Corporate Taxes and Securitization





    Search for more papers by this author
    • JoongHo Han is at Sungkyunkwan University, Kwangwoo Park is at the Korea Advanced Institute of Science and Technology, and George Pennacchi is at the University of Illinois. We are grateful for valuable comments from an anonymous referee, Adam Ashcraft, Mark Flannery, the Editor Campbell Harvey, Edward Kane, Hayne Leland, Greg Nini, and participants of the 2010 Financial Intermediation Research Society Conference and of seminars at Bocconi University, the Federal Deposit Insurance Corporation, the Federal Reserve Banks of Chicago and New York, KAIST, KDI School, Seoul National University, Sungkyunkwan University, Tilburg University, and the University of Venice. Hakkon Kim and Hyun-Dong Kim provided excellent research assistance.


Most banks pay corporate income taxes, but securitization vehicles do not. Our model shows that, when a bank faces strong loan demand but limited deposit market power, this tax asymmetry creates an incentive to sell loans despite less-efficient screening and monitoring of sold loans. Moreover, loan-selling increases as a bank's corporate income tax rate and capital requirement rise. Our empirical tests show that U.S. commercial banks sell more of their mortgages when they operate in states that impose higher corporate income taxes. A policy implication is that tax-induced loan-selling will rise if banks’ required equity capital increases.

Prior to the 2007 to 2009 financial crisis, securitization grew rapidly for several decades.1 Securitization is a process whereby banks and nonbank lenders sell mortgages or other loans to special purpose vehicles that issue mortgage- or asset-backed securities. Securitization permits a bank to originate loans and then transfer their interest rate and credit risks to mortgage- and asset-backed security investors. A potential benefit to the banking system is reduced exposure to risks that threaten its stability. For example, banks that securitize long-duration fixed-rate mortgages can avoid the extreme interest rate risk that decimated U.S. thrift institutions during the 1980s. However, the recent revelation of mortgage-backed securities’ poor credit quality has highlighted problems with the securitization process.2

Academic research has long recognized that securitization can have detrimental side effects. Models such as those in Diamond (1984), Ramakrishnan and Thakor (1984), and Rajan (1992) explain why a financial contract resembling a bank is the most efficient means of funding borrowers whose creditworthiness is not public information. The contract requires that a bank's owner-manager bear its loans’ credit risks to have the incentive to efficiently screen loan applicants and monitor borrowers’ actions. If this credit risk is transferred by selling (securitizing) a loan, the bank's incentive to credit-screen and monitor declines. While Pennacchi (1988) and Gorton and Pennacchi (1995) show that loan sales contracts can be structured to mitigate the moral hazard of reduced credit-screening/monitoring, whenever some risk is transferred the bank's equilibrium credit-screening/monitoring remains suboptimal. Consistent with this theoretical prediction, Keys et al. (2010) find that securitized pools of subprime mortgages had default rates that were 10% to 25% higher than similar mortgages that were not securitized. Elul (2011) and Jiang, Nelson, and Vytlacil (2014) present additional evidence consistent with reduced screening of securitized mortgages.

If investors in mortgage- and asset-backed securities recognize the loan-selling bank's suboptimal credit-screening/monitoring, they should rationally discount the value of securitized loans, forcing the bank to internalize this inefficiency. Consequently, a bank's decision to sell or retain a loan would weigh the economic benefits of risk transfer against the economic costs of inefficient screening/monitoring. However, as our paper points out, other factors largely overlooked in the academic literature can play major roles in a bank's loan-selling decision.

Our paper is the first to model and empirically analyze how corporate taxation of a bank's income and the competitive conditions in the bank's loan and deposit markets affect its incentives to securitize. The model assumes that a bank can improve the returns on the loans that it originates by screening and monitoring them. The bank can also invest in marketable securities and fund loans and securities “on balance sheet” by issuing equity and inelastically supplied deposits. Alternatively, it can fund loans by selling them to rational loan buyers who recognize the bank's equilibrium disincentive to screen/monitor securitized loans.

The model produces novel results by deriving conditions under which a bank will sell loans. It shows that, if a bank has limited loan origination opportunities but significant deposit market power, it will not sell loans but instead choose to invest its inexpensive excess deposits in securities. In contrast, a bank with significant loan opportunities but limited deposit market power will not invest in securities and may have an incentive to sell loans. Importantly, this bank's loan-selling incentive rises as its corporate income tax rate increases. Furthermore, since the bank pays more corporate taxes when its equity capital ratio is higher, raising regulatory capital requirements also increases loan sales. These greater loan sales come at the cost of less efficient screening and monitoring of sold loans.

Though it is well known that corporate taxes create an incentive for excess leverage, it is not commonly recognized that corporate taxes promote securitization.3 We show that tax-induced securitization occurs because of the asymmetric tax treatment of on- and off-balance sheet financing.4 While a bank is subject to corporate income taxes, a special purpose vehicle that purchases loans and issues mortgage- or asset-backed securities is corporate tax-exempt.5 Thus, corporate tax avoidance can lead to loan-selling despite the cost of inefficient screening and monitoring of sold loans.6 We discuss how this corporate tax wedge, along with greater deposit market competition, may have contributed to the growth in securitizations and the defaults on securitized loans.

This paper also provides empirical evidence on corporate taxes and loan sales. We analyze sales of mortgages by U.S. commercial banks using Home Mortgage Disclosure Act (HMDA) filings for the period 2001 to 2008. The key to identifying how corporate taxes affect mortgage sales is differences in corporate tax rates imposed by state governments. State corporate tax rates vary from zero to over 10%, and we use this variation to examine whether banks operating in high-tax states tend to sell a greater proportion of the mortgages that they originate.7 We also refine this analysis by identifying those banks that are likely to face high loan demand but limited deposit market opportunities, which are the banks that our model predicts would be candidates to sell loans. This is done using the proportion of a Metropolitan Statistical Area's (MSA's) population that is below versus above age 65. As shown by Becker (2007), MSAs with a younger population are markets with high loan demand and low deposit savings. The reverse is true for MSAs that have a relatively high proportion of seniors, who tend to have low loan demand but invest heavily in retail deposits.

Our empirical results support the model's predictions. A bank operating in a state that imposes high corporate income taxes tends to sell more of the mortgages that it originates. This is particularly true if the bank operates in an MSA with a young population and hence tends to experience substantial loan demand but limited deposit funding opportunities. For such banks operating in MSAs with populations younger than the median, we estimate that a one-standard-deviation (1.88%) increase in the corporate income tax rate raises mortgage sales by 24.6%. This estimate suggests that corporate taxes are a first-order determinant of mortgage sales.

The paper's findings contribute to the recent debate on the costs of raising banks’ equity capital requirements. Under Basel III, the Basel Committee for Banking Supervision will raise the minimum common equity requirement from 2% to 4.5% and the Tier 1 capital requirement from 4% to 6%. In addition, a mandatory capital conservation buffer of 2.5% and a countercyclical buffer of zero to 2.5% (at the discretion of national regulators) are required. For global systemically important banks (G-SIBs), up to an additional 3.5% of common equity can be mandated. Some academics believe that Basel III's reforms are inadequate and that even greater capital should be required because bank equity capital is not socially costly (e.g., Admati et al. (2013)). In contrast, our theory and empirical findings reveal a potentially important social cost to higher bank equity that will occur unless banks’ corporate taxes are greatly reduced or eliminated. Solely raising bank capital requirements creates incentives to expand “shadow” banking, including loan securitization, thereby reducing equilibrium credit-screening and monitoring and raising average loan losses above the socially optimal level.8

The remainder of the paper is organized as follows. The next section presents a model of a corporate tax-paying bank that has loan origination opportunities that can be funded on- or off-balance sheet. We solve for the conditions that must be satisfied for the bank to have an incentive to sell loans. Section II provides empirical tests of the model's predictions using data on U.S. commercial banks’ mortgage originations and sales during the years 2001 to 2008. Section III concludes.

I. A Model of Bank Loan Sales

The model extends the state-pricing framework in Pennacchi (1988) to consider banks whose costs of deposits may be inelastic and whose assets may include securities as well as loans. Because of imperfect competition in retail deposit markets and some banks’ lack of easy access to wholesale funding, the interest rate that a bank pays on insured retail deposits and wholesale funds may differ from perfectly competitive rates, leading to a rising marginal cost of deposit financing. Market power in retail deposit markets can also motivate banks’ purchase of securities when loan opportunities are limited.

We first discuss the model's assumptions and the bank's maximization problem, and then state propositions that give the model's implications for how corporate taxes and capital requirements affect a bank's loan origination and sale choices. Detailed proofs of the propositions are given in the Appendix.

A. Assumptions

  1. Each period a bank faces multiple lending opportunities. Potential borrowers are heterogeneous, and a one-unit loan made to borrower i returns the cash flow xi(s,ai) at the end of the period, where s is the end-of-period state of nature and ai is the amount of credit-screening and/or monitoring that the bank performs on the borrower at the beginning of the period. A possibly infinite number of states are indexed to an interval of the real line, S, that is, sS.
  2. Credit-screening/monitoring improves loan payments such that xi(s,ai) is a weakly increasing and concave function of ai. The bank's cost of credit-screening/monitoring equals c×ai, where c is a constant marginal cost incurred at the end of the period. Moreover, the amount of credit-screening/monitoring, ai, is not verifiable by a court and hence not contractible.
  3. The bank's income is subject to taxation at the corporate tax rate τ. Also, regulation requires a minimum equity capital-to-deposit ratio. Denoting by E and D the bank's total amounts of equity (capital) and deposit (debt) financing, respectively, this leverage constraint is κDE, where κ is the minimum equity-to-deposit ratio.
  4. Due to differences in personal taxation or investor clienteles, the competitive state prices of equity and debt claims may differ. Let pe(s) and pd(s) be the beginning-of-period prices (densities) of unit end-of-period payments in state s for securities that are equity and debt, respectively. Then, the certainty-equivalent competitive returns to equity holders and debt holders are defined as re and rd, respectively, where 1/1+rjSpjsds, j = e, d. The ratio pe(s)/pd(s) is assumed to be constant across states, and therefore equal to (1+rd)/(1+re). Furthermore, rd(1−τ) < re.
  5. In addition to making loans, a bank can invest in debt securities that have state-contingent returns. Securities differ from loans in that screening/monitoring by the bank cannot increase the securities’ returns. Securities’ certainty-equivalent rate of return equals the competitive rate rd.
  6. Due to imperfect competition, the interest rate that the bank pays on its deposits can differ from the competitive certainty-equivalent rate rd. Let rD = rD(D) be the bank's actual certainty-equivalent deposit rate, with ∂rD/∂D ≥ 0 since a higher rate is needed to attract additional deposits.
  7. A bank's objective is to maximize the after-tax return to shareholders’ equity. Deposits are assumed to be government-insured at a premium that fairly reflects the deposit insurer's losses due to the bank's risk of failure.

In our model, banks can improve the returns on loans by credit-screening and monitoring borrowers (Assumption 1). While credit-screening/monitoring is costly (Assumption 2), models such as Diamond (1984) predict that banks are the most cost efficient because they can avoid duplication and free-rider problems. The assumption that the levels of credit-screening/monitoring are not verifiable and contractible by banks and loan buyers allows us to study the moral hazard implications of loan securitization.9

The model is meant to examine the impact of corporate taxes (Assumption 3) on the incentive to securitize loans. Banks cannot completely avoid corporate taxes because, consistent with regulatory policy, they face a minimum equity capital ratio that limits their tax shield from issuing debt (deposits). Since income taxes are paid at both corporate and personal levels, the model allows for different competitive returns on debt and equity securities arising from differences in personal taxation or investor clienteles (Assumption 4). The assumption that the ratio of equity-to-debt state prices, pe(s)/pd(s), is constant across states simplifies our analysis.10 The assumption rd(1−τ) < re is supported by empirical evidence, such as Graham (2000), showing that there is a net tax advantage to debt financing, even when differences in the personal taxation of debt and equity are taken into account.

Besides making loans, banks may invest in competitively priced securities (Assumption 5). As we show below, a bank's decision to purchase securities depends on its loan-making and deposit-issuing opportunities. Several empirical studies show that many banks exert market power when issuing FDIC-insured deposits.11 This inelasticity of deposit supply by retail depositors is modeled by supply's positive relation with the deposit interest rate (Assumption 6).

Lastly, Assumption 7 assumes fairly priced deposit insurance. While, in practice, deposit insurance is often subsidized, we wish to focus our model on the impact of taxes and not insurance distortions. Moreover, insurance subsidies are unlikely to be systematically linked to corporate tax differences across banks. A deposit insurance subsidy may add an additional incentive to fund loans on-balance sheet, but it would not change our qualitative results regarding the effect of taxes on loan securitization.12

B. The Bank's Optimization Problem

The bank makes funding, investment, and credit-screening/monitoring decisions. Funding involves the choice of deposits and equity and possibly which loans to sell, with any profits from loan sales also being available to fund investments. Investment decisions involve originating loans and possibly purchasing securities.

We assume that, if a bank sells loans, it does so without recourse, that is, it retains none of the risk of the loan that is sold. Pennacchi (1988) and Gorton and Pennacchi (1995) examine situations in which banks may wish to retain some of the sold loan's risk. However, for many loan types, such as mortgages, the originating bank often retains no risk of the loan it sells.13 This commonly occurs when small- or medium-sized banks sell loans purchased by larger bank underwriters who securitize them by combining them with loans that they, themselves, originated or with loans purchased from other banks. It also occurs when banks sell loans to government-sponsored enterprises (GSEs), such as Fannie Mae and Freddie Mac. To more closely match this institutional practice, we assume that a bank's choice is to either fully retain or fully sell a loan.14

Let Nh be the number of unit-sized loans that a bank originates and holds to maturity, and define xihs,ai as the end-of-period cash flow from such a loan paid by borrower i in state s. Similarly, let Nm be the number of loans that a bank originates and sells (markets) to loan buyers, and define xims,0 as the cash flow from such a loan paid by borrower i in state s. Since the bank has no incentive to perform costly credit-screening/monitoring for loans that it sells, ai = 0. Also, define B to be the amount of securities (bonds) purchased by the bank.

If the bank's objective is to maximize the after-tax return to shareholders’ equity, and its insured deposits are fairly priced, then the Appendix shows that the bank's objective function is


subject to the financing and leverage constraints


The expression i=1Nh1+reSpesxihs,aids1cai in objective function (1) is the shareholders’ end-of-period value of the bank's Nh retained loans net of the unit amount lent and the bank's cost of screening/monitoring the loan. The term i=1NmSpdsxims,0ds1 is the bank's profit earned at the beginning of the period from selling Nm loans. Because the state-contingent cash flows from the loans are taxed at the personal level as debt, the competitive price paid by loan buyers reflects this tax status. Also, since credit-screening/monitoring is not contractible, the bank does not perform these services when selling the loans and the competitive price paid by loan buyers reflects this moral hazard.15 After adding in competitive securities returns of rdB, and subtracting total deposit interest expense of rDD, shareholders’ after-tax return is then maximized relative to the certainty-equivalent opportunity cost of capital, reE. Constraint (2) shows that deposits, equity, and the bank's profit from loan sales are used to fund the loans and securities that the bank holds on its balance sheet.

C. Model Results

To better understand the factors determining the incentive to sell loans, let us begin by considering a bank's optimal investment, funding, and credit-screening/monitoring decisions in the absence of a loan sales (securitization) market. We then analyze how opening a loan sales market changes a bank's decisions.

C.1. Equilibria without a Securitization Market

When Nm is restricted to equal zero so that all loans are funded with deposits and equity, the bank optimally credit-screens/monitors borrower i at a level ai such that the marginal increase in value from monitoring equals the present value of the end-of-period marginal cost:


Now let λf and λk be the Lagrange multipliers on the financing constraint (2) and the capital constraint (3), respectively. Then, as shown in the Appendix, the bank's optimal decisions lead to one of the following three possible equilibria:

E1. Loan Poor, Deposit Rich: In this equilibrium, the functions characterizing a bank's loans and deposits, xi(s, ai), i = 1,…, Nh, and rD(D), are such that the bank has limited lending opportunities but substantial deposit market power. The bank makes loans to the point where the least profitable loan's return equals the certainty-equivalent return to investing in securities:


The bank's excess of “cheap” deposits is invested in securities and, because equity is relatively expensive, the bank's capital constraint binds, λk > 0. Thus, deposits are issued until the marginal weighted cost of funding equals the certainty-equivalent security return:


E2. Loan Rich, Deposit Poor: In this equilibrium, the bank has many profitable loans but little deposit market power. Deposits are issued to the point where their marginal cost equals the tax-adjusted cost of equity, and any additional funding is done with equity so that the bank's capital constraint does not bind, λk = 0. In this case, the bank's tax-adjusted marginal cost of financing is


and the amount of loans originated, Nh, is such that the least profitable (marginal) loan satisfies


Furthermore, a bank in this situation would not invest in securities since their certainty-equivalent return is less than the marginal cost of funding: rd < re/(1−τ) by Assumption 4.

E3 Loan and Deposit Compatibility: In this equilibrium, the bank's lending opportunities and deposit funding costs are compatible such that its capital constraint binds λk > 0 and its marginal cost of funding is between that of E1 and E2: rd < λf/(1−τ) < re/(1−τ), implying that it is not profitable to invest in securities. A special case is a bank that has access to a perfectly elastic supply of competitively priced brokered or wholesale deposits costing rd, which leads to


C.2. Equilibria with a Securitization Market

Now consider opening a loan-selling market such that Nm may be nonzero. The profit from retaining on the bank's balance sheet a loan that is funded with deposits and equity is


where ai*is the efficient level of credit-screening/monitoring that satisfies (4). Instead, if this loan were sold, its profit to the bank is


Combining (10) and (11), the excess profit from retaining versus selling the loan can be written as


The first term in (12) is unambiguously nonnegative, since it equals the increase in value from efficient credit-screening/monitoring. Considering the second term, if a bank faces an E1 loan poor, deposit rich equilibrium, and thus from (6) λf/(1−τ) = rd, then the second term in (12) is zero. Consequently, for such a bank, it is never profitable to sell loans and a marginal increase in its corporate tax rate (τ) or capital requirement (κ) has no effect on its loan-selling incentive.

In contrast, suppose that a bank faces an E2 loan rich, deposit poor equilibrium. In this case with λf/(1−τ) = re/(1−τ), the second term in (12) equals −[re/(1−τ) − rd]/(1+rd), which is negative. Thus, if the loss from inefficient monitoring is sufficiently small, the sum of the two terms in (12) could be negative, leading to a profitable loan sale. Since λf/(1−τ) > rd for a bank in the E3 loan and deposit compatibility equilibrium, a similar loan-selling incentive is possible. Thus, banks with sufficient loan-making relative to deposit funding opportunities, reflected in the fact that their cost of funding exceeds rd, are candidates to sell loans. Importantly, developments over the last few decades suggest that more banks face an E2 or E3 equilibrium, rather than an E1 equilibrium. Much research, as reviewed by Berger, Kashyap, et al. (1995) and Berger, Demirgüç-Kunt et al. (2004) discusses how entry by money market mutual funds and the elimination of bank branching restrictions have raised competition for retail deposits. Consequently, a greater cost of deposit funding, along with the corporate tax disadvantage of on-balance sheet funding, has made loan-selling relatively more attractive. Therefore, while the size of this tax wedge may not have changed substantially, increasingly competitive deposit funding could explain the growth in securitization.

Several observations for banks subject to equilibria E2 and E3 can now be made. First, with the option of selling loans, the bank may originate a greater number of loans relative to when a loan-selling market does not exist. This occurs when there are one or more loans for which condition (10) with λf/(1−τ) > rd is negative but condition (12) is positive. In other words, it is unprofitable to make the loan if it were funded on-balance sheet at a cost exceeding rd but it would be profitable to sell the loan, essentially funding it at rd, even though it would not be efficiently credit-screened/monitored. Such loans are likely to be made to borrowers requiring relatively little screening and monitoring.

Second, a marginal increase in the corporate tax rate, and possibly capital requirements, makes it more likely that loans will be sold. From condition (7), banks facing equilibrium E2 have a cost of on-balance sheet funding of λf/(1−τ) = re/(1−τ), which is increasing in its corporate tax rate. Moreover, from condition (9), banks facing equilibrium E3 have a cost of on-balance sheet funding of λf/(1−τ) = rd/(1+κ) + κre/[(1−τ)(1+κ)], which is increasing in both its corporate tax rate and its capital requirement. Because the profits from loan-selling in (12) are independent of the tax rate and capital requirement while the profits from holding them on-balance sheet are not, loan-selling becomes relatively more attractive as the tax rate and capital requirement rise.

Third, a related implication is that, as the marginal tax rate increases and more loans are shifted from on-balance sheet to be sold, the equilibrium E2 loan rich, deposit poor bank's excess capital decreases so that its leverage must increase. Still, relative to the E1 loan poor, deposit rich and the E3 loan and deposit compatibility banks that hold no excess capital, the E2 bank could, in equilibrium, continue to have some excess capital. Such a prediction is consistent with the empirical evidence in Minton, Sanders, and Strahan (2004), who find that better capitalized banks are more likely to securitize loans.

To summarize this section, we can state the two following propositions.

Proposition 1. A bank that has limited lending opportunities and substantial deposit market power will invest in securities, and a marginal increase in its corporate tax rate has no effect on leverage or its incentive to sell loans. A marginally higher tax rate or equity capital requirement decreases its securities purchased but not the quantity of loans held on its balance sheet.

Proof: See the Appendix.

Proposition 2. A bank that has substantial lending opportunities and limited deposit market power chooses not to invest in securities, and a marginal increase in its corporate tax rate increases its incentive to sell loans and can raise its leverage. A marginal increase in the bank's equity capital requirement also raises its incentive to sell loans.

Proof: See the Appendix.

The next section tests whether differences in U.S. banks’ corporate tax rates are linked to loan-selling in the manner predicted by the above propositions. In particular, we examine whether banks located in markets with substantial lending opportunities but limited deposit market power sell a greater proportion of the mortgages that they originate when they face a higher corporate income tax rate. Variation in corporate income tax rates comes from differences in corporate tax rates imposed by state governments. Variation in lending versus deposit-issuing opportunities comes from the demographic characteristics of the MSA in which each bank operates.

II. Empirical Evidence

We first describe our data and then discuss different empirical tests and their results.

A. Data Sources

Our empirical work uses information on individual banks’ mortgage originations and sales from HMDA filings.16 Of all commercial banks that filed FDIC Call Reports over the 2001 to 2008 sample period (59,636 bank-year observations), 49.2% of them filed an HMDA report. Banks are not required to submit a report if their total assets are below a threshold and if all of their offices and branches are outside an MSA.17 Thus, our sample excludes small rural banks.

To accurately identify the state corporate income tax rate that each bank faces, we limit our sample to banks with single-state operations, defined as a bank having over 90% of its deposits issued by branches in the same state during a given year.18 Branch-level deposit data for each bank come from the FDIC's Summary of Deposits. Exclusion of multistate banks further reduces the sample by 5.2%. We also restrict our sample to banks that are subject to corporate income taxes, that is, those organized as C-corporations. Banks organized as S-corporations, which are exempt from federal and some state corporate income taxes, were eliminated, further reducing the sample by 22.9% to 21,452 bank-year observations.19 Finally, most of our analysis uses the demographic characteristics of the market in which a bank operates, and so we also restrict the sample to those banks having at least 90% of their deposits from branches in one MSA. This final sample consists of 12,175 bank-year observations.

The HMDA data reports the individual one- to four-family residential mortgages that a bank originated during each calendar year. It also reports whether the mortgage was sold in that calendar year. Consequently, we can construct an annual flow of mortgage sales to mortgages originated by defining a bank's mortgage sales ratio (MSR) during a particular year as20


The MSR that we calculate may understate a bank's true ratio because HMDA only records mortgage sales that occur during the calendar year in which the mortgage was originated, thereby missing mortgage sales that occur after the year's end. If there is a lag between the time when a bank originates a mortgage and the time when it is sold, mortgage sales will be unreported for originations near the end of a calendar year.21 While this underreporting of sales downward-biases our MSR, there is no apparent rationale for the bias to be correlated with state corporate income tax rates and influence our empirical tests.

Some of our analysis differentiates between jumbo and nonjumbo mortgages, where jumbo mortgages have a loan amount that exceeds the conforming limits of U.S. GSEs, such as Fannie Mae and Freddie Mac, for the MSA in which the bank operates. Since GSEs do not purchase jumbo mortgages, it may be less easy to sell them (Loutskina and Strahan (2009)). Hence, some of our tests include the proportion of nonjumbo to total mortgages originated as a control variable. Other tests calculate MSR using only nonjumbo or jumbo mortgages.

In addition to HMDA mortgage information, we collect data from several sources to construct control variables for our empirical tests. From Call Reports, we calculate each bank's total operating costs based on annual averages of quarterly data.22 For our analysis, a bank's operating costs, rather than (on-balance sheet) assets, may be a better proxy for size because assets will be directly affected by the mortgage selling decision. Call Reports are also used to determine if a bank reported negative net income (losses), since loss-making banks may have less of a corporate income tax incentive to sell loans. A proxy for the average interest rate paid on savings and time deposits is calculated by dividing annual nontransactions deposit interest expense by annual average nontransactions deposit balances.23

State corporate income tax rates are obtained from the Tax Foundation.24 A few states permit a proportion of federal income taxes to be deducted from pretax income prior to calculating state corporate income taxes. Deductibility of federal taxes lowers the effective state income tax rate. If tF is the federal corporate income tax rate while p is the deductible proportion of federal taxes, then, when tS is the state corporate tax rate applied to income net of federal tax deductions, the effective pretax state income tax rate is (1 – p tF)tS. Similarly, federal tax rules permit deductions of state income taxes. Combining federal and state deductions implies a total effective corporate income tax rate of approximately τ = tF + (1 – p tF)tStF tS [1 − p(tF + tS)]. Our empirical work uses this effective corporate income tax rate assuming tF = 35%.

To control for characteristics of a bank's local market, we collect a variety of MSA-level demographic, geographic, and economic data. Following Becker (2007), we proxy for a banking market's relative deposit supply to loan demand using the proportion of an MSA's population that is aged 65 and above. This proportion of “seniors” comes from Census Bureau projections for each year based on the 2000 census. Also, we calculate each bank's deposit-weighted Herfindahl-Hirschman Index (HHI) from MSA and county deposit data obtained from the FDIC's Summary of Deposits. To control for land availability that may influence mortgage demand, we obtain from Saiz (2010) a housing supply elasticity for each MSA that is constructed from satellite-based geographic data and measures the availability of developable land.25 We also obtain annual MSA-level data on population growth, growth in personal income, and the rate of unemployment. The sources of this data are the U.S. Department of Commerce's Bureau of the Census and Bureau of Economic Analysis and the U.S. Department of Labor's Bureau of Labor Statistics.

B. Summary Statistics

Summary statistics are given in Tables I and II. The first row of Table I, Panel A, shows the total numbers of banks in our sample for each year from 2001 to 2008. We find that, for 35.4% of all bank-year observations, at least some sales of mortgages were originated in the same calendar year, with 34.7% of the banks reporting sales of nonjumbo mortgages and 19.1% reporting sales of jumbo mortgages. The next rows of Panel A give the average MSR across all banks for each year. Over the entire sample, the average annual MSR was 16.7%. Hence, conditional on a bank selling mortgages, its ratio of mortgages sold to mortgages originated was 16.7 ÷ 35.4 = 47.1%. Because some of our tests analyze mortgage sales based on whether the mortgage was jumbo or not, Panel A also calculates the average MSRs when only nonjumbo and only jumbo mortgages are used. As expected, the MSR is higher for nonjumbo mortgages (18.4%) compared to jumbo mortgages (11.3%).

Table I. Summary Statistics for Single State, Single MSA Commercial Banks, 2001 to 2008
This table presents summary statistics for our sample of banks that comprise commercial banks organized as C-corporations and that have at least 90% of their deposits issued by branches located in a single MSA and a single state. Mortgage data come from HMDA filings. Jumbo mortgages have amounts exceeding the conventional conforming loan limits of GSEs. MSRs equal the dollar amount of mortgages both originated and sold during a calendar year divided by the dollar amount of all mortgages originated during that calendar year. Operating costs equal the sum of salaries and employee benefits, expenses of premises and fixed assets, and other noninterest expenses. Deposit-weighted HHI is calculated from MSA or county deposit data obtained from the FDIC's Summary of Deposits.
A. Numbers of Banks and Average Mortgage Sales Ratio by Year
Total Number of Single-State/MSA Banks1,6331,6111,5651,5961,5231,4631,4051,37912,175
Proportion of Banks with Mortgage Sales (%)34.5435.3236.2937.2236.6435.2734.3833.4335.43
Nonjumbo mortgages33.8034.7635.5336.4735.7234.4533.8832.9934.74
Jumbo mortgages16.5321.8524.7922.1823.3716.6116.589.0619.06
MSR (average%)16.9618.6420.1316.6317.0114.8114.9413.6716.69
Nonjumbo MSR (average%)18.3420.0521.7218.2218.5516.3216.9316.6218.42
Jumbo MSR (average%)10.4213.6314.6712.5814.7010.218.833.8311.26
B. Bank Characteristics
 Obs.MeanStd. Dev.P(0.01)P(0.25)MedianP(0.75)P(0.99)
Total assets ($ Millions)12,175727.768,037.1636.4890.05168.87343.425,645.99
Operating costs ($ Millions)12,17528.39497.090.982.694.689.25154.22
Interest expense/Deposits (%)12,1752.913.740.712.042.753.635.49
Dummy for net income < 0 in current or previous year12,1750.150.350.
Proportion of nonjumbo mortgages (%)12,17570.8728.770.0051.9080.1295.43100.00
Deposit-weighted HHI12,1751,084.33607.18410.73695.67980.391,327.433,443.80
Table II. Summary Statistics for Population, Economic Conditions, and Effective Corporate Tax Rates by State, 2001 to 2008
This table reports summary statistics on MSA-level population, economic conditions, and effective corporate tax rates within the same state. Data on population aged 65 and above and population growth by state come from the Commerce Department's Census Bureau. CPI inflation-adjusted personal income growth and unemployment rates by state are obtained from the Commerce Department's Bureau of Economic Analysis and the Department of Labor's Bureau of Labor Statistics. Housing supply elasticity, an MSA-level measure of the availability of developable land for the year 2000, is from Saiz (2010) and is not available for Alaska, Hawaii, and Wyoming. State corporate income tax rates are from the Tax Foundation.
 Population ≥ Real Personal Housing 
StateAge 65 (ProportionPopulationIncomeUnemploymentSupply 
 of Seniors (%))GrowthGrowthRateElasticityEffective State Tax Rate, 2001 to 2008
New Hampshire11.220.570.544.000.868.528.718.00
New Jersey12.830.551.435.241.408.679.007.45
New Mexico11.011.823.655.112.125.605.605.60
New York14.450.101.354.951.877.528.007.10
North Carolina12.311.722.005.762.506.906.906.90
North Dakota11.921.323.352.964.628.009.007.00
Rhode Island14.640.121.565.601.619.009.009.00
South Carolina11.641.752.176.082.505.005.005.00
South Dakota11.101.853.542.904.480.000.000.00
West Virginia14.190.421.784.952.128.989.008.75
District of Columbia9.071.772.193.541.619.979.979.97

Panel B of Table I gives the distribution of bank characteristics for our sample of 12,175 bank-year observations. The average and median total asset sizes for our sample banks are $728 million and $169 million, respectively, while the sample banks’ average and median operating costs are $28.4 million and $4.7 million, respectively.26 For 15% of the bank-year observations, a bank reported a loss for the current or previous year. The median proportion of nonjumbo to total mortgages originated is 80%. The median HHI for the markets in which banks operate is 980.

Table II reports summary statistics by state averaged over the years 2001 to 2008. The first column displays the state population's proportion of individuals aged 65 and above.27 The next four columns show each state's average rates of MSA-level population growth, CPI inflation-adjusted personal income growth, unemployment, and housing supply elasticity. The last three columns give each state's average, minimum, and maximum effective corporate income tax rate over the sample period. There is significant cross-sectional variation, with eight states having average effective rates of at least 9% and six states having rates below 2.25%. There is much less variation in tax rates over time.

Figure 1 graphically displays the 2001 to 2008 average proportion of seniors (aged 65 and above) in the population and corporate income tax rate by state. The states in black (white) have a senior population below (above) the median and a state corporate income tax rate above (below) the median. The states in darker (lighter) gray have a senior population above (below) the median and a state corporate income tax rate above (below) the median. In parentheses are the percentages of seniors and the effective state corporate income tax rate. Although there are several exceptions, northeastern states generally have older populations and high tax rates, while midwestern states have older populations and low tax rates. Much of the south has younger populations and low tax rates. Notably, Florida has an old population and low tax rates while California has a young population and high tax rates. Our model predicts that banks operating in states like California, with high taxes and loan demand that is high relative to deposits, have the greatest incentive to sell loans.

Figure 1.

Proportion of seniors and corporate income tax rate by state.

This figure graphically displays the 2001 to 2008 average proportion of seniors (aged 65 and above) in the population and corporate income tax rate by state. The states in black (white) have a senior population below (above) the median and have a state corporate income tax rate above (below) the median. The states in darker (lighter) gray have a senior population above (below) the median and a state corporate income tax rate above (below) the median. In parentheses are the proportion of seniors (in %) and the effective state corporate income tax rate (in %).

C. Univariate Tests

As summarized in our model's Proposition 1, banks that operate in MSAs with limited lending opportunities but substantial deposit market power, which are proxied by MSAs with a high proportion of seniors, will not have a tax incentive to sell loans. In contrast, Proposition 2 predicts that banks operating in MSAs with substantial lending opportunities but limited deposit market power, typically MSAs with relatively few seniors, are candidates for loan-selling, and the higher the corporate income tax rate that they face, the greater is their incentive to sell loans.

Clearly, our model is stylized—it considers only a tax-related motive for selling loans. More generally, there can be risk-management reasons for loan sales. Retaining long-duration loans, such as fixed-rate mortgages, exposes a bank to interest rate risk when the bank's primary source of funding is short-duration deposits. In addition, our sample's banks operate in a single MSA so that the credit risk of their mortgage portfolio may lack geographic diversification and be sensitive to local economic conditions. The ability to transfer interest rate risk and local-economy credit risk to loan buyers is an additional rationale for mortgage sales, and higher tax rates may tip the balance to transferring these risks off-balance sheet rather than retaining them, particularly if these on-balance sheet risks require higher equity capital.

While risk management may also generate a desire to sell loans, it is unlikely to negate our model's basic prediction that the corporate tax asymmetry between banks and special purpose vehicles will lead higher taxed banks to sell more loans. The model's additional insight should also hold, namely, that loan-selling will be more prevalent when a bank does not enjoy low cost deposit funding, as when its market is populated with younger individuals rather than seniors.

Table III provides a first look at whether the data roughly coincide with our model's predictions. The first row compares the average MSR of banks operating in high tax states versus those operating in low tax states, where our sample of bank-year observations is split into halves based on whether a bank's effective corporate tax rate is lower versus higher than the median. The results show that banks operating in high tax states tend to sell a significantly larger fraction of the mortgages that they originate (18.8%) compared to banks in low tax states (14.2%). The difference is statistically significant with a p-value less than 1%. As our theory predicts, taxes raise the cost of equity financing and, all else equal, provide an incentive to fund loans off-balance sheet.

Table III. Univariate Tests for Mortgage Sales Ratios by State Corporate Income Tax Rate and Senior Population, 2001 to 2008
This table tests for differences in MSRs of banks whose state corporate income tax rate is higher than the sample median versus banks whose state corporate income tax rate is lower than the sample median. The table also compares the MSRs of banks whose MSA's proportion of seniors (proportion of the population aged 65 or above) is higher than the sample median versus lower than the sample median. The MSR is defined as the ratio of mortgages originated and sold during the calendar year to total mortgages originated during the calendar year.
 2001 to 2008 Average ofDifferenceRank-sum Z
 Mortgage Sales Ratio(High-Low)statistics
 Low TaxHigh Tax  
All banks14.2418.814.5712.55
Banks in MSAs with a high proportion of seniors15.7216.630.915.05
Banks in MSAs with a low proportion of seniors13.1821.858.6711.99
p-value of Ha:MSR¯HighTaxLowSeniorsMSR¯LowTaxLowSeniors>MSR¯HighTaxHighSeniorsMSR¯LowTaxHighSeniors< 0.001

A somewhat sharper examination of our model's implications is given in the second and third rows of Table III, which report MSRs for our sample split in quarters. Row 2 (row 3) examines the half of observations associated with banks that operate in MSAs with a proportion of seniors that is higher (lower) than the sample median. Then, for each of these subsamples, bank-year observations are split into those for which the bank's tax rate is above (below) the subsample's median. The result is a division of the sample into quarters composed of banks with high senior/low tax, high senior/high tax, low senior/low tax, and low senior/high tax characteristics.

The results in row 2 of Table III are consistent with our model: a bank's incentive to sell mortgages is relatively insensitive to its corporate tax rate when it has limited loan origination opportunities but substantial deposit market power, a situation that occurs when there is a high proportion of seniors. The results in row 3 also strongly support the model. MSRs are significantly higher for the quarter of banks that have the lowest proportion of seniors and are located in the highest tax states (21.9%). The difference in MSRs of banks facing high versus low taxes is 8.7% for low senior banks and is statistically significant at a p-value less than 1%. Hence, this preliminary examination of the data is consistent with banks selling more loans when they operate in high tax states, particularly when they operate in MSAs with significant lending opportunities but limited deposit funding opportunities.

D. Multivariate Tests

The previous section's analysis does not control for other potential factors affecting the decision to sell mortgages. In this section, we investigate the relationship between corporate taxes and mortgage sales by incorporating other bank and state-level factors that may influence sales. Because mortgage loan sales ratios are bounded between zero and unity, we conduct a Tobit regression of the form


where MSR is the mortgage sales ratio defined in (13), TaxRate is an individual bank's total corporate tax in a given year, and ISeniors<Median is an indicator variable that equals one if the proportion of seniors in the bank's MSA is lower than the median for all bank-years and zero otherwise. Our model predicts that the sum of the coefficients a1 plus a2 should be positive, since higher taxes should lead to greater mortgage sales primarily for banks operating in markets with substantial loan origination opportunities but limited deposit market power, which are MSAs characterized by a low proportion of seniors. The coefficient a1 by itself is the sensitivity of loan sales of high securities banks to tax rates. As discussed earlier, it need not be the case that these banks’ mortgage sales respond to tax rates.

Our tests consider the following bank-level controls. First, we include bank size, as proxied by the natural log of operating costs, as it may influence the MSR if fixed costs of initiating mortgage sales generate economies of scale. Second, we include two proxies for competition in the bank's deposit market that should increase the incentive to sell mortgages: the average interest rate on deposits, and HHI, the deposit-weighted Herfindahl-Hirschman Index for the markets in which each bank operates. Third, we include the indicator variable INetIncome<0, equal to one if the bank reported losses (negative net income) in the current or previous year, since a loss-making bank is expected to have a smaller tax incentive to sell mortgages. Fourth, we include the proportion of nonjumbo to total mortgages that the bank originated, which should be positively related to the MSR.

We also include controls for local economic, demographic, and geographic factors, in particular, the MSA's rates of unemployment, personal income growth, and population growth, as well as the MSA's housing supply elasticity. These local conditions may affect housing demand and thereby determine whether a bank faces substantial or limited mortgage loan origination opportunities. In addition, the regressions control for year fixed effects, such as macroeconomic conditions or financial innovations that could affect mortgage sales. The bank and local market variables used in the regressions are winsorized at the 1% and 99% levels, though our results are not sensitive to this adjustment of outliers.

Tobit regression results are given in Table IV. All estimates are calculated as the variables’ marginal effects at their means and the reported p-values in parentheses are adjusted for clustering of errors at the state level. Column 1 reports results for a regression where a1, a2, and a3 are all restricted to zero so that determinants of mortgage sales due to taxes and the proportion of seniors are ignored. One sees that most of the remaining control variables have the expected signs. In particular, the log of operating costs is positive and statistically significant, consistent with economies of scale in mortgage selling. The bank's deposit interest rate, a proxy for the cost of on-balance sheet funding, is also a significantly positive determinant of mortgage sales. Banks with negative net income are less likely to sell loans while those originating mostly nonjumbo mortgages are more likely to sell them.

Table IV. Tobit Analysis of Mortgage Sales Ratios
This table reports the estimated marginal effects from Tobit regressions of MSR=a1TaxRate+a2ISeniors<Median×TaxRate+a3ISeniors<Median+BankControls+MSAControls+TimeEffects. The dependent variable, MSR, is the ratio of a bank's mortgages originated and sold during the year to total mortgages originated during the year. The sample is over the 2001 to 2008 period. TaxRate is the federal plus state corporate income tax rates adjusted for deductions. ISeniors<Median equals one if the bank is located in an MSA whose population's proportion of seniors is below the sample median for a given year, and zero otherwise. Bank Size is captured by operating costs. Deposit-weighted HHI is calculated from MSA or county deposit data obtained from the FDIC's Summary of Deposits. INet Income < 0 equals one if the bank made losses in the current or previous year. Nonjumbo mortgages have amounts that do not exceed GSE conforming loan limits. Housing Supply Elasticity, an MSA-level measure of the availability of developable land for 2000, is from Saiz (2010). Personal Income Growth comes from the Bureau of Economic Analysis. Unemployment Rate and Population Growth Rate by MSA are obtained from the Bureau of Labor Statistics and the Census Bureau, respectively. Year fixed effects are included and t-statistics are adjusted for clustering at the state level. p-values are in parentheses. All estimates are calculated as the variables’ marginal effects at their means. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively.
TaxRate 1.087**−0.561
ISeniors<Median ×TaxRate  2.796***
ISeniors<Median  −0.913***
ln(Bank Size)0.075***0.075***0.075***
Interest Rate on Deposits1.965**2.058***1.827**
INet Income < 0−0.009−0.009−0.009
Nonjumbo Mortgage Proportion0.269***0.266***0.271***
Housing Supply Elasticity0.0050.0080.007
Population Growth Rate−1.660−1.025−1.099
Personal Income Growth Rate0.0410.0970.096
Unemployment Rate−0.605−0.129−0.139
p-value of Ha: a1+a2 > 0  < 0.001
Pseudo R20.0910.0940.100

Column 2 adds TaxRate to the regression. Its coefficient, a1 = 1.09, is positive and significant with a p-value of 2%. The regression in column 3 adds controls for the lending and deposit-taking opportunities in the bank's MSA by including ISeniors<Median and its interaction with TaxRate. Here, one sees that the independent effect of taxes in MSAs with younger loan rich, deposit poor populations, the sum of a1 and a2, equals 2.24 and is highly statistically significant. The implication is that a 1% increase in the corporate tax rate increases the proportion of mortgages sold by 2.24%. This model prediction appears economically significant for banks operating in MSAs whose populations are younger than the median. Relative to the average MSR for these banks (17.1%), a one-standard-deviation increase in the corporate income tax rate of 1.88% raises mortgage sales by 2.24×1.88/17.1 = 24.6%.

Table V examines the robustness of these Tobit regressions by considering four modifications. Column 1 (Column 2) reports regression results using only nonjumbo (jumbo) mortgages to construct the MSR. In both cases, the effect of taxes in loan rich, deposit poor markets is positive and statistically significant, but as one might expect, the effect is larger for the easier-to-sell nonjumbo mortgages (a1+a2 = 2.46) compared to jumbo mortgages (a1+a2 = 1.53). Because the housing supply elasticity variable of Saiz (2010) is not available for some MSAs, column 3 of Table V drops this variable and includes all single-state, single-MSA banks. This procedure increases the number of bank-year observations from 12,175 to 13,701 but does not have a sizable effect on the results. Finally, column 4 runs a similar Tobit regression but does not restrict banks to have 90% of their deposits issued from branches in a single MSA; rather, it restricts banks to have 90% of their deposits in branches in a single state. In essence, this test assumes that each bank's relevant “market” is at the state, rather than MSA, level. By including multi-MSA banks as well as banks with significant deposits from rural (non-MSA) branches, the sample size increases to 21,452 bank-year observations. The explanatory variables in this regression, including the proportion of seniors in the population, are now measured at the state, rather than MSA, level. Broadening the definition of the market and the number of bank-year observations does not change the results. Higher corporate taxes increase mortgage sales only in loan rich, deposit poor states, as proxied by a below-median proportion of seniors in the state. The magnitude of the effect (a1+a2 = 1.45) is somewhat lower than when we use MSA-level measures (a1+a2 = 2.24), perhaps because the state is less precise than the MSA as a measure of most banks’ relevant loan and deposit market.

Table V. Tobit Analysis of Mortgage Sales Ratios: Robustness
This table reports Tobit estimation results of the following model for various subsamples: MSR=a1TaxRate+a2ISeniors<Median×TaxRate+a3ISeniors<Median+BankControls+MSA(orState)Controls+TimeEffects. The dependent variable, MSR, is the ratio of a bank's mortgages originated and sold during the year to total mortgages originated during the year. Definitions of the explanatory variables are given in Table IV. The sample is over the 2001 to 2008 period. Year fixed effects are included and t-statistics are adjusted for clustering at the state level. p-values are in parentheses. All estimates are calculated as the variables’ marginal effects at their means. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively.
   WithoutSingle-State Banks
 NonjumboJumboHousing Supplyand State Proportion
 Loans SalesLoan SalesElasticityof Seniors
ISeniors<Median ×TaxRate2.629***2.233***2.366***1.939**
ln(Bank Size)0.070***0.047***0.076***0.073***
Interest Rate on Deposits1.939**1.735**1.890***1.477*
INet Income < 0−0.034**0.006−0.005−0.014
Nonjumbo Mortgage Proportion0.030***0.015**0.279***0.330***
Housing Supply Elasticity−1.324−0.206  
Population Growth Rate−0.1100.280−1.266*−1.087
Personal Income Growth Rate0.0190.2110.177−0.135
Unemployment Rate−0.167−0.708−0.4460.458
p-value of Ha: a1+a2 > 0< 0.0010.004< 0.0010.007
Pseudo R20.0580.0570.0970.107

The Tobit regressions in Tables IV and V analyze the sensitivity of mortgage sales to tax rates based on whether a bank's market has below- or above-median seniors as a proxy for loan-making and deposit-funding differences. Essentially, these tests separate sample banks into two halves but constrain all regression coefficients to be equal except for the TaxRate. As a further robustness test, columns 1 and 2 of Table VI report separate Tobit regressions run on subsamples of the two halves of banks so that none of the regression coefficients need be equal.

Table VI. Tobit Analysis of Mortgage Sales Ratios: Senior Population Subsamples
This table reports Tobit estimation results of the following model for subsamples based on the MSA population's proportion of seniors: MSR=a1TaxRate+BankControls+MSAControls+TimeEffects. The dependent variable, MSR, is the ratio of a bank's mortgages originated and sold during the year to total mortgages originated during the year. Definitions of the explanatory variables are given in Table IV. The sample is over the 2001 to 2008 period. Year fixed effects are included and t-statistics are adjusted for clustering at the state level. p-values are in parentheses. All estimates are calculated as the variables’ marginal effects at their means. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively.
 Banks inBanks inBanks inBanks inBanks in
 MSAsMSAsMSAs withMSAs withMSAs with
 with Seniorswith Seniorsthe Lowestthe Middlethe Highest
 Below theabove theTercile ofTercile ofTercile of
ln(Bank Size)0.072***0.077***0.074***0.068***0.079***
Interest Rate on Deposits1.951*1.534**1.9212.119**0.791
INet Income< 0−0.0260.008−0.016−0.0220.020
Nonjumbo Mortgage Proportion0.262***0.269***0.235***0.266***0.298***
Housing Supply Elasticity−0.0080.015*−0.0130.0080.016*
Population Growth Rate−1.773*0.239−1.343−1.4300.924
Personal Income Growth Rate0.2540.1150.243−0.0130.139
Unemployment Rate−2.0821.956**−2.843*0.5612.440*
p-value of Ha : a1LowSeniors>a1HighSeniors0.005 0.046 
Pseudo R20.1090.1010.1060.0940.125

Consistent with theory, TaxRate is only statistically significant for the regression run on the subsample of banks operating in MSAs that have a low proportion of seniors, that is, MSAs characterized by substantial lending opportunities but limited deposit market power. Columns 3, 4, and 5 of Table VI report the results of a similar exercise where the sample of banks is divided into thirds based on whether a bank operates in an MSA having a proportion of seniors in the lowest, middle, or highest third of all bank-year observations. Notably, the effect of tax rates on mortgage sales is greatest for the banks operating in MSAs having the lowest third of seniors. For these banks, a 1% increase in the corporate tax rate leads to a statistically significant 1.8% increase in the MSR. The coefficients on the tax rate variable decline to statistically insignificant values of 0.55 and −0.10 for the regressions using the subsamples of banks operating in MSAs having the middle and highest thirds of seniors, respectively. The last row of Table VI reports whether the coefficient on TaxRate for the lowest senior subsample of banks is greater than that for the highest senior subsample and shows that such an inequality can be established at a p-value of less than 5%.

We consider an additional specification in the form of Fama and MacBeth (1973) regressions. The results of these year-by-year regressions for banks operating in MSAs with a below-median proportion of seniors are given in Table VII. The coefficients on TaxRate are all positive and statistically significant at p-values of 1% or less for six out of the eight years, and the average coefficient is also statistically significant at a p-value below 1%. When we repeat these Fama and MacBeth (1973) regressions for the other half of banks operating in MSAs with an above-median proportion of seniors, the average coefficient on TaxRate is −0.03 and insignificantly different from zero.

Table VII. Fama-MacBeth Tobit Regressions of Mortgage Sales Ratios
This table reports Fama and MacBeth (1973) Tobit regressions of the following model for the subsample of banks in MSAs whose population's proportion of seniors was below the median for the given year: MSR=a1TaxRate+BankControls+MSAControls. The dependent variable, MSR, is the ratio of a bank's mortgages originated and sold during the year to total mortgages originated during the year. The definitions of the explanatory variables are given in Table IV. The sample is over the 2001 to 2008 period. p-values are in parentheses. All estimates are calculated as the variables’ marginal effects at their means. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively.
ln(Bank Size)0.072***0.080***0.087***0.064***0.080***0.067***0.061***0.061***0.072***
Interest Rate on Deposits3.208***2.5364.290**0.7862.7090.8871.3430.5312.036
INet Income<00.0130.0110.004−0.065**−0.082**−0.016−0.028−0.014−0.022
Nonjumbo Mortgage Proportion0.296***0.240***0.293***0.250***0.242***0.233***0.257***0.265***0.260***
Housing Supply Elasticity0.003−0.005−0.010−0.019*−0.023**−0.005−0.009−0.002−0.009
Population Growth Rate−1.537−0.096−0.0470.7161.136−4.299***−3.626***−1.609−1.170
Personal Income Growth Rate−0.3652.386***1.284−0.458−1.504***2.147*1.0380.1220.581
Unemployment Rate−1.0860.259−0.581−3.537**−2.870**−2.509−2.164−1.971−1.807

As shown in Table II, there is more cross-sectional variation than time-series variation in states’ corporate income tax rates. Still, in one last set of tests, we analyze the impact on mortgage sales of changes in a given state's tax policy. Table VIII reports regressions in which the dependent variable is the annual change in a bank's MSR, ΔMSR, and the explanatory tax rate variables are the annual change in the effective state tax rate for the contemporaneous year as well as one lag. Similar to the previous level regressions, these first-difference regressions show a statistically significant positive relationship between changes in MSRs and changes in state tax rates for banks operating in MSAs having a relatively low proportion of seniors. As shown at the bottom of the table, the sum of the coefficients on the contemporaneous and lagged tax rate variables is statistically different from zero, though the point estimate of 1.10 is somewhat less than the tax effect for the level regressions.28 However, the hypothesis that this sum is greater for banks in low senior MSAs versus banks in high-senior MSAs is established at a p-value of 7.9%.

Table VIII. Regressions of Annual Changes in Mortgage Sales Ratios
This table reports regressions of the following model for annual changes in MSRs: ΔMSRt=a0+j=01a1jΔTaxRatetj+ΔControlst+TimeEffects. MSR is the ratio of a bank's mortgages originated and sold during the year to total mortgages originated during the year. Definitions of the explanatory variables are given in Table IV. The sample is over the 2001 to 2008 period. Year fixed effects are included and t-statistics are adjusted for clustering at the state level. p-values are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively.
 Banks in MSAsBanks in MSAs
 with Seniorswith Seniors
 Below the MedianAbove the Median
Δln(Bank Sizet)0.048**0.060***
ΔDeposit Interest Ratet−0.886−0.315
ΔNet Income/Bank Sizet0.0230.003
ΔNonjumbo Mortgage Proportiont0.109***0.121***
ΔPopulation Growth Ratet−0.386−0.196
ΔPersonal Income Growth Ratet0.1400.062
ΔUnemployment Ratet0.131−0.269
p-value of Ha :aLowSeniors>aHighSeniors 0.079
Adjusted R20.0250.028

The paper's Internet Appendix contains additional robustness tests. The Internet Appendix is available in the online version of the article on the Journal of Finance website. It shows that the results are insensitive to excluding banks in large states such as California and New York or to excluding banks experiencing losses (negative net income). The results also continue to hold when we include a continuous variable for net income rather than the dummy indicating losses or when we use a continuous variable for the proportion of an MSA's population below the age of 65 rather than the above- and below-median dummy variable.

III. Conclusion

The recent financial crisis has highlighted the critical role that securitization plays in global credit markets. As policy makers reexamine the structure of securitization and propose reforms, it is important to understand banks’ motives for selling and securitizing loans. This paper emphasizes the impact of corporate income taxes in creating incentives to move loans off banks’ balance sheets. The model shows that banks with rich loan origination opportunities but limited deposit market power have the greatest incentive to sell or securitize loans in response to higher corporate taxes. The paper's empirical findings confirm the model's prediction. Banks operating in states that levy higher corporate income taxes are more likely to sell mortgages.

To identify the corporate tax rate that banks face, we limit our empirical tests to banks that operate in a single state and MSA. This sample selection, while necessary to conduct a precise test of our model's predictions, excludes larger, multistate banks that typically are the most prolific loan sellers. In all likelihood, the impact of corporate taxes on loan sales and securitization is even greater for these larger banks. Large multistate and international banks tend to have substantial loan origination opportunities that outstrip their limited retail deposit base. Their marginal cost of debt financing is a competitive wholesale deposit rate. Given that these banks are required to also use equity capital whose tax-adjusted cost exceeds that of debt, their relatively expensive on-balance sheet financing creates an incentive to sell loans to reduce funding costs. A similar analysis of nonbank lenders that depend on wholesale funding, such as major finance companies, also predicts that corporate taxation creates incentives for them to sell loans.

In the past, banks’ tax disadvantage relative to special purpose vehicles may have been partially offset by subsidized bank deposit insurance and lower-than-fair bank capital standards.29 While reforms to deposit insurance and capital standards are needed, the asymmetric taxation of banks versus special purpose vehicles needs to be taken into account. Neglecting this tax distortion, while strengthening deposit insurance and capital standards, may drive even more loans off-balance sheet and lead to even lower credit-screening and monitoring by banks.30 Other proposals to levy new taxes on banks for the purpose of funding past or future government assistance also may have this detrimental effect of encouraging excessive securitization. Tax reforms need to be designed carefully in conjunction with regulatory reforms to avoid worsening financial stability.

Editor: Campbell Harvey


A. Bank's Objective Function

This section shows that, if the bank's objective is to maximize the value of the after-tax return to shareholders’ equity, and the bank's deposits are fairly insured, then the bank's objective function takes the form of function (1) in the text.

Let b(s) be the payoff in state s of the bank's portfolio of securities that has the initial value of B. Also, note that the initial value of shareholders’ equity is E, which is also the bank's initial net worth. Define W(s) as the end-of-period after-tax net worth of the bank when state s occurs; W(s) equals


The terms in the curly brackets in (A1) represent the bank's before-tax income net of interest expenses. This includes the profits from on-balance sheet loans, loan sales, and security holdings, less the deduction for interest expense on deposits. The bank keeps (1−τ) of this income with the rest going to tax authorities. It is assumed that the bank pays an end-of-period deposit insurance premium of ϕD when W(s) is at least as big as this premium. When W(s) > ϕD, so that the bank has positive net worth after paying its premium, the end-of-period value of shareholders’ equity equals W(s) – ϕD. A bankruptcy state occurs if W(s) < ϕD, in which case shareholders receive zero and the deposit insurer suffers the loss −W(s). Let p(s) be the primitive security price that the deposit insurer uses to value its end-of-period cash flow when state s occurs. Then, if S+ denotes the set of positive net worth states and S notes the set of bankruptcy states, the fair premium satisfies


The present value of shareholders’ end-of-period payoff equalsS+pesWsϕDds. Substituting for ϕD in this expression using (A2) gives


It is assumed that p(s) is a constant proportion of pe(s) for all s. Therefore, from Assumption (4) in the text, p(s) is also a constant proportion of pd(s). This implies that (A3) equals


Also, note that, for the present value of the competitive security return given in W(s), we have


Therefore, the present value of the payoff to shareholders’ equity,SpesWsds, equals


The present value of the return on shareholders’ equity is therefore (A6) less initial equity, E. Subtracting E from (A6) and multiplying the entire expression by (1+re), which does not change the bank's optimal decisions, gives the objective function (1) in the text.

B. Model First-Order Conditions

Without loss of generality, let the loans that the bank originates and holds to maturity be ordered from highest to lowest based on their net present values, so that loan i = Nh is the “marginal” loan held by the bank. Similarly, let loans that the bank sells be ordered from highest to lowest based on their net present values, so that Nm is the marginal loan sold by the bank. Then, the first-order Kuhn-Tucker conditions for the general problem (1) are


C. Equilibria with No Securitization

With Nm = 0 and all loans that a bank originates being retained, equation (4) determines the optimal ai* for each retained loan. Then, condition (A7) determines the optimal quantity of loans by specifying the “marginal” (last) loan made by the bank. The term in square brackets in condition (A7) is the value of the return on the marginal loan net of monitoring costs. At the optimum, this equals the tax-adjusted marginal cost of financing, λf /(1−τ). This financing cost depends on the equilibrium supply of deposits relative to loans and whether the bank's capital constraint binds. Assume throughout that D > 0 so that (A10) binds. Then, if the capital constraint also binds, so that E = κD and λk > 0, the first equality in (6) holds. Moreover, if in equilibrium the bank purchases securities (B > 0), then, from (A12), it follows that the second equality in (6) holds. This implies the E1 loan poor, deposit rich equilibrium.

Instead, if the capital constraint does not bind, so that E > κD and λk = 0, equations (A10) and (A11) imply that the tax-adjusted cost of financing, λf/(1−τ), satisfies (7). Further, B > 0 could not be an equilibrium since from condition (A12) λf/(1−τ) = rd contradicts (7) due to Assumption (4). These circumstances then determine the E2 loan rich, deposit poor equilibrium.

If the capital constraint binds, then the first equality in (6) holds. However, if (A12) does not bind so that B = 0, then (A12) and Assumption (4) implies rd < λf/(1−τ) < re/(1−τ). This situation determines the E3 loan and deposit compatibility equilibrium.

D. Equilibria with Securitization and Proofs of Propositions 1 and 2

Recall from Assumption (4) that the ratio pe(s)/pd(s) is constant across states, and, therefore, equals (1+rd)/(1+re). Thus, if pe(s) = pd(s)(1+rd)/(1+re) is substituted into (A7) and one divides through by (1+rd), condition (10) results. The subsequent arguments made in the text prove Propositions 1 and 2.

  1. 1

    The total value of U.S. agency- and government-sponsored enterprise-backed mortgage pools and private issue mortgage, consumer, and trade credit loan pools grew at average annual continuously compounded rates of 33.5%, 25.2%, 12.9%, and 11.6% during the decades 1967 to 1977, 1977 to 1987, 1987 to 1997, and 1997 to 2007, respectively. Source: Flow of Funds Accounts of the United States, Board of Governors of the Federal Reserve System.

  2. 2

    Fender and Mitchell (2009) review the recent collapse of global securitization markets. Mian and Sufi (2009) and Keys et al. (2010) provide additional discussion of securitization with a focus on the U.S. subprime mortgage crisis.

  3. 3

    For example, Shaviro (2009) reviews how tax rules contributed to the 2008 financial crisis by encouraging excessive corporate debt, derivative transactions, housing leverage, and poorly designed incentive compensation schemes. But the tax incentive for excessive securitization is not mentioned.

  4. 4

    Leland (2007) shows that unused tax deductions when net operating losses occur create a corporate tax-minimizing incentive to place assets with different risks in particular corporate entities. This incentive can generate mergers, spin-offs, or asset securitizations. His analysis assumes that different entities are taxed symmetrically, whereas our model accounts for the corporate tax-exempt status of special purpose vehicles that hold securitized loans.

  5. 5

    Most U.S. special purpose vehicles are organized as limited liability corporations (LLCs). If the LLC passes through all loan income to the mortgage- or asset-backed security investors, it is exempt from corporate taxes. Of course, the income received by investors is subject to personal taxation, but in a symmetric fashion so is the income received by a bank's depositors and equity holders. See Bank for International Settlements (2009).

  6. 6

    Lawyers recognize the importance of structuring securitizations so that special purpose vehicles do not generate taxes. Peaslee and Nirenberg (2001, p. 2) state “a securitization transaction almost certainly would not be viable if passing cash through the (special purpose vehicle) issuer resulted in significant additional tax burdens. One of the main goals of tax planning in this area—indeed the sine qua non—is to ensure that no such tax costs are incurred.”

  7. 7

    Ashcraft (2008) uses variation in state corporate tax rates to identify a bank's choice of capital instruments.

  8. 8

    In “French Banks Try an Import from the U.S.,” Wall Street Journal, November 6, 2012, Basel III's higher capital requirements are claimed to be the reason why French banks only recently began to securitize loans. Another example is the insurance firm MetLife, one of the top 15 U.S. mortgage lenders, which will continue to originate mortgage loans but, due to Basel III and the Dodd-Frank Act, is selling its bank that previously funded many of its mortgages. See “Regs Push MetLife Out of Banking, into Shadow System,” American Banker, July 21, 2011. In February 2011, the property and life insurer Allstate Corporation also announced that, due to financial regulations, it is winding down its Allstate Bank and will cancel its charter.

  9. 9

    That credit-screening/monitoring is not contractible or verifiable is a standard assumption (e.g., Rajan (1992)). If part of the optimal level of credit-screening/monitoring can be contracted upon by loan buyers, we can redefine xi(s,0) as the loan's return net of screening/monitoring costs for this partial level. The variable ai then can be redefined as the amount of screening/monitoring in excess of this partial level.

  10. 10

    If there are different investor personal tax clienteles, as assumed in Miller (1977), then Litzenberger and Van Horne (1978) show that pe(s)/pd(s) would be constant across states if the personal tax bracket of the marginal investor indifferent between holding equity and debt were constant across states. This assumption is also made in DeAngelo and Masulis (1980) and would hold with investor risk-neutrality.

  11. 11

    A large literature, beginning with Berger and Hannan (1989), documents that banks in more concentrated markets pay lower retail deposit rates. Despite liberalization of branching and greater competition from mutual funds over the past few decades, studies such as Park and Pennacchi (2009) continue to find evidence of deposit market power by many banks.

  12. 12

    Our qualitative results extend to the case in which some deposits are uninsured and the bank pays a fair credit spread on uninsured deposits. The only difference relates to tax deductibility because deposit insurance premiums are not tax deductible but credit spreads on uninsured deposits are. Such deductibility could provide an incentive for banks to issue uninsured deposits with high credit spreads, that is, take excessive on-balance sheet risk.

  13. 13

    Risk retention may be more likely for securitizations of revolving credits, such as credit card receivables. See Gorton and Souleles (2006), Higgins and Mason (2004), and Calomiris and Mason (2004) for evidence.

  14. 14

    Partial risk retention by the loan-selling bank would reduce, but not eliminate, the moral hazard problem of suboptimal credit-screening and monitoring.

  15. 15

    Kiff and Mills (2007) and Jiang, Nelson, and Vytlacil (2014), among others, have argued that investors who bought mortgage-backed securities (MBS) did not realize the moral hazard of deficient screening and monitoring. Investors relied on credit ratings, but rating agencies did not reveal the low quality of the mortgages due to the lucrative fees they were paid for rating MBS. Even if credit ratings accurately gauged the (physical) probability of MBS defaults, Coval, Jurek, and Stafford (2009) argue that investors overpaid for structured financial securities because they did not realize the extreme systematic risk (risk-neutral probability of default) of the securities (though Collin-Dufresne, Goldstein, and Yang (2012) provide counter evidence). If, indeed, it had been the case that investors overpaid for securitized loans, there would have been a greater incentive to sell loans than what our model predicts.

  16. 16

    An earlier draft used Call Report loan servicing information as a proxy for the stock of existing loans that a bank had originated and sold, similar to Jiangli, Pritsker, and Raupach (2007). This proxy assumes that banks retained the servicing rights of sold loans. Test results based on this Call Report data are qualitatively similar to those using HMDA data. However, a bank-by-bank comparison of HMDA and Call Report mortgage sales data show that, on average, the Call Report data miss the majority of HMDA mortgage sales, implying that many banks do not retain servicing rights. We report only tests using HMDA data due to its much greater accuracy.

  17. 17

    The total asset threshold was $31 million in 2001, rising to $37 million in 2008. In addition, banks that originate no mortgages are exempt from an HMDA filing.

  18. 18

    A bank with multistate operations is subject to several state tax rates, and data limitations do not permit identification of the relevant state tax rate that a bank faces when deciding to retain or sell a particular loan.

  19. 19

    The paper's Internet Appendix (which can be found in the online version of the article) analyzes mortgage sales by S-corporation banks.

  20. 20

    For mortgages that are sold, HMDA identifies various categories of mortgage purchasers. We define a mortgage as sold if the purchaser is Fannie Mae, Ginnie Mae, Freddie Mac, Farmer Mac, a private securitization, another commercial bank, savings bank, savings association, insurance company, credit union, mortgage bank, finance company, or “other” purchaser. We do not consider a mortgage as sold if the purchaser is affiliated with the originating bank.

  21. 21

    Rosen (2010) estimates that the average time between origination and sale of a HMDA mortgage is 39 days.

  22. 22

    Operating costs are calculated as the sum of salaries and employee benefits, expenses of premises and fixed assets, and other noninterest expenses (Call Report codes riad4135+ riad4217+riad4092).

  23. 23

    Nontransactions interest expense is riad0093 + riadA517 + riadA518. Nontransactions balances equal rcon6810 + rcon0352 + rcon6648 + rcon2604.

  24. 24

    State corporate tax rates are available at For some states, marginal tax rates vary depending on a corporation's income bracket. Following Ashcraft (2008), we use the tax rate corresponding to $1 million in income, which is almost always the top income bracket.

  25. 25

    The population-weighted average elasticity for all MSAs is 1.75, but land-constrained MSAs such as San Francisco, Miami, New York, Boston, and Chicago have elasticities below one.

  26. 26

    Total assets and operating costs are inflation-adjusted, given in year 2000 dollars.

  27. 27

    This figure for seniors is the average for the state. However, most of our tests use the proportion of seniors at the level of each MSA.

  28. 28

    A possible explanation may be longer lag times to adjusting mortgage sales following tax rate changes. In particular, there may be fixed costs to initiating mortgage sales for banks that previously sold none.

  29. 29

    Pennacchi (2006) discusses the inadequacies of deposit insurance and capital standards and demonstrates why they produced incentives for banks to take excessive systematic risk.

  30. 30

    An approach to raising capital standards that would mitigate the greater tax incentive to securitize is to require that banks issue contingent capital. See, for example, Flannery (2014). Appropriately designed contingent capital could provide the tax advantages of debt during normal conditions but provide the loss absorption of common equity capital during times of financial distress.