History of Savings Institutions

Savings institutions first appeared in the mid-to-late nineteenth century.^{Footnote 9} They were typically set-up by local or central governments to encourage thrift among lower social classes. Yet, they soon started attracting funds from higher social classes and even from small businesses (Vogler Reference Vogler1991; Brück Reference Brück1995; Mura Reference Mura1996; Schuster, Jaremski, and Perlman 2020; Monnet, Riva, and Ungaro Reference Monnet, Riva and Ungaro2021). They were primarily set up to promote savings, in contrast to commercial and cooperative banks, whose first aim was to develop credit. This implied specific regulations and the holding of safer assets.^{Footnote 10} Savings institutions’ deposits thus had three main advantages: they were safe (usually due to state guarantees), they were widely accessible (unlike commercial banks in most countries, savings institutions established branches in rural and sparsely populated areas), and they paid interest (unlike cash and other hoarded funds). The special status of savings institutions progressively disappeared in the second half of the twentieth century, as deposit insurance was extended to commercial banks. Starting in the late 1970s and early 1980s, most savings institutions were privatized or merged with commercial banks (Carletti, Hakenes, and Schnabel Reference Carletti, Hakenes and Schnabel2005; Bülbül, Schmidt, and Schüwer Reference Bülbül, Schmidt and Schüwer2013). In the 1920s and 1930s, however, commercial banks were essentially unregulated (U.S. banking regulation being an exception), and savings institutions’ deposits were a unique haven for precautionary savings.

Savings institutions enjoyed a privileged relationship with the State. Still, the degree of state involvement varied widely between (and sometimes within) countries. Two broad groups of institutions can be distinguished. In the first group, savings institutions were set up as state-backed institutions under the direct responsibility of the Ministry of Finance. Their deposits were explicitly guaranteed by the State. Postal savings systems are included in this category. They were created by European states at the end of the nineteenth century, based on the expansion of the postal network, and transplanted to the United States in 1911 (National Monetary Commission 1910; Kemmerer Reference Kemmerer1911). As a rule, funds deposited at these institutions were invested in government securities or deposited at the Treasury Department.^{Footnote 11} Some private institutions’ deposits were also centralized by the state. The U.K.’s Trustee Savings Banks were private institutions, but their deposits were collected by the Bank of England and invested in government bonds (Horne Reference Horne1947). Likewise, the French Caisses ordinaires d’épargne were privately owned, but, from 1837 onward, their assets were nevertheless managed by the Caisses des dépôts et consignations, a government-sponsored financial institution, and fully invested in government securities or deposits at the Treasury (Monnet, Riva, and Ungaro Reference Monnet, Riva and Ungaro2021).

The second group includes private savings institutions, which enjoyed more autonomy to manage their portfolio. They were however strictly regulated by the State, contrary to commercial banks at that time (with the exception of U.S. banking regulation). In Austria, savings banks had to report annually on their activities, and an imperial commissioner sat on their board (Lepelletier Reference Lepelletier1911). In Denmark and Norway, a supervision authority was created in 1880 and 1887 to monitor private savings banks and liquidate them if they lost more than 5 percent of deposits over a year.^{Footnote 12} In return, private savings institutions’ deposits enjoyed an explicit or implicit state guarantee. The founding act of the Spanish cajas de ahorros of June 1880, for example, stated that the cajas were “under the protection” of the government (Mura Reference Mura1996; Casals Reference Casals1991). Funds deposited at private savings institutions were primarily invested in government securities or in long-term mortgage loans. Discounting of commercial bills or other short-term loans similar to commercial bank loans were not their main business activity (see the later discussion and Table A2 in the Appendix). Starting in the 1920s, some private savings institutions became involved in commercial discounting, especially in Germany (Proettel Reference Proettel2017), but short-term loans to businesses remained the prerogative of commercial banks.

Table 2 BANKING PANICS, PRECAUTIONARY SAVINGS, AND TOTAL SAVINGS

* p<0.1 ** p<0.05 *** p<0.01

Notes: The dependent variable is the growth rate of total savings. The estimation period is 1920–1936. Except in Column (4), total savings is calculated as the sum of savings deposits, cash in circulation, and life insurance policies. In Column (4), we exclude life insurance policies from total savings. Column (2) controls for the return on equity, and Column (3) controls for the growth rate of housing prices. In the first four specifications, we use an updated version of Bernanke and James (Reference Bernanke, James and Glenn Hubbard1991) definition of banking crises (see footnote 18 and Online Appendix). In Column (5), we combine Bernanke and James’ coding with the coding of Baron, Verner, and Xiong (Reference Baron, Verner and Xiong2021). We exclude the 1922 “quiet” crisis in Japan, which was mostly a crisis of the small savings banks (Shizume Reference Shizume and Ögren2012). All estimations include country-fixed and year-fixed effects, and standard errors are clustered at the country level.

Source: See Online Appendix.

The ability to set the interest rate on deposits also varied across countries and institutions. In public savings institutions (and sometimes also in private ones), the interest rate on deposits was set directly by the Ministry of Finance, usually below market interest rates. A higher interest rate would have put considerable pressure on commercial banks. During the Great Depression, however, many governments failed to keep the rate on savings deposits in line with falling market rates (and deposit rate at commercial banks). More importantly, the risk-adjusted interest rate was clearly in favor of the savings banks once the commercial banks started to face deposit withdrawals. This generated intense criticisms on the part of commercial banks and considerable debate among the political class in many countries. Savings institutions were accused of unfair competition by commercial banks (see, e.g., Skulic (Reference Skulic1936) on Yugoslavia, Monnet, Riva, and Ungaro (Reference Monnet, Riva and Ungaro2021) on France, de Vicq and Peeters (Reference de Vicq and Peeters2022) on the Netherlands, Sissman (Reference Sissman1938), O’Hara and Easley (Reference O’Hara and Easley1979) and Shaw (Reference Shaw2018) on the United States, and Horne (Reference Horne1947) on the United Kingdom).^{Footnote 13}

New Data on Deposits in Savings Institutions

We built a new international database of deposits at savings institutions in 23 countries, covering the 1920–1936 period. Our sample includes the richest part of the world and the countries hardest hit by banking crises during the Great Depression: North America, Japan, and most European countries. Overall, the 23 countries included in our sample accounted for 65 percent of world real GDP in 1930 (Bolt et al. Reference Bolt, Inklaar, De Jong and Luiten van Zanden2018).^{Footnote 14} Data are annual. We collected the data from national statistical yearbooks and, in a few instances, from the League of Nations statistical yearbooks. Whenever possible, we corrected and improved these series with more recent estimates built by other scholars or by going back to the original annual reports published by savings institutions (see Online Appendix). Both the League of Nations and national yearbooks clearly distinguished commercial banks from savings institutions. We relied on these categories built by contemporaries, rather than on our own assessment to construct our database (see Online Appendix for a country-by-country list of savings institutions). Data on commercial banks are from similar sources or from recent scholarship when available. Other data sources for financial institutions and macroeconomic variables are listed in the Online Appendix.

For the majority of countries, higher-frequency data were not available. Data on the flow of deposits are also absent, so we have to rely on the end-of-year balance sheets. In the absence of banking regulation in most countries (Grossman Reference Grossman2010), it is also impossible to find a continuous and representative series of deposit interest rates for commercial banks.

The Evolution of Savings during the Great Depression

Savings institutions thrived during the Great Depression.^{Footnote 15} Table A1 in the Appendix shows the growth rate of deposits between 1930 and 1932—the years of the great banking panics (country-by-country plots are presented in the Online Appendix). Commercial bank deposits declined everywhere, whereas most countries experienced a strong increase in savings institutions’ deposits. As we will see later, the exceptions (6 out of 23 countries) were either countries with a major sovereign debt crisis or countries without a banking crisis. The increase in nominal deposits is all the more striking given that, as is well known, the years 1930–1932 were characterized by global deflation and falling output.

Figure 1 plots the average ratio between savings institutions’ deposits and commercial bank deposits over the 1920–1936 period for 23 countries. The average ratio increases from 71.4 percent in 1928 to 115.8 percent in 1932. This means that, on average, in our sample of countries, the early years of the Great Depression saw savings institutions’ deposits become the dominant form of deposit. Figure 1 also suggests that the increase in the ratio started before the Great Depression, as early as 1926. This increase is driven by a few countries (notably Japan and Scandinavian countries), which experienced banking instability in the mid-to-late 1920s. Excluding these countries, we find that the ratio was stable at around 35 percent from 1925 to 1928 and then increased to 45 percent in 1929, reaching 84 percent in 1932 and 95 percent in 1935.

Figure 1 RATIO OF DEPOSITS IN SAVINGS INSTITUTIONS TO DEPOSITS IN COMMERCIAL BANKS, 1920–1936

Note: Unweighted average of the data for the 23 countries in our sample.

Source: See Online Appendix.

The 1928–1933 increase in the ratio is not only due to a fall in commercial bank deposits, but also to a sharp increase in savings deposits. On average, bank deposits decreased by 14.5 percent between 1928 and 1933, while savings institutions’ deposits increased by 111 percent.

Figure 2 RATIO OF BANK DEPOSITS, SAVINGS INSTITUTIONS DEPOSITS, AND CASH IN CIRCULATION TO NOMINAL GDP, 1926–1936

Note: Unweighted average of the data for the 19 countries for which we have nominal GDP data.

Source: See Online Appendix.

In terms of GDP, the figures are equally striking (Figure 2). Note, however, that our sample is reduced to 19 countries in this case because of the lack of GDP data for several countries. Deposits in savings institutions increased from 13 to 20 percent of nominal GDP, while the share of commercial bank deposits remained constant. The cash-to-GDP ratio increased more modestly, and the increase is mostly driven by the drop in GDP. The nominal quantity of cash increased in only 6 countries out of 23.^{Footnote 16} Cash hoarding explains little of the rise in savings (contrary to the widely held belief that precautionary savings were put “under the mattress,” Fisher (Reference Fisher1932)).^{Footnote 17} As can be seen from Table A1 and from the Online Appendix (which provides country-by-country graphs), the increase in savings institutions’ deposits is especially visible in countries that experienced a banking crisis, although its extent varied from country to country.

Savings Institutions and Banking Crises

The link between the occurrence of banking crises and the rise of deposits at savings institutions is confirmed by descriptive statistics. On average, in each country in our sample, the growth rate of savings deposits was 6.5 percentage points higher during banking crisis years. If we look at the growth rate of the savings deposits to GDP ratio, the difference is 10.7 percentage points.

To code banking crises, we started with the definition of banking panics provided by Bernanke and James (Reference Bernanke, James and Glenn Hubbard1991) (also used by Grossman (Reference Grossman1994)). It covers the largest share of countries in our sample, and it is the most documented and consistent with scholarship on specific countries. In a few instances, we supplement their data using more recent studies. For France, we coded the year 1932 as a “non-crisis year” (contrary to 1930 and 1931) based on recent research by Baubeau et al. (Reference Baubeau, Monnet, Riva and Ungaro2021). For Spain, we coded the year 1931 as “crisis year” based on the work of Jorge-Sotelo (Reference Jorge-Sotelo2020). Portugal and Bulgaria are not covered by Bernanke and James, so we instead rely on Reinhart and Rogoff (Reference Reinhart and Rogoff2009) for Portugal, and on Kossev (Reference Kossev2008) for Bulgaria.

The difference between crisis and non-crisis years holds if we also include episodes of banking crises that were not associated with banking panics or bank failures (see Table 1, bottom line). These episodes were defined and named “quiet crises” by Baron, Verner, and Xiong (Reference Baron, Verner and Xiong2021). These are years when the equity prices of banks underperformed markedly compared to the rest of the stock market. Considering “quiet crises” in our sample is important since waves of bank failures were not always necessary to trigger a reallocation of savings from commercial banks to savings institutions. The Netherlands is a clear case of an increase in savings deposits without major banking panic (see Table A1 in the Appendix and de Vicq and Peeters (Reference de Vicq and Peeters2022)), but with a “quiet banking crisis” identified by Baron, Verner, and Xiong (Reference Baron, Verner and Xiong2021). So is Greece. By contrast, we see no flight-to-safety in Canada, a country that is known as a notable exception during the Great Depression because banks remained strong despite the economy being also affected by the international slump in trade and production (Bordo, Redish, and Rockoff 2015).

Table 1 shows that, in each country in the database, the growth rate of savings deposits was 6.3 percentage points higher during years of banking crises (including quiet crises), while the growth rate of the savings deposits to GDP ratio is higher by 11 percentage points.^{Footnote 18} Focusing only on the Great Depression crises (1929–1933), we find similar results.

Although not recorded in international accounts of the Great Depression (Kindleberger Reference Kindleberger1973; Bernanke and James Reference Bernanke, James and Glenn Hubbard1991; Eichengreen Reference Eichengreen1992), the increase in savings institutions’ deposits during banking crises is not surprising given the lack of financial insurance in the interwar period. The absence of public unemployment insurance and, more importantly, of financial insurance meant that consumers had to self-insure against risk. When credit dried up, financially unconstrained consumers had a strong incentive to accumulate savings as a buffer against future shocks.^{Footnote 19}

To be sure, part of the increase in savings deposits during a banking crisis was driven by portfolio reallocation (by a flight-to-safety from commercial bank deposits to safe savings institutions), rather than by an increase in precautionary savings. The last section of the article will discuss this issue extensively. This potential reallocation is almost impossible to track precisely, however. Aggregate capital stock data such as those compiled in Piketty and Zucman (Reference Piketty and Zucman2014) are not well suited to address these issues because most of the changes in financial wealth may be driven by a price effect rather than by savings flows. If stock market prices decrease and individuals do not sell their stocks, their nominal savings decrease. This decrease should not be interpreted as a reallocation toward other forms of savings. At this stage, it is sufficient to say that, for this reason, we cannot include in our study data on the stock of housings, bonds, and stocks.

Last, it is worth remembering that interwar banking crises were often independent from stock market crises (see Grossman and Meissner (Reference Grossman and Meissner2010) for a recent survey). After the U.S. stock market crash of 1929, the Great Depression was characterized by a series of banking crises in the early 1930s, not by stock market crashes. Our first graphical and statistical investigations suggest that savings institutions’ deposits strongly reacted to the uncertainty surrounding banking crises, rather than being a consequence of the 1929 international stock market crash.

Savings institutions that were privately owned often invested in mortgage loans. Contrary to business and commercial loans, they were considered very safe. We still lack a comparative history of housing and mortgage markets during the Great Depression, but—except in the United States—we are not aware of any major mortgage crisis that would have destabilized savings institutions. Starting from a similar observation, Eichengreen and Mitchener (Reference Eichengreen and Mitchener2004) hypothesized that countries where the mortgage market was primarily financed by savings institutions and where rent controls were prevalent in the 1920s escaped a construction and mortgage boom. The mortgage crisis in the United States affected Building and Loan associations, which were not considered saving banks because they were financed by equity rather than deposits. Their equity decreased when borrowers defaulted on their loans. The unique Building and Loan share installment contract encouraged borrowing members to postpone dissolution and made it almost impossible for investors to recover their funds (Fleitas, Fishback, and Snowden Reference Fleitas, Fishback and Snowden2018). At the same time, U.S. savings institutions (mutual savings and especially postal savings) attracted new funds from savers that, otherwise, could have gone to Building and Loan funds (O’Hara and Easley Reference O’Hara and Easley1979; Schuster, Jaremski, and Perlman 2020). The 1930 crisis of Building and Loan associations led to the creation of Savings and Loan institutions during the New Deal era and of the Home Owners’ Loan Corporation (Rose and Snowden Reference Rose and Snowden2013).^{Footnote 20}

Sovereign debt crises, on the contrary, had more important consequences for savings institutions since they turned government securities into risky assets. In a few countries (see Table A1 in the Appendix), a banking crisis hit, but deposits in savings institutions did not increase (in nominal terms). These were the three countries that suffered very strong public debt crises starting in 1931: Austria, Germany, and Hungary. Note, however, that the growth rate of nominal savings deposits was still much less negative than that of commercial bank deposits in these countries. The situation was especially difficult for savings institutions that had invested in government securities. Austria’s postal savings bank lost 13 percent of its deposits during the year 1931. Savings institutions that invested a lower share of their assets in public securities suffered less from debt crisis. Key examples include Germany’s public savings banks (Sparkassen), which invested a large fraction of their assets in short-term loans to the economy and mortgages (Lehmann-Hasemeyer and Wahl 2021).

Number of Depositors

How many depositors increased their savings during the crisis? If the increase in savings was driven by a small share of the population, then the increase in savings deposits might not be associated with an increase in total savings, or in the average saving rate. To discuss if the increase in savings deposits was a widespread phenomenon within the population, we collected data on the number of savings accounts for a sub-sample of 7 countries. First, it is worth looking at how many accounts were open. As can be seen from Panel (a) in Figure 3, the number of savings accounts per inhabitant in 1933 was well below 10 percent in Greece and Romania and barely above 10 percent in the United States. By contrast, the ratio of savings accounts per inhabitant was equal to more than half in Belgium and France. The differences between these countries can be explained quite easily. Greece and Romania are typical examples of countries that were still poor at that time. Savings were unlikely to be widespread in the population despite state policies to import financial institutions from the wealthiest European countries. The low number of accounts in the United States—already a rich country—can be explained by the fact that government policies to develop savings institutions have been more modest than in Western Europe. Postal savings were created there in 1911 only, and mutual savings banks were usually created through private initiatives. On the other side of the spectrum, Belgium, France, and the Netherlands were rich countries where the state had pushed the development of savings institutions since the nineteenth century, whereas commercial banks were not regulated. Italy could be included in this group, but the lower number of accounts is probably due to weaker economic and financial development than in the other three countries.

Figure 3 NUMBER OF SAVINGS ACCOUNTS AND AVERAGE SAVINGS DEPOSIT

Notes: Panel (b) focuses on the 1928–1936 banking crises. We calculate—for each country—the growth rate between the year before the first banking crisis and the year after the last banking crisis. For countries that did not have a banking crisis (here, only Greece), we take the growth rate between 1930 and 1933.

Source: See Online Appendix.

These accounts were not limited to individuals and could also be used by firms. Thus, it cannot be interpreted strictly as the number of accounts per inhabitant. Considering the existence of accounts opened by firms (although we cannot estimate their exact share), it is safe to state that less than half of the population had a savings account, even in countries like Belgium and France where savings accounts were widespread. Moreover, except in Greece, the growth of the volume of deposits per account was larger than the growth of new accounts during the crisis (see Panel (b) in Figure 3). Although there were some new entrants, the increase in savings deposits occurred significantly at the intensive margin: people who already had an account repatriated their funds from other institutions or saved more. This type of behavior shows that these people were not over-indebted or in hand-to-mouth households. The increase in savings during the Great Depression was unequal. Poor households, all the more those who faced debt repayment and/or unemployment, were unlikely to increase their deposit accounts.

Life Insurance

Life insurance policies were also an important vehicle for savings in the interwar period. They had emerged later than savings institutions but became prominent in the late nineteenth century (Radice Reference Radice1939; Goldsmith Reference Goldsmith1969; Hautcoeur 2004). We collected data on life insurance companies for 16 countries in our sample (in the remaining 7 countries, life insurance companies were either non-existent or not sufficiently organized to report aggregated data). A list of the sources used is given in the Online Appendix.

In these 16 countries, life insurance policies increased on average by 42 percent in nominal terms between 1928 and 1933, while savings deposits increased by 48 percent. Figure 4 shows that, in these 16 countries, the ratio between life insurance policies and nominal GDP increased in the early 1930s, but less than the ratio between savings deposits and nominal GDP.

Figure 4 LIFE INSURANCE POLICIES TO NOMINAL GDP (1929=100)

Notes: Unweighted average of the data for the 16 countries for which we have data on the assets of life insurers. The base year is 1929.

Source: See Online Appendix.

Caution should be exercised when using data on life insurance in international comparisons. Life insurance policies often took the form of investment accounts. The value of an investment account depends on the value of stocks and bonds in which the capitals are invested. Since both the composition of life insurance companies’ assets (e.g., government securities, corporate bonds, or stocks) and the fluctuation of the prices of these assets differed widely from one country to another during the Depression (Snowden Reference Snowden, Naomi and Daniel1995; Baker and Collins Reference Baker and Collins2003; Hautcoeur 2004), it is quite difficult to present reliable cross-country comparisons that would capture the true increase in savings and exclude valuation effects. We therefore prefer to focus mainly on savings institutions in the rest of our analysis, while checking that our main conclusions do not differ if we also consider life insurance.

Still, it is remarkable that life insurance policies increased on average by 42 percent between 1928 and 1933, while the price of stocks and junk bonds decreased and the price of safest government and corporate bonds remained stable (for data on the United States, see Basile et al. (Reference Basile, Won Kang, Landon-Lane and Rockoff2017)).

The Assets of Savings Institutions

Savings institutions lent less to private businesses than commercial banks. A larger share of their assets was invested in safe government bonds or Treasury deposits.^{Footnote 21} Nevertheless, some savings banks were involved in short-term credit markets (such as the Sparkassen in Germany, see Lehmann-Hasemeyer and Wahl (2021)), and could potentially have taken over the role of commercial banks during the crisis. Whether the flight-to-safety led to a decline in the credit multiplier therefore remains an empirical question.

To evaluate this claim, we assembled data on savings institutions’ assets using annual balance sheets, which we then compared to commercial banks’ assets. We managed to collect this data for the 23 countries in our sample, covering the 1923–1936 period. For most countries, we rely on national statistical yearbooks (see Online Appendix). These sources generally report the yearly balance sheets of each type of savings institution. We focus on two categories of assets: loans and securities. Loans include discounts of commercial paper, advances, overdraft credit, and long-term loans (mortgages, long-term loans to the State, long-term loans to agriculture, etc.). Securities include stocks and bonds (importantly, we consider sight deposits at the central bank or at the Treasury as short-term securities, to reflect their high degree of liquidity). For each savings institution, we thus compute two annual series: loans and securities. Then, we split each category into two sub-categories: private and public. The purpose of these two categories is to isolate the lending activities of savings institutions that were potentially similar to commercial banks in that they could finance private businesses.

“Private loans” are comparable to the loans granted by commercial banks (short-term loans to businesses or individuals), while “public loans” are mostly granted to central or local governments or as mortgages. We choose to include mortgages in this second category, because in most countries, the mortgage market was highly organized by the State (with specific regulations and guarantees) while commercial banks were little involved in it.^{Footnote 22} Private securities are stocks and bonds issued by businesses, while public securities are essentially government (central or regional) bonds and short-term claims on government institutions. So, according to our definition, deposits invested in public loans and securities could not replace bank credit to businesses.

For a few institutions, balance sheets are not available. This is often the case for postal savings and for savings institutions that were required by law to hold mostly or only government securities. We then rely on available institutional information to characterize their assets. For example, funds collected by the French Caisse Nationale d’Epargne (CNE) had to be deposited at the Treasury or invested in government bonds. In this case, we compute loans as equal to zero and securities as equal to the deposits of the CNE. Here, all securities are public securities (private securities are equal to zero). A country-by-country description of the choices made to construct asset series for savings institutions is provided in the Online Appendix.

For commercial banks, a standardized series of assets were published by the League of Nations.^{Footnote 23} As for deposits, we also used more recent historical studies when possible. Assets are classified into five different categories: cash, commercial bills discounted and bought, investment and securities, participations, and loans and advances. We classify bills discounted and bought, loans and advances under “loans,” and investment and securities and participations under “securities.” All banking loans are classified as “private” loans, and all securities are assumed to be private securities. This assumption cannot be verified by our sources and is very likely to be false in some countries.^{Footnote 24} Yet, it will only underestimate the private credit crunch if banks purchased government securities during the crisis.

The Credit Crunch

We expect that the total loan-to-deposit ratio (covering both commercial banks and savings institutions) decreased during the early 1930s when depositors shifted their money from banks to savings institutions. Panels (a) and (b) in Figure 5 compare the credit multiplier (loan-to-deposit ratio) of commercial banks to the aggregate credit multiplier (which also includes savings institutions). The latter is calculated as the sum of “private” loans by commercial banks and by savings institutions divided by the sum of deposits at both types of institutions.

Figure 5 FLIGHT-TO-SAFETY AND THE CREDIT MULTIPLIER

Notes: Unweighted average of the data. Panels (a) and (b) include the 23 countries in our sample. Panels (c) and (d) include the 16 countries for which we are able to distinguish between private and public securities held by savings institutions.

Source: See the text for the definition of private loans and securities.

Bernanke (Reference Bernanke1983) looked at the loan-to-deposit ratio of U.S. commercial banks to conclude that the fall in bank loans was not simply a balance sheet reflection of the decline in deposits, but was also driven by the rise in information asymmetries after bank failures.^{Footnote 25} In the same spirit, Panel (a) in Figure 5 displays the mean of this ratio in our sample of countries. By contrast, a fall in the aggregate credit multiplier (Panel (b)) also reflects a decrease in loans to private businesses driven by the flight-to-safety from commercial banks to savings institutions.

Two comments are in order. First, the pre-crisis aggregate multiplier is significantly below the commercial banks’ multiplier (80 vs. 110 percent). Second, the drop in the aggregate credit multiplier during the Great Depression outweighs the drop in the commercial banks’ multiplier. The aggregate multiplier indeed drops from 80 percent in 1930 to 65 percent in 1935 (an absolute decrease of 15 percent), while the commercial bank multiplier decreases from 110 to 100 percent (an absolute decrease of 10 percent). In relative terms, the 1935 aggregate multiplier is 19 percent below its 1930 value, while the commercial bank multiplier is only reduced by 9 percent. As shown in Panels (c) and (d) in Figure 5, this result is even stronger if we take into account private securities. This time, the aggregate multiplier drops from 92 percent in 1930 to 77 percent in 1935 (–15 percent in absolute terms), while the commercial bank multiplier only decreases from 129 to 124 percent (–5 percent in absolute terms).^{Footnote 26} The differences between multipliers are even stronger if we consider 1936 rather than 1935. Thus, a key message of Figure 5 is that the drop in the national loans-to-deposits ratio is larger if we include savings institutions.

By switching their funds from commercial banks to savings institutions, depositors therefore impaired the ability of the financial system to create credit. This is because, indeed, savings institutions provided less credit to businesses than commercial banks. Figure 6 reminds us of this fundamental difference, consistent with the pattern of the ratios in Figure 5. The average share of “private loans” in total assets of savings institutions was around 20 percent and had even slightly decreased during the 1920s, reaching 19 percent in 1929 (Figure 6, Panel (a)). This was in striking contrast with commercial banks, whose loan-to-deposit ratio was above 100 percent (loans exceeded deposits). Not only was the loan-to-deposit ratio of savings institutions low but it continued to decrease slightly during the crisis. The pattern is similar when we include private securities in Panel (b). The savings institutions did not attempt to replace the commercial banks by increasing the share of “private loans” in their balance sheet. This explains why the aggregate credit multiplier fell so much. As can be seen in Table A2, only 33 percent of the assets of savings institutions were invested in private loans similar to those of commercial banks.

Figure 6 COMPARISON OF LOAN-TO-DEPOSIT RATIOS: COMMERCIAL BANKS VS. SAVINGS INSTITUTIONS

Notes: Unweighted average of the data. Panel (a) includes the 23 countries in our sample. Panel (b) includes the 16 countries for which we are able to distinguish between private and public securities held by savings institutions.

Source: See Online Appendix.

Figure 7 highlights the heterogeneity across countries in response to banking crises. It displays the first difference between both the aggregate multiplier (black) and the commercial bank multiplier (grey) before and after banking crises (in percentage points). The first difference is thus calculated over a specific period for each country, depending on when it was hit by a banking crisis (in the case of several banking crises, the last one is considered). For countries that did not have a banking crisis, we take the growth rate between 1930 and 1933 (the years when most banking crises were concentrated in other countries).

Figure 7 CHANGE IN THE LOAN-TO-DEPOSIT RATIO DURING BANKING CRISES

Notes: Difference between the year before the first banking crisis and the year after the last banking crisis (in pp). We focus on the 1928–1936 banking crises. We calculate the difference of both ratios between the year before the first banking crisis and the year after the last banking crisis. For countries that did not have a banking crisis, we take the growth rate between 1930 and 1933. Data for Bulgaria is not available before 1932; hence, Bulgaria does not appear in the graph (Bulgaria experiences one banking crisis in 1931). For Italy, asset data for the casse di risparmio ordinarie are not available for year 1936, so we calculate the growth rate between the years 1930 and 1935 instead (Italy experiences two crises, one in 1931 and one in 1935). Finally, Romania appears as a clear outlier, with an increase in both multipliers of more than 30 percent (driven by a 50 percent fall in commercial bank deposits). In the Appendix, we present the same graph, including Romania.

Source: See Online Appendix.

It is important to note that, since the loan-to-deposit ratio was always higher for commercial banks, a decline in both ratios of the same magnitude (in pp) means that the decline in the aggregate ratio was caused by more than the decline in the commercial bank ratio.^{Footnote 27} In the U.S. case, where the fall in both ratios is of similar magnitude, a flight-to-safety to savings institutions was also at work (and the loan-to-deposit ratio of these institutions plunged), in addition to the fall in the commercial bank credit multiplier observed by Bernanke. Figure 7 shows that the change in the commercial bank multiplier was sometimes very small (or even positive), which did not prevent a large fall in the aggregate credit multiplier (Netherlands, Italy, Poland, and France). This fall was driven by the flight-to-safety. In fact, the commercial bank credit multiplier can remain stable (or even slightly increase) even if there is a commercial banking crisis. Friedman and Schwartz’s argument, for example, did not require a decrease in this ratio, but simply that loans and deposits plunged together. For several countries (Hungary, Yugoslavia, and Norway), the commercial bank multiplier even increased while commercial bank deposits fell. The aggregate multiplier also increased in such cases. Consistent with the observation and motivation of Bernanke (Reference Bernanke1983), the United States—together with Belgium—experienced the largest fall in the commercial bank credit multiplier. In all other countries, the aggregate multiplier decreased more (or increased less) than the commercial bank multiplier.

Comprehensive data on life insurance’s assets for all countries in our sample are altogether missing for the interwar years. However, the investment strategy of life insurance companies in a limited number of countries was studied by Baker and Collins (Reference Baker and Collins2003), Hautcoeur (2004), and Stalson (Reference Stalson1942). Much like savings institutions, life insurance companies invested primarily in safe assets. In France, on the eve of the Great Depression, life insurance companies, for example, held nearly 70 percent of their assets in state-guaranteed bonds and in real estate. In the United Kingdom, public sector investments accounted for 43 percent of total assets (the rest being mainly held in mortgages and shares). This share remained broadly constant during the Great Depression. Data for the United States show that about one-third of the assets of life insurers were invested in bonds in 1930 (no detail is available on the types of bonds held) and only 3 percent in stocks (see Goldsmith (Reference Goldsmith1969, vol. 1, table I5, p. 450)). The largest share was mortgage loans (40 percent). These shares remained stable in the early 1930s. It is therefore unlikely that the absence of data on life insurance companies’ assets biases our conclusion on the drop of the aggregate credit multiplier.

Why Did Savings Institutions Not Increase Lending to the Private Sector?

Most public savings institutions were required by law to invest exclusively in safe public assets (long-term government debt or deposits at the Treasury) and could therefore not replace banks as lenders to businesses and households. Thus, only a change in their legal statutes or government policy could have allowed them to lend to the private economy. However, political support for such reforms was lacking. Governments believed that using savings deposits to invest in the economy would damage the credibility of the State and those of savings institutions themselves (see, e.g., Tournié (Reference Tournié2011) and Monnet, Riva, and Ungaro (Reference Monnet, Riva and Ungaro2021) on the French debates, de Vicq and Peeters (Reference de Vicq and Peeters2022) on the Dutch ones, and Shaw (Reference Shaw2018) on those about the assets of postal savings in the United States). Parliamentarians and governments feared that giving more lending power to savings institutions would trigger runs on these institutions, similar to those experienced by commercial banks. They were also sometimes reluctant to increase the competition with the commercial banks. In the United States, after debates on whether the money deposited in postal savings should be used to finance the New Deal, the government eventually decided to create Federal lending programs mostly financed by bond issuance, rather than by savings accounts, and reinforce the banking sector by organizing deposit insurance (Shaw Reference Shaw2018).^{Footnote 28} It was only in the late 1930s, and most of the time during and after WWII, that governments in Europe (Bülbül, Schmidt, and Schüwer Reference Bülbül, Schmidt and Schüwer2013; Monnet Reference Monnet2018) or Japan (Park Reference Park2011) started to use savings institutions to direct credit to some specific sectors or firms. This post-war development coincided with a greater role for government in industrial policy and economic planning, as well as strict banking regulations that reduced the lending role of commercial banks (Monnet Reference Monnet2023).

Portugal is the only country where the government openly asked the public savings institution to take over the role of banks and lend to businesses during the Great Depression. As a result, the decline in the aggregate multiplier was small compared to the commercial bank multiplier. Starting in 1929, Portugal’s national savings bank, the Caixa Geral de Depositos, was involved in a government policy designed to modernize agriculture and industry (Lains Reference Lains2008). This is the only case in our sample where the loan-to-deposit ratio of savings institutions increased during a banking crisis. Between 1930 and 1932, the Caixa increased its credit to the economy by 58 percent, even more than the 46 percent increase in deposits. Still, counter-cyclical lending by public savings institutions remained the exception rather than the rule.

Private savings institutions enjoyed more autonomy from the state to manage their portfolios. Yet, their business model crucially depended on being perceived as a safe haven by depositors. Unlike public savings institutions, they could not always count on an explicit state guarantee to elicit depositors’ confidence. For this reason, they followed prudent lending policies (Andersson and Rodriguez Reference Andersson and Rodriguez2013; Martin-Aceña Reference Martin-Aceña2013), by keeping a relatively large share of their assets in cash or public bonds. In France, some members of the private savings institutions asked in 1931 (but did not succeed) to be given more freedom regarding asset management. But they asked for being granted the right to finance long-term safe investments, especially mortgage bonds, rather than to start lending short-term to businesses (Tournié Reference Tournié2011). From the point of view of private savings institutions, not increasing lending to businesses was therefore justified. From a macroeconomic perspective, however, it had disastrous consequences.

Attempting an Evaluation of Precautionary Savings

Figure 8 presents the evolution of total savings, calculated as the sum of commercial bank deposits, savings institution deposits, cash in circulation, and life insurance policies. This is intended to measure approximately the financial wealth of households and firms (excluding securities). The key feature is to include commercial bank deposits that decreased during the period. Thus, if all the increase in savings deposits, cash, and life insurance was driven by the fall in commercial bank deposits, this measure of total savings should not grow. If it grew at a faster rate than total income, we interpret this as evidence of precautionary savings.^{Footnote 32} This method deliberately understates precautionary savings because it assumes that the entirety of the decrease in bank deposits was driven by withdrawals. Moreover, looking at the growth of real savings also understates precautionary savings because of the “paradox of thrift” emphasized by Keynes (Chamley Reference Chamley2012; Eggertsson and Krugman Reference Eggertsson and Krugman2012): the initial increase in savings causes a fall in GDP so that both real savings and real GDP eventually grow less than their potential.

Figure 8 SAVINGS DURING THE GREAT DEPRESSION (DEPOSITS, CASH, AND LIFE INSURANCE)

Notes: Total savings is calculated as the sum of savings institution deposits, commercial bank deposits, cash in circulation, and life insurance policies. Panel (a) shows the unweighted average of the ratio total savings to nominal GDP across the 16 countries for which we have life insurance data. Panel (b) presents the evolution of the ratio of global savings to global nominal GDP. To obtain this ratio, we first convert each country’s total savings in dollars using yearly exchange rate data (see Appendix). We do the same for national GDPs. Then we add up savings to create a yearly measure of world savings, which we divide by the sum of national GDPs. The sample is the same as in Panel (a). Panel (c) plots the cross-country average of the ratio of real total savings to GDP. Unlike in Panel (a), the GDP is held constant (at its 1929 value). Panel (d) looks at real global savings. We use the same measure of world dollar savings as in Panel (b), but instead of scaling it by nominal GDP, we deflate it with the U.S. price index (constant 1929 prices). The sample is the same as in Panels (a), (b), and (c).

Source: See Online Appendix.

We first look at the ratio of savings to GDP in order to check if economic agents decreased their savings to compensate fully for the decrease in income. It is clear that they did not. In Figure 8, Panel (a) presents the cross-country average of the ratio of total savings to nominal GDP. The sample is limited to the 16 countries for which we have data on life insurance. Starting in 1929, the ratio increased very sharply. The ratio indeed jumps from 69 percent in 1929 to 79 percent in 1930, finally reaching a maximum of 102 percent in 1934. To rule out the possibility that outliers are driving this result, we also calculate the ratio of savings to GDP at the global level, that is, for all countries in our sample. To do so, we convert national series into dollars using yearly exchange rate data. We then compute a series of total savings, which we divide by total nominal GDP. This method gives larger economies a larger weight. This property is particularly interesting to capture how precautionary saving may have affected aggregate demand at the international level during the Great Depression. Panel (b) plots the ratio of global savings to global GDP. Once again, the ratio jumps from 60 percent in 1929 to 94 percent in 1932 and 117 percent in 1935.

We now look at the real growth rates of our measure of savings. Panels (c) and (d) confirm that the evolution presented in Panels (a) and (b) is not only driven by the fall in GDP during the Great Depression. While GDP and prices collapsed, total savings accumulated in financial institutions continued to grow. Panel (c) looks at the cross-country average of the real savings to GDP ratio with GDP kept constant (at its 1929 value). We still scale savings by GDP in order to obtain comparable values across countries. Yet, as savings are divided by the value of GDP in 1929 for each year, the increase in the series after 1929 cannot be due to the fall in GDP. Like in Panel (a), the ratio increased markedly in 1930 (from 69 to 84 percent) to finally reach 103 percent in 1934. It is not surprising to see that real savings (expressed as a share of 1929 GDP at constant 1929 prices) increased in the 1920s in line with real GDP growth. It is striking to see that it continued to rise in the 1930s while the world entered an economic depression and real GDP fell.

Panel (d) provides a similar picture looking at the evolution of the world’s real savings (in dollars). To obtain this series, we calculate total nominal savings by year in dollars (like in Panel (a)), which we then deflate using U.S. prices. The series is thus expressed at constant 1929 prices. As Panel (d) shows, world real savings more than doubled between 1929 and 1935 (930 to 2150 billion constant U.S. dollars). Consistent with studies that have underlined the deflationary international context (Eichengreen Reference Eichengreen1992), we thus show that the increase in precautionary savings was a global phenomenon, which means that it could affect aggregate demand across countries. All series in Figure 8 decreased markedly in 1936, when the last European countries (Gold bloc) left the gold standard.

Banking Crises and Precautionary Savings

We now test directly whether a rise in savings was associated with a banking crisis. Following the literature, the usual method is simply to regress a measure of precautionary savings on a measure of banking crisis (Mody, Ohnsorge, and Sandri Reference Mody, Ohnsorge and Sandri2012). In accordance with the previous discussion, it is important in our case to control for reallocation between different types of savings. Otherwise, we could wrongly interpret the positive coefficient on the banking crisis dummy as evidence of a positive correlation between precautionary savings and crisis, whereas it in fact captures the reallocation between different forms of savings during banking crises. Thus, we use the growth rate of the sum of savings deposits, cash, and life insurance policies on the left-hand side, and we include the growth rate of commercial bank deposits, equity return, and the growth rate of housing prices on the right-hand side. If a banking crisis triggers only a reallocation of funds from the variables on the right-hand side to those on the left-hand side, then the correlation between these variables should be negative, and the coefficient on the banking crisis dummy should equal zero. By contrast, if the coefficient on the banking crisis is positive despite the aforementioned control variables, we conclude that a crisis was associated with a rise in precautionary savings, at least for a part of the population. Finally, since our dependent variable is the nominal growth rate of savings, we control for the nominal growth rate of GDP on the right-hand side. This prevents the correlation between banking crises and savings from being driven by the direct effect of banking crises on output and prices.

(1) $$ Saving{s_{i,t}} = {\beta _0} + {\beta _{\rm{1}}}BankPanic{s_{i,t}} + {\beta _{\rm{2}}}Ban{k_{it}} + {\beta _{\rm{3}}}GD{P_{it}} + {\beta _{\rm{4}}}Equit{y_{it}} \\ + {\beta _{\rm{5}}}Housin{g_{it}} + {y_t} + {d_i} + {\varepsilon _{i,t}} $$

Savings is the growth rate of our measure of total savings (savings deposits + cash + life insurance). An alternative specification will exclude life insurance from this measure. BankPanics is a banking crisis dummy; Bank is the growth rate of commercial bank deposits; GDP is the growth rate of nominal GDP; Equity is the return on equity (calculated as the growth rate of the stock market index by Baron, Verner, and Xiong (Reference Baron, Verner and Xiong2021)); and Housing is the growth rate of housing prices.^{Footnote 33} Bank controls the reallocation away from commercial bank deposits, while Equity controls the reallocation from stocks, and Housing controls the reallocation away from real estate. All specifications include country-fixed and year-fixed effects. Standard errors are clustered at the country level. Year-fixed effects capture common global shocks, including the international contagion of the U.S. stock market crash of 1929.

Column (1) estimates our model without the control variables Equity and Housing. The sample is limited to 16 countries because of the availability of GDP and life insurance data. In Column (2), we add Equity as control variable. In Column (3), we estimate the full equation by adding Housing as control variable. Adding housing prices to the equation decreases the sample to 12 countries. Column (4) replicates the specification from Column (3), but it excludes life insurance policies from our measure of total savings. The first four specifications are estimated using an updated version of Bernanke and James’ definition of banking panics. In Column (5), we look at the effects of banking crises, which include both panics and “quiet” crises.

According to the benchmark specification, banking panics lead to a significant increase of 2.5 percentage points in the growth rate of our measure of total savings (Column (1)). The coefficient is larger (3.1 pp) when adding control variables (Columns (2) and (3)) and increases again (4.5 pp) when excluding life insurance policies from total savings (Column (4)).^{Footnote 34}

In the fifth column, we combine Bernanke and James’ coding of banking panics with the recent coding of Baron, Verner, and Xiong (Reference Baron, Verner and Xiong2021), which includes banking crises without panics. Compared to Column (3), the coefficient is slightly larger (3.5 pp vs. 3.1 pp). In line with previous discussions, considering such episodes confirms that bank runs and bank failures were not a necessary condition for banking crises to produce an increase in precautionary savings.

Did Precautionary Savings Predict Banking Crises?

A different interpretation of the results displayed in Table 2 would be that banking crises were themselves due to a drop in aggregate demand in the preceding years, driven by precautionary savings. This could have weakened the economy and hence the health of the banking system, as argued by Temin (Reference Temin1976).^{Footnote 35}

To account for this potential problem, we run a regression with a banking crisis dummy as an independent variable and the lagged value of the growth of savings deposits as an explanatory variable. If Temin’s argument were valid, a banking crisis at date T could be accurately predicted by the increase in savings at date T–1. The results presented in Table 3 invalidate this hypothesis. The coefficient on lagged savings deposits is never significant (and usually negative). In Columns (1) and (3), we consider the logarithm of savings deposits and, in Columns (2) and (4), their growth rates. In Columns (3) and (4), we add control variables (commercial bank deposits, cash, and GDP). In the Online Appendix, we consider alternative specifications where we also include life insurance with a smaller sample size or use the extended definition of banking crises. We still find no evidence that the previous values of the level or growth rate of savings predicted banking crises. This invalidates the Keynesian interpretation that the banking crises of the Great Depression were the consequence of a decline in aggregate demand in the preceding years.

Table 3 THE INCREASE IN SAVINGS DOES NOT PREDICT BANKING CRISES

* p<0.1 ** p<0.05 *** p<0.01

Notes: The dependent variable is our banking panic dummy. The estimation period is 1920–1936. All variables are lagged by one year. Columns (1) and (2) do not include any controls. In Column (3), we control by the lagged value of log bank deposits, log banknote circulation, and log nominal GDP. In Column (4), we use the growth rate of the same variables as control. All estimations include country-fixed and year-fixed effects, and standard errors are clustered at the country level.

Source: See Online Appendix.