1. Quarterly Payout and Deposit Flow

Figure 3 shows that the quarterly deposit flow of the household sector is positively correlated with the net equity payout of the nonfinancial corporate business sector during the period of 1987 to 2019. This positive correlation is not obvious a priori, as corporate equity payouts tend to be procyclical (Jermann and Quadrini (Reference Jermann and Quadrini2012), Begenau and Salomao (Reference Begenau and Salomao2018)), while deposit growth does not have a significant correlation with GDP growth and tends to increase during stock market downturns (Lin (Reference Lin2020)). To estimate the relationship between the two variables, I regress changes in household sector deposits on net payout, both scaled by lagged GDP, along with other macroeconomic factors:

(1) $$ \varDelta Deposit{s}_t/ GD{P}_{t-1}=\alpha +\beta Net\ payou{t}_t/ GD{P}_{t-1}+\gamma {X}_t+{\epsilon}_t. $$

Figure 3Quarterly Payouts and Deposit Flow, 1987–2019

Figure 3 plots the quarterly flow of household deposits (FA153020005 + FA153030005) against net corporate payout (FA106121075  $ - $  FA103164105). All values are in 2010 dollars. Data source: Financial Accounts of the United States (FAUS).

Figure 3 Long description

The X-axis represents Net payout in billion, ranging from 0 to 300. The Y-axis represents Deposit flow in billion, ranging from negative 100 to 300.

* Data Distribution: The scatter plot consists of numerous blue circular data points. While there is significant variance, the points generally cluster in an upward-sloping pattern from the bottom-left toward the top-right.

* Low Range: Between 0 and 100 on the X-axis, most deposit flows are concentrated between negative 50 and 100, with a few outliers reaching near 200.

* High Range: Between 200 and 300 on the X-axis, the density of points shifts higher, with many values falling between 50 and 250.

* Trend Line: A solid dark red linear regression line starts at approximately 10 on the Y-axis when X is near 0 and rises steadily to approximately 150 on the Y-axis when X reaches 300, indicating a positive linear relationship between the two variables.

Table 2 reports the estimation results. Column 1 reports that, without any controls, the coefficient of net payout is 0.49, suggesting that a $1 increase in net payout is associated with a $0.49 increase in deposits. The R ² of 16.4% from this univariate regression also suggests an important role of payout in explaining changes in deposits. I next add various measures of macroeconomic conditions as control variables. These include contemporaneous GDP growth, changes in stock values (scaled by lagged GDP), changes in federal funds rate, changes in credit spread, changes in inflation, changes in unemployment rate, and changes in term spread. Column 2 reports that the coefficient of net payout remains similar with the addition of these controls. Therefore, the positive relationship between corporate equity payout and deposits observed in Figure 3 cannot be attributed to fluctuations in economic, monetary, or financial conditions.Footnote ¹⁴

Table 2Quarterly Aggregate Net Payout and Deposit Flow

Table 2 Long description

A table with 5 columns labeled 1 through 5. The rows represent different financial variables and their coefficients.

* Row 1: Net payout sub t all over G D P sub t minus 1. Column 1: 0.49 (0.10). Column 2: 0.46 (0.11). Column 3: 0.49 (0.18). Column 4: 0.61 (0.16).

* Row 2: Net payout sub Comp sub t all over G D P sub t minus 1. Column 5: 0.61 (0.17).

* Row 3: Delta Stock value sub t all over G D P sub t minus 1. Column 2: -0.00 (0.01). Column 3: -0.00 (0.01). Column 4: 0.00 (0.01). Column 5: -0.00 (0.01).

* Row 4: G D P sub t all over G D P sub t minus 1. Column 2: -0.10 (0.08). Column 3: -0.12 (0.09). Column 4: 0.20 (0.11). Column 5: -0.05 (0.09).

* Row 5: Delta Fed funds rate sub t. Column 2: 0.06 (0.14). Column 3: 0.07 (0.14). Column 4: -0.18 (0.27). Column 5: 0.07 (0.14).

* Row 6: Delta Credit spread sub t. Column 2: 0.16 (0.14). Column 3: 0.18 (0.14). Column 4: 0.28 (0.24). Column 5: 0.23 (0.14).

* Row 7: Delta Inflation sub t. Column 2: -0.04 (0.08). Column 3: -0.03 (0.08). Column 4: -0.22 (0.08). Column 5: -0.01 (0.09).

* Row 8: Delta Unemployment rate sub t. Column 2: -0.17 (0.20). Column 3: -0.19 (0.20). Column 4: 0.46 (0.27). Column 5: -0.06 (0.21).

* Row 9: Delta Term spread sub t. Column 2: 0.14 (0.13). Column 3: 0.15 (0.13). Column 4: -0.25 (0.16). Column 5: 0.14 (0.13).

* Row 10: Net payout sub t minus 1 all over G D P sub t minus 2. Column 3: -0.11 (0.19).

* Row 11: Delta Deposit sub t minus 1 all over G D P sub t minus 2. Column 3: 0.11 (0.10).

* Row 12: R-squared. Values range from 0.164 to 0.337.

* Row 13: N (Sample size). Values are 132 for most columns and 131 for column 3.

The deposit measure used in the analysis above includes both categories reported in the FAUS data: i) checkable deposits and currency and ii) time and savings deposits. I next examine these two components separately. The last 2 columns of Supplementary Material Table A2 report that the positive relationship between payout and deposit flow is concentrated on time and savings deposits, which are more likely to serve as substitutes for stock investment. In contrast, the coefficient on net payout is close to zero when checkable deposits and currency are examined.Footnote ¹⁵

I next examine the dynamics of deposits following changes in payout. In column 3, the lagged payout scaled by GDP and the lagged dependent variable are added to the estimation. Deposit growth does not exhibit a significant first-order autocorrelation at the quarterly frequency. Lagged payout during quarter $ t-1 $ has an insignificant effect on deposit growth in quarter $ t $ , whereas the coefficient of payout in quarter $ t $ increases slightly. In addition, untabulated results show that when net payout from quarter $ t-4 $ to quarter $ t $ are all included, time $ t $ payout has a coefficient of 0.46 and a p-value of 0.021, whereas the coefficients of all four lagged payouts are close to zero and statistically insignificant. This indicates that deposit growth is significantly associated only with contemporaneous net corporate payout, with no subsequent reversal in the effect.

To provide further evidence on the persistence of the impact of payout on deposits, I estimate the impulse responses of deposits to innovations in payout using the local projections approach of Jordà (Reference Jordà2005). Specifically, at each horizon $ h=0,\dots, 8 $ , the log change in deposits from quarter $ t-1 $ to $ t+h $ is regressed on payout scaled by GDP in quarter $ t $ as well as its 2 lags. Figure 4 plots the coefficients of quarter $ t $ payout. An increase in payout leads to an immediate increase in deposits, and the effect persists for up to 2 years. A 1-percentage-point increase in the payout-to-GDP ratio is associated with a roughly 2% increase in deposits.

Figure 4Impulse Response of Deposits to Innovations in Corporate Equity Payout

Figure 4 plots the coefficients $ {\beta}_0^h $ s and the 90% confidence intervals from the regression, $ Ln{(deposit)}_{t+h}- Ln{(deposit)}_{t-1}=\alpha +\sum \limits_{i=0}^2{\beta}_i^h\Delta {\frac{Payout}{GDP}}_{t-i}+{\unicode{x025B}}_t $ , for $ h=0,\dots, 8 $ , where $ t $ indexes quarters. Newey–West standard errors with $ h $ order of autocorrelation are used. The sample period is from 1987 to 2019.

Figure 4 Long description

The horizontal X axis is labeled Quarter and ranges from 0 to 8 in increments of 2. The vertical Y axis is labeled L n (deposits) and ranges from 0 to 6 in increments of 2. A horizontal black baseline is set at 0.

* The primary data is represented by a thick blue dashed line starting at approximately 1.1 at quarter 0.

* The line rises to 2.2 at quarter 1 and remains relatively flat through quarter 4.

* It peaks at approximately 3.4 at quarter 6 before slightly declining to 2.6 by quarter 8.

* A light gray shaded area representing the 90% confidence interval surrounds the dashed line. This interval widens significantly over time. At quarter 0, it spans from 0.2 to 2.0. By quarter 8, the interval expands from negative 1.5 to 6.8, crossing the zero baseline around quarter 7.

The results so far show a strong positive association between net corporate equity payout and household holdings of deposits. One might wonder whether the rise in corporate payout is associated with a decline in corporate deposits, to the extent that firms use their deposit holding to fund some of the payout.Footnote ¹⁶ To address this possibility, in column 4, I replace FAUS household deposits with FDIC total domestic deposits, which include both household and corporate deposits at all FDIC-insured institutions (and are on average about 14% larger than FAUS household deposits). The coefficient of net payout is 0.61 and highly statistically significant. This finding suggests that any decline in corporate deposits is small relative to the increase in household deposits as corporate equity payout rises.Footnote ¹⁷

Lastly, the aggregate payout of nonfinancial firms in Compustat is used as an alternative to FAUS payout. Column 5 reports that the coefficient of Compustat payout scaled by GDP is 0.61 and significant at the 1% level.

2. IV Estimation Based on Large Idiosyncratic Payouts

The results in the previous section show a robust positive association between net corporate equity payout and the quantity of deposits in the economy. Of course, firms do not adjust payouts randomly and one naturally worries that the factors that lead firms to adjust payouts (cash flows, investment opportunities, taxation, etc.) could be correlated with deposits through some alternative channels. While it is difficult to establish causality using aggregate data, this section seeks to partially address concerns about omitted variables by exploiting large, idiosyncratic firm-level payouts as an instrumental variable for aggregate payouts, in the spirit of the granular instrumental variable approach of Gabaix and Koijen (Reference Gabaix and Koijen2024).

As a prominent example, in December 2004, Microsoft distributed a dividend of $33.5 billion, of which $32 billion was a one-time special dividend. This accounted for more than 50% of all dividends paid out by CRSP firms that month. Thus, Microsoft’s special dividend payout represents a significant shock to the aggregate dividend in December 2004, and it can be considered plausibly exogenous to aggregate deposit growth during that month. Importantly, not all firms increased dividends significantly that month. In fact, the median firm maintained the same amount of dividend between September and December, and the 75th percentile dividend growth rate is merely 1%.

To begin, I define a simple measure of abnormal changes in dividend payments at the firm level:

(2) $$ \Delta {Div}_{i,t}={Div}_{i,t}-{Div}_{i,t-3}\times Median{\left(\frac{\Delta {Div}_{j,t}}{Div_{j,t-3}}\right)}_t, $$

where $ Median{\left(\frac{\Delta {Div}_{j,t}}{Div_{j,t-3}}\right)}_t $ is the median dividend growth rate from month $ t-3 $ to month $ t $ of all dividend-paying firms, and is used as the benchmark for the “normal” dividend growth rate. (As noted below, the results are robust to using dividend growth at the 75th or 90th percentile as the benchmark.) I then aggregate the abnormal dividend payments of the top payers every month and use it as an instrument for the actual dividend payout in that month. In what follows, I report the results based on the top 3% abnormal dividend payments in a given month but note below that the results are robust to alternative cutoffs.Footnote ¹⁸ The total abnormal dividend payment in a given month is large when some major dividend payers raise their dividends substantially relative to other dividend-paying firms in the same month. If these large increases in dividends are idiosyncratic and uncorrelated with other determinants of aggregate deposits, as appears to be the case for Microsoft’s special dividend in December 2004, the exclusion condition will be satisfied.

Figure 5 plots seasonally adjusted aggregate monthly dividend payments by publicly traded firms in CRSP, along with total abnormal dividend payments in the top 3% for a given month, both scaled by lagged GDP. It shows that sharp increases in aggregate dividend payments, such as those in December 2004 and 2012, are frequently the result of large idiosyncratic increases in dividends by individual firms. The sharp increase in December 2004 was almost entirely driven by Microsoft’s special dividend payment discussed above, whereas the increase in December 2012 was driven by some firms’ decision to raise dividends ahead of a potential dividend income tax rate hike in 2013. However, there are also times when a sharp rise in total dividends is not accompanied by large idiosyncratic dividend payments. The precise timing of dividend payments at the monthly frequency also helps rule out alternative explanations. For example, while changes in dividend income tax rate could affect the amount of bank deposits through other channels, there is no obvious reason to expect such impacts to coincide exactly with the months when large special dividends are paid.

Figure 5Aggregate Monthly Dividend Payments and Aggregate Abnormal Large Dividend Payments

Figure 5 plots the aggregate monthly dividend payment by firms traded on NYSE, AMEX, or NASDAQ with CRSP shares codes 10 or 11 and first 2-digit distribution code (distcd) 12, and the total top 3% abnormal increases in monthly dividend payments, both scaled by lagged GDP. Abnormal change in monthly dividend payment is defined as $ \Delta {Div}_{i,t}={Div}_{i,t}-{Div}_{i,t-3}\times Median{\left(\frac{\Delta {Div}_{j,t}}{Div_{j,t-3}}\right)}_t $ , where $ Median{\left(\frac{\Delta {Div}_{j,t}}{Div_{j,t-3}}\right)}_t $ is the median dividend growth rate from month $ t-3 $ to month $ t $ of all dividend-paying firms.

Figure 5 Long description

The x-axis is labeled y m and represents time from 1987 m 1 to 2020 m 1. The y-axis ranges from 0 to 0.004.

Two data series are plotted:

* Dividend sub C R S P forward slash G D P (blue line): This series fluctuates between 0.001 and 0.002 for most of the period. It shows a significant vertical spike reaching approximately 0.0034 around 2004. After 2010, the baseline trends upward toward 0.002 with periodic sharp peaks.

* Delta Dividend sub 3% forward slash G D P (maroon line): This series remains near the 0 baseline with frequent small spikes. It features two prominent outliers: a major spike around 2004 reaching 0.0024 and a secondary spike around 2013 reaching 0.001.

The 2004 spike is the most significant feature where both lines peak simultaneously.

The model estimated is

(3) $$ {\displaystyle \begin{array}{c}\Delta {Deposits}_t/{GDP}_{t-1}=\alpha +{\beta}_1{Dividend}_t/{GDP}_{t-1}\\ {}\hskip1.5em +{\beta}_2 Other\;{Income}_t/{GDP}_{t-1}+\gamma {X}_t+{\unicode{x025B}}_t,\end{array}} $$

where $ \Delta {Deposits}_t $ is the change in monthly deposits held by all commercial banks from the Federal Reserve’s H8 releases and the main independent variable of interest $ {Dividend}_t $ is the aggregate monthly dividend income from BEA NIPA. $ Other\;{Income}_t $ is the personal income minus dividend income from the BEA. As the GDP data are not available at the monthly frequency, GDP in the previous quarter is used as the scaling variable (the results are robust to using alternative scaling variables such as lagged total monthly personal income from the BEA.) $ X $ includes other control variables, including monthly stock returns, changes in federal funds rate, credit spread, inflation, unemployment rate, and term spread, as well as month-of-the-year dummies.

Before proceeding to the IV estimation, I first report the results of the OLS estimation of equation (3). The first column of Panel A of Table 3 reports that the coefficient of dividend income is 0.74 and highly statistically significant. The magnitude of the effect is larger than the estimates from Table 2, which use aggregate quarterly payout data.

Table 3Aggregate Payout and Deposit Flow: IV Estimation

Table 3 Long description

The table is divided into two sections.

Panel A. Monthly Dividend:

* Row 1 (Dividends sub t all over G D P sub t minus 1): O L S is 0.74 with three asterisks; Second stage is 1.21 with three asterisks.

* Row 2 (Standard errors for Row 1): O L S is 0.13; Second stage is 0.36.

* Row 3 (Delta Dividend sub 3 percent comma t all over G D P sub t minus 1): First stage is 1.88 with three asterisks.

* Row 4 (Standard error for Row 3): First stage is 0.70.

* Controls: Yes for all three columns.

* R-squared: O L S is 0.161; First stage is 0.101; Second stage is 0.122.

* N: 396 for all three columns.

Panel B. Quarterly Payout:

* Row 1 (Delta Payout sub 3 percent comma t all over G D P sub t minus 1): First stage is 1.69 with two asterisks.

* Row 2 (Standard error for Row 1): First stage is 0.67.

* Row 3 (Payout sub t all over G D P sub t minus 1): Second stage is 0.92 with two asterisks.

* Row 4 (Standard error for Row 3): Second stage is 0.38.

* Controls: Yes for both columns.

* R-squared: First stage is 0.150; Second stage is 0.063.

* N: 131 for both columns.

The second column reports the first stage result of the IV estimation. It shows a significantly positive relationship between the abnormal monthly dividend payments of the top 3% and aggregate monthly dividends. The third column presents the second stage results. The instrumented dividend payout has a significantly positive effect on deposit growth. A one-dollar incremental dividend is associated with a contemporaneous $1.21 change in aggregate deposits.Footnote ¹⁹ To examine the dynamics of this effect, I replace the change in deposits from month $ t-1 $ to $ t $ on the left-hand side of equation (3) with the change in deposits from month $ t-1 $ to up to 5 months after $ t $ and rerun the IV estimation using the same right-hand side variables and the same instrument. The impulse responses plotted in Supplementary Material Figure A3 show no evidence of a reversal in the effect in subsequent months.

I next apply the same IV strategy to estimate the relationship between quarterly net payout and deposit flow, which is explored using OLS estimation in Section III.A.1. The IV is the aggregate abnormal payouts that are in the top 3% among all nonfinancial publicly traded firms in the Compustat/CRSP universe, scaled by lagged GDP. The only modification to equation (2) is the use of the 75th percentile of payout growth as the benchmark for normal payout growth, instead of the median, due to mostly negative median values. Panel B reports the first- and second-stage results, where the instrumented variable is the net payout from FAUS divided by lagged GDP. Column 2 reports that the estimated coefficient on payout is 0.92 and significant at the 5% level.Footnote ²⁰

Overall, the analysis using aggregate payout and deposit data reveals that a large fraction of the payout funds flows into the banking sector in the form of deposits. An advantage of using macro-level data is that they capture the net aggregate effects of individual behaviors and various adjustments in general equilibrium. Di Maggio et al. (Reference Di Maggio, Kermani and Majlesi2020) reported a small response of bank accounts (deposits) to changes in dividend income and a large fraction of dividends reinvested in the stock market, based on household level data.Footnote ²¹ While household-level data are useful to understand the behavior of individuals, they do not necessarily map directly into aggregate outcomes. For example, when someone reinvests dividends in the stock market—whether by purchasing shares of the same firm or of a different firm—there must be a counterparty who sells shares and receives cash, which may ultimately be deposited in the banking system. As a result, estimates based on individual responses to dividend income shocks can understate the aggregate flow into bank deposits and overstate the extent to which dividends are reinvested in equities. In fact, shareholders as a whole cannot reinvest dividends into stocks, absent net share issuances by the corporate sector.

3. Additional Results

The previous sections present evidence on how deposit quantities change with corporate equity payouts. A natural follow-up question is whether deposit rates also adjust to payout-driven fluctuations in deposit demand. To investigate this, I calculate the average interest rates banks pay on their deposits using the call report data (Acharya and Mora (Reference Acharya and Mora2015), Drechsler et al. (Reference Drechsler, Savov and Schnabl2017)) and examine the relationship between aggregate payout and the deposit spread—measured as either the difference between the federal funds rate and the average deposit rate or the difference between the 1-year Treasury yield and the deposit rate.

Supplementary Material Table A5 reports the results. Columns 1 and 2 report results from level specifications, where deposit spread is regressed on net payout (scaled by GDP) and the level of control variables. The coefficient of net payout is positive and statistically significant at the 1% level when the federal funds rate is used as the benchmark, suggesting that deposit spreads tend to be elevated during periods of high payouts. However, when the 1-year Treasury yield is used as the benchmark instead, the coefficient becomes substantially smaller and statistically insignificant.

Columns 3 and 4 report results from first-difference specifications, where changes in deposit spread are regressed on changes in net payout and controls. In both cases, the coefficient on net payout is close to zero and statistically insignificant. These findings suggest that deposit rates do not adjust quickly in response to payout-driven fluctuations in deposit demand. Overall, the response of deposit rates to corporate payouts appears muted, a pattern consistent with the well-documented stickiness of deposit rates, particularly in response to short-term fluctuations in deposit flows.

This article mainly focuses on the flow of payout funds into the banking sector in the form of deposits. However, the money could also flow into other types of financial assets, such as non-deposit fixed-income securities. To explore this possibility, I examine the correlation between net equity payout and changes in household holding of non-deposit fixed-income assets at the quarterly frequency. Supplementary Material Figure A4 plots the quarterly flow of household holding of non-deposit fixed income assets (including debt securities and money market funds) against quarterly net equity payout,Footnote ²² indicating a weakly positive relationship between the two variables.

Supplementary Material Table A6 reports the results of regressing changes in households’ holdings of non-deposit fixed-income securities on net corporate equity payout, controlling for the same variables as in Table 2. In column 1, which uses the FAUS payout measure, the coefficient on net payout is positive (0.25) but not statistically significant at conventional levels (p-value $ = $ 0.174). In column 2, which uses payout data from Compustat, the coefficient is smaller (0.15) and also statistically insignificant. These results suggest limited evidence that household holdings of debt securities respond to corporate equity payouts. The stronger and more robust relationship between corporate payouts and deposits likely reflects the larger share and broader prevalence of deposits in household financial portfolios, as discussed in Section II.C.

The focus of this article is on corporate equity payout, which has received the most attention from researchers and policymakers. However, prior research finds that aggregate equity payout and debt payout are negatively correlated, as total equity payout tends to be procyclical while debt payout is countercyclical (Jermann and Quadrini (Reference Jermann and Quadrini2012), Begenau and Salomao (Reference Begenau and Salomao2018), and Ma (Reference Ma2019)). One might wonder whether corporate debt payout exhibits any correlation with deposit flow at the aggregate level. To explore this, I construct measures of net debt issuance using data from both Compustat and FAUS. I first follow Begenau and Salomao (Reference Begenau and Salomao2018) in measuring firm-level net debt issuance as the sum of changes in quarterly long- and short-term debt from Compustat and then aggregate debt issuances of all nonfinancial firms incorporated in the United States. The quarterly total net debt issuance is then scaled by lagged GDP and added to equation (1). Column 1 of Supplementary Material Table A7 reports that the coefficient of net debt issuance is close to zero and statistically insignificant. Meanwhile, the coefficient of net equity payout remains virtually unchanged from column 2 of Table 2.

I next construct alternative measures of net debt payout using the FAUS data set. Specifically, I use the net change in total debt (including both debt securities and loans) for the nonfinancial corporate business sector. Column 2 reports that, consistent with the Compustat-based results, the coefficient on total net debt issuance remains close to zero and statistically insignificant. I next examine net debt security issuance and net loan issuance separately. Column 3 reports that net debt security issuance has a coefficient of −0.40, significant at the 10% level, suggesting that reductions in corporate bond issuance (i.e., payouts to bondholders) are associated with increases in deposits, similar to the effect observed for equity payouts. In contrast, in column 4, net loan issuance has a positive coefficient of 0.19 with a p-value of 0.179.

Lastly, as discussed in Section II.A, the main analysis ends in 2019, prior to the onset of the COVID-19 pandemic and the policy responses that introduced substantial volatility in the banking sector. Here, I extend the sample through 2024 to assess the robustness of the main findings. Column 1 of Supplementary Material Table A8 reports that when net payout data from FAUS are used, the coefficient on net payout declines from 0.46 (as reported in column 2 of Table 2) to 0.29 ( $ p $ -value = 0.052). This attenuation reflects the fact that deposit growth and corporate equity payouts—measured using FAUS data—tended to move in opposite directions during the extended period. In 2020, corporate payouts declined modestly, likely due to heightened uncertainty, whereas deposits surged in response to large-scale fiscal stimulus, unconventional monetary policy, and precautionary savings. In subsequent years, corporate payouts rebounded, but deposits fell sharply, driven by rising interest rates and outflows during the 2023 banking turmoil.

When net payouts are measured using Compustat data, however, the extended-sample results remain largely consistent with the baseline findings. Column 2 reports that the coefficient on net payout is 0.65—slightly larger than the estimate of 0.60 reported in column 5 of Table 2. This difference arises mainly because net equity payouts measured using Compustat data did not exhibit the sharp increase observed in the FAUS series during the post-pandemic period, when deposits were flowing out of the banking system.

1. OLS Estimation

As in the estimation using aggregate data, I regress the change in deposits on the amount of dividend income. Both variables are scaled by lagged county deposits, but as noted below, the results are robust to using alternative scaling variables. The model estimated is

(4) $$ \varDelta Deposit{s}_{i,t}/ Deposit{s}_{i,t-1}={\displaystyle \begin{array}{l}{\alpha}_i+{\beta}_1 Dividen{d}_{i,t}/ Deposit{s}_{i,t-1}\\ {}+\hskip2px {\beta}_2 Otherincom{e}_{i,t}/ Deposit{s}_{i,t-1}\\ {}+\gamma {X}_{i,t}+{\mu}_t+\hskip2px {\epsilon}_{i,t}.\end{array}} $$

Since deposits are measured as of June 30 of each year in the SOD data, I use the average IRS dividend income in years $ t $ and $ t-1 $ to approximate the amount of dividend received by each county between July of year $ t-1 $ and June of year $ t $ in the main independent variable. Other income is the total income from the BEA minus the IRS dividend income (averaged over years $ t-1 $ and $ t $ and scaled by lagged county deposits). Other control variables include population growth and the interaction between stock returns and the average county-level ratio of dividend income to adjusted gross income.Footnote ²³ This interaction term is included to account for the finding by Lin (Reference Lin2020) that there is a stronger negative association between deposit growth and stock returns in counties with greater stock participation (as proxied by the dividend income ratio). Controlling for this effect is potentially important because corporate equity financing and payout activities are correlated with market valuation (Baker and Wurgler (Reference Baker and Wurgler2002)). County fixed effects, $ {\alpha}_i $ , are included to control for the average county-level dividend income and deposit growth rates. Standard errors are clustered by county to allow for correlations across years within a county, as in Drechsler et al. (Reference Drechsler, Savov and Schnabl2017).

Table 4 reports the results. In the first column, where counties are weighted equally, the coefficient of dividend income is 0.29 and significant at the 1% level, indicating that a $1 increase in dividend income is associated with a $0.29 increase in deposits at the county level.Footnote ²⁴ When the regression is weighted by a county’s average population in column 2, the point estimate increases notably to 0.60. The marked difference between the weighted and unweighted estimates indicates potential heterogeneity in the effect across counties of different sizes (Solon, Haider, and Wooldridge (Reference Solon, Haider and Wooldridge2015)). To explore this, the estimation is performed separately for counties in the largest 5% and the remaining 95% based on the average population during the sample period. As reported in columns 3 and 4, the effect is indeed more pronounced in large counties, with a point estimate of 0.81, compared with 0.25 for smaller counties, and this difference is significant at the 5% level. One possible explanation for the stronger impact in larger counties is that non-reinvested dividend income is more likely to flow into bank branches located in major cities.

Table 4Dividend Income and Deposit Flow at the County Level

Table 4 Long description

The table consists of 6 columns labeled 1 through 6. Column 3 is sub-headed Top 5 percent and Column 4 is sub-headed Bottom 95 percent.

Row 1: Dividends sub t all over Deposits sub t minus 1. Coefficients are 0.29 in column 1, 0.60 in column 2, 0.81 in column 3, 0.25 in column 4, and 0.65 in column 6. Standard errors in parentheses are 0.07, 0.15, 0.22, 0.07, and 0.13 respectively.

Row 2: Dividends sub t all over Pop sub t minus 1. Column 5 shows a coefficient of 0.71 with a standard error of 0.17.

Row 3: Other income sub t all over Deposits sub t minus 1. Coefficients are 0.03 in column 1, 0.02 in column 2, 0.03 in column 3, 0.03 in column 4, and 0.02 in column 6. Standard errors are 0.00 or 0.01.

Row 4: Other income sub t all over Pop sub t minus 1. Column 5 shows a coefficient of 0.02 with a standard error of 0.01.

Row 5: Population growth sub t. Coefficients range from 0.52 to 1.24 in columns 1 through 4 and 6, with a much higher 14.38 in column 5.

Row 6: Div ratio times ret sub t. Coefficients are negative across all columns: minus 1.13, minus 1.61, minus 1.83, minus 1.01, minus 47.81, and minus 1.33.

Fixed Effects and Controls:

* Year F E: Yes for columns 1 through 5, No for column 6.

* County F E: Yes for all columns.

* State times Year F E: No for columns 1 through 5, Yes for column 6.

* Population weighted: Yes for columns 2, 5, and 6.

Summary Statistics:

* R-squared: Ranges from 0.161 in column 4 to 0.352 in column 6.

* N: Approximately 76,485 for most columns, except column 3 which has 3,845 and column 4 which has 72,641.

The coefficient of non-dividend income is also statistically significant, ranging from 0.02 to 0.03. This suggests that a $1 increase in non-dividend income is associated with an increase in deposits by 2–3 cents, which is consistent with the personal savings rate during this period and the share of savings in deposits. In addition, the interaction between the dividend income ratio and stock returns is significantly negative, confirming the findings in Lin (Reference Lin2020).

I next show the robustness of the results to alternative scaling variables. Column 5 reports the results where deposit flow, dividend income, and non-dividend income are scaled by lagged county population. The coefficient of dividend income is 0.71, larger than but not statistically different from the estimate in column 2.Footnote ²⁵

In column 6, state $ \times $ year fixed effects are included to account for any unobserved common shocks at the state level. The point estimate increases slightly to 0.65 from 0.60 in column 2, suggesting that within the same state, deposits also grow faster in places where residents receive more dividends. Overall, the evidence from the county-level estimation provides a strong indication that the positive relationship between aggregate payout and deposit flow documented in Section III.A is not simply driven by confounding macro factors omitted from the estimation.

Lastly, one might wonder about the timing of the impact of dividend income on deposits at the county level. To explore this, I examine deposit growth around instances of large and sudden changes in county dividend income.Footnote ²⁶ Specifically, for each county, I calculate the mean and standard deviation of changes in the amount of dividends (scaled by population), and define a large increase in dividends as any change exceeding three standard deviations above the county average.Footnote ²⁷ A total of 133 such large changes are identified, which spread across the entire sample period with a higher concentration in the periods of 2004–2006 and 2010–2012. I then regress the growth rate of deposits on indicators for such large increases, as well as 2 leads and lags of the indicator variable. The estimation includes the same control variables as in equation (4) and county and year fixed effects. Figure 6 plots the coefficient estimates and the 95% confidence intervals. Counties with large dividend increases experience significantly higher growth in deposits during the year of the dividend increase. However, their deposit growth rates are statistically similar to other counties both before and after the large dividend increase. The results suggest that a sharp increase in dividends leads to a contemporaneous increase in deposits, with no evidence of subsequent reversal.

Figure 6Dynamic Effects of Large Dividend Increases on Deposit Growth

Figure 6 plots the coefficients and the 95% confidence intervals of a dummy variable indicating large increases in county dividend ( $ T=0 $ ) and its 2 leads and lags from a regression of county deposit growth on these indicator variables, control variables, and county and year fixed effects.

Figure 6 Long description

A line graph with a vertical Y-axis labeled Deposit growth ranging from negative 0.02 to 0.03 in increments of 0.01. The horizontal X-axis is labeled Year to large dividend increases and contains five discrete time points: negative 2, negative 1, T equals 0, 1, and 2.

Data is represented by blue circular points with vertical error bars representing 95 percent confidence intervals.

* At year negative 2, the point is slightly above 0 with error bars extending from approximately negative 0.005 to 0.01.

* At year negative 1, the point is slightly below 0 with error bars extending from approximately negative 0.01 to 0.006.

* At year T equals 0, the point shows a sharp increase to approximately 0.013, with error bars extending from 0.002 to 0.025.

* At year 1, the point drops back to slightly below 0, with error bars from negative 0.01 to 0.008.

* At year 2, the point rises slightly above 0, with error bars from negative 0.006 to 0.015.

2. IV Estimation

One concern with the OLS estimation is that county-level dividend income may be correlated with unobserved local economic shocks that also affect deposit growth. For instance, a positive local economic shock, not captured by non-dividend income or population growth, could simultaneously increase both deposits and dividend income (e.g., from locally headquartered public companies and private C corporations), resulting in an upward bias in the OLS estimates. To address this concern, I adopt an IV estimation strategy, which resembles a shift-share approach. The instrument for the actual dividend income is the projected dividend income, calculated based on a county’s lagged dividend income and the aggregate dividend income growth:

(5) $$ Dividend\_{proj}_{i,t}={Dividends}_{i,t-2}\times \frac{\sum {Dividends}_{i,t}}{\sum {Dividends}_{i,t-2}}. $$

Similar to the measurement of dividend income in equation (4), the amount of total dividend in equation (5), $ \sum {Dividends}_{i,t} $ , is the average of the total dividends in year $ t $ and $ t-1 $ . County dividend share is lagged by 2 years in equation (5) to ensure that the IV does not contain dividends from year $ t-1 $ , which is part of the instrumented variable. As in the OLS estimation, county fixed effects are included to control for the cross-county variation in dividend income share. The intertemporal variation in projected dividends within counties is largely driven by variations in the amount of aggregate dividend payments, which can be considered exogenous to a given county. As discussed below, the results are robust to using a county’s average dividend share to calculate projected dividends in equation (5), in which case all of the within-county variation in projected dividends is driven by variations in the amount of aggregate dividend income.

The key identifying assumption of the IV estimation is the absence of confounding macroeconomic shocks that are correlated with aggregate payouts and affect deposit flows in the same heterogeneous manner as dividend payouts. Therefore, the main threat to identification is the presence of such shocks. For example, there may be concerns that counties with large exposure to dividend payments are also more sensitive to overall economic conditions. Given that aggregate corporate equity payouts are typically procyclical, the faster deposit growth in these counties during high-payout years could potentially stem from their greater exposure to aggregate economic activities, which would represent a violation of the share exogeneity assumption in shift-share designs (Goldsmith-Pinkham, Sorkin, and Swift (Reference Goldsmith-Pinkham, Sorkin and Swift2020)). While this is a valid concern, it is important to note that the estimation controls for both non-dividend income and the interaction between stock market exposure and stock returns. In fact, as shown in Lin (Reference Lin2020), this interaction term is significantly negative, indicating a slower deposit growth in high-exposure counties during good times.Footnote ²⁸

Columns 1–3 of Table 5 report the first-stage results. Column 1 reports that with county and year fixed effects and no other controls, the coefficient of projected dividends is 0.61 and highly statistically significant, suggesting that a $1 increase in projected dividends is associated with a $0.61 increase in actual dividend income at the county level. The R ² of 92.7% indicates that nearly all of the variation in county dividend income can be explained by the projected dividend income and county and year fixed effects. In column 2, including control variables reduces the coefficient of projected dividend to 0.55, but it remains highly statistically significant. In column 3, when counties are weighted by population, the coefficient increases slightly to 0.63.

Table 5Dividend Income and Deposit Flow at the County Level: IV Estimation

Table 5 Long description

The table is divided into two main sections: First Stage (columns 1 to 3) and Second Stage (columns 4 and 5).

Variables and coefficients:

* Dividends sub Proj t all over Deposits sub t minus 1: First stage coefficients are 0.61, 0.55, and 0.63, all significant at the 1 percent level. Standard errors are 0.01, 0.01, and 0.02.

* Dividends sub t all over Deposits sub t minus 1: Second stage coefficients are 0.45 and 0.67, both significant at the 1 percent level. Standard errors are 0.10 and 0.17.

* Other income sub t all over Deposits sub t minus 1: Coefficients range from 0.00 in the first stage to 0.03 and 0.02 in the second stage.

* Population growth: Coefficients are 0.01 and 0.04 in the first stage, and 0.55 and 0.77 in the second stage.

* Div ratio times ret: Coefficients are 0.16 and 0.26 in the first stage, and minus 1.17 and minus 1.65 in the second stage.

Fixed Effects and Controls:

* Year F E: Yes for all columns.

* County F E: Yes for all columns.

* Population weighted: No for columns 1, 2, and 4; Yes for columns 3 and 5.

Summary Statistics:

* R-squared: 0.927, 0.932, and 0.949 for the first stage; 0.058 and 0.045 for the second stage.

* N: 76,464 for the first stage and 76,463 for the second stage.

Columns 4 and 5 report the second-stage results for the unweighted and population-weighted estimations, respectively, incorporating the same controls as in Table 4. The coefficients of dividend income are 0.45 and 0.67, respectively, both slightly larger than the corresponding OLS estimates.Footnote ²⁹ These results further substantiate a positive causal effect of dividend income on deposits at the county level.

3. Within-Bank Estimation

I next conduct a within-bank estimation of the relationship between dividend income and deposit growth across a bank’s branches located in different counties. This estimation helps further rule out some alternative explanations of the results, such as that dividend income is positively correlated with unobserved local demand shocks. This is because banks respond to loan demand shocks at the institutional level, which can lead to relatively uniform deposit growth across branches in different counties, depending on where it is most cost-effective to raise funding. In contrast, differential shocks to household deposit demand in counties where a bank operates will cause observed branch deposits to grow differentially across counties.

Because I do not observe the amount of dividend income at the bank-county level, I create a proxy measure by multiplying a county’s total dividend income by a bank’s average share of deposits in the county. The average deposit share is used to ensure that the amount of dividend assigned to a bank is not affected by changes in a bank’s market share in a county, thus preventing any mechanical correlation between the dividend measure and the dependent variable. Similarly, non-dividend income is calculated by multiplying the total non-dividend income of a county by a bank’s deposit share.

Panel A of Table 6 reports the results. To facilitate the comparison of the effects, column 1 first presents the estimation results without controlling for bank $ \times $ year fixed effects but using only multicounty banks. The point estimate is 0.5 and significant at the 1% level. In column 2, when bank $ \times $ year fixed effects are added to remove common shocks at the bank level, the point estimate becomes 0.46 and remains highly significant. The results suggest that, within the same bank, deposits also grow faster at branches located in areas receiving large dividend income. This within-bank identification strategy provides further evidence that payouts influence deposit growth through a shift in deposit demand.

Table 6Dividend Income and Deposit Flow: Within-Bank and Zip-Code-Level Estimation

Table 6 Long description

Panel A presents Within-Bank Estimation with two columns of results.

* The first row shows the variable Dividends sub i,b,t all over Deposits sub i,t-1 with coefficients of 0.50 and 0.46, both significant at the 1 percent level. Standard errors are 0.09 for both.

* The second row shows Other income all over Deposits sub t-1 with coefficients of 0.04 and 0.04, both significant at the 1 percent level. Standard errors are 0.00.

* The third row shows Population growth sub t with coefficients of 0.06 and 0.02, significant at the 1 percent and 10 percent levels respectively. Standard errors are 0.02 and 0.01.

* Fixed effects include Year F E, Bank times County F E, and Bank times Year F E.

* R-squared values are 0.173 and 0.363. N is 403,574 and 403,570.

Panel B presents Zip-Code-Level Estimation with two columns of results.

* The first row shows Dividends sub t all over Deposits sub t-1 with coefficients of 0.29 and 0.34, both significant at the 1 percent level. Standard errors are 0.03.

* The second row shows Other income all over Deposits sub t-1 with coefficients of 0.02 and 0.02, both significant at the 1 percent level. Standard errors are 0.00.

* The third row shows Div ratio times ret with coefficients of negative 0.76 and negative 0.67, both significant at the 1 percent level. Standard errors are 0.18 and 0.20.

* Fixed effects include Year F E, Zip code F E, and County times Year F E.

* R-squared values are 0.267 and 0.358. N is 317,503 and 309,128.

4. Zip-Code-Level Estimation

In this section, I conduct a within-county estimation exploiting data on deposit growth and dividend income at the zip-code level. To do this, I match zip-code branch deposits from the FDIC SOD to zip-code level dividend income data from the IRS SOI.Footnote ³⁰ The estimation uses county $ \times $ year fixed effects to control for common shocks at the county level and compares how deposit growth in different zip codes of the same county varies as a function of the zip-code level dividend income. The identification assumption is that, while deposits tend to be localized (people tend to put money in nearby branches), any effects of confounding local shocks, such as local productivity or loan demand shocks, should be more spread out across the whole county. Because the zip-code level dividend income data are only available since 2004, the sample period is from 2004 to 2019.

Panel B of Table 6 reports the results of the estimation. Column 1 reports that, without $ county\times year $ fixed effects, the coefficient of dividend income is 0.29 and is significant at the 1% level. With $ county\times year $ fixed effects in column 2, the estimate increases slightly to 0.34, suggesting that deposit growth is faster when a zip code receives more dividend income, compared with other zip codes located in the same county. This finding helps further alleviate concerns that unobserved local economic shocks are driving the results at the county level.

1. Bank-Level Results

A caveat to this analysis is that the exact amount of dividends received by the depositors of each bank is not directly observed and must be inferred from the bank’s deposit share within local areas where dividend income data are available. This imputation could introduce nontrivial measurement errors due to varying clientele of stock investors across banks. To partially alleviate this issue, I use bank deposit share and dividend data at the zip code level to infer the amount of dividends received by depositors of each bank, based on the presumption that residents in the same zip codes are more homogeneous in terms of income levels and stock market exposure. I calculate the imputed dividends using each bank’s average deposit share within each zip code to prevent any mechanical correlation between the imputed dividend and changes in bank balance sheet variables. The estimated dividends are then aggregated across all zip codes where the bank has branches. The estimation controls for the growth in the number of zip codes where a bank has depository branches.

The model estimated is

(6) $$ {\displaystyle \begin{array}{c}\Delta {Y}_{i,t}/{Assets}_{i,t-1}={\alpha}_i+{\beta}_1{Dividend}_{i,t}/{Assets}_{i,t-1}+{\beta}_2 Other\;{income}_{i,t}/{Assets}_{i,t-1}\\ {}\hskip1.5em +\hskip.4em {\beta}_3{X}_{i,t}+{\beta}_4{Z}_{i,t-1}+{\mu}_t+{\unicode{x025B}}_{i,t},\end{array}} $$

where the dependent variable is the change in the amount of deposits or loans from year $ t-1 $ to year $ t $ , scaled by lagged total assets. $ Dividend $ and $ Other\ income $ are the imputed dividend income and non-dividend income, respectively, scaled by lagged assets. $ X $ is a vector of region-level control variables, including lagged logs of deposit-weighted income per capita and population and the growth in the number of zip codes where a bank has branches. $ Z $ is a vector of lagged bank-level controls, including log total assets, the income-to-asset ratio, the equity-to-asset ratio, and the deposit-to-asset ratio. $ {\mu}_t $ denotes year fixed effects and $ {\alpha}_i $ denotes bank fixed effects.

Panel A of Table 7 reports the results. Column 1 confirms that the positive impact of dividend income on deposits is also observed at the bank level.Footnote ³² Column 2 reports the estimation results for total loans. The coefficient estimate is 0.66 and significant at the 1% level, indicating that the increase in deposits due to rising depositor dividend income is associated with an expansion in bank lending.

Table 7Dividend Income, Deposits, and Bank Lending

Table 7 Long description

Panel A: Bank-Level Estimation.

Columns 1 through 5 represent Delta Deposits, Delta Loan, Delta R E Loan, Delta C I Loan, and Delta Securities.

- Row 1: Dividends sub t all over Assets sub t minus 1. Coefficients are 0.52, 0.66, 0.52, and 0.12 (all significant at the 1 percent level) for columns 1-4, and negative 0.05 for column 5. Standard errors in parentheses are 0.10, 0.09, 0.07, 0.02, and 0.04 respectively.

- Row 2: Other income sub t all over Assets sub t minus 1. Coefficients are 0.02, 0.01, 0.01, 0.00, and 0.01 (all significant at the 1 percent level). Standard errors are all 0.00.

- Fixed Effects and Controls: Controls, Year F E, and Bank F E are all marked Yes for all columns.

- Statistics: R-squared values range from 0.137 to 0.426. N equals 90,515 for all columns.

Panel B: Bank-County-Level Estimation.

Columns 1-2 represent Mortgages; columns 3-4 represent Small Business Loans.

- Row 1: Dividends sub t all over Assets sub t minus 1. Coefficients are 0.0044, 0.0049, 0.0121, and 0.0120 (all significant at the 10 percent level). Standard errors are 0.0026, 0.0028, 0.0062, and 0.0063.

- Row 2: Other income sub t all over Assets sub t minus 1. Coefficients for Mortgages are 0.0004 and 0.0003 (significant at the 1 percent level); for Small Business Loans, coefficients are 0.0000. Standard errors are all 0.0001.

- Fixed Effects: Bank times County F E is Yes for all. County times Year F E is Yes for columns 2 and 4. Year F E is Yes for columns 1 and 3.

- Statistics: R-squared values range from 0.750 to 0.910. N ranges from 469,065 to 734,957.

It is important to note that any effect on lending may reflect both credit supply and credit demand. Loan demand may also rise in areas experiencing an increase in dividend income through local general equilibrium effects. For example, beyond the direct flow-of-funds channel emphasized in this article, dividend income may influence the real economy through other mechanisms—such as a propensity to consume out of dividend income (e.g., Baker et al. (Reference Baker, Nagel and Wurgler2007), Di Maggio et al. (Reference Di Maggio, Kermani and Majlesi2020)). Such spending responses could increase loan demand, prompting banks to meet funding needs by raising deposits. While the within-bank and within-county specifications in Table 6 and the results in the next section suggest that loan demand is unlikely to be the primary driver of the relationship between dividend income and changes in bank deposits and loans, the bank-level estimates may partially reflect this effect.

I next examine different loan types separately. Column 3 reports that the majority of the increase in lending is attributed to loans secured by real estate, which includes commercial and residential mortgages, as well as business loans secured by real estate. Column 4 reports that a smaller but still significant increase is observed for other commercial and industrial loans.

Lastly, column 5 examines banks’ holdings of securities. The results show that an increase in dividends is associated with a modest decline in securities that is marginally insignificant (p-value $ = $ 0.105).

2. Bank-County-Level Results

As discussed above, the positive association between dividends and bank lending may reflect the influence of both credit demand and credit supply. This section uses bank-county-level lending data and a within-county estimation approach to account for credit demand. By including county-by-year fixed effects, the analysis compares lending across banks operating in the same county, thereby controlling for local demand shocks.

I begin by examining mortgage lending using HMDA data. The model estimated is analogous to equation (6):

(7) $$ {\displaystyle \begin{array}{c}{Y}_{i,c,t}/{Assets}_{i,t-1}={\alpha}_{i,c}+{\beta}_1{Dividend}_{i,t}/{Assets}_{i,t-1}+{\beta}_2 Other\;{income}_{i,t}/{Assets}_{i,t-1}\\ {}\hskip1.5em +\hskip.4em {\beta}_3{X}_{i,t}+{\beta}_4{Z}_{i,t-1}+{\mu}_{c,t}+{\unicode{x025B}}_{i,c,t},\end{array}} $$

where the dependent variable is the amount of originated mortgages in year $ t $ , scaled by lagged bank assets. The specification includes county-by-year fixed effects $ {\mu}_{c,t} $ and bank-county fixed effects $ {\alpha}_{i,c} $ . All other variables are the same as those in equation (6).

To facilitate comparison, column 1 of Panel B of Table 7 first reports the results without county-by-year fixed effects. The estimated coefficient is 0.0044 and is statistically significant at the 10% level. Column 2 includes county-by-year fixed effects to control for local credit demand shocks. The coefficient increases slightly to 0.0049 and remains statistically significant at the 10% level.

I next repeat the analysis using small business loans data, replacing the dependent variable with the amount of originated small business loans (scaled by lagged bank assets) and re-estimate equation (7). In columns 3 and 4, the point estimates are nearly identical without and with county  $ \times $  year fixed effects and are statistically significant at the 10% level in both cases.

Taken together, while the evidence is modest, the similarity of point estimates with and without county-year fixed effects suggests that the relationship between dividends and lending is unlikely to be explained solely by differential demand shocks across banks.

Overall, the findings from the bank-level analysis suggest that funds paid out by corporations with excess capital are partially channeled through the banking system to its borrowers. Many of these borrowers, including households and small businesses, rely on banks to fund their consumption and investment activities. Restrictions on payout could potentially exacerbate the financial constraints faced by these economic agents.