normally distributed, we follow previous research and apply the inverse hyperbolic sine transformation (e.g., Burbidge, Magee, and Robb 1988; Kale, Reis, and Venkateswaran 2009; Karlan et al. 2016).¹⁹Figure 3 plots the inverse hyperbolic sine of monthly inflation of all 256 towns in our sample over time.

Fig. 3

Local inflation

This figure shows the inverse hyperbolic sine of monthly local inflation for towns in which at least one client lives or at least one firm was headquartered between January 1920 and December 1924. Each dot represents the monthly local inflation rate of one town. We assign clients and firms to the closest town for which we have inflation data within a 25-km radius based on the place of residence and the location of headquarters, respectively. Internet Appendix A describes all variables used throughout the study in detail.

Figure IA8 in the Internet Appendix shows the locations of clients and firms in our final sample. We plot the map of Germany in 1920 and the map of Germany today (in gray). We mark the 256 towns for which we have inflation data and to which we assign at least one client (dots) or at least one firm (crosses). The figure reveals that both investors and firms are quite evenly distributed across Germany, with clusters that broadly follow the distribution of the population.

4.5 Determinants of local inflation

The cross-sectional and time-series variation in monthly local inflation that we exploit in our analysis is likely not random. Thus, we next explore the determinants of local inflation. We estimate ordinary least squares (OLS) regressions and use the inverse hyperbolic sine of monthly inflation of towns in our sample as dependent variable. The sample starts in January 1920 and ends in September 1923, shortly before Germany reformed its currency. As explanatory variables, we use the natural logarithm of the local population in 1919, a dummy variable that equals one for territories occupied by the French or Belgian troops, the monthly local unemployment rate, a dummy variable that equals one for towns with a branch of the central bank, and the fraction of local employees working in the paper industry in 1921.²⁰

Table 2 reports the results. We find that inflation is higher in towns that are larger (column 1), that were occupied (column 2), that have lower unemployment (column 3), that are home to a branch of the central bank (column 4), and where a higher fraction of employees works in the paper industry (column 5). The latter result lends support to the notion that the German Central Bank relied on local firms to produce bank notes.²¹ In column 6, we include all explanatory variables simultaneously. We still find the occupied dummy variable, the unemployment rate, and the fraction of employees working in the paper industry to be significantly related to inflation. Finally, in column 7, we add year-month and town fixed effects, which control for the overall time trend and all time-invariant determinants of local inflation. We find that the dummy variable indicating the occupation of a town remains significantly related to inflation. Taken together, we document that local inflation is not randomly distributed, but correlated with other local factors. Thus, regressions that follow include year-month and client fixed effects as well as our two time-varying determinants of local inflation.

Table 2

Determinants of local inflation

	Local inflation${}_{c,t}$
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
log(local population)${}_{c,1919}^{}$	0.021*					0.016
	(1.69)					(1.67)
Occupied (d)${}_{c,t}^{}$		0.081**				0.073*	0.029*
		(2.20)				(1.86)	(1.89)
Local unemployment rate${}_{c,t}^{}$			–6.387***			–6.471***	–0.084
			(–3.28)			(–3.17)	(–0.56)
German Central Bank (d)${}_{c,1920}^{}$				0.065*		0.013
				(1.73)		(0.48)
% local employees in paper${}_{c,1921}^{}$					0.116***	0.162*
					(3.28)	(1.81)
Year-month fixed effects	No	No	No	No	No	No	Yes
Town fixed effects	No	No	No	No	No	No	Yes
Adj. R2	.001	.001	.008	.001	–.000	.010	.986
N	10,634	10,634	9,629	10,634	10,634	9,629	9,629

This table presents the results from panel regressions with year-month and town fixed effects. The dependent variable is the inverse hyperbolic sine of local inflation of town c in month t. We focus on the time period from January 1920 to September 1923 and on towns in which at least one client lives or at least one firm is headquartered. We assign clients and firms to the closest town for which we have inflation data within a 25-km radius based on the place of residence and the location of headquarters, respectively. Internet Appendix A describes all variables used throughout the study in detail. Standard errors are double-clustered by town and month. t-statistics are provided in parentheses.

*

$p<$⁠.1;

**

$p<$ .05;

***

$p<$ .01.

5 Empirical Results

This section contains our empirical results. First, we investigate the relationship between local inflation and clients’ investment behavior in stocks (Section 5.1). Second, we analyze whether there is heterogeneity in this relationship (Section 5.2). Next, we study the association between local inflation and returns following stock sales (Section 5.3). We then run numerous tests to rule out alternative explanations (Section 5.4). Finally, we investigate the second form of money illusion (Section 5.5).

5.1 Local inflation and stock trades

To test for money illusion, we regress monthly buy-sell imbalances for stock trades of clients on the monthly local inflation rate, as outlined in Equation (2). Depending on the estimated specification, we include controls and fixed effects. In all our regressions, we double-cluster standard errors by town and month.²²

We present the results in Table 3. In column 1, we include year-month fixed effects to control for the overall time trend. The coefficient estimate is negative and statistically significant at the 5% level. In column 2, when adding client fixed effects that control for all time-invariant investor characteristics, the documented effect becomes statistically and economically stronger. The negative coefficient estimate suggests that investors buy fewer (sell more) stocks when facing higher local inflation. We find that a 1% increase in local inflation is associated with a 3.5% decline in buy-sell imbalances for stocks. In column 3, local inflation is measured over the current month and the previous month (rather than over the current month only). Consistent with the idea that not only contemporaneous inflation affects clients’ inflation expectations but also inflation experienced in the recent past, we again find a strong negative relation between local inflation and buy-sell imbalances. In column 4, we augment the regression with the two time-varying control variables from Table 2, a dummy variable that equals one if a town was in occupied territory in a given month and the local unemployment rate. Including them, however, does not materially change our findings. In columns 5 and 6, we split the sample period into two subperiods. In the first subperiod, from January 1920 to June 1922, the German economy did relatively well and experienced comparably low inflation.²³ The second subperiod, from July 1922 to September 1923, is characterized by the hyperinflation. We find a negative and statistically significant relationship between inflation and buy-sell imbalances in both subperiods. Taken together, we document a strong negative relation between local inflation and buy-sell imbalances for stocks, which supports the money illusion hypothesis, but not the hedging hypothesis.²⁴

Table 3

Local inflation and stock trades

	Buy-sell imbalance for stocks${}_{i,t}$
					Jan. 1920–	Jul. 1922–
					Jun. 1922	Sep. 1923
	(1)	(2)	(3)	(4)	(5)	(6)
Local inflation${}_{i,t}^{}$	–0.536**	–0.650**		–0.548**	–0.990*	–0.584**
	(–2.48)	(–2.63)		(–2.07)	(–1.83)	(–2.42)
Local inflation${}_{i,t-1,t}^{}$			–0.353**
			(–2.57)
Occupied (d)${}_{i,t}^{}$				–0.484*
				(–1.93)
Local unemployment rate${}_{i,t}^{}$				–2.188
				(–0.74)
Year-month fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Client fixed effects	No	Yes	Yes	Yes	Yes	Yes
Adj. R2	.036	.036	.035	.037	.021	.055
N	8,057	8,057	7,961	7,962	3,394	4,663

This table presents the results from panel regressions with year-month and client fixed effects. The dependent variable is the buy-sell imbalance for stocks of client i in month t. We focus on the time period from January 1920 to September 1923. In column 5 (column 6), we restrict the sample to the time period from January 1920 to June 1922 (July 1922 to September 1923). The variable Local inflation is the inverse hyperbolic sine of local inflation of the town where the client lives. Internet Appendix A describes all variables used throughout the study in detail. Standard errors are double-clustered by town and month. t-statistics are provided in parentheses.

*

p <.1;

**

$p<$ .05;

***

$p<$ .01.

To shed additional light on whether contemporaneous or lagged inflation matters for investors, we replicate our analysis at the weekly level, including several lags of inflation. Weekly inflation data are available from July 12, 1923, onward. We compute weekly buy-sell imbalances over the same time period over which weekly inflation is measured. Results are presented in Table IA2 in the Internet Appendix. In column 1, we include time fixed effects. Column 2 additionally contains client fixed effects. In column 1, we find the relationship between local inflation and buy-sell imbalances to be negative up to the fourth lag, albeit not statistically significant. In column 2, the relationship is significantly negative up to the fifth lag. This finding suggests that it is local inflation in the last few weeks that matters for investors, and not local inflation in the more distant past, supporting the use of contemporaneous inflation as our main explanatory variable.

So far, we have restricted our analysis to the period from January 1920 to September 1923, which is the time period characterized by rising prices. Next, we test whether we find consistent results when we explore a reverse inflation shock. In particular, we investigate trading patterns in a 12-month window around October 1923, when the government successfully stabilized the currency. Within a few weeks, inflation dropped sharply (see Figure 3). In principle, a reduction in inflation should produce the opposite effect of what we showed in Table 3. As inflation declined close to zero, nominal and real discount rates converged. Hence, investors subject to money illusion no longer make a valuation error and we expect them to increase their demand for stocks after the reform. The effect should be greater for clients who experienced higher inflation right before the reform as they made greater errors. We identify these investors in two ways. First, we take the cumulative local inflation rate over the 6 months preceding the currency reform. Second, we compare clients living in Germany with clients living in neighboring countries.²⁵ To test this conjecture, we adapt Equation (2) in the following ways:

$$\mathit{Buy}-\mathit{sell}\hspace{0.17em}\mathit{imbalanc}{e}_{i,t}={\alpha }_t+{\alpha }_i$$

(4)

$$\begin{array}{c}\hspace{1em}+\beta \mathit{Local}\hspace{0.17em}\mathit{inflatio}{n}_{i,\mathit{Apr}.-\mathit{Sep}.1923}\times \mathit{Post}-\mathit{reform}{(d)}_t+{ϵ}_{i,t},\\ Buy-\mathit{sell}\hspace{0.17em}\mathit{imbalanc}{e}_{i,t}={\alpha }_t+{\alpha }_i+\beta \mathit{Germany}{(d)}_i\times \mathit{Post}-\mathit{reform}{(d)}_t+{ϵ}_{i,t},\end{array}$$

(5)

where $\mathit{Local}\hspace{0.17em}\mathit{inflatio}{n}_{i,\mathit{Apr}.-\mathit{Sep}.1923}$ is the cumulative inflation rate in the 6-month period preceding the currency reform (from April to September 1923) of the town where investor i lives. $\mathit{Post}-\mathit{reform}{(d)}_t$ is a dummy variable that takes the value of one in the 6-month period following the currency reform (from October 1923 to March 1924), and zero otherwise. $\mathit{Germany}{(d)}_i$ is a dummy variable that equals one for investors who live in Germany and zero for investors who live in neighboring countries. For both regressions, the money illusion hypothesis predicts a positive β.

We present the results in Table 4. Estimates from Equation (4) are shown in column 1. We find that clients living in towns with higher prestabilization inflation buy more (sell fewer) stocks after the stabilization compared to clients in towns with lower prestabilization inflation. Estimates for Equation (5) are presented in column 2. We document that clients living in Germany increase their demand for stocks after the currency reform compared to clients living abroad. Figure IA10 in the Internet Appendix graphically illustrates the increase in buy-sell imbalances of clients living in Germany around the currency reform relative to clients living in neighboring countries. Hence, the analysis of the currency reform of October 1923 confirms that there is a negative relation between local inflation experienced by investors and investors’ demand for stocks.

Table 4

Local inflation and stock trades around the currency reform

	Buy-sell imbalance for stocks${}_{i,t}$
	(1)	(2)
Local inflation${}_{i,\mathit{Apr}.-\mathit{Sep}. 1923}^{}\times$ Post-reform (d)${}_t^{}$	0.365*
	(1.89)
Germany (d)${}_i^{}\times$ Post-reform (d)${}_t^{}$		0.339***
		(3.53)
Year-month fixed effects	Yes	Yes
Client fixed effects	Yes	Yes
Adj. R2	.081	.088
N	3,544	3,891

This table presents the results from panel regressions with year-month and client fixed effects. The dependent variable is the buy-sell imbalance for stocks of client i in month t. We focus on the time period starting 6 months prior to the currency reform and ending 6 months after the currency reform. In column 2, the sample includes all clients who live in Germany and all clients who live in neighboring countries. The variable Local inflation is the inverse hyperbolic sine of cumulative local inflation of the town where the client lives over the 6 months preceding the currency reform. The variable Post-reform (d) equals one after Germany reforms its currency (October 1923 onward), and zero otherwise. The variable Germany (d) equals one for clients who live in Germany, and zero otherwise. Internet Appendix A describes all variables used throughout the study in detail. Standard errors are double-clustered by town and month. t-statistics are provided in parentheses.

*

$p<$⁠.1;

**

$p<$ .05;

***

$p<$ .01.

5.2 Cross-sectional results

Next, we analyze the heterogeneity in the relation between local inflation and stock trades. We begin by investigating whether our results differ across investor types. Existing research shows that sophisticated investors are less prone to behavioral biases (e.g., Feng and Seasholes 2005; Locke and Mann 2005; Grinblatt et al. 2016). Moreover, anecdotal evidence suggests that sophisticated investors bought large amounts of stocks during our sample period (e.g., Bresciani-Turroni 1937, pp. 290–8). Hence, we investigate whether sophistication reduces the documented effect. We use four different measures to capture individual investors’ sophistication. Following existing studies, we take the portfolio value as a proxy for sophistication (e.g., Hirshleifer et al. 2008; Barber, Huang, and Odean 2016). We create a dummy variable that takes the value of one (zero) if clients’ portfolio market value in January 1920 is above (equal to or below) the median. The second sophistication proxy is a dummy variable that equals one (zero) if clients’ number of different stocks in the portfolio in January 1920 is above (equal to or below) the median. This measure captures investors’ degree of diversification (e.g., Feng and Seasholes 2005). The third sophistication measure is a dummy variable that equals one for clients who are employees of our bank. Prior research shows that financial professionals tend to be more sophisticated than retail traders (e.g., Locke and Mann 2005). The fourth sophistication proxy is a dummy variable that equals one for investors who traded on margin. As highlighted by Bresciani-Turroni, 1937, p. 294, sophisticated investors quickly realized during the German hyperinflation that trading with borrowed money increased profits as debt depreciated quickly because of rising prices. To test the conjecture that sophistication reduces money illusion, we use our main specification from column 2 of Table 3 and interact the local inflation variable with our sophistication measures.

Table 5 shows the results. We continue to find a negative and statistically significant coefficient on the local inflation variable across all four columns. However, the significantly positive coefficient on the interaction term implies that the negative relationship between local inflation and buy-sell imbalances is weaker for more sophisticated investors. For instance, the coefficient estimates in columns 2 and 3 suggest that the effect is more than 10% weaker for more diversified investors and employees of the bank. This is in line with our conjecture that sophisticated investors are less prone to money illusion.²⁶

Table 5

Local inflation, client characteristics, and stock trades

	Buy-sell imbalance for stocks${}_{i,t}$
	(1)	(2)	(3)	(4)
Local inflation${}_{i,t}^{}$	–0.748***	–0.784***	–0.674***	–0.683***
	(–2.82)	(–3.04)	(–2.72)	(–2.80)
Local inflation${}_{i,t}^{}\times$ Wealthy (d)${}_{i,\mathit{Jan}. 1920}^{}$	0.035***
	(4.29)
Local inflation${}_{i,t}^{}\times$ Diversified (d)${}_{i,\mathit{Jan}. 1920}^{}$		0.095***
		(5.83)
Local inflation${}_{i,t}^{}\times$ Bank employee (d)${}_i^{}$			0.085***
			(6.61)
Local inflation${}_{i,t}^{}\times$ Levered (d)${}_i^{}$				0.053***
				(3.86)
Year-month fixed effects	Yes	Yes	Yes	Yes
Client fixed effects	Yes	Yes	Yes	Yes
Adj. R2	.078	.080	.038	.036
N	3,561	3,561	8,057	8,057

This table presents the results from panel regressions with year-month and client fixed effects. The dependent variable is the buy-sell imbalance for stocks of client i in month t. We focus on the time period from January 1920 to September 1923. The variable Local inflation is the inverse hyperbolic sine of local inflation of the town where the client lives. The variable Wealthy (d) equals one (zero) if clients’ portfolio market value in January 1920 is above (equal to or below) the median. The variable Diversified (d) equals one (zero) if clients’ number of different stocks in the portfolio in January 1920 is above (equal to or below) the median. The variable Bank employee (d) equals one for clients who are employees of the bank, and zero otherwise. The variable Levered (d) equals one for clients with a levered portfolio, and zero otherwise. Internet Appendix A describes all variables used throughout the study in detail. Standard errors are double-clustered by town and month. t-statistics are provided in parentheses.

*

$p<$⁠.1;

**

$p<$ .05;

***

$p<$ .01.

To shed additional light on the behavior of sophisticated investors, we rerun our analysis for institutional clients. Our bank served as a broker not only for private clients but also for institutional clients, such as banks, insurance companies, and pension funds. Professional investors are typically considered more sophisticated than individual investors (e.g., Locke and Mann 2005). We hand-collect security portfolio data of 172 institutional investors who execute 5,426 stock trades between January 1920 and September 1923. We then replicate the main regression specifications from Table 3 using these institutional transactions. We report the results in Table IA3 in the Internet Appendix. The relationship between local inflation and buy-sell imbalances for stocks is positive across all specifications. It is not statistically significant at conventional levels in columns 1 and 4 but is statistically significant at the 10% level in columns 2 and 3. This suggests that institutional investors are not subject to money illusion. If anything, they buy more (sell fewer) stocks when facing higher local inflation, which is consistent with institutional investors hedging against local price increases.

We also analyze whether our results vary across stocks. Since investors suffering from money illusion do not properly adjust the growth rate of firms’ cash flows to inflation, we expect the documented effect to be stronger for stocks where growth prospects are more important than current cash flows. To proxy for current cash flows and future growth prospects, we use stocks’ dividend yield and firms’ industry classification. Stocks with low dividend yield and stocks of high-tech firms are likely more susceptible to money illusion. We classify stocks as low-yield (high-yield) stocks if their average dividend yield over the past 3 months is equal to or below (above) the median. Moreover, we classify chemical companies, electric power companies, and machine building companies as high-tech firms and companies active in other industries (e.g., mining, iron and steel works, timber) as low-tech firms (e.g., Bresciani-Turroni 1937, p. 410; White 1990). We then rerun our main specification from column 2 of Table 3 separately for low-yield and high-yield stocks and for high-tech and low-tech stocks.

Columns 1 to 4 of Table 6 report the results. The coefficient estimates on local inflation are more negative for low-yield stocks (column 1) than for high-yield stocks (column 2) and for high-tech stocks (column 3) than for low-tech stocks (column 4). Differences in coefficient estimates between the groups are statistically significant, as indicated by the results of tests for differences reported at the bottom of the table. Hence, we indeed find that the effect is more pronounced for stocks where money illusion is more likely.

Table 6

Local inflation, stock characteristics, and stock trades

	Buy-sell imbalance for stocks${}_{i,t}$
	Low-yield	High-yield	High-tech	Low-tech	Low-volatility	High-volatility	Low-beta	High-beta
	stocks	stocks	stocks	stocks	stocks	stocks	stocks	stocks
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Local inflation${}_{i,t}^{}$	–1.645***	–0.397	–1.420***	–0.450	–0.912**	–0.792**	–0.346	–0.413
	(–3.85)	(–1.10)	(–3.72)	(–1.24)	(–2.03)	(–2.14)	(–0.40)	(–0.79)
Year-month fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Client fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Adj. R2	–.019	–.004	–.013	–.021	.046	.021	–.094	–.009
N	2,671	3,107	2,809	4,035	2,269	2,602	1,215	1,886
F-statistic (p-value) of	5.037		3.453		0.044		0.005
difference in coefficients	(.027)		(.066)		(.835)		(.946)

This table presents the results from panel regressions with year-month and client fixed effects. The dependent variable is the buy-sell imbalance for stocks of client i in month t. We focus on the time period from January 1920 to September 1923. In column 1 (column 2), we restrict the sample to trades in low-yield (high-yield) stocks. We classify stocks as low-yield (high-yield) stocks if their average dividend yield over the past 3 months is equal to or below (above) the median. In column 3 (column 4), we restrict the sample to trades in high-tech (low-tech) stocks. We classify chemical companies, electric power companies, and machine building companies as high-tech firms and companies active in other industries as low-tech firms. In column 5 (column 6), we restrict the sample to trades in low-volatility (high-volatility) stocks. We classify stocks as low-volatility (high-volatility) stocks if their return volatility over the past 6 months is equal to or below (above) the median. In column 7 (column 8), we restrict the sample to trades in low-beta (high-beta) stocks. We classify stocks as low-beta (high-beta) stocks if their market beta over the past 3 years is equal to or below (above) the median. The variable Local inflation is the inverse hyperbolic sine of local inflation of the town where the client lives. Internet Appendix A describes all variables used throughout the study in detail. Standard errors are double-clustered by town and month. t-statistics are provided in parentheses.

*

$p<$⁠.1;

**

$p<$ .05;

***

$p<$ .01.

5.3 Local inflation and the performance of stock sales

Next, we analyze the relation between local inflation and the performance of stock sales. In inflationary periods, investors subject to money illusion are more likely to sell stocks since they perceive them to be overvalued. If these stocks were truly overvalued, we should observe negative real returns following inflation-induced stock sales. To test this conjecture, we investigate whether foregone profits following stock sales are correlated with local inflation experienced at the time of the sale. We estimate the following regression:

$$r_{i,j,t+1,t+\tau }={\alpha }_t+{\alpha }_i+{\alpha }_j+\beta \mathit{Local}\hspace{0.17em}\mathit{inflatio}{n}_{i,t}+{ϵ}_{i,j,t},$$

(6)

where $r_{i,j,t+1,t+\tau }$ is the real return of stock j over the window $t+1,t+\tau$ sold by investor i in month t. α_t are year-month fixed effects, α_i correspond to client fixed effects, and α_j are firm fixed effects. Including year-month fixed effects has an effect similar to computing market-adjusted returns because we compare returns of trades conducted in the same month over the same post-trade time window. $\mathit{Local}\hspace{0.17em}\mathit{inflatio}{n}_{i,t}$ is the inflation rate experienced by client i in month t. The money illusion hypothesis predicts β to be zero or positive.

We present the results in Table 7. In columns 1 to 3 (column 4), we measure real returns of stock sales over a 3-month (6-month) period following the sales. Across all specifications, we find a positive relationship between local inflation in the month of the sale and real returns in the following months. We find the relationship to be statistically significant at the 10% level in column 1, where we include time fixed effects, and in column 2, where we add firm fixed effects. Results are not statistically significant at conventional levels in columns 3 and 4, when adding client fixed effects and when investigating the performance over a 6-month period. The positive coefficient suggests that sales in periods of high local inflation deliver higher real returns than sales in periods of low local inflation. Thus, stocks sold by investors facing high inflation tend to be undervalued, rather than overvalued, which is again in line with investors suffering from money illusion.

Table 7

Local inflation and the performance of stock sales

	Real return of individual			Real return of individual
	stock sale${}_{i,j,t+1,t+3}$			stock sale${}_{i,j,t+1,t+6}$
	(1)	(2)	(3)	(4)
Local inflation${}_{i,t}^{}$	1.459*	1.656*	1.159	0.261
	(1.83)	(1.94)	(1.26)	(0.17)
Year-month fixed effects	Yes	Yes	Yes	Yes
Firm fixed effects	No	Yes	Yes	Yes
Client fixed effects	No	No	Yes	Yes
Adj. R2	.307	.457	.482	.374
N	4,585	4,585	4,585	4,569

This table presents the results from panel regressions with year-month, firm, and client fixed effects. The dependent variable is either the 3-month real return following the sale of stock j by client i in month t (columns 1 to 3) or the 6-month real return following the sale of stock j by client i in month t (column 4). We focus on trades executed between January 1920 and September 1923. The variable Local inflation is the inverse hyperbolic sine of local inflation of the town where the client lives. Internet Appendix A describes all variables used throughout the study in detail. Standard errors are double-clustered by town and month. t-statistics are provided in parentheses.

*

$p<$⁠.1;

**

$p<$ .05;

***

$p<$ .01.

5.4 Alternative explanations

Up to this point, we have established that investors buy fewer (sell more) stocks when local inflation increases. To credibly claim that these results are consistent with money illusion as in Modigliani and Cohn (1979), we need to rule out a number of alternative explanations.

5.4.1 Do investors shy away from stocks because inflation reveals information about the gloomy economic prospects of firms?

As argued by Fama (1981), inflation could proxy for economic prospects. Thus, in our setting, investors might reduce their demand for stocks with rising inflation because they expect lower growth in firms’ future cash flows. We already presented evidence inconsistent with this hypothesis. In particular, in column 5 of Table 3, we found a negative relationship between local inflation and buy-sell imbalances for stocks for the beginning of our sample period, when inflation was comparably low and economic prospects were good.

We run two additional tests to rule out this alternative explanation. First, we rerun the analysis from Table 3 but change the unit of observation. Recall that, in Table 3, the dependent variable is the buy-sell imbalance for all stocks traded by investor i in month t. As our raw data are at the transaction level, we can also compute the buy-sell imbalance for each stock j traded by investor i in month t. This enables us to saturate the regression with security-year-month fixed effects, which control for all time-varying characteristics of the security, such as changes in cash flows. We present the results in Table 8. The coefficient estimates on local inflation have magnitudes similar to those in Table 3 and stronger statistical significances. Hence, these results suggest that local inflation does not proxy for economic conditions at the firm level.

Table 8

Local inflation and individual stock trades

	Buy-sell imbalance for individual stocks${}_{i,j,t}$
	(1)	(2)	(3)	(4)	(5)	(6)
Local inflation${}_{i,t}^{}$	–0.364**	–0.570***	–0.614***	–0.514***		–0.492**
	(–2.10)	(–3.48)	(–3.90)	(–3.08)		(–2.44)
Local inflation${}_{i,t-1,t}^{}$					–0.296**
					(–2.30)
Occupied (d)${}_{i,t}^{}$						–0.312
						(–1.10)
Local unemployment rate${}_{i,t}^{}$						–7.510**
						(–2.30)
Year-month fixed effects	Yes	Yes	Yes	No	No	No
Client fixed effects	No	Yes	Yes	Yes	Yes	Yes
Security fixed effects	No	No	Yes	No	No	No
Security-year-month fixed effects	No	No	No	Yes	Yes	Yes
Adj. R2	.026	.032	.038	.330	.331	.329
N	15,189	15,189	15,189	15,189	14,986	15,051

This table presents the results from panel regressions with client and security-year-month fixed effects. The dependent variable is the buy-sell imbalance for stock j of client i in month t. We focus on the time period from January 1920 to September 1923. The variable Local inflation is the inverse hyperbolic sine of local inflation of the town where the client lives. Internet Appendix A describes all variables used throughout the study in detail. Standard errors are double-clustered by town and month. t-statistics are provided in parentheses.

*

$p<$⁠.1;

**

$p<$ .05;

***

$p<$ .01.

In a second test to examine whether our results are driven by inflation making investors more pessimistic about future economic conditions, we investigate the relation between local inflation and investment behavior around economic and political events that likely had a negative effect on economic prospects. If local inflation negatively affects clients’ beliefs about future economic conditions, we expect clients located in high-inflation areas to be more pessimistic than clients located in low-inflation areas. As a result, clients located in high-inflation areas should react less to bad news about future economic prospects than clients located in low-inflation areas, as expectations of the former are already low. Inspired by Bittlingmayer (1998), we analyze investors’ behavior around four important events between January 1920 and September 1923. Specifically, we study the announcement of the reparation amount to be paid by Germany in May 1921, the assassinations of finance minister Matthias Erzberger in August 1921 and foreign minister Walther Rathenau in June 1922, and the invasion of the Ruhr region in January 1923. We employ regression specifications similar to Equation (4). We compare investors’ response to inflation in the 6 months prior to each event to investors’ response to inflation in the 6 months after each event. Local inflation is measured as the cumulative inflation rate over the 6 months preceding the respective event. For each event, we create a dummy variable that equals one after the event. Each of the four dummy variables is interacted with the respective cumulative local inflation variable. The coefficients of interest are the ones on the interaction terms as they capture investors’ differential response to inflation around bad news. We report results in Table IA4 in the Internet Appendix. Across all specifications, the coefficient estimates on the interaction terms are never statistically significant, indicating that investors’ response to inflation is uncorrelated with these events. Taken together, the results in this subsection suggest that our main findings are not driven by increases in investors’ fear of deteriorating economic prospects of firms.

5.4.2 Do investors shy away from stocks because inflation increases risk aversion?

Next, we investigate whether clients buy fewer (sell more) stocks because higher inflation increases their risk aversion (e.g., Brandt and Wang 2003; Cohen, Polk, and Vuolteenaho 2005). The tests presented in Table IA4 in the Internet Appendix and discussed above enable us to also address this concern. Bad news most likely not only affected investors’ expectations about economic prospects but also their risk aversion. If high local inflation results in higher risk aversion, investors located in these areas should react less to bad news as their risk aversion is already high. However, as discussed above, we do not find a differential response to inflation around these events.

We perform an additional test to rule out that changes in risk aversion explain our findings. Inspired by Cohen, Polk, and Vuolteenaho (2005), we analyze whether results vary across more and less risky stocks. If inflation increases risk aversion, we expect clients to primarily divest risky stocks. We capture stocks’ riskiness by their return volatility and their market beta. We classify stocks as low-volatility (high-volatility) stocks if their return volatility over the past 6 months is equal to or below (above) the median. Moreover, we classify stocks as low-beta (high-beta) stocks if their market beta over the past 3 years is equal to or below (above) the median. We then rerun our main specification from column 2 of Table 3 separately for low-volatility and high-volatility stocks and for low-beta and high-beta stocks. Columns 5 to 8 of Table 6 present the results. Coefficient estimates neither differ significantly for low-volatility and high-volatility stocks nor for low-beta and high-beta stocks. Thus, changes in risk aversion are unlikely to explain our results.

5.4.3 Do investors shy away from stocks to finance consumption?

Next, we consider the potential concern that investors sell stocks to finance consumption. Under this alternative explanation, clients are less likely to buy (more likely to sell) stocks if local inflation increases because goods for daily consumption become more expensive.

We address this concern in two ways. First, we investigate the relation between local inflation and clients’ trades in bonds. If clients were to reduce their demand for stocks because of consumption needs, we would also expect them to reduce their demand for bonds, since bonds are inferior to stocks as a hedge against inflation. Like stocks, bonds protect against expected inflation, but unlike stocks, they do not protect against unexpected inflation. However, as suggested by Cohen, Polk, and Vuolteenaho (2005), investors suffering from money illusion do not make the same valuation mistake when they value bonds. Thus, relative to stocks, bonds become more attractive for investors subject to money illusion. To test this conjecture, we replicate the main specifications from Tables 3 and 8 for bond trades. Results are reported in Table 9. Across all specifications, we find a positive relationship between local inflation and the buy-sell imbalance. In three specifications, the effect is also statistically significant at least at the 10% level. These results suggest that clients are more likely to buy (less likely to sell) bonds in periods of high inflation. This pattern is not consistent with investors reducing their demand for stocks to finance consumption in times of rising prices. Rather, it suggests that clients reallocate funds from stocks to bonds in inflationary periods.²⁷

Table 9

Local inflation and bond trades

	Buy-sell imbalance for bonds${}_{i,t}$				Buy-sell imbalance for individual bonds${}_{i,j,t}$
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Local inflation${}_{i,t}^{}$	0.085	0.413		0.391	0.836*		1.134**
	(0.21)	(1.11)		(1.01)	(1.98)		(2.22)
Local inflation${}_{i,t-1,t}^{}$			0.473			0.759**
			(1.66)			(2.56)
Occupied (d)${}_{i,t}^{}$				0.579***			–0.147
				(7.84)			(–0.83)
Local unemployment rate${}_{i,t}^{}$				1.023			–0.176
				(0.23)			(–0.03)
Year-month fixed effects	Yes	Yes	Yes	Yes	No	No	No
Client fixed effects	No	Yes	Yes	Yes	Yes	Yes	Yes
Security-year-month fixed effects	No	No	No	No	Yes	Yes	Yes
Adj. R2	.026	.065	.075	.068	.424	.433	.424
N	4,406	4,406	4,321	4,296	5,191	5,056	5,076

This table presents the results from panel regressions with year-month and client fixed effects. In columns 1 to 4, the dependent variable is the buy-sell imbalance for bonds of client i in month t. In columns 5 to 7, the dependent variable is the buy-sell imbalance for bond j of client i in month t. We focus on the time period from January 1920 to September 1923. The variable Local inflation is the inverse hyperbolic sine of local inflation of the town where the client lives. Internet Appendix A describes all variables used throughout the study in detail. Standard errors are double-clustered by town and month. t-statistics are provided in parentheses.

*

$p<$⁠.1;

**

$p<$ .05;

***

$p<$ .01.

Second, we compare clients’ stock trading behavior in months in which they receive dividends with months in which they do not receive any dividends. If clients liquidated stocks to finance their consumption when local prices rise, we would expect them to be less likely to reduce their stock exposure when they receive dividends. To test this hypothesis, we construct a dummy variable that equals one in months in which at least one stock in a client’s portfolio pays a dividend, and zero otherwise. We then interact this dummy variable with our local inflation variable. If dividend payments alleviate financial constraints and reduce the need to sell stocks to finance consumption, the coefficient on the interaction term should be positive. We present the results in Table IA6 in the Internet Appendix. Across all specifications, the coefficient on the interaction term is negative and statistically significant. In line with previous results, this suggests that clients do not reduce their demand for stocks to finance consumption when local prices rise.

5.4.4 Do investors shy away from stocks because they invest in other asset classes?

We also consider the possibility that the negative association between inflation and buy-sell imbalances is due to clients shifting their funds from stocks into other asset classes that potentially offer a hedge against inflation. We first evaluate potential investments in foreign exchange. Trading in foreign currencies was severely restricted during our sample period. This suggests that foreign exchange most likely did not offer a viable alternative to hedge against inflation.

Nevertheless, we run two tests to rule out this alternative explanation. First, we explore the relationship between local inflation and the buy-sell imbalance for securities denominated in foreign currencies. Even though trading in foreign exchange was restricted, we observe some trades in foreign securities. We replicate the main specifications from Tables 3 and 8 using the buy-sell imbalance for foreign securities as dependent variable. Table 10 shows the results. We predominantly find negative coefficients on the local inflation variable that are never statistically significant at conventional levels. Hence, there is no evidence that clients reallocate funds from stocks to foreign securities.

Table 10

Local inflation and trades in securities denominated in foreign currencies

	Buy-sell imbalance for foreign exchange${}_{i,t}$				Buy-sell imbalance for individual foreign exchange${}_{i,j,t}$
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Local inflation${}_{i,t}^{}$	–0.301	–0.486		–0.501	–0.937		–0.966
	(–0.50)	(–0.50)		(–0.52)	(–1.12)		(–1.15)
Local inflation${}_{i,t-1,t}^{}$			0.214			–0.461
			(0.30)			(–0.56)
Occupied (d)${}_{i,t}^{}$				–0.645*			0.000
				(–1.97)			(0.00)
Local unemployment rate${}_{i,t}^{}$				2.334			6.313
				(0.32)			(0.52)
Year-month fixed effects	Yes	Yes	Yes	Yes	No	No	No
Client fixed effects	No	Yes	Yes	Yes	Yes	Yes	Yes
Security-year-month fixed effects	No	No	No	No	Yes	Yes	Yes
Adj. R2	.060	–.058	–.062	–.061	.194	.197	.196
N	1,868	1,868	1,837	1,855	1,550	1,527	1,542

This table presents the results from panel regressions with year-month and client fixed effects. In columns 1 to 4, the dependent variable is the buy-sell imbalance for securities denominated in foreign currencies of client i in month t. In columns 5 to 7, the dependent variable is the buy-sell imbalance for security j denominated in foreign currency of client i in month t. We focus on the time period from January 1920 to September 1923. The variable Local inflation is the inverse hyperbolic sine of local inflation of the town where the client lives. Internet Appendix A describes all variables used throughout the study in detail. Standard errors are double-clustered by town and month. t-statistics are provided in parentheses.

*

$p<$⁠.1;

**

$p<$ .05;

***

$p<$⁠.01.

Second, we investigate the relation between local inflation and the investment behavior of our clients around a regulatory change introduced on October 12, 1922, that essentially outlawed the use of any currency other than the Mark for all types of transactions (Verordnung gegen die Spekulation in ausländischen Zahlungsmitteln, Reichsgesetzblatt 1922, p. 796). Hence, transactions in foreign currencies became significantly more difficult, thereby reducing the set of investment opportunities and making stocks a relatively more attractive hedging instrument. If investors were actively trading foreign exchange to hedge against inflation, they should buy more (sell fewer) stocks after the regulatory change, and we would expect a more positive association between local inflation and buy-sell imbalances for stocks. This effect should be stronger for clients living in towns with higher local inflation. To test this conjecture, we again employ a regression specification similar to Equation (4). We present the results in column 1 of Table IA7 in the Internet Appendix. We do not find a significant change in the investment behavior of clients around October 1922, suggesting that clients’ trading in stocks did not change following the restrictions to trade foreign currencies.

Next, we evaluate whether clients shift assets from stocks to real estate to protect against inflation. The housing market was also highly regulated during our sample period. This resulted in negative real returns, suggesting that real estate investments did not offer protection against inflation. Nevertheless, we also run a test to rule out that investors sold stocks to acquire real estate. In March 1922, the German government introduced a new law that softened the cap on rents and increased the relative attractiveness of real estate investments (Reichsmietengesetz, Reichsgesetzblatt 1922, p. 273). If investors actively invested in real estate to hedge against inflation, they should buy fewer (sell more) stocks after the deregulation of the housing market, and we would expect a more negative association between local inflation and buy-sell imbalances for stocks after March 1922. This effect should be stronger for clients with higher inflation expectations. To test this prediction, we again employ a regression specification similar to Equation (4). Column 2 of Table IA7 in the Internet Appendix presents the results. We do not find a significant change in the investment behavior of clients around this regulatory change, suggesting that investments in real estate were also not a viable alternative to hedge against inflation. Hence, the results in this subsection do not support the conjecture that investors reduce their exposure to stocks to invest in other asset classes that offer a hedge against inflation.

5.4.5 Instrumental variables regressions

In a final test to address the concern that local inflation may be correlated with unobservable determinants of stockholdings, we run instrumental variables regressions that exploit quasi-exogenous variation in local inflation. Results from these regressions lend support to a causal interpretation of our findings. We discuss this test in detail in Internet Appendix D.

5.5 The second money illusion hypothesis

We also test for the second form of money illusion of Modigliani and Cohn (1979). This form of money illusion predicts that investors reduce their demand for stocks of firms that are exposed to increasing inflation and increasing net leverage. As discussed in detail in Internet Appendix E, we find results consistent with this prediction.

6 Conclusion

In this paper, we study the relationship between inflation and individual investors’ decision-making. There are conflicting theories on how inflation affects investors’ behavior. We test these competing hypotheses using a unique data set containing all trades of private clients of a German bank between 1920 and 1924, covering the hyperinflation. We find that investors buy fewer (sell more) stocks when local inflation rises. This effect is more pronounced for investors considered unsophisticated by the extant literature and for stocks where money illusion is more likely. Moreover, we find a positive relation between local inflation and foregone returns following stock sales. Overall, our results are in line with individual investors suffering from money illusion as in Modigliani and Cohn (1979). Additional tests indicate that our findings are unlikely to be driven by investors using local inflation as a proxy for future economic outcomes, by investors’ risk aversion increasing with local inflation, by investors liquidating stocks to meet consumption needs, or by investors shifting to other asset classes also offering a hedge against inflation.

To the best of our knowledge, our paper is the first to document empirically that individual investors’ behavior is consistent with money illusion. Thus, our results are of particular importance in light of the ongoing debate on the financial literacy of individuals. As highlighted in the introduction, individuals might not be financially literate enough to respond appropriately to the resurfacing inflation currently observed.

Acknowledgements

We would like to thank an anonymous bank for providing the data. We are grateful to Tarun Ramadorai (the editor), two anonymous referees, Rainer Baule, Juliane Begenau, Martin Brown, Carsten Burhop, Constantin Charles, Zhi Da, Ralf Fendel, Alexander Hillert, Xing Huang, Peter Koudijs, Sung Kwan Lee, Lyndon Moore, Abhiroop Mukherjee, Theo Nijman, Larissa Schäfer, Andrew Sinclair, Oliver Spalt, and Michael Weber; conference participants at the 2021 annual meeting of the Swiss Society for Financial Market Research (SGF), the 2021 Swiss Winter Conference on Financial Intermediation, the 6th European Retail Investment Conference (ERIC), the 2021 SFS Cavalcade North America, the 27th annual meeting of the German Finance Association (DGF), the 3rd European Macrohistory Workshop, the 2022 European Winter Finance Conference (EWFC), the 9th annual conference of the Asian Bureau of Finance and Economic Research (ABFER), the 2022 annual meeting of the Western Finance Association (WFA), the 10th Helsinki Finance Summit, the 2022 annual meeting of the European Finance Association (EFA), the 2022 SFS Cavalcade Asia-Pacific, and the 2023 Inquire Europe Joint Spring Seminar; and seminar participants at the Finance and History Workshop, the Quantitative History Webinar Series, the CEPR International Macro History Online Seminar, the University of St. Gallen, Tilburg University, the EUR-Economic History Group, the Dutch National Bank, Goethe University Frankfurt, the University of Bristol, Peking University HSBC Business School, the University of Zurich, and the University of Mannheim for helpful comments. Nicolas Bopp, Ellen Hazeleger, Nicolai Stübiger, Nikolaus Veith, and Alena Willi provided excellent research assistance. Financial support from the Basic Research Fund of the University of St. Gallen and Inquire Europe is gratefully acknowledged. Braggion is a CEPR fellow and an ECGI member.

Footnotes

1

See, for example, Alderman (2022), Giles (2022), and Guilford (2022).

2

See, for example, Osipovich (2020), Wursthorn, Frankl-Duval, and Zuckerman (2020), and Martin and Wigglesworth (2021).

3

In periods of low inflation, investors may not react to inflation because of rational inattention (e.g., Mankiw and Reis 2002; Sims 2003; Katz, Lustig, and Larsen 2017).

4

Existing empirical work shows that personally experienced inflation is a crucial determinant of individuals’ inflation expectations (e.g., Malmendier and Nagel 2016; D’Acunto et al. 2021).

5

Many companies issued new equity during our sample period, providing an explanation for why buy-sell imbalances are positive on average. This was driven by firms’ capital needs following the war (e.g., Aron 1927; Bresciani-Turroni 1937, p. 255).

6

Barber and Odean (2013) provide a review of the literature on individual investors’ behavior.

7

See, for example, Johnson (2021), Reinicke (2021), and McGuinness (2022).

8

There was little trading in derivatives on German stock exchanges. Official trading in derivatives was stopped completely prior to the First World War and was not resumed until the currency had stabilized (e.g., Buchwald 1924, p. 233; Schütze 1925, p. 507).

9

According to the Statistical Office of Berlin, 63% of house purchases were made by individuals living outside of Germany between September 1922 and January 1923.

10

The data on real estate prices come from Jordà et al. (2019).

11

Our buy-sell imbalance measure is based on the number of purchases and sales. In robustness tests reported in Table IA1 in the Internet Appendix, we replicate our main analysis using buy-sell imbalances based on the face value of stock trades and using clients’ stock holdings. Doing so does not materially change our findings.

12

In a robustness test reported in Table IA1 in the Internet Appendix, we replicate our main analysis using an extended sample that also includes accounts for which we cannot clearly identify the person responsible for the investment decisions. We find effects that are only marginally weaker, suggesting that the filtering of our sample does not have a meaningful impact on our findings.

13

A paying agent makes dividend, coupon, and principal payments to investors on behalf of security issuers.

14

Germany collected a wealth tax not only in 1913 but also in other years. However, there is no or only limited data available on wealth taxes collected in other years. Data on the wealth tax in 1913 come from the German Statistical Office.

15

D’Acunto et al. (2021) show that when forming inflation expectations, individuals strongly rely on experienced grocery price changes. Groceries are the most important category in the basket of goods used by the statistical office. According to a sample calculation from 1920, groceries make up approximately 80% of the basket.

16

For 92% (64%) of clients (firms), we have inflation data for the town where they live (have their headquarters). For the remaining clients (firms), the average distance between their place of residence (headquarters) and the town for which we have inflation data is 9 (5) km.

17

Stocks denominated in foreign currencies account for 18% of the holdings in foreign securities, bonds denominated in foreign currencies account for 74%, and foreign bills account for 7%. Holding and purchasing foreign exchange was difficult for German investors during our sample period. Some foreign securities were still available to them. These were mainly securities of issuers located in countries that were allied with Germany during the First World War (e.g., the countries that were part of the Austro-Hungarian Empire and the Ottoman Empire).

18

If we exclude stocks purchased in equity issues, the average buy-sell imbalance for stocks decreases from 0.18 to 0.05.

19

The inverse hyperbolic sine is an alternative to a log-transformation when a variable takes on zero or negative values. In robustness tests shown in Table IA1 in the Internet Appendix, we rerun our main analysis using raw inflation, the natural logarithm of inflation (setting months with negative inflation to zero), and inflation deciles. Our findings remain qualitatively unchanged.

20

The German Statistical Office provides data on the population of towns in 1919. Data on monthly local unemployment come from the German Employment Agency (Reichsamt für Arbeitsvermittlung) and the German Department of Labor. The German Department of Labor also provides information on the number of employees working in the paper industry in 1921. Finally, the German Central Bank reports the locations of its branches in its annual reports.

21

At the end of 1923, more than 300 paper mills worked continuously to produce bank notes (Braun 1990, p. 39). In an additional test, we run instrumental variables regressions in which we use the local availability of raw paper, proxied by the fraction of local employees working in the paper industry, as an instrument for local inflation.

22

We also estimate our main regressions using the spatial correction proposed by Conley (1999), with different thresholds (25 km, 50 km, 75 km, 100 km, 125 km, and 150 km). Results remain virtually unchanged.

23

In Figure IA9 in the Internet Appendix, we show the monthly number of applicants per 100 open positions in Germany between January 1920 and December 1924. These data come from the German Statistical Office. Unemployment only started to rise toward the end of 1922, providing evidence that the Germany economy did well in the first part of our sample period.

24

In robustness tests reported in Table IA1 in the Internet Appendix, we replicate our main analysis using alternative measures for local inflation and individual investors’ trading response as well as an extended sample. The details of these tests are described in Internet Appendix C. Across all specifications we find a negative and statistically significant relation between local inflation and investors’ demand for stocks.

25

In 1923, Germany’s neighboring countries were Austria, Belgium, Czechoslovakia, Denmark, France, the Free City of Danzig, Lithuania, Netherlands, Poland, Switzerland, and the Territory of the Saar Basin (Saargebiet). Germany had not only the highest inflation rate among its neighbors but also the highest inflation rate in the world (e.g., Hanke and Krus 2013).

26

We also investigate whether the documented effect is different for clients typically considered having sticky wages, such as civil servants and retirees. Hedging motives might be more pronounced when wages are stickier. However, we do not find evidence that our results differ for these investor types.

27

In theory, investors could also buy bonds to protect against inflation. For instance, short-term bonds can provide a hedge against inflation if interest rates adjust quickly to changes in inflation. In Table IA5 in the Internet Appendix, we replicate Table 9 separately for short-term German government bonds and all other bonds, which tend to be longer term. We find most coefficient estimates to be positive. The results are statistically stronger for longer-term bonds than for short-term bonds.