Dividend signaling and bank payouts in the Great Financial Crisis

Juelsrud, Ragnar Enger; Nenov, Plamen T. Working Paper Dividend signaling and bank payouts in the Great Financial Crisis Working Paper, No. 9/2022 Provided in Cooperation with: Norges Bank, Oslo Suggested Citation: Juelsrud, Ragnar Enger; Nenov, Plamen T. (2022) : Dividend signaling and bank payouts in the Great Financial Crisis, Working Paper, No. 9/2022, ISBN 978-82-8379-250-8, Norges Bank, Oslo, https://hdl.handle.net/11250/3031785 This Version is available at: https://hdl.handle.net/10419/298430 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/4.0/ Dividend Signaling and Bank Payouts in the Great Financial Crisis NORGES BANK RESEARCH 9 | 2022 RAGNAR E. JUELSRUD PLAMEN T. NENOV WORKING PAPER NORGES BANK WORKING PAPER XX | 2014 RAPPORTNAVN 2 Working papers fra Norges Bank, fra 1992/1 til 2009/2 kan bestilles over e-post: [email protected] Fra 1999 og senere er publikasjonene tilgjengelige på www.norges-bank.no Working papers inneholder forskningsarbeider og utredninger som vanligvis ikke har fått sin endelige form. Hensikten er blant annet at forfatteren kan motta kommentarer fra kolleger og andre interesserte. Synspunkter og konklusjoner i arbeidene står for forfatternes regning. Working papers from Norges Bank, from 1992/1 to 2009/2 can be ordered by e-mail: [email protected] Working papers from 1999 onwards are available on www.norges-bank.no Norges Bank’s working papers present research projects and reports (not usually in their final form) and are intended inter alia to enable the author to benefit from the comments of colleagues and other interested parties. Views and conclusions expressed in working papers are the responsibility of the authors alone. ISSN 1502-8190 (online) ISBN 978-82-8379-250-8 (online) Dividend Signaling and Bank Payouts in the Great Financial Crisis * Ragnar E. Juelsrud  Plamen T. Nenov  Abstract We study the dividend payouts of U.S. banks during the 2008 financial crisis. Using a difference-in-differences methodology, we shows that banks with higher share of short-term liabilities to total liabilities, which were thus more exposed to the rollover crisis that took place in 2008, increased their dividend payouts relative to less exposed banks. This relative increase in dividend payouts is concentrated in relatively cash-rich banks. The dividend payout increase was associated with a short-run increase in stock valuations. We argue that this front-loading of dividends of more exposed banks is consistent with a theory of dividend payouts, in which the payout policy has a (short-run) stabilizing role on the bank’s liquidity position by signaling information to short-term lenders about the bank’s available liquidity. Key words: payout policies, dividend signaling, rollover crises, bank runs, liquidity crises. JEL Codes: G01, G21, G35 * This paper should not be reported as representing the views of Norges Bank. The views expressed are those of the authors and do not necessarily reflect those of Norges Bank. We thank seminar participants at Norges Bank and at NBRs Workshop on Financial Intermediation for useful comments and suggestions.  Norges Bank, e-mail: [email protected].  Norges Bank, e-mail: plamen.neno[email protected]. 1 Introduction The dividend payout policies of banks during the Great Financial Crisis of 2007-2008 have served to provide important insights to both researchers and policy-makers about bank behavior under severe liquidity stress. The large increase in payouts in the midst of the crisis, was contemporaneously viewed as excessive by many commentators, and has shaped postcrisis financial stability policies, tying payouts to hard measures of bank balance sheet health, assessed via objective stress test criteria set forward by regulating authorities. The main explanation for the observed bank payout policies during the financial crisis has been an agency problem between bank insiders and equity holders and outside debt holders (see, for example, Scharfstein and Stein (2008)). The misaligned incentives between these two groups became exacerbated after the arrival of bad news about the banks’ loan portfolios, tied to the housing market, which in turn lead to insiders shifting risk onto debt holders by reducing bank equity via increased payouts. A complementary story focuses on the severe debt rollover problems that some banks experienced during the crisis, and in particular during its most acute phase in 2008. In that period many banks and non-bank financial institutions were exposed to counterparties either refusing to roll-over maturing short-term debt or terminating their trading relationships. In that situation, a bank’s dividend payout policy may start having an important (short-run) stabilizing role on the bank’s liquidity position by conveying valuable information to short-term lenders about the bank’s available liquidity (Juelsrud and Nenov, 2020). In this paper we empirically investigate the role this dividend signaling channel may have played in U.S. banks’ dividend payout policies during and after the Great Financial Crisis. This episode is unique for understanding the underlying incentives of banks and their attempts to manage a liquidity crisis, since bank dividend payouts have been heavily regulated in its aftermath, so any analysis of subsequent crisis episodes is made substantially harder by the regulatory framework in place. Consistent with the signaling channel, we show, using a difference-in-differences methodology, that banks with higher reliance on shortterm debt (which we also refer to as “short-term leverage” below) increased their dividend payouts relative to banks with lower short-term leverage during 2008, and moreover, this relative increase in dividend payouts was concentrated in relatively cash-rich banks. This was associated with a relative short-run increase in stock valuations of these cash-rich banks. However, this behavior around the crisis was associated with a persistent decline in dividend payouts and stock prices in the post-crisis period. We start our analysis with a stylized model of intertemporal dividend payout choice in the midst of a rollover crisis with a short-run signaling role of dividends as in Juelsrud and 1 Nenov, 2020, which tends to elevate the optimal dividend payout rate. In the model a bank (owner) decides on the optimal duration of maintaining elevated dividend payouts by trading off the increase in expected payoff from“weathering the storm”(i.e. withstanding the rollover crisis) against the long-run impact a large draw-down of liquidity has on the bank’s franchise value. We show that, relative to banks with lower short-term leverage, banks with higher short-term leverage front-load their dividend payouts – paying out relatively more in the beginning of the rollover crisis, and relatively less thereafter. Moreover, this front-loading occurs only for banks with sufficiently high values of initial available liquidity. We then move on to our empirical analysis. We examine data on U.S. bank holding companies over the period 2004q1-2011q4. Specifically, we combine balance sheet and income statement information from banks FR-Y9C reports with data on dividend payouts and stock prices from CRSP. Using a flexible difference-in-differences framework, we identify the effects of the rollover crisis in 2008 by considering banks with different pre-crisis exposure to shortterm debt defined as the ratio of 2006q4 short-term debt to total liabilities. Our baseline specification accounts for a systematic relation between short-term leverage and pre-crisis dividend payout capacity, proxied by the 2006q4 return on assets, as well as with differential regional exposure to the housing market and general economic downturn via state-by-quarter fixed effects. Our main identifying assumption is thus one of “parallel trends”, namely that irrespective of short-term leverage, absent the liquidity and rollover crisis, banks would have had a similar evolution of dividend payouts. Our main finding is that exposure to rollover risk given by short-term leverage has a significant and quantitatively large impact on bank dividend payouts in 2008. For instance, focusing on 2008q3 – which may represent a peak in the liquidity and rollover crisis in the U.S. banking system – we find that variation in exposure to rollover risk explains between 80% and 100% of the variation in dividend growth, conditional on a set of controls. In the aggregate, back-of-the-envelope calculations suggest that dividend signaling due to rollover risk increased aggregate dividend outflows of the U.S banking system by about 10 - 26 % (1 - 2.5 USD billion) from 2008q2 to 2008q3, relative to scenarios where U.S banks were relatively less exposed to rollover risk. Moreover, the relative increase in dividends of more exposed banks is driven by banks with high initial cash holdings. Both of these findings are consistent with our conceptual framework. We also find that more exposed banks experienced higher stock returns in the midst of the crisis, most likely as a direct effect of having higher dividend payouts. A key identification concern is that exposure is correlated with agency problems resulting from differential loading on bad news about future returns on assets due to differences in the initial portfolio composition. As a robustness exercise, we therefore augment our baseline 2 specification and explicitly incorporate portfolio differences by allowing for a time-varying impact of the 2006q4 share of mortgages to total assets, other loans to total assets, and securities to total assets. We also allow initial leverage to have time-varying effects on the subsequent growth in dividend payouts. Our results are qualitatively and quantitatively robust to these extensions. Related literature Our paper contributes to an empirical literature on bank dividend payouts during the financial crisis (Acharya, Gujral, Kulkarni, and Shin (2011), Floyd, Li, and Skinner (2015), Acharya, Le, and Shin (2017), Hirtle (2014), Cziraki, Laux, and Loranth (forthcoming)). Floyd, Li, and Skinner (2015) compare the payout policies of non-financial firms and banks before and during the financial crisis of 2007-2008. They document that unlike non-financial firms, banks have relied on dividends over repurchases as the primary payout to shareholders since the 1980s. They also document the notable stickiness in bank dividend payouts during the financial crisis, with aggregate dividends exceeding aggregate earnings by 30% in 2008, and argue that this behavior is consistent with banks signaling financial strength to short-term lenders and depositors. Acharya, Gujral, Kulkarni, and Shin (2011) and Acharya, Le, and Shin (2017) show a similar pattern but also document substantial heterogeneity in the dividend payouts of large banks in 2008. Hirtle (2014) and Cziraki, Laux, and Loranth (forthcoming) compare the evolution of dividend payouts and share repurchases by U.S. banks prior to and during the crisis. While dividends and share repurchases followed similar patterns prior to the crisis, they diverged strongly in 2007 and 2008, with banks cutting their share repurchase programs substantially, while maintaining their dividend payouts. Furthermore, Cziraki, Laux, and Loranth (forthcoming) also argue that the cross-sectional evidence is not consistent with risk shifting, by examining the dependence of dividend payouts on bank characteristics associated with higher propensity towards risk shifting, such as leverage and higher risk-taking prior to the crisis. Relative to this literature, we provide new evidence in favor of the dividend signaling channel by showing that banks with higher short-term leverage tended to front-load their dividend payouts during the financial crisis and that this behavior was concentrated among cash-rich banks. Our paper is also related to a broader literature that studies dividend payouts by firms. Given the generally less favorable tax treatment of dividend income, the persistent use of dividends as a way of paying shareholders has been a puzzle in economics and finance (Black, 1976).1One important explanation for this puzzle is that dividends are a signal of future profitability (Bhattacharya (1979), Miller and Rock (1985), John and Williams (1985), Bernheim 1See Allen and Michaely (2003), Frankfurter, Wood, and Wansley (2003), Baker (2009), DeAngelo, DeAngelo, and Skinner (2009), and Farre-Mensa, Michaely, and Schmalz (2014) for surveys of this literature. 3 (1991), Hausch and Seward (1993), Guttman, Kadan, and Kandel (2010), Baker, Mendel, and Wurgler (2016)).2In that literature paying dividends is relatively less costly for firms with higher future profitability, and is therefore used by such firms in equilibrium. However, the empirical evidence on this channel has been mixed. For example, Bernheim and Wantz (1995) argue in favor of dividends serving as a signal of future profitability by showing that higher dividend taxation is associated with a stronger response of stock prices to news about dividends. On the other hand, Benartzi, Michaely, and Thaler (1997), and Grullon, Michaely, and Swaminathan (2002) find limited support for dividend payouts forecasting future earnings. Instead dividend increases tend to correlate with past earnings growth.3In contrast to the literature that argues that dividends signal future profitability, Juelsrud and Nenov (2020) argue that dividends may be used to signal available liquidity to short-term lenders that are considering rolling over their debt. As we argue in this paper using a simple dynamic extension of Juelsrud and Nenov (2020), our findings of a crisis-induced front-loading of dividend payouts by more exposed cash-rich banks are consistent with the dividend signaling channel. 2 Conceptual framework In this section we present a simple dynamic model of dividend payout choice in a liquidity crisis triggered by a rollover crisis. The model illustrates the dynamic incentives of banks in an environment where dividend payouts can impact the severity of the rollover crisis via a signaling effect as in Juelsrud and Nenov (2020). The key trade-off underpinning these dynamic incentives is the balancing of the increase in expected payoff from withstanding the rollover crisis against the long-run impact a large and prolonged draw-down of liquidity has on the bank’s franchise value. We show that, relative to banks with lower short-term leverage, banks with higher short-term leverage front-load their dividend payouts – paying out relatively more in the beginning of the rollover crisis, and relatively less thereafter. Moreover, this front-loading occurs only for banks with sufficiently high values of initial available liquidity. 2The other main explanations for firms paying out dividends is that dividend payouts resolve the free cashflow agency problem between firm managers and shareholders (Jensen and Meckling (1976), Jensen (1986)) or that dividends are more valuable to a subset of investors (Shefrin and Statman, 1984). 3There is however a large empirical literature that documents a positive relationship between dividend updates and stock price responses. See, for example, Charest (1978), Aharony and Swary (1980), Brickley (1983), Asquith and Mullins Jr (1983), Bajaj and Vijh (1990), Denis, Denis, and Sarin (1994), Michaely, Thaler, and Womack (1995). Our findings of stock prices tracking the relative dividend payout profile of more exposed cash-rich banks is consistent with this literature. 4 Model set-up Time is continuous and runs forever. There is a single bank that is facing a liquidity crisis due to lenders refusing to roll over their debt holdings. We assume as in He and Xiong (2012) that the bank has a staggered debt expiration schedule that is distributed uniformly over time. The bank’s debt is held by a continuum of small creditors. In each instant, a measure b of debt matures with the corresponding creditors making a roll-over decision. Let a(t, d (t)) denote the share of creditors who refuse to roll over their maturing debt at time t, where d(t) is the bank’s dividend payout rate at t. We do not model explicitly the lenders’ rollover decisions but instead make the following two reduced-form assumptions about a. First, we assume that the rollover episode has an uncertain duration that ends with Poisson rate λ, that is the liquidity crisis ends at a random time X∼Exp (λ).4 5 Second, following Juelsrud and Nenov (2020), we assume that the run size aresponds to ddue to a static signaling effect of the dividend payout at time t, so that the dividend payout rate, d(t), affects the magnitude of the (instantaneous) liquidity outflow but not the expected duration of the liquidity crisis.6 The total liquidity outflow rate (including the direct payment of dividends) is then l=d(t) + ba (t, d (t)) .(1) As Juelsrud and Nenov (2020) show, lis non-monotone in din the unique equilibrium of the static rollover game. Specifically, for sufficiently large values of b, there is a strictly positive value of the dividend payout, dmin >0, such that lis minimized at dmin. Moreover, as lenders observe arbitrarily precise signals about the dividend payout rate, then a(t, d (t)) converges to a step function with a= 0 for d≥dmin, and a= 1 for d < dmin. Additionally, in that framework, dmin is increasing in the nominal value of maturing debt b(see Proposition 8 in Juelsrud and Nenov (2020)). Below we will utilize these observations directly and assume 4In a standard rollover global game (e.g. Morris and Shin (2004)), the roll-over decision of a lender depends on the ratio of the (expected) gain from rolling over given that the bank survives the rollover crisis and repays the lenders fully over the (expected) loss from rolling over given that the bank fails. He and Xiong (2012) generalize this trade-off in a dynamic context. Therefore, any shock that lowers the gain given bank survival or that increases the loss given bank failure weakens the incentives of lenders to roll over. In our rollover crisis episode we have in mind such a temporary shock which increases the share of lenders that choose to not roll over in a given period. After this shock dissipates, the economy reverts back to a normal time where the share of lenders that do not roll over goes back down to some steady state value. For simplicity, we assume that this value is 0. 5Note that the liquidity crisis may end on its own or it may end because of a government intervention. Therefore, expectations about the duration of the liquidity crisis may also reflect expectations of a future government intervention. 6The signaling effect arises whenever creditors observe (dispersed) signals about the dividend payout and make inference about the bank fundamental based on these signals. See Juelsrud and Nenov (2020) for a discussion about why dividends may be used to signal available liquidity relative to other available actions. 5 0.00 1.00 2.00 3.00 4.00 5.00 % 01jan2006 01jan2007 01jan2008 01jan2009 01jan2010 01jan2011 Figure 4: Rollover risk during the financial crisis, as captured by the TED-spread. Notes: This figure shows the evolution of the TED-spread, defined as the difference between the 3m USD LIBOR rate and the 3m Treasury rate. 3.4 Methodology 3.4.1 Baseline regression To isolate the effect of the rollover crisis on dividend payouts, we compare dividend-related outcomes for banks with a high reliance on short-term debt (i.e. banks with a high exposure) relative to less reliant banks. We compare banks according to their 2006q4 exposure.15 This comparison allows us to isolate the impact of the rollover crisis from other, aggregate factors affecting the dividend payouts of the US banking system. Our baseline specification addresses two potential confounding factors. First, to ensure that high-exposure banks are not banks with a particularly high dividend payout capacity, we estimate the dynamic effects of having a high exposure to the rollover risk on dividend payouts conditional on the dynamic effects of the 2006q4 Return on Assets (RoA). Second, to ensure that high-exposure banks are not differentially affected by the contemporaneous downturn in the US-economy due to different regional exposures, we employ state×time fixed effects. Thus, the estimated impact of rollover risk on dividend payouts is driven by within-state×time variation. Our identifying assumptions is thus that high- and low-exposure banks would have had a similar evolution of dividend payouts during 2007-2008 absent any rollover risk, conditional on state×time fixed effects and initial profitability as captured by the RoA (net income dividended by total assets) in 2006q4. In Section 3.4.2 we discuss further threats to identification and how our results are robust to those. Our baseline specification is outlined in equation (6). The main outcome variable Yi,t 15The capital structure of US banks is fairly sticky at short- and medium-run horizons, as illustrated by Table 4 in the Appendix. Our results are robust to defining exposure based on other time periods. 12 is the change in log(dividends) for a bank ibetween period t−1 and t. We also consider alternative outcomes variables, such as measures capturing the extensive margin of dividend payouts (i.e. a dummy for whether a bank initiates dividend payouts or discontinues its dividend payouts), as well as stock prices. Exposurei,2006q4is the exposure measure for bank imeasured in 2006q4, while RoAi,2006q4captures RoA for bank imeasured in 2006q4. As discussed above, the specification allows for state×quarter (γs(i),t) fixed effects, in addition to a bank fixed effect (αi). We cluster standard errors on the bank-level to account for within-bank autocorrelation in the un-modeled shocks ϵi,t. The coefficients of interests are the β′ ks, which capture the impact of the interaction term between exposure to the rollover risk and the time variable. The dynamic treatment effects are captured by β′ ksfrom 2007q1 and onward. An important test of the validity of our research design is whether βk= 0 before the financial crisis, as that would indicate that there are systematic differences in the evolution of dividend payouts also before the crisis. Yi,t =αi+X k βkExposurei,2006q4×1t=k+X k ηk(RoAi,2006q4×1t=k) + γs(i),t +ϵi,t (6) 3.4.2 Threats to identification In this section, we discuss three potential threats to our interpretation of βkas mapping out the effect of the rollover crisis on dividend payouts, namely risk-shifting due to agency problems and bad news about portfolios, optimism about the future franchise value, and government interventions as a response to the stress in financial markets. Below we discuss each of these possible confounders, while in Section 4.4 we show that our baseline results are robust to them in a number of robustness exercises. Risk-shifting One explanation for the increase in dividend payouts observed during the financial crisis is that they reflected a form of moral hazard. According to Scharfstein and Stein (2008), during 2007-2008 banks were realizing that latent losses were large and the franchise value low, and so increased their dividend payouts in “... an attempt by shareholders to beat creditors out the door”. To the extent that high-exposure banks faced larger ex post losses compared to other banks or had higher total leverage, the risk-shifting hypothesis and the rollover crisis hypothesis would have similar predictions for dividend growth. To isolate the effect of the rollover crisis, we therefore include several measures of differential exposure to asset losses as well as leverage. Specifically, we control for portfolio heterogeneity via the 2006q4 share of real estate loans to total assets, other loans to total assets and securities to total assets. These variables capture, in a parsimonious way, the differential exposure of 13 banks in our sample to the decline in the U.S housing market and the associated recession, as well as the general decline in international asset markets. We allow these controls to have a time-varying impact on our outcome variables, i.e. we augment equation (6) to include them multiplied with year-quarter dummies. Similarly, we control for a time-varying impact of bank leverage (defined as total assets over equity). Finally, we also perform a robustness exercise where we exclude banks that failed the 2009 Supervisory Capital Assessment Program (SCAP) stress test, i.e. banks that in 2009 were deemed by the Federal Reserve as having too low capital. While this may capture banks that were more prone to risk-shifting during the crisis, it has the drawback that it might also exclude the banks that were most exposed to the rollover crisis to the extent that higher exposure to the crisis implies reduced overall capitalization post-crisis (for example, precisely because of front-loading of dividends). Optimism about future franchise value Another possible explanation is that high-exposure banks are banks that have lower latent losses due to exposure to fundamentally different sectors compared to other banks, higher franchise values, and therefore have a higher dividend growth compared to other banks. Note, however, that by controlling for the different initial portfolio shares in robustness, we also implicitly control for such a confounding factor. Moreover, our analysis of dividend payouts in the medium-run in Section 4.3 also does not provide support for this channel. Troubled-Asset Relief Program U.S authorities implemented the Troubled-Asset Relief Program (TARP) towards the end of 2008. By and large, TARP entailed a recapitalization of the participating banks which left them with additional funds. Importantly, as Scharfstein and Stein (2008) highlight, TARP was not associated with a cap on dividends. The newly acquired funds via TARP could therefore, in principle, be used to pay dividends. To the extent that banks with a higher exposure were more likely to participate in TARP and that funds obtained in TARP were used for dividends due to risk-shifting, it could be a confounding factor when interpreting βk. To ensure that this is not driving our results, we follow two alternative approaches. First, we use the high frequency of the data on dividend announcments and payouts to verify that changes in the dividend payouts observed is not related to the announcement and implementation of the TARP, but rather focused on the initial phases of increased rollover risk in the summer of 2008. Second, we augment our baseline specification by also including a dummy for whether the bank was one of the 18 banks obtaining additional funds via TARP. 14 4 Results In this section, we present the main results of the paper. We start by considering the impact of being exposed to the rollover crisis on dividend payouts in Section 4.1. We show that banks that were more exposed in terms of higher short-term leverage had higher dividend growth compared to other banks during the crisis. We find limited effects on both dividend initiations and dividend discontinuations. Consistent with our model, we document that this increase in dividends is driven by banks with high ratios of initial liquidity. Next, we show in Section 4.2 that the same set of banks experienced more favorable stock returns during the crisis. 4.1 Dividends 4.1.1 Intensive margin We start by investigating the intensive margin response of dividend payouts to rollover risk. Specifically, in Figure 5 we plot the evolution of average dividend growth for banks with an exposure below and above the median. While the evolution is fairly similar prior to the crisis, there is a sharp relative increase in dividend growth by high-exposure banks in 2008q2 and 2008q3, which reverses in 2008q4. The relative increase at the height of the rollover crisis is consistent with rollover risk causing upward pressures on dividends, in line with our theoretical framework.16 However, Figure 5 only plots the unconditional evolution of dividend growth. In Figure 6 we, therefore, plot the estimated sequence of βkestimated from equation (6). These coefficients captures the dynamic impact of exposure to rollover risk as measured by 2006q4. Note now that, in contrast to Figure 5, we now measure exposure as a continuous measure. 16Note that our theoretical framework abstracts from contemporaneous shocks to “fundamentals” that would result in a general decline in dividends and instead only makes predictions about the relative behavior of high vs. low short-term leverage banks and high vs. low cash banks. 15 -.6 -.4 -.2 0 .2 Average prct. change in dividends 2005q1 2006q1 2007q1 2008q1 2009q1 Low exposure High exposure Figure 5: Evolution of dividends Notes: This figure shows the evolution of the (size-weighted) average ∆ log(dividends) for banks with an exposure above the median and banks with an exposure below the median. -5 0 5 10 Time x Exposure coefficient 2006. 1 .2 2006. 3 .4 2007. 1 .2 2007. 3 .4 2008. 1 .2 2008. 3 .4 Outcome: ∆Log(dividends) Figure 6: Rollover risk and dividend payouts Notes: This figure shows the estimated βkfrom equation (6) using ∆ log(dividends) as outcome variable. Confidence intervals are at the 95% level, and standard errors are clustered at the bank level. The figure illustrates that there are essentially no significant differences in the growth of dividends according to the 2006q4 exposure measure up to and including 2008q1 – a reflection of the well-known stickiness in bank dividend payouts. However, in 2008q2 and 2008q3 – 16 N Number of clusters R2 5,176 318 0.662 Table 2: Summary statistics for Figure 6 Notes: This table contain accompanying summary statistics for Figure 6 precisely at the height of the rollover crisis – bank exposure has a significant positive impact on the growth of dividend payouts. This is particularly clear in 2008q3. In that quarter, a 10 percentage point higher exposure measure is associated with a 50 percentage points higher dividend growth. Quantitatively, a standard deviation change in the exposure measures explains nearly 100% of the dispersion in dividend growth in 2008q3 (conditional on bank and state x year fixed effects, as well as the dynamic effect of RoAb,2006q4), suggesting that exposures to the ongoing rollover crisis was key to understand dividend payout policies. In line with our conceptual framework, we would expect more exposed banks with higher initial cash holdings to frontload their dividend payouts more strongly. To test whether this holds empirically, we partition the sample into cash-rich vs. other banks and estimate equation (6) for these two groups. Specifically, we measure cash as the sum of cash and trading assets, scale it by total liabilities, and define banks with a cash-ratio above the median in 2006q4 as cash-rich. We then estimate equation (6) for cash-rich and other banks. The results are shown in Figure 7. The estimated coefficients illustrates that all of the effect of rollover risk exposure on dividend growth is driven by banks with high initial cash holdings. 4.1.2 Extensive margin Next we consider whether exposure to the rollover crises affected banks initiation or discontinuiton of dividend payouts. On average, dividend initiations were fairly unchanged in 2008 compared to previous years, as illustrated in Figure 8a. However, dividend discontinuiations became substantially more likely as the financial crisis unfolded as illustrated in Figure 8b. In that figure, we plot the fraction of banks discontinuing dividend payouts from t−1 to t. On average, the fraction is a little bit higher than 2% in the period prior to the financial crisis. From 2008q1, however, the fraction of banks discontinuing dividend payouts increase substantially, and reaches a peak in 2008q4. It is unclear, however, whether the changes in dividend discontinuiations are driven by banks exposed to the rollover crises or not. Moreover, even though there is no increase in average dividend initiations, it could for instance be the case for high exposure banks. To investigate whether the rollover crisis affected banks dividend payouts along the extensive margin, we, therefore, estimate equation (6) with either a dummy for whether the bank 17 -10 0 10 20 Time x Exposure coefficient 2006. 1 .2 2006. 3 .4 2007. 1 .2 2007. 3 .4 2008. 1 .2 2008. 3 .4 Below median cash holdings Above median cash holdings Outcome: ∆Log(dividends) Figure 7: Rollover risk and dividend payouts, according to initial cash position. Notes: This figure shows the estimated βkfrom equation (6) using ∆ log(dividends) as outcome variable. Confidence intervals are at the 95% level, and standard errors are clustered at the bank-level. “Above median cash holdings” refers to a sample where banks had cash to total liabilities above the median, where cash is defined as the sum of cash and trading assets. “Below median cash holdings” refers to a sample of all other banks. .005 .01 .015 .02 .025 .03 Fraction of banks initiating dividend payouts 2004q1 2005q1 2006q1 2007q1 2008q1 2009q1 (a) Fraction of dividend initiations 0 .02 .04 .06 Fraction of banks stopping dividend payouts 2004q1 2005q1 2006q1 2007q1 2008q1 2009q1 (b) Fraction of dividend discontinuiations Figure 8: Dividend initiations and discontinuiations. Notes: This figure shows the fraction of banks in our sample which initiated dividend payouts (a) and discontinuied dividends (b). Dividend initiations is defined as events where a bank holding company paid zero dividends in t−1 and a positive amount of dividends in t. Dividend discontinuiations is defined as events where a bank holding company paid positive amount of dividends in t−1 and paid zero dividends in t. 18 -.1 0 .1 .2 Time x Exposure coefficient 2006. 1 .2 2006. 3 .4 2007. 1 .2 2007. 3 .4 2008. 1 .2 2008. 3 .4 Outcome: Dummy for dividend initiation (a) Exposure the rollover crises and dividend initiations -.2 -.1 0 .1 .2 .3 Time x Exposure coefficient 2006. 1 .2 2006. 3 .4 2007. 1 .2 2007. 3 .4 2008. 1 .2 2008. 3 .4 Outcome: Dummy for dividend discontinuiation (b) Exposure to the rollover crises and dividend discontinuations. Figure 9: Exposure to the rollover crises and dividend payouts along the extensive margin. Notes: This figure shows the estimated βkfrom equation (6) using a dummy for dvidend initiations (a) or dividend discontinuiations (b) as outcome variable. Confidence intervals are at the 95% level, and standard errors are clustered at the bank-level. Dividend initiations is defined as events where a bank holding company paid zero dividends in t−1 and a positive amount of dividends in t. Dividend discontinuiations is defined as events where a bank holding company paid positive amount of dividends in t−1 and paid zero dividends in t. initiates dividend payouts or discontinues dividend payouts. Figures 9a and 9b show the estimated effect of exposure to the rollover crisis on the propensity to initiate or discontinue dividend payouts. All estimated coefficients are relative to 2007q1. As is clear from both figures, exposure to the rollover crises does not have a significant impact on the propensity to initiate or discontinue dividends, suggesting that any potential adjustment of dividend payouts to rollover risk primarily operates along the intensive margin. 4.2 Stock price effects In this section we investigate the financial markets impact of increasing dividend payouts in response to the rollover crisis. We focus on the effect on stock returns. Figure 10 shows the evolution of βkusing quarterly stock returns (ex dividend) as outcome variable. Interestingly, stock returns increase for banks with a higher exposure in 2008q3, which is also the quarter with highest growth in dividends for these banks. Quantitatively, dispersion in exposure explains approximately 80% of the dispersion in stock returns in that quarter with a 10 percentage point higher exposure being associated with around 7 percentage points higher stock return. In line with Figure 7, we find that cash-rich banks also are the drivers of the differential stock return response to rollover risk, as documented in Figure 11 below. Specifically, among the cash-rich banks, a 10 percentage point higher exposure associated with around 20 percentage points higher stock return – a sizable effect. 19 -.5 0 .5 1 1.5 Time x Exposure coefficient 2006. 1 .2 2006. 3 .4 2007. 1 .2 2007. 3 .4 2008. 1 .2 2008. 3 .4 Outcome: Stock returns Figure 10: Stock returns Notes: This figure shows the estimated βkfrom equation (6) using stock returns (ex. dividend) as outcome variable. Confidence intervals are at the 95% level, and standard errors are clustered at the bank-level. -1 0 1 2 3 4 Time x Exposure coefficient 2006. 1 .2 2006. 3 .4 2007. 1 .2 2007. 3 .4 2008. 1 .2 2008. 3 .4 Below median cash holdings Above median cash holdings Outcome: Returns Figure 11: Stock returns and initial cash position Notes: This figure shows the estimated βkfrom equation (6) using stock returns (ex. dividend) as outcome variable. Confidence intervals are at the 95% level, and standard errors are clustered at the bank-level.“Above median cash holdings” refers to a sample where banks had cash to total liabilities above the median, where cash is defined as the sum of cash and trading assets. “Below median cash holdings” refers to a sample of all other banks. 20 Therefore, we find a clear positive link between dividend payout changes and stock price responses that took place in 2008q3, consistent with a large empirical literature that has documented this relationship in other settings. Since exposed banks tend to only front-load dividends rather than permanently increase dividends (as we further document in Section 4.3 below), it is somewhat challenging to understand the mechanism through which stock prices respond to dividends.17 One possible explanation is that given banks’ reputation for stable dividend payouts any changes in dividend payouts tend to be interpreted by market participants as persistent and thus lead to changes in stock prices. A complementary explanation is that some investors may overreact to dividends because they consider dividend income separately from capital gains due to behavioral frictions like mental accounting (Shefrin and Statman, 1984, Baker, Nagel, and Wurgler (2007), Hartzmark and Solomon (2019), Di Maggio, Kermani, and Majlesi (2020), Hartzmark and Solomon (2021)). 4.3 Medium-run effects We also examine the dividend payout and stock price effects in the quarters beyond the rollover crisis of 2008. Figure 12 shows the dynamic response of dividends and stock prices of more exposed banks through 2011q4.18 As the Figure shows the relative increase in dividend payouts of more exposed banks was only short-lived and only during the rollover crisis. After that episode more exposed banks tended to permanently lower their dividend payouts. This pattern is consistent with the frontloading behavior of more exposed banks in our theoretical framework (cf. Figure 2). Figure 13 plots the dynamic response of stock prices. Stock prices track dividend payouts very closely, with the initial sharp (relative) increase in 2008q3 followed by the same persistent decline post 2009q1. It is important to mention that there were substantial regulatory changes post-2008 (e.g. the Dodd-Frank act) which may have had a differential impact on profitability and hence dividend payouts and stock prices of banks with different levels of pre-crisis short-term leverage. Moreover, the SCAP stress test in 2009q1 may have played an important effect on post-crisis dividend payouts as well. In the Appendix we investigate this further by looking only at the medium-term response of banks that did not fail the 2009 stress test. We show that our results are qualitatively robust to excluding these banks though they are smaller in magnitude. 17In our theoretical framework, a rollover crisis leaves banks with higher short-term debt worse off compared to banks with lower short-term debt, irrespective of their deividend payouts. Therefore, if one equates stock prices to the payoffs of the bank owner in the model, one should expect a relative decline in stock prices for banks with higher short-term debt, not an increase. 18To estimate these responses we estimate Equation (6) in levels, i.e. we use log(dividends) and log(price) as outcome variables rather than the one-quarter log difference. 21 Brickley, J. A. (1983): “Shareholder wealth, information signaling and the specially designated dividend: An empirical study,” Journal of Financial Economics, 12(2), 187–209. Charest, G. (1978): “Dividend information, stock returns and market efficiency-II,”Journal of financial Economics, 6(2-3), 297–330. Cziraki, P., C. Laux, and G. Loranth (forthcoming): “Understanding bank payouts during the crisis of 2007-2009,” Review of Corporate Finance Studies. DeAngelo, H., L. DeAngelo, and D. J. Skinner (2009): Corporate payout policy. Now Publishers Inc. DeAngelo, H., L. DeAngelo, and R. M. Stulz (2006): “Dividend policy and the earned/contributed capital mix: a test of the life-cycle theory,” Journal of Financial economics, 81(2), 227–254. Denis, D. J., D. K. Denis, and A. Sarin (1994): “The information content of dividend changes: Cash flow signaling, overinvestment, and dividend clienteles,”Journal of Financial and Quantitative Analysis, 29(4), 567–587. Di Maggio, M., A. Kermani, and K. Majlesi (2020): “Stock market returns and consumption,” The Journal of Finance, 75(6), 3175–3219. Duffie, D. (2010): How big banks fail and what to do about it. Princeton University Press. Farre-Mensa, J., R. Michaely, and M. Schmalz (2014): “Payout policy,” Annu. Rev. Financ. Econ., 6(1), 75–134. Floyd, E., N. Li, and D. J. Skinner (2015): “Payout policy through the financial crisis: The growth of repurchases and the resilience of dividends,”Journal of Financial Economics, 118(2), 299–316. Frankfurter, G., B. G. Wood, and J. Wansley (2003): Dividend policy: Theory and practice. Elsevier. Grullon, G., R. Michaely, and B. Swaminathan (2002): “Are dividend changes a sign of firm maturity?,” The journal of Business, 75(3), 387–424. Guttman, I., O. Kadan, and E. Kandel (2010): “Dividend stickiness and strategic pooling,” Review of Financial Studies, 23(12), 4455–4495. Hartzmark, S. M., and D. H. Solomon (2019): “The dividend disconnect,” The Journal of Finance, 74(5), 2153–2199. 28 (2021): “Predictable price pressure,” Available at SSRN 3853096. Hausch, D. B., and J. K. Seward (1993): “Signaling with dividends and share repurchases: A choice between deterministic and stochastic cash disbursements,” Review of Financial Studies, 6(1), 121–154. He, Z., and A. Manela (2016): “Information acquisition in rumor-based bank runs,” The Journal of Finance, 71(3), 1113–1158. He, Z., and W. Xiong (2012): “Dynamic debt runs,” The Review of Financial Studies, 25(6), 1799–1843. Hirtle, B. (2014): “Bank holding company dividends and repurchases during the financial crisis,” FRB of New York Staff Report, (666). Jensen, M. C. (1986): “Agency costs of free cash flow, corporate finance, and takeovers,” The American economic review, 76(2), 323–329. Jensen, M. C., and W. H. Meckling (1976): “Theory of the firm: Managerial behavior, agency costs and ownership structure,” Journal of financial economics, 3(4), 305–360. John, K., and J. Williams (1985): “Dividends, dilution, and taxes: A signalling equilibrium,” Journal of Finance, 40(4), 1053–1070. Juelsrud, R. E., and P. T. Nenov (2020): “Dividend payouts and rollover crises,” The Review of Financial Studies, 33(9), 4139–4185. Michaely, R., R. H. Thaler, and K. L. Womack (1995): “Price reactions to dividend initiations and omissions: Overreaction or drift?,” the Journal of Finance, 50(2), 573–608. Miller, M. H., and K. Rock (1985): “Dividend policy under asymmetric information,” Journal of Finance, 40(4), 1031–1051. Morris, S., and H. Shin (2004): “Coordination Risk and the Price of Debt,” European Economic Review, 48(1), 133–153. Scharfstein, D. S., and J. C. Stein (2008): “This bailout doesnt pay dividends,” New York Times, 20. Shefrin, H. M., and M. Statman (1984): “Explaining investor preference for cash dividends,” Journal of financial economics, 13(2), 253–282. 29 Appendix Characterizing the bank owner’s problem from Section 2 The bank (owner) solves W(L(0)) = max {d(t)}∞ 0,0≤τ,˜ L≥0nPr {X < τ}V+ Pr {X > τ}D˜ Lo, s.t. ˜ L=L(0) −Zτ 0 l(t)dt. We have that Pr {X > τ}= exp {−λτ}. Moreover, by assumptions on l(Eq. (2)), it follows that d(t)≥dmin, and so l(t) = d(t),∀t. It then follows that Rτ 0l(t)≥dminτor equivalently ˜ L≤L(0) −dminτ. Moreover, we can simplify the objective to W(L(0)) = max 0≤τ,˜ L≥0nV−exp (−λτ)hV−D˜ Lio, s.t.˜ L≤L(0) −dminτ. Note that ˜ L > 0 by the Inada conditions, so we can disregard this constraint. Let µdenote the multiplier on the inequality constraint. The first-order conditions for this problem, assuming an interior value of τ, are ˜ L: exp (−λτ)D′˜ L−µ= 0,(9) and τ:λexp (λτ)hV−D˜ Li−dminµ= 0,(10) together with the complementary slackness condition µL(0) −dminτ−˜ L= 0.(11) Note that condition (10) implies that µ > 0, so that the constraint is binding at the optimum. Intuitively, since overcoming the liquidity crisis is always preferred to giving up, the bank would prefer to survive the longest possible time for any L(0). This requires choosing the lowest liquidity outflow rate possible, which is dmin. 30 Combining Eq. (9)-(11), we get λ[V−D(L(0) −dminτ)] = [D′(L(0) −dminτ)] dmin.(12) Finally, note that the optimal value of τis at an interior point, provided that λ[V−D(L(0))] > D′(L(0)) dmin. This inequality holds for sufficiently high values of L(0), i.e. for L(0) > L, where Lsolves λV−DL=D′Ldmin.(13) For values of L(0) ≤L, the bank optimally chooses to“Give up”immediately, so that τ= 0.19 Note that we can also express the optimality condition in Eq (12) in terms of the available liquidity at τ,˜ L=L(0) −dminτ, so that λhV−D˜ Li=D′˜ Ldmin.(14) Observe that Eq. (13) and (14) coincide. Therefore, banks with different values of L(0) choose to “Continue” until they reach the same amount of available liquidity L. If they start with lower available liquidity than L, then they choose to “Give up” immediately. Robustness to a wider set of controls In this section, we augment equation (6) to also include the 2006q4 mortgage share, corporate loan share, securities share, other loans share as well as leverage.20 Figure 15 and 16 considers the impact of rollover risk exposure on dividend growth, on average and by initial cash holdings, respectively. Figure (17) and (18) investigates the impact on stock returns. In sum, the results are qualitatively and quantitatively robust to the inclusion of these additional controls. Once controlling for these other factors, the dispersion in exposure can explain roughly 80% of the dispersion in dividend growth in 2008q3. 19The assumption that λ < −D′′ L/D′Ldmin ensures that Lis unique. 20Controlling for leverage is also captures a broader measure of payout capacity, as banks also can pay out dividends based on retained earnings, see DeAngelo, DeAngelo, and Stulz (2006). 31 -2 0 2 4 6 8 Time x Exposure coefficient 2006. 1 .2 2006. 3 .4 2007. 1 .2 2007. 3 .4 2008. 1 .2 2008. 3 .4 Outcome: ∆Log(dividends) Figure 15: Rollover risk and dividend payouts Notes: This figure shows the estimated βkfrom equation (6) using ∆ log(dividends) as outcome variable, and where we also include the 2006q4 share of mortgages to total assets, non-mortgage lending to total assets, securities to total assets, and leverage. Confidence intervals are at the 95% level, and standard errors are clustered at the bank-level. Controlling for participation in the Troubled-Asset Relief Program In this section, we show the results from estimating the same regression as in the last section but where we also allow participation in the Troubled-Asset Relief Program (TARP) to have a time-varying impact on the dividends paid. This is potentially important, as participation in TARP could influence the funds available and the incentives to pay dividends. Figure 19 and 20 considers the impact of rollover risk exposure on dividend growth, on average and by initial cash holdings, respectively. Figure (21) and (22) investigates the impact on stock returns. In sum, the results are qualitatively and quantitatively robust to the inclusion of these additional controls. 32 -10 0 10 20 Time x Exposure coefficient 2006. 1 .2 2006. 3 .4 2007. 1 .2 2007. 3 .4 2008. 1 .2 2008. 3 .4 Below median cash holdings Above median cash holdings Outcome: ∆Log(dividends) Figure 16: Rollover risk and dividend payouts, according to initial cash position. Notes: This figure shows the estimated βkfrom equation (6) using ∆ log(dividends) as outcome variable, and where we also include the 2006q4 share of mortgages to total assets, non-mortgage lending to total assets, securities to total assets, and leverage. Confidence intervals are at the 95% level, and standard errors are clustered at the bank-level. “Above median cash holdings” refers to a sample where banks had cash to total liabilities above the median, where cash is defined as the sum of cash and trading assets. “Below median cash holdings” refers to a sample of all other banks. -2 0 2 4 6 8 Time x Exposure coefficient 2006. 1 .2 2006. 3 .4 2007. 1 .2 2007. 3 .4 2008. 1 .2 2008. 3 .4 Outcome: ∆Log(dividends) Figure 19: Rollover risk and dividend payouts Notes: This figure shows the estimated βkfrom equation (6) using ∆ log(dividends) as outcome variable, and where we also include the 2006q4 share of mortgages to total assets, non-mortgage lending to total assets, securities to total assets, leverage and a dummy for whether the bank was a participant in the TARP program or not. Confidence intervals are at the 95% level, and standard errors are clustered at the bank-level. 33 Figure 17: Stock returns Notes: This figure shows the estimated βkfrom equation (6) using stock returns (ex. dividend) as outcome variable, and where we also include the 2006q4 share of mortgages to total assets, non-mortgage lending to total assets, securities to total assets, and leverage. Confidence intervals are at the 95% level, and standard errors are clustered at the bank-level. 34 -.5 0 .5 1 1.5 2 Time x Exposure coefficient 2006. 1 .2 2006. 3 .4 2007. 1 .2 2007. 3 .4 2008. 1 .2 2008. 3 .4 Below median cash holdings Above median cash holdings Outcome: Returns Figure 18: Stock returns and initial cash position Notes: This figure shows the estimated βkfrom equation (6) using stock returns (ex. dividend) as outcome variable, and where we also include the 2006q4 share of mortgages to total assets, non-mortgage lending to total assets, securities to total assets, and leverage. Confidence intervals are at the 95% level, and standard errors are clustered at the bank-level.“Above median cash holdings” refers to a sample where banks had cash to total liabilities above the median, where cash is defined as the sum of cash and trading assets. “Below median cash holdings” refers to a sample of all other banks. Excluding banks in the Supervisory Capital Assessment Program In this section, we show the main (average) results using ∆ log (dividends) and stock returns (ex. dividends) as outcome variables. For comparison, we also include the coefficient estimates obtained on the full sample. Our results are qualitatively robust to excluding the low-capitalized banks based on the stress tests, but the responses are somewhat muted. A potential interpretation of this, is that banks that failed the stress test of 2009 did so in particular because their dividend signaling in 2008 was particularly strong. 35 -10 0 10 20 Time x Exposure coefficient 2006. 1 .2 2006. 3 .4 2007. 1 .2 2007. 3 .4 2008. 1 .2 2008. 3 .4 Below median cash holdings Above median cash holdings Outcome: ∆Log(dividends) Figure 20: Rollover risk and dividend payouts, according to initial cash position. Notes: This figure shows the estimated βkfrom equation (6) using ∆ log(dividends) as outcome variable, and where we also include the 2006q4 share of mortgages to total assets, non-mortgage lending to total assets, securities to total assets, leverage and a dummy for whether the bank was a participant in the TARP program or not. Confidence intervals are at the 95% level, and standard errors are clustered at the bank-level. “Above median cash holdings” refers to a sample where banks had cash to total liabilities above the median, where cash is defined as the sum of cash and trading assets. “Below median cash holdings” refers to a sample of all other banks. 36 Figure 21: Stock returns Notes: This figure shows the estimated βkfrom equation (6) using stock returns (ex. dividend) as outcome variable, and where we also include the 2006q4 share of mortgages to total assets, non-mortgage lending to total assets, securities to total assets, leverage and a dummy for whether the bank was a participant in the TARP program or not. Confidence intervals are at the 95% level, and standard errors are clustered at the bank-level. 37