The Federal Reserve Board eagle logo links to home page
Skip to: [Printable Version (PDF)] [Bibliography] [Footnotes]
Finance and Economics Discussion Series: 2007-31 Screen Reader version ♣

Federal Home Loan Bank Advances and Commercial Bank Portfolio Composition

W. Scott Frame*
Financial Economist & Associate Policy Advisor
Federal Reserve Bank of Atlanta
Atlanta, GA 30309 USA.
[email protected]
Diana Hancock*
Assistant Director and Chief, Financial Studies, Research & Statistics
Board of Governors of the Federal Reserve System
Washington, DC 20551 USA.
[email protected]
Wayne Passmore*
Deputy Associate Director, Research & Statistics
Board of Governors of the Federal Reserve System
Washington, DC 20551 USA.
[email protected]

Abstract:

The primary mission of the 12 cooperatively owned Federal Home Loan Banks (FHLBs) is to provide their members financial products and services to assist and enhance member housing finance. In this paper, we consider the role of the FHLBs' traditional product - "advances," or collateralized loans to members - in stabilizing commercial bank members' residential mortgage lending activities.

Our theoretical model shows that using membership criteria (such as a minimum of 10 percent of the portfolio being in mortgage-related assets) or using mortgage-related assets as collateral does not ensure that FHLB advances will be put to use for stabilizing members' financing of housing. Indeed, our model demonstrates that advances - a relatively low cost managed liability - are most likely to influence lending only when such liabilities are used to finance "relationship" loans (i.e., loans to bank-dependent borrowers) that will be held on a bank's balance sheet and are least likely to influence lending for loans where the loan rate is heavily influenced by securitization activities, like mortgages.

Using panel vector autoregression (VAR) techniques, we estimate recent dynamic responses of U.S. bank portfolios to FHLB advance shocks, to bank lending shocks, and to macroeconomic shocks. Our empirical findings are consistent with the predictions of our theoretical model. First, recent bank portfolio responses to FHLB advance shocks are of similar magnitude for mortgages, for commercial and industrial loans, and for other real estate loans. This suggests that advances are just as likely to fund other types of bank credit as to fund single-family mortgages. Second, unexpected changes in all types of bank lending are accommodated using FHLB advances. Third, FHLB advances do not appear to reduce variability in bank residential mortgage lending resulting from macroeconomic shocks. However, some banks appear to have used FHLB advances to reduce variability in commercial and industrial lending in response to such macroeconomic shocks. Thus, relatively low cost managed liabilities may be used to finance "relationship" borrowers (which are typically business borrowers, rather than residential mortgage borrowers), although this use for advances appears to have diminished over time.

Keywords: Advances, Government-Sponsored Enterprises, GSE, Portfolio shocks, Panel-VAR

JEL Classification Numbers: G21: Banks, Depository Institutions, Mortgages
G18, G38: Government Policy and Regulation

*The views expressed do not necessarily reflect those of the Board of Governors of the Federal Reserve System, the Federal Reserve Bank of Atlanta, or their respective staffs. We thank Brent Ambrose, Joseph McKenzie, Joe Peek, Larry Wall, and various seminar participants for helpful comments on previous drafts.


1. Introduction

Government-sponsored enterprises (GSEs) represent an unusual intervention by the federal government into private capital markets. GSEs are financial institutions that are individually chartered by Congress, but owned by private shareholders (cooperative members or outside investors depending on the ownership arrangement). The Congressional charters, extraordinary ongoing interactions between these institutions and government officials, and past government actions have created a perception in financial markets that GSE debt obligations are implicitly guaranteed by the federal government. This perception allows each institution to borrow at favorable interest rates and then pass some of these savings on to consumers. Hence, by chartering a specific GSE, the federal government can target benefits toward a specific sector of the economy without recognizing the attendant costs in the federal budget. The three most prominent GSEs are those serving housing: the Federal Home Loan Bank (FHLB) System, the Federal National Mortgage Association (Fannie Mae), and the Federal Home Loan Mortgage Corporation (Freddie Mac).1^{,} 2

Measuring the extent to which a GSE's primary business activities provide gross social benefits - as defined by its statutory mission - is a critical first step in understanding whether such interventions are desirable. (Of course, even then, one has not accounted for costs, including general equilibrium distortions.) With respect to Fannie Mae and Freddie Mac, a large literature has emerged that attempts to estimate the effect of their activities on mortgage interest rates.3 Remarkably, as noted by McCool (2005), there has been little attempt to examine similar questions for the FHLB System.4 This is the aim of our paper.

The FHLB System is a collection of 12 cooperatively owned wholesale banks. The statutory mission of this GSE is to provide their members financial products and services, including collateralized loans (advances), to assist and enhance such members' financing of (1) housing and (2) community lending.5 Membership is open to all depository institutions with more than 10 percent mortgage assets and also to community financial institutions (i.e., those with less than $587 million in total assets as of December 2005). Over 8,000 financial institutions are currently members of the FHLB System.

In this paper, we focus on the role of FHLB advances in stabilizing commercial bank members' financing of housing.6 We specifically consider three questions. First, are unexpected changes in advances correlated with changes in residential mortgage lending and other forms of bank lending? Second, are unexpected changes in bank loan portfolios, including residential mortgages, accommodated using FHLB advances? Third, do FHLB advances help to insulate bank portfolios from macroeconomic shocks (e.g., unexpected changes in the federal funds rate, the yield curve, or GDP) and do these shocks have less of an effect on residential mortgage lending than on other forms of bank lending?

Toward answering these questions, we first develop a loan pricing model that provides the conditions for relatively cheap, or more steadily priced, FHLB advances to influence loan rates and thereby affect bank credit and, in particular, mortgage credit. As we demonstrate later, if mortgage markets are heavily influenced by securitization or other forms of market-based financing, then a bank might hold a mortgage portfolio (because it "cherry picks" from the flow of mortgage originations) but still have no influence on mortgage pricing or mortgage credit availability.

As suggested by our theory, a better measure of a bank's influence on mortgage markets, relative to other forms of lending, is to compare changes in mortgage supply after either a change in the bank's cost of funds or a change in an exogenous factor that could be offset by a change in a bank's cost of funds. If FHLBs are providing funds that ultimately create more mortgage credit or if they stabilize members' financing of housing -- rather than simply funding all bank assets -- then it is through innovations in supply or demand across loan types that this relationship might be observed.

We use the predictions of the theoretical model to interpret recent dynamic responses of U.S. commercial bank portfolios to FHLB advance shocks, to unexpected loan demand shocks, and to macroeconomic shocks using a panel-VAR. We present the following results. First, bank portfolio responses to FHLB advance shocks are of similar magnitude for mortgages, for commercial and industrial loans, and for other real estate loans. Hence, advances are just as likely to fund other types of bank credit as to fund single-family mortgages. Second, unexpected changes in lending, due to changes in loan demand for example, are accommodated using advances by active FHLB members. Mortgage lending is not unique in this respect. Third, FHLB advances do not appear to have reduced the variability in residential mortgage lending by banks that resulted from either federal funds rate shocks or GDP shocks. However, some banks appear to have used FHLB advances to smooth commercial and industrial lending in response to such macroeconomic shocks, although this use appears to have diminished over time. Therefore, FHLB advances do not appear to be stabilizing commercial bank members' financing of housing. Overall, we find that commercial banks are increasingly relying on FHLB advances as a wholesale funding source and - because money is fungible - advances are being used to fund all types of commercial bank assets, not just residential mortgages.

The rest of the paper is organized as follows: Section 2 provides some background information on the FHLB System to lay the foundation for understanding our theoretical and empirical modeling strategies. Section 3 provides a theory for considering the effects of FHLB advances on bank portfolio lending in the context of modern capital markets. Section 4 describes our empirical approach, while Section 5 discusses our findings. The last section provides a summary.

2. Background: The Federal Home Loan Bank (FHLB) System and FHLB Advances

The FHLB System was created in 1932 and consists of 12 regional wholesale banks (FHLBs) and an Office of Finance that acts as the FHLBs' gateway to the capital markets.7 Each FHLB is a separate legal entity, cooperatively owned by its member financial institutions (commercial banks, thrifts, credit unions, and insurance companies), that has its own management, employees, and board of directors. The individual FHLBs do not generally compete for members as each institution is assigned a distinct geographic area to serve.8 However, the FHLB System is often viewed as a whole because most of the FHLBs' financing takes the form of debt for which the 12 institutions are jointly and severally liable ("consolidated obligations").9 Flannery and Frame (2006) provide a detailed overview of the structure, activities, and risks of the FHLB System.

FHLB System assets totaled just over $1 trillion at year-end 2006.10 Advances comprise the majority of assets ($641 billion, or about 63 percent of total assets). The FHLBs also maintain portfolios of investments ($271 billion on a consolidated basis) and residential mortgage loans purchased from their members ($98 billion on a consolidated basis).11 Around 95 percent of the consolidated asset portfolio of the FHLB System is funded with debt, almost all of which takes the form of the consolidated obligations issued by the Office of Finance. The FHLB System also funds roughly about 5 percent of their assets through equity capital, most of which is derived from mandatory member stock subscriptions.12

Advances are historically the dominant activity conducted by the FHLB System and hence the most natural place to look for an effect of FHLBs on mortgage markets. By law, these collateralized borrowings are to be used only for residential housing finance.13 The most common forms of advance collateral are mortgage-related assets (whole loans and mortgage-backed securities) and U.S. Treasury and Federal Agency securities.14

Beyond the explicit collateral, the FHLBs also have priority over the claims of depositors and almost all other creditors (including depositors and the Federal Deposit Insurance Corporation) in the event of a member's default - known as a "super-lien."15 Importantly, the super-lien may have the effect of muting FHLB incentives to underwrite and price member credit risk.16

Any link between FHLB advance activity and mortgage funding by the GSEs' member financial institutions must be made through the collateral posted on advances. However, this link is likely to have markedly weakened over the past 75 years due to changes in the legal environment, information technology, and financial practice.

During its first 50 years or so of existence, the FHLB System primarily acted as a reliable supplier of long-term funding for thrift industry mortgage lending by making collateralized advances to these depository institutions. During this time, Congress imposed asset limitations on thrifts that resulted in balance sheets almost entirely comprised of residential mortgage-related assets. All depository institutions were also subject to limitations on the interest rates that they paid depositors (under Regulation Q), which periodically resulted in liquidity pinches. Hence, the availability of FHLB advances to thrifts for the purpose of funding mortgages during deposit shortages. Below, we refer to this smoothing of deposit funding for the purpose of originating mortgages as the "mortgage funding view" of FHLB advance activity.17

A series of changes since 1980 significantly altered the U.S. mortgage funding system. First, the Depository Institutions Deregulation and Monetary Control Act of 1980 and the Garn-St. Germain Depository Institutions Act of 1982 terminated the Regulation Q ceiling on savings account interest rates and gave thrifts expanded investment powers. Second, the Financial Institutions Recovery and Reform Act of 1989 opened FHLB membership to all depository institutions with more than 10 percent of their portfolios in residential mortgage-related assets (whole mortgages and mortgage-backed securities). Most recently, the Financial Services Modernization Act of 1999 expanded the mission of the FHLB System to act as a general source of liquidity to "community financial institutions" and lifted the requirement that thrifts be members of this GSE.18

Today, all types of depository institutions are eligible for FHLB membership. Moreover, few depository institutions maintain portfolios heavily concentrated in mortgages, like the thrifts of yesteryear. In addition, mortgage markets are now national in scope because of improvements in information technology, the growth in mortgage securitization, and the related investment in mortgage-backed assets by a wide-variety of domestic and international investors. Hence, any link between FHLB advances and mortgage lending is likely to be much weaker today than in the past. Indeed, given the modest constraint on FHLB membership related to residential mortgage activity, the portfolio composition of most FHLB members (especially the largest members which dominate advance activity), and the simple fact that money is fungible; FHLB advances could be funding virtually any type of asset. Below, we will refer to the view that FHLB advances are but one of many sources of wholesale funds that are not linked to any particular asset type as the "wholesale funding view" of FHLB advance activity.

3. FHLB Advances, Modern Capital Markets, and Bank Lending

This section presents a theoretical model in order to formalize the prior notions about the relationship between FHLB advances and bank lending in general and residential mortgage lending in particular. Our model is based on Heuson, Passmore, and Sparks (2001), who provide a rigorous treatment of the theory as it relates to mortgage markets, as well as Hancock, Lehnert, Passmore and Sherlund (2005), who customized the model for bank capital requirements.

3.1 The Model

Banks, in our model, have a choice between funding their lending on balance sheet using a mix of deposits and FHLB advances or alternatively via securitization. As we described above, the FHLBs are nominally collateralized lenders but because of the super-lien, they can lay claim to all bank assets. This legal right is important because it means that FHLBs, unlike other wholesale liability providers to banks (or equivalently securitizers), do not necessarily need to worry about adverse selection or "cherry picking" of collateral by FHLB members. In other words, the FHLBs may not need a different (and tighter) underwriting standard than the bank itself.

Figure 1 provides a graphical representation of the industry supply curve for a given bank loan market. On the horizontal axis is the probability that borrower will not default, q, in a given market segment, which ranges from 0 to 1. Borrowers with higher probabilities of not defaulting (i.e., those closer to 1 in the right corner of the figure) have the lowest credit risks. The marginal cost of bearing borrower credit risks declines as qincreases, so the lowest rate that a lender is willing to accept falls as the probability of not defaulting on a loan rises.19

The purple line (solid and dashed) represents the locus of zero economic profit combinations of loan rates (r) and credit risks (q) from using bank liabilities (including core deposits and FHLB advances) to fund loans directly.20 The bank is willing to use its liabilities to fund all loans with credit risks equal to, or less than, the credit risks represented by this line (denoted as the set of all points to the right of B(r,q)). The red line (solid and dashed) is the locus of zero economic profit combinations of loan rates and credit risks if the bank uses its liabilities to fund securities backed by loans rather than to fund loans directly (denoted S(r,q)). Again, the bank is willing to fund all securities with credits risks equal to, or less than, the credit risks represented by this line. Because securities yield a liquidity benefit to the bank, the bank would prefer to use its liabilities to fund securities, all things equal.

However, all things are not equal. Market securities provide a liquidity benefit because they have credit risk properties that can be easily communicated to market investors. Generally, this credit risk is communicated by a high credit rating, a third-party guarantee, and/or a debt structure that shields investors from credit losses. Market investors also know that banks will tend to keep the best loans and, because only banks can originate loans, they must guard themselves against cherry picking. The loans that a bank will keep are all those with credit risks that are equal to, or less than, those to the right of the blue dashed line (denoted CP(r,q)).

Market investors must offset the loan originators' first mover advantage (the "cherry picking") to earn a competitive rate of return, and thus they have a higher credit risk standard than does the bank. This standard is given by the green line -- market investors will only purchase, securitize, or rate, loans with credit risks equal to, or less than, those represented by the green line (denoted SU(r,q)).

To summarize, when the loan rate is r_{1}, loan originators (which are almost always depositories) only want to sell loans with credit risks between 0 and q_{2 }because of cherry picking.21 Moreover, because of this cherry picking activity, market participants only want to use loans with credit risks between q_{1} and 1 to create marketable securities. Loans originated by banks with credit risks lower than q_{2 } are placed in the bank's investment portfolio. Therefore, the effective industry supply curve for loan credit risks of a given product type is represented by the solid segments of the purple and red lines.

Figure 2 illustrates that equilibrium loan rates are determined by the intersection of supply and demand. The demand curve in this model ranks borrowers by the maximum interest rate they are willing to pay for a loan, suggesting that borrowers with a high probability of paying back their loan are more willing to pay higher interest rates, all other things equal. Loan default is assumed to be costly for borrowers, so that when high interest rates prevail, only borrowers with low odds of default stay in the loan applicant pool.22 This means that the demand curve slopes upward when the probability of not defaulting on a loan is used on the horizontal axis. The equilibrium loan rate for a loan market segment is determined where the demand curve for that segment crosses the industry supply curve.23

3.2 Necessary Conditions for FHLB Advances to Influence Mortgage Interest Rates.

Two necessary conditions must be met in order for FHLB advances to influence mortgage interest rates: (1) the marginal borrower is funded by a bank's liabilities and not by the capital markets, and (2) FHLB advances must be among the bank's lowest cost funding sources. For the mortgage market to respond uniquely, a third condition is also required - that only mortgage loans collateralize FHLB advances and that the banks do not maintain excess collateral.

In the context of the model, FHLBs have the potential to moderate loan rates only if a loan market segment (e.g., mortgage, consumer, commercial and industrial, etc.) demand curve crosses the industry supply curve in the areas where loans, and not securities, are funded by the FHLB member's liabilities (as shown by the orange line in the top panel of figure 2). Here, the demand curves D_{1} crosses the supply curve between q_{0 }and q_{1}. Loans to borrowers with credit risks from [q_{0,}q_{1}] are funded directly with bank liabilities (which include FHLB advances). In contrast, loans to borrowers with credit risks from [q_{1,}q_{2 }] are effectively market-priced because the bank swaps the loans for securities at the market rate and then funds the securities by using its liabilities.24 In addition, loans to borrowers who are very good credit risks [q_{2,}1] are "cherry picked" and held in the bank's investment portfolio and funded with bank liabilities.

The bottom panel of figure 2 illustrates an alternative scenario in which the demand curve D_{2 } cuts across the portion of the industry supply curve where credit risks are borne by the market. This implies that marginal borrowers are not borrowers with a bank relationship and therefore the bank (and thus FHLB advances) cannot influence loan rates. Instead, loans to borrowers with credit risks from [q_{1,}q_{2 }] are effectively funded by the market through asset-backed securitization, with the bank's liabilities being used to purchase the securities. Again, loans to borrowers from [q_{2 }, 1] are "too good" to market fund; these loans are funded directly by bank liabilities but only because they yield a competitive return for the bank's investment portfolio.

As shown in the top panel of figure 3, which portrays a loan market where borrowers are funded on the margin by bank liabilities, rates on FHLB advances can potentially influence loan rates. However, in order for this to occur, FHLB advances must be among a bank's lower cost funding alternatives, so as to lower the bank's cost of funds (shown by the move from curve B to curve B\prime in figure 3). This lower cost of funds would cause loan rates to fall because the bank provides funds to the marginal borrower in the loan market (and banking markets are assumed to be competitive). In other words, the FHLB advance must actually be a lower cost of funds than other wholesale liability alternatives.

The bottom panel of figure 3 alternatively portrays a loan market where the bank's borrowers are funded on the margin by market-priced funding. In this case, lowering the bank's own cost of funds does not influence loan rates. Thus, when securitization plays an important role in a loan market, a lower cost for FHLB advances would be unlikely to influence loan rates. Because mortgage lending is especially influenced by securitization in practice, and because most banks have substantial excess potential collateral, FHLB advances seem unlikely to influence mortgage rates and mortgage lending. However, this is ultimately an empirical question to which we will now turn.

4. Estimation of Bank Portfolio Dynamics

This section describes data and the specification we use to estimate the dynamic interactions between the various components of banks' balance sheets and aggregate economic conditions.

4.1 FHLB Advances Data

Information on FHLB advances held by FHLB members (ADV) is available on a quarterly basis from the Federal Housing Finance Board.25 Table 1 provides data on advances outstanding and FHLB capital stock ownership as of 2006:Q3 for thrifts and commercial banks stratified into three asset size groups (less than $100 million, $100 million to $1 billion, and greater than $1 billion). Strikingly, commercial bank borrowers greatly out number - more than four to one - thrift borrowers (column (2)). Indeed, about half of FHLB advances outstanding are to commercial bank members (columns (3) and (4)), which also own almost half of the FHLB System's capital stock (column (5)). And these advances are concentrated in the largest entities: Commercial banks with greater than $1 billion in total assets, institutions that have many sources of wholesale funding, hold more than one-third of FHLB advances outstanding (column (4)). Below, we focus our study on commercial banks given their increasing importance to the FHLB System and because these lenders have greater opportunity to employ advances for many different types of loans -- not only for residential mortgages.

Because asset-liability management is typically centralized within a banking organization, we constructed asset and liability data at the "top holder" level. For example, a bank holding company, which is comprised of a lead bank and several subsidiary banks, would be the top holder of the banking organization. We aggregated individual bank asset and liability information to the domestic top holder level using information from the National Information Center (NIC), which is the central repository containing information about all U.S. banking organizations and their domestic and foreign affiliates. A bank that is unaffiliated with any other bank is considered to be its own top holder organization.

Bank top holder entities were stratified into three size groups in each quarter: (1) Small top holders have total assets at or below the 50th percentile of the distribution of total assets; (2) medium top holders have total assets between the 50th and 95th percentiles of the distribution of total assets; and (3) large top holders have total assets at or above the 95th percentile of the distribution of total assets. These percentile cutoffs for the three top holder size groups allocate top holders such that 43 percent of the sample is considered "small", 51 percent of the sample is "medium", and 6 percent of the sample is "large" as of 2006:Q3 (table 2). These top holder percentage allocations across the three size groups roughly correspond to those for commercial banks in table 1: small banks (less than $100 million in assets), 38 percent; medium banks ($100 million to $1 billion in total assets), 55 percent; and, large commercial banks (greater than $1 billion in total assets), 7 percent.

As of quarter-end 2006:Q3, only 66 percent of the smallest top holder members borrowed from a FHLB and together they borrowed just $6.4 billion (table 2). In contrast, about 83 (92) percent of medium (large) top holders borrowed from their FHLB and together these entities borrowed $56..1 billion ($304.5 billion). Thus, actual FHLB borrowings are heavily skewed towards the largest top holder banking organizations.

Figure 4 presents quarterly time-series information on the percent of total FHLB advances outstanding and on the percent of advances-to-borrower assets for each of three top holder size groups for 1994:Q1 through 2006:Q3, inclusive. Over this period, large top holder FHLB members, who have many sources of wholesale funding, have steadily increased their share of total FHLB advances outstanding (top panel) as the proportion of their assets funded by FHLB advances has risen and kept pace with advance usage by smaller top holders (bottom panel).

FHLB System members generally have a stock of eligible advance collateral that far exceeds their actual advance borrowings. Figure 5 presents histograms for advances-to-eligible asset ratios (in percent) for small top holders (top panel), medium top holders (middle panel), and large top holders (bottom panel) at quarters-end, 1997:Q3 (left panel), 2001:Q1 (middle panel) and 2006:Q3 (right panel).26 Looking across the top and middle panels of Figure 5, it is clear that a fairly high proportion of small- and medium-sized top holder FHLB members respectively had no advances at all. This pattern did not hold true for the large top holder FHLB members (bottom panel). Regardless of top holder size, virtually all top holder FHLB members used much less than 50 percent of their eligible collateral for FHLB advances, suggesting that collateral is not a binding constraint for commercial banks.

Figure 6 presents aggregate time-series information on the number of FHLB members and their advances outstanding for the three top holder size groups during 1994:Q1 through 2006:Q3. Each series was normalized by its respective time-series mean during the sample period so changes in growth can more easily be discerned.

Interestingly, there appear to be three distinct time periods for the growth rates of FHLB membership and advance usage. During the first period (1994:Q1 - 1997:Q2), membership grew rapidly, but advance usage grew only modestly. This difference in growth patterns across the two indices suggests that new FHLB members were not actively using FHLB advances to fund their asset portfolios during this first period. Consequently, we do not use data from this first period to consider the dynamic interactions between bank balance sheets and aggregate economic conditions. In contrast, during the second period (1997:Q3 - 2000:Q4), which is shaded in figure 6, advance usage grew at least as rapidly as did FHLB membership. In addition, advance usage growth was most rapid for the largest top holders. This difference in growth patterns is consistent with FHLB members (new and old alike) more actively using FHLB advances to fund their portfolios. Because this is likely a learning period for banks not familiar with using FHLB advances, we consider data from this period separately from the later period (2001:Q1-2006:Q3) when FHLB membership is stable and FHLB advance usage appears to be responding to other factors. Indeed, we consider this later period the most useful for understanding whether FHLB advances are influencing mortgage rates or stabilizing members' financing of housing.

4.2 Bank Portfolio Data

Call Reports for individual, federally-insured, domestically chartered commercial banks were used to construct quarterly data for five balance sheet components - residential mortgages (MORT), other real estate loans (OREL), securities (SEC), commercial and industrial loans (C&I), and domestic deposits (DEP).27 Call Reports generally include book values, rather than market values, for each balance sheet component.28

Data were constructed for four asset categories. When feasible, only domestic loans were included in each of these asset categories. Residential mortgages (MORT) include (1) the amount of all permanent loans secured by first liens on 1-to-4 family residential properties, (2) the amount of all permanent loans secured by junior (i.e., other than first) liens on 1-to-4 family residential properties, and (3) the amount outstanding of "home equity lines."29^{,}30 Other real estate loans (OREL) consist of (1) construction and land development loans, (2) loans secured by farmland, (3) loans secured by multi-family (5 or more unit) residential properties, and (4) loans secured by nonfarm nonresidential properties.31 Securities (SEC) equaled the sum of the amortized cost for "held-to-maturity" securities and fair value for "available-for-sale" securities.32 Lastly, the amount of commercial and industrial loans (C&I) includes loans to borrowers domiciled in both the U.S. and abroad.33

In addition, data were constructed for domestic deposits (DEP). These deposits include transaction accounts, non-transaction savings deposits, and total time deposits less than $100,000.34

4.3 Time-Series Information Concerning Commercial Bank Portfolios

For each top holder size group, quarterly aggregate portfolio share data were constructed for entities without advances and for two types of FHLB members. 35^{,}36 The "active FHLB members" had a ratio of FHLB advances-to-total assets greater than or equal to 2.5 percent. In contrast, "passive FHLB members" had a ratio of FHLB advances that was both greater than zero and less than 2.5 percent. Both top holder size groups and FHLB advance usage status were determined on a quarter-by-quarter basis.

The time-series information on portfolio composition for the top holders stratified by asset size and by FHLB advance usage status is presented in figure 7. Panel A contains time-series on two liabilities: domestic deposits and FHLB advances. Panel B contains time-series on two types of real estate asset categories: mortgage-related assets, which consist of mortgages and mortgage-backed securities, and other real estate (OREL) loans. Panel C contains time-series on commercial and industrial (C&I) loans and on securities, excluding mortgage-backed securities. In each of these three panels, aggregate time-series for small top holders is presented on the left, aggregate time-series for medium top holders is presented in the middle, and aggregate time-series for large top holders is presented on the right. (Note that shading is used in each panel to distinguish the three distinct periods that were described above for FHLB membership and advance usage where only the second period (1997:Q3 - 2000:Q4) is shaded.)

Focusing on the liability side of the portfolio (panel A), the domestic deposit portfolio shares for small top holders and for medium top holders were quite similar for active FHLB members, passive FHLB members, and entities that did not use advances by the end of the period (2006:Q3). This pattern, however, did not emerge for the largest top holders. The large active FHLB member top holders continued to rely more heavily on domestic deposits than did other large top holders. By construction, active FHLB members had a higher proportion of their total assets funded by FHLB advances than did passive FHLB members (panel A, bottom).

Turning to real estate lending (panel B), active FHLB members held higher proportions of their total assets in mortgage-related assets than did passive FHLB members, regardless of their top holder size. Moreover, passive FHLB members held a higher proportion of their assets in mortgage-related assets than did entities without advances, regardless of their top holder size. These patterns are not surprising given that most FHLB members have at least 10 percent of their portfolio in mortgages due to FHLB membership requirements. Moreover, as suggested by the theory presented above, entities that specialize in mortgage origination may hold a higher proportion of their portfolio in mortgages even if they have no influence on mortgage pricing or mortgage credit availability because they can "cherry pick" the highest quality mortgages along the credit risk continuum.

Interestingly, the time-series patterns for the portfolio shares of other real estate lending (panel B, bottom) were quite different across the three top holder size groups: Over the period, small top holder FHLB members increased their other real estate portfolio share modestly, medium top holder FHLB members increased their other real estate portfolio share rapidly, and large top holder FHLB members, who had the smallest such portfolio share at the beginning of the period, had a modest increase in their other real estate portfolio. Small top holder FHLB members with advances held similar proportions of other real estate loans in each quarter regardless of whether they were active or passive members. This similarity in the time-series data across active and passive FHLB members for other real estate loan portfolio shares was also apparent for medium top holder FHLB members. In contrast, large top holder active members held substantially higher proportions of other real estate loans in their portfolios than was held by large top holder passive members.

The time-series for commercial and industrial (C&I) lending portfolio shares (panel C, top) indicate that small- and medium-sized top holder FHLB members with advances did not have as dramatic of a run-off in their commercial and industrial portfolio after the turn of the century as did entities without advances. For the largest top holders, active FHLB advance users appear to have been able to mitigate the run-off in the commercial and industrial lending portfolio whereas passive FHLB advance users had a steep decline in this lending activity. Nevertheless, the passive FHLB advance users among the largest top holders held about the same proportion of their portfolio in commercial and industrial loans as did small- and medium-sized top holders by the end of the sample period.

Small- and medium-sized top holders that did not rely on FHLB advances tended to hold a higher proportion of their portfolio in securities excluding mortgage-backed assets (and a correspondingly lower proportion of their portfolio in the three lending categories (i.e., mortgage-related assets, other real estate loans, and commercial real estate loans)) than did comparably-sized FHLB members (panel C, bottom). Since these top holders did not employ FHLB advances, this higher portfolio share for securities excluding mortgage-backed assets may have been held either for liquidity purposes or to meet (unexpected) increases in loan demands. In contrast, the largest top holders without FHLB advances held similar proportions of securities (excluding mortgage-backed assets) as did comparably-sized FHLB advance users. Unlike other top holders, the largest top holders would have been more likely able to tap managed liability markets to meet their liquidity needs or to fund unexpected increases in loan demands.37 Alternatively, the high proportions of securities in the portfolio for small- and medium-sized top holders may have resulted from fairly modest lending opportunities relative to the availability of domestic deposit funding for these entities.

It is also informative to compare the liability structures of top holders with advances to the top holders without advances for each top holder size group (table 3).38 The liability structures are quite similar for small and medium top holders once one controls for FHLB advance usage, but large top holders have a different liability structure: Large top holders fund a higher proportion of their assets with managed liabilities, subordinated debt, and other liabilities than do smaller top holders. Focusing on the highlighted rows, small and medium top holders with advances tend to fund a slightly higher proportion of their assets with advances than do large top holders with advances. For the small and medium top holder size categories, top holders without advances funded a greater proportion of their assets with equity capital and with core deposits than did top holders with advances. In contrast, for the largest top holder size category, those with FHLB advances funded a greater proportion of their assets with core deposits than did those without advances.

4.4 Macroeconomic Conditions Data

Several data series were constructed at a quarterly frequency to gauge aggregate economic conditions. Our measure of aggregate output - quarterly gross domestic product (GDP) - is measured in real time (i.e., without any subsequent revisions).
These data were obtained from the Federal Reserve Bank of Philadelphia. Our measure of the short-term interest rate - the quarter-end daily federal funds rate (FFR) - was collected from the Federal Reserve Economic Database (FRED) maintained by the Federal Reserve Bank of St. Louis. Our measure of the slope of the yield curve - the difference between quarter-end 10-year and 1-year Treasury rates (YIELD) - was computed using constant maturity Treasury yields, which are also available from the FRED.39

4.5 Econometric Model

There is considerable evidence that banks typically make portfolio-wide but gradual, adjustments to their holdings of both financial assets and liabilities in response to unexpected events. For example, using aggregate data, Bernanke and Blinder (1992) and Den Haan, Sumner, and Yamashiro (2004) estimate that interest rate shocks affect the size and composition of banks' portfolios for more than two years. Analogously, Hancock and Wilcox (1995) use individual bank data to estimate that portfolio adjustments can take two to three years to complete after a bank capital shock. There are several explanations for why bank portfolio adjustments are gradual and have differing speeds across balance sheet categories, including: the complexity of loan documentation, the difficulty of judging the quality of loan applicants, the speed with which loan applicants alter their loan demand in response to changing circumstances, and the relative liquidity of secondary markets for the different portfolio components.

We use a panel-VAR technique to obtain banks' dynamic responses to portfolio and macroeconomic shocks because of the ability of this type of model to approximate complicated, interdependent adjustment paths with fairly short time-series information. Our first-order nine-equation VAR system takes into account the dynamic effects on individual banks of unexpected changes in their own balance sheets (i.e., residential mortgage loans (MORT), other real estate loans (OREL), commercial and industrial loans (C&I), deposits (DEP), securities, (SEC), and advances (ADV) and of the relatively more exogenous economic conditions (the short rate measured by FFR, YIELD measured by the difference between the 10-year Treasury rate and the 1-year Treasury rate, and GDP). We also allow for individual heterogeneity in the levels of the variables by introducing fixed effects, f_{i}. In notational terms, our panel-VAR model is:

\begin{displaymath} x_{{it} }=\alpha _{0 }+f_{i }+Ax_{{t-1} }+\varepsilon _{t } \end{displaymath} (1)

where x_{t} is the vector {FFR, YIELD, GDP, MORT, C&I, OREL, DEP, SEC, ADV}. Since the fixed effects are correlated with the regressors due to lags of dependent variables, the mean differencing procedure commonly used to eliminate fixed effects will create biased coefficients. To avoid this problem, we use a forward mean-differencing procedure (the Helmert procedure described in Arellano and Bover, 1995). This transformation preserves the orthogonality between transformed variables and lagged regressors. We use lagged regressors as instruments and estimate coefficients by a system of Generalized Method of Moments (GMM).40

The six bank balance sheet variables are measured in logs. A log specification ameliorates the error-term heteroskedasticity that un-logged variables would almost certainly entail. It also has a significant advantage over a portfolio shares specification because it permits a bank's size to change in response to shocks. In contrast, the aggregate economic condition variables - GDP, FFR, and YIELD - are measured in levels with output in nominal dollars and interest rates in percent.

Impulse-response functions, which are reported below, are based on VARs with variables in the flowing order: (1) FFR, (2) YIELD, (3) GDP, (4) MORT, (5) C&I, (6) OREL, (7) DEP,
(8) SEC, and (9) ADV.41 This order is ranked from the most exogenous to the most endogenous of the variables. Therefore, when one considers the impulse-response function matrix, the federal funds rate, FFR, does not respond to any of the other variable in the system in the first (i.e., contemporaneous) period, but FHLB advances, ADV, respond to everything in that period. In our theory, advances are extended in response to the demands of profit-maximizing banks; these banks base their decision on the price of advances relative to other funding choices and on the availability of advances based on whether advances are uniquely suited to mortgage lending. This suggests that advances should be treated as the most endogenous of the variables in the VAR model, relative to macroeconomic variables (which are the most exogenous to the bank) and relative to forms of lending (which are simultaneously determined conditional on the cost of funding).42

Within this model structure, we can test the response of advances to unexpected changes in all types of lending. As suggested by our theory, if advances are uniquely suited for mortgage funding and are critical for funding the borrower on the margin, then advances should not rapidly respond to shocks to other forms of lending (i.e., the "mortgage funding view.") But if advances are just one of many sources of wholesale funding or do not influence the marginal borrower, then the response of advances to, say, an unexpected increase in mortgage loan demand, should be similar to an unexpected increase in demand for other types of loans (i.e., the "wholesale funding" view.)

These distinctions can be illustrated in figure 2, where the demand curve moves upward and to the left. As shown in the top panel, a profit-maximizing bank would demand more advances in response to a positive loan shock. If advances were uniquely suited to mortgages, then only a mortgage loan shock would generate this response. However, as shown in the bottom panel, a loan shock does not influence a bank's demand for funds if securitization funds the borrower on the margin. Thus, if we order the VAR so that advances are influenced quickly by unexpected loan shocks, but can only influence loans themselves with a lag, we can examine both the uniqueness of mortgage funding in generating advance demand and the responsiveness of advances to various loan shocks. In order words, as articulated in our theory, this ordering is consistent with the demand for funding by a profit-maximizing bank.

Along similar lines, if advances have a role in encouraging mortgage funding, then our variable ordering is also consistent with our theory. If the FHLBs put advances on "sale," the shift in a bank's cost of funds (as illustrated in the top panel of figure 3) would influence loan demand when advances played a significant role in a bank's cost of funds, assuming the bank was key in funding marginal borrowers in that loan market (the contrast between the top and bottom panels of figure 3). Using our VAR model, and assuming that FHLB pricing influences the cost of funds with some delay, we can examine how responsive loan demand is to lower prices for FHLB advances.

Finally, using the structured VAR proposed above, we can examine whether or not FHLB advances help to smooth the response of member mortgage lending to unexpected macroeconomic fluctuations (such fluctuations, according to our theory, could be mitigated if advance pricing was adjusted to offset unexpected loan demands in either mortgage or other forms of lending, assuming advances were important in funding the marginal borrower). If the smoothing role is unique to mortgage funding, it would be readily apparent in a VAR model with this structure, when applied to both banking organizations with and without FHLB advances. 43

  1. Empirical Results
The VAR system was estimated for two time periods. As discussed above, the first time period (1997:Q3 through 2000:Q4, inclusive) is one that is likely to be a learning period for banks not familiar with using FHLB advances. In that period, regardless of top holder group size, both FHLB membership and advance usage growth were quite rapid. In contrast, during the second time period (2001:Q1 through 2006:Q3, inclusive) FHLB membership was relatively stable in each top holder size group and their advance usage appeared to respond to other factors than membership growth (figure 6).

We consider impulse responses that trace out the response of current and future values of top holders' portfolio components (e.g., mortgages, C&I loans, and other real estate loans) to a one-standard deviation increase in the current value of various VAR errors, assuming that each error returns to zero in subsequent periods and that all other errors are equal to zero. More specifically, we consider top holders' responses to one-standard deviation shocks to (1) FHLB advances, (2) three lending categories (mortgages, C&I loans, and other real estate loans) and (3) two measures of macroeconomic conditions (federal funds rate and GDP). We consider each in turn.

5.1 Bank Loan Responses to Advance Shocks: Are Mortgages Different?

Figure 8 presents impulse-response functions of mortgages (MORT), commercial and industrial (C&I) loans, and other real estate loans (OREL) for a standardized one-standard deviation shock to FHLB advances for small top holder members (top panel) and for large top holder members (bottom panel) for the first period (panel A) and second period (panel B), respectively.44

The top panels show that small top holder members have a statistically significant positive mortgage loan response to a one standard deviation FHLB advance shock in both periods. The shaded regions indicate the 5 percent and 95 percent confidence intervals. The confidence intervals are above zero for the small top holder member mortgage response in each period, though the lower bound of the shaded region is barely above zero in the latest period. This is not the case in both periods for either the C&I or OREL loan responses for these small top holder members, which are not statistically different from zero.

Notably, the positive and significant small top holder member mortgage responses are within the confidence intervals for the small top holder member C&I responses (OREL responses) provided in the top left (right) panel of figures 8A and 8B. Thus, a one standard-deviation advance shock has statistically similar effects on mortgages, on C&I loans, and on other real estate loans for these member groups, suggesting that mortgages are not unique in their response to a shock in FHLB advances. (Portfolio responses of medium top holder members -- not shown -- to a one-standard deviation advance shock had similar patterns to those shown for small top holder members.)

The bottom panels of figures 8A and 8B show that large top holder members do not significantly change their mortgage, C&I, or other real estate lending in response to a one-standard deviation FHLB advance shock. Also, as was the case for small top holder members, an advance shock has statistically similar effects on mortgages, on C&I loans, and on other real estate loans for large top holder members.

Overall, the confidence intervals for C&I lending responses and for mortgage lending responses overlap one another for the advance shocks in both periods. This suggests that advance shocks change C&I and mortgage lending in a similar fashion.

5.2 FHLB Advances Response to Bank Loan Shocks

Bank loan shocks, perhaps due to an increase in the demand for loans of a specific type, could potentially be accommodated by FHLB members using advances. To ascertain whether FHLB members employ advances in this manner, we consider how (unexpected) changes in lending affected FHLB advance usage.

Figures 9A and 9B present for the first and second periods respectively impulse-responses of FHLB advances (ADV) for standardized one-standard deviation shocks to mortgage (MORT) loans, to commercial and industrial (C&I) loans, and to other real estate loans (OREL) for small top holder members (top panel), medium top holder members (middle panel), and large top holder members (bottom panel). On the left side of each figure A and B, the response of advances to a mortgage loan shock is compared to the response of advances to a C&I loan shock. And on the right side of each figure A and B, the response of advances to a mortgage loan shock is compared to the response of advances to a shock in other real estate loans (OREL).

Interestingly, during both periods, the estimated response of current and future values of FHLB advances to a positive (standardized) one-standard deviation mortgage loan shock results in the current value of advances rising; the percent change from the base is positive and statistically significantly different from zero based on the 5 percent and 95 percent confidence intervals for small, medium, and large top holders with advances. Moreover, the estimated percent change from the base for the current value is larger as one peruses down the figure from small to large top holders with advances.

Comparing the magnitudes of the advance responses to mortgage shocks with the advance responses to C&I shocks (left side, figures 9A and 9B), it is apparent that the advance responses are smaller for C&I shocks than for mortgage shocks for both small- and medium-sized top holders with advances in both the first and second periods. This is not the case for the large top holders with advances. Such entities have estimated advance responses that are of statistically similar magnitudes for a C&I shock and for a mortgage shock in the second period.

In the first period, the responses of advances to mortgage shocks is larger than the responses of advances to other real estate lending shocks for top holders with advances regardless of size group (right side, figure 9A). This statistically larger response, however, is short-lived lasting about one-year for small- and medium-sized top holders and about two-years for large top holders.

By the second period, however, the distinction between the responses of advances to mortgage shocks and to other real estate lending shocks is not material (right side, figure 9B). Strikingly, the 5 percent and 95 percent confidence bands for these impulse-response functions for FHLB advances are intertwined. Although the responses of advances to a mortgage loan shock or to an other real estate loan shock are each (positive and) statistically significant, the confidence intervals together indicate that neither response of advances is significantly different from the other.

Overall, the positive and statistically significant estimated responses of FHLB advances to shocks in all three lending categories considered for both estimation periods suggests that FHLB members of all sizes (with advances already) do use this funding source to accommodate unanticipated changes in various types of lending - not only to accommodate unanticipated changes in mortgage lending.

5.3 Portfolio Responses to Macroeconomic Shocks: The Role of FHLB Advances

Federal Funds Shocks. For the first period (left side) and for the second period (right side), the estimated responses (with the shaded 5 percent and 95 percent confidence intervals) of mortgages (top panel), of commercial and industrial loans (middle panel) and of other real estate loans (bottom panel) to a federal funds shock are presented in figure 10 for small top holders (panel A), medium top holders (panel B) and large top holders (panel C).45 In each panel of figure 10, the responses of top holders with advances (for a size class) are compared to the responses of top holders (both members and non-members) without advances (for the same size class). These responses are measured using the percent change from the base value for the portfolio category type.

First, consider the portfolio responses of small top holders to a one-standard deviation federal funds shock during the first period (left side, panel A). Regardless of the loan category, the response to a one standard-deviation federal funds rate increase was statistically less negative for small top holders with advances than the corresponding response for small top holders without advances. Therefore, in the first period, the estimated responses for the loan portfolio are consistent with small top holder FHLB members using FHLB advances to reduce the impact of a (positive) federal funds rate shock on their loan customers.

Turning to the second period, an unexpected one standard-deviation federal funds increase (0.29 percent) resulted in a significantly more positive response in mortgage lending by small top holders with advances than by small top holders without advances (top right side, panel A). In fact, the initial increase in mortgage lending for small top holders without advances was not significantly different from zero and within a year of the federal funds shock this response became significantly negative. With respect to commercial and industrial lending and other real estate lending, the estimated responses to a one-standard deviation federal funds shock were not statistically different from one another for small top holders with and without advances. That is, the 5 percent and 95 percent confidence bands for the impulse-response function of small top holders with advances overlays the 5 percent and 95 percent confidence bands for the response of the respective loan type of small top holders without advances. These findings are consistent with a federal funds rate shock having the same effect on the non-mortgage lending portfolios of small top holders with and without advances in the second period.

The responses of medium top holder lending portfolios (figure 10, panel B) to a one-standard deviation federal funds rate shock are quite similar to the responses of small top holder lending portfolios to such a shock. In the first period, the estimated responses for the loan portfolio are consistent with medium top holder FHLB members using FHLB advances to reduce the long-term (negative) impact of a (positive) federal funds rate shock on their loan customers. But in the second period, the estimated portfolio lending responses to a federal funds rate shock are statistically indistinguishable across medium top holders with and without advances.

The portfolio lending responses of large top holders to a federal funds shock are presented in panel C. Because there are so few large top holders without advances, the 5 percent and 95 percent confidence bands around each of the estimated portfolio lending responses are quite wide for this group. As a result, the portfolio responses of large top holders with and without FHLB advances to a federal funds shock are generally statistically indistinguishable.46 Nevertheless, the estimated impulse-response functions for mortgages in the first period are statistically less positive during the first four quarters after a one-standard deviation federal funds shock for the large top holders with advances than for the large top holders without advances. The smaller magnitude of the response by large top holders with advances is not consistent with FHLB members stabilizing housing finance by using FHLB advances, but it may be the case that large top holders without advances lend to mortgage customers that are less sensitive to interest rate shocks than are the mortgage customers that borrow from large top holders with advances.

Overall, our findings suggest that recent portfolio lending responses to a federal funds shock are similar for top holders with and without FHLB advances. The smaller (negative) mortgage, commercial and industrial, and other real estate lending responses to a one-standard deviation federal funds shock for small- and medium-sized top holders with FHLB advances than for small- and medium-sized top holders without advances during the 1997:Q3- 2000:Q4 period suggests that small-and medium-sized FHLB members who use advances employed them to dampen the effects of interest rate shocks on their loan customers including their "bank-dependent" borrowers.

GDP Shocks. Estimated responses of mortgages (top panel), commercial and industrial loans (middle panel) and other real estate loans (bottom panel) to a (standardized) one-standard deviation gross domestic product shock are presented in figure 11. This figure has the same layout as figure 10, so responses for the first period (1997:Q3-200:Q4) are presented on the left side and responses for the second period (2001:Q1-2006:Q3) are presented on the right side with panels A, B, and C corresponding to small-, medium-, and large-sized top holders, respectively. Each estimated portfolio response for each top holder size group is measured using the percent change from its respective base.

Panel A of figure 11 presents the estimated portfolio responses to a negative gross domestic product (GDP) shock for small top holders with and without advances. Looking across the three lending types (mortgage, commercial and industrial, and other real estate lending) in each period considered, there is a dampened (negative) response to a GDP shock of small top holders with advances (red solid line) compared to the mortgage response to a GDP shock of small top holders without advances (blue solid line). Moreover, this different response is statistically significant since the shaded confidence intervals for lending responses of small top holders with advances (red shaded area) are not always intertwined with the confidence interval for lending responses of small top holders without advances (blue shaded area). Particularly for the first period, the impulse-response functions (and their associated confidence intervals) are consistent with FHLB members using advances not only to stabilize mortgage lending, but also to smooth fluctuations in their lending to bank-dependent borrowers.

In panel B, the portfolio responses of medium top holders to a one-standard deviation GDP shock are presented. Focusing on the first period (left panel), the estimated longer-term (more than four quarter out) medium top holder responses to a (negative) GDP shock are less negative for medium top holders with advances (red line) than for medium top holders without advances (blue line) for all three lending types - mortgages, commercial and industrial loans, and other real estate loans. Moreover, the non-overlapping confidence intervals that correspond to these estimated responses imply that these responses tended to be statistically different from one another. These findings are consistent with the wholesale funding view that money is fungible. In the second period, regardless of the lending type, the confidence intervals around the impulse-response functions are intertwined for medium top holders with and without advances. In this later period, the portfolio responses to a GDP shock for medium top holders with advances are statistically indistinguishable from the responses to a GDP shock for medium top holders without advances. These results are consistent with advances not playing a special role with regard to stabilizing mortgage lending over the business cycle.

As was the case with the confidence bands around the portfolio responses to a federal funds rate shock, the confidence bands around the portfolio responses to a GDP shock are wide for large top holders without advances (panel C). There are simply not enough degrees of freedom to concisely estimate such responses since most large top holders are FHLB members.47 Regardless, the estimated impulse-response functions for each of the three lending types presented are quite similar for large top holders with and without advances. Moreover, none of the three portfolio responses to a GDP shock for large top holders with advances are statistically different from the respective portfolio responses to a GDP shock for large top holders without advances. These findings are consistent with the portfolio responses to GDP shocks of large top holders with advances being the same as the portfolio responses to GDP shocks of large top holders without advances. Interestingly, the statistically indistinguishable responses of commercial and industrial lending to a one-standard deviation GDP shock for large top holders with and without advances is consistent with the view that these lenders tend to make loans to less bank-dependent borrowers (i.e., borrowers with more collateral or higher net worth) than smaller top holders who are more likely to specialize in relationship-based loans (e.g., Frame, Srinivasan, and Woosley 2001; Berger 2003).

6. Summary of Findings and Conclusion

In principle, a relatively low cost for FHLB advances does not guarantee that loan rates for borrowers will be lower. Moreover, using membership criteria (such as a minimum of 10 percent of the portfolio being in mortgage-related assets) or using mortgage-related assets as primary collateral does not ensure that FHLB advances will be put to use for stabilizing members' financing of housing. Indeed, our theoretical model shows that subsidized funding is most likely to be used for "relationship" loans (i.e., loans to bank-dependent borrowers) that will be held on a bank's balance sheet and are least likely to be used for loans where the loan rate is heavily influenced by securitization activities. Thus, it is an empirical question whether FHLB advances result in mortgage credit being more available or result in more stable mortgage credit markets.

Using a panel VAR approach, we estimate commercial bank top holders' responses to unexpected FHLB advances, to unanticipated changes in their portfolio, and to shocks in macroeconomic conditions. With regard to shocks to FHLB advances, confidence intervals for C&I lending responses and for mortgage lending responses overlapped one another in both periods considered. This implies that advance shocks change C&I and mortgage lending in a similar fashion.

Loan shocks, perhaps due to an increase in demand for loans of a specific type, appear to be accommodated by FHLB members by using advances, regardless of the loan type. A one-standard deviation (positive) loan shock (for mortgages, for commercial and industrial loans, or for other real estate loans) resulted in (positive) statistically significant changes in FHLB advances in both periods considered, regardless of the top holder size group. Interestingly, advance responses were larger for mortgages than for C&I lending and other real estate lending for all top holder size groups in the 1997-2000 period, but these differences were generally not material in the 2001-2006 period. These findings suggest that FHLB advances are used to accommodate changes in the demand for all types of loans across bank's portfolios. That is, our findings support the view that FHLB advances are like other forms of non-deposit bank funding and will be put to use to increase the overall return for a banking organization.

With respect to macroeconomic shocks (i.e., federal funds rate shocks and GDP shocks), smaller institutions with advances have smaller (negative) responses than do smaller institutions without advances. This is true for both mortgage and C&I lending. These findings are consistent with smaller institutions using FHLB advances for their bank dependent (relationship-based) borrowers. Across time, there has been a diminished difference in responses to macroeconomic shocks across smaller institutions with and without advances. This finding is consistent with smaller institutions having a wider availability of wholesale funding options more recently. In contrast, large top holders with and without advances had similar responses of their loan portfolio to such macroeconomic shocks. This finding is consistent with the view that these lenders tend to make loans to less bank-dependent borrowers (i.e., borrowers with more collateral or higher net worth) than smaller top holders who are more likely to specialize in relationship-based loans

Overall, our findings are consistent with the view that FHLB advances are not special, but rather are a general source of liquidity. The bulk of the empirical evidence suggests that FHLB advances are not connected to mortgage funding in the sense of uniquely funding mortgages or stabilizing mortgage funding. In other words, FHLB advances are fungible.


Figure 1: Bank Funding of Loan Portfolios. This figure shows the theoretical relationship between mortgage rates (vertical axis, r) and credit quality (horizontal axis, q). Higher risk mortgages are to the left and lower risk mortgages are to the right. The panel has four curves and is horizontally divided into four regions. All four curves have lines implying shading to the right of every curve until the next curve is met. The first curve, B(r,q), shows where a bank is willing to fund loans directly. It starts at a high vertical level on the left hand side and slopes downward as the curve moves to the right. The second curve is S(r,q), which is below B(r,q). Furthermore, S(r,q) does not extend fully to the left of the chart, ending when it intersects with the curve SU(r,q). S(r,q) is the level at which a bank is willing to fund securities. The third curve is CP(r,q), which is above B(r,q). This is the level at which loans are cherrypicked by the bank. Like B(r,q), S(r,q) and CP(r,q) slope downward. The fourth line is SU(r,q), which starts low on the left hand side and slopes upward. This represents the level at which the market is willing to securitize. SU(r,q) intersects S(r,q) at the horizontal point q0. The horizontal region between the highest risk loans and q0 are where loans are not extended to borrowers. There is also an equilibrium loan rate, r1, a horizontal dashed line slightly above the middle of the vertical axis. In Figure 1, it is arbitrarily imposed as there is no demand curve. Q1 is the horizontal point where SU(r,q) reaches r1 on the vertical axis. Q2 is the horizontal point where CP(r,q) reaches r1 on the vertical axis. The horizontal area between q0 and q1 is where loans will be funded by the bank. The horizontal area between q1 and q2 is where loans will be securitized by the bank. The horizontal area right of q2 is where loans are cherrypicked and funded by the bank. Thus the supply curve for loans is B(r,q) to the left of q1 and to the right of q2 and is S(r,q) between q1 and q2. The supply curve line is solid, all other curves are dashed.

Figure 2: Equilibrium Loan Rates Depend on Whether the Marginal Loan is Funded by the Primary or Secondary Markets. This is a two panel figure. Both panels look similar to figure 1. The top panel demonstrates where banks' liabilities set the loan rate. It appears just like Figure 1, with an upward sloping demand curve, D1, that intersects the supply curve toward the left of the figure. It does not intersect with S(r,q) as that line does not extend fully to the left of the chart; instead it first intersects with B(r,q). The point at which D1 intersects with B(r,q) is the equilibrium loan rate r1. In all other respects, the top panel in Figure 2 looks like Figure 1. Because the equilibrium loan rate is set where D1 intersects B(r,q), the equilbrium loan rate is set by the bank's liabilities (where banks are willing to fund loans directly). The lower panel depicts a situation where the secondary markets set the loan rate. In this panel, the demand curve, D2, is again upward sloping, but now does intersect S(r,q) at q1 on the horizontal axis. The equilibrium loan rate, r2, is set at the intersection of D2 and S(r,q). As in the Figure 1, q2 is the risk level where the equilibrium loan rate r2 intersects CP(r,q). In this panel, there are only three horizontal regions depicting if loans will be extended and how they will be funded. To the left of q1, loans are not extended. Between q1 and q2, loans are securitized. To the right of q2, loans are funded by the bank. Because the equilibrium loan rate is set at the intersection of the demand curve and S(r,q), the secondary markets set the loan rate.

Figure 3: A Change in the Cost of Bank Liabilties May Not Change the Loan Rate. This figure appears very similar to Figure 2. It is a two panel chart with the same content as Figure 2. The difference is that in both panels, B(r,q) is shifted downward to B'(r,q). In the top panel, the equilibrium loan rate falls as the equilibrium loan rate was determined by the intersection of the demand curve D1 and B (or B', after the shift). The falling equilibrium loan rate also causes q0 to shift to the left, q1 to shift to the left and q2 to shift to the right, with corresponding changes to the regions where loans will not be extended, funded by the bank, or securitized. In the lower panel, B(r,q) is also shifted down to B'(r,q), however since here the equilibrium loan rate is determined by the intersection of the demand curve D2 and S(r,q), the equilibrium loan rate is not affected. This demonstrates that if the equilibrium loan rate is set by secondary markets, a change in the cost of bank liabilities (depicted by the shift in B(r,q)) may not change the loan rate.

Figure 4: Percent of Total FHLB Advances Outstanding and Percent of AdvancestoBorrower Assets By Bank Top Holder Size Category (Small, Medium and large): 1994:Q1 2006:Q3. This is a two panel chart. The top panel charts the percent of total FHLB Advances outstanding held by small, medium, and large top holders from 1994:Q1 to 2006:Q3. The lower panel charts the percent of advancestoborrower assets for the same three groups over the same time period. Data points are quarterly. Small top holders are those with assets at or below the 50th percentile of the distribution of total assets. Medium top holders are those with assets between the 50th and 95th percentiles of the distribution of total assets. Large top holders are those with assets at or above the 95th percentile of the distribution of total assets. Size determinations are made on a quarterly basis. For the top panel, small top holders hold close to zero percent of the total FHLB advances outstanding; this does not noticably change over the time period. Medium top holders start by holding around 5% of the total FHLB advances outstanding and gradually rise to a little under 10% by 2006:Q3. Large top holders hold about 20% of the total FHLB advances outstanding at the beginning of 1994 and their share of the total gradually rises to about 50% by 2006:Q3. In the bottom panel examining percent of advancestoborrower assets, the three lines follow paths that are generally similar to each other, beginning around 3% to 4% at the beginning of 1994:Q1, holding steady until around 1997, when all three rise to around 6% to 7% by 2000. The percent of advancestoborrower assets holds steady at about this level for the remainder of the time period, diverging slightly from about 2003 to early 2005 with large top holders at about 5%, small top holders staying constant, and medium top holders rising to around 7.5%. All three top holder size categories reconverge in late 2005 at about 7%.

Figure 5: Ratio of FHLB AdvancestoEligible Assets By Bank Top Holder Size Category (Small, Medium and Large) and Time Period (1997:Q3, 2001:Q1 and 2006:Q3). This figure consists of nine histograms in a 3 by 3 grid. From left to right, the columns use data from 1997:Q3, 2001:Q1 and 2006:Q3, respectively, while from top to bottom, the rows use data from Small Top Holder FHLB Members, Medium Top Holder FHLB Members, and Large Top Holder FHLB Members. Small top holders are those with assets at or below the 50th percentile of the distribution of total assets. Medium top holders are those with assets between the 50th and 95th percentiles of the distribution of total assets. Large top holders are those with assets at or above the 95th percentile of the distribution of total assets. Each histogram depicts the frequency of the ratio of advancestoeligible assets (in percent) in 14 buckets. The first bucket is 0%. The cutoffs for the next 12 buckets are in five percent intervals (i.e. 05% not including zero, 510%, 1015% etc.). The final bucket is for greater than 60%. In all nine histograms, the four highest frequency buckets are the first four (0%, 05%, 510%, 1015%). For the small and medium banks across time periods, having no advances at all is the highest frequency bucket, with the exception of medium banks in 2006:Q3, where is is the third most common frequency bucket. After the first four buckets, in all nine histograms, the frequency of observations is monotonically decreasing bucketbybucket, reaching relatively very small levels by the 3540% bucket. Only three histograms have any observations at all in the 60+% bucket (Small 2001:Q1, Small 2006:Q3 and Large 2001:Q1).

Figure 6: Trends in FHLB Membership and Advance Usage By Bank Top Holder Size Category (Small Medium and Large): 1992:Q4 2006:Q3. This figure has three panels, each showing time series data on two indicies: a membership index and an advances index. From top to bottom, the panels are for small top holders, medium top holders and large top holders. Each panel has quarterly data from 1992:Q4 to 2006:Q3. Size determinations for top holders are made on a quarterly basis. Each series was normalized by its respective mean for the 1992:Q4 to 2006:Q3 period to create the indices of membership and advances usage. Small top holders are those with assets at or below the 50th percentile of the distribution of total assets. Medium top holders are those with assets between the 50th and 95th percentiles of the distribution of total assets. Large top holders are those with assets at or above the 95th percentile of the distribution of total assets. Each panel has three periods demarcated: 1992:Q3 to 1997:Q2, 1997:Q3 to 2000:Q4 and 2001:Q1 to 2006:Q3. These different periods correspond to what appear to be three distinct time periods for the growth rates of FHLB membership and advance usage. During the first period (1994:Q1 ? 1997:Q2), membership grew rapidly, but advance usage grew only modestly. In contrast, during the second period (1997:Q3 ? 2000:Q4) advance usage grew at least as rapidly as did FHLB membership. In addition, advance usage growth was most rapid for the largest top holders. In the final period (2001:Q12006:Q3), FHLB membership is stable and FHLB advance usage appears to be responding to other factors.

Figure 7a: Portfolio Composition: Liabilities Ratios of Domestic Deposits (Percent Total Assets) and FHLB Advances (Percent Total Assets) By FHLB Advance User Status and Bank Top Holde Size Category (Small, Medium and Large): 1994 2006. This is a sixpanel chart showing how the portfolio composition of bank top holders has changed over the time period 1994 to 2006. The first column is for small top holders, the middle column is for medium top holders and the final column is for large top holders. The first row tracks domestic deposits as a percentage of total assets and the second row tracks FHLB advances as a percentage of total assets. The domestic deposit portfolio shares for small top holders and for medium top holders were quite similar for active FHLB members, passive FHLB members, and entities that did not use advances by the end of the period (2006:Q3). This pattern, however, did not emerge for the largest top holders. domestic deposits as a percent of total assets fell at a similar rate between 1994 and 2006 for all three FHLB user status groups for small and medium bank top holders. For small and medium top holders, the fall was from around 65% to 70% to around 40% to 50%. The large active FHLB member top holders continued to rely more heavily on domestic deposits than did other large top holders, with a domestic deposits as a percent of total assets about five percentage points higher than passive users. Large top holders with no advances had domestic deposits as a percentage of total assets about ten percentage points lower than passive users in 1994, though that gap fell to less than five percentage points by the mid2000s. By construction, active FHLB members had a higher proportion of their total assets funded by FHLB advances than did passive FHLB members. Passive users for all size categories had FHLB advances equivalent to about one percent of their total assets across the entire time period. FHLB advance holdings of small and medium top holder active users rose from about six percent to eight percent. Large top holder active users rose from about four and a half percent to slightly under eight percent. Top holders without advances held no FHLB advances as a percentage of total assets by definition.

Figure 7b: Portfolio Composition: Real Estate Lending Ratios of Domestic Deposits (Percent Total Assets) and FHLB Advances (Percent Total Assets) By FHLB Advance User Status and Bank Top Holde Size Category (Small, Medium and Large): 1994 2006. This is a sixpanel chart showing how the portfolio composition of bank top holders has changed over the time period 1994 to 2006. The first column is for small top holders, the middle column is for medium top holders and the final column is for large top holders. The first row tracks mortgagerelated assets as a percentage of total assets and the second row tracks other real estate loans as a percentage of total assets. Active FHLB members held higher proportions of their total assets in mortgagerelated assets than did passive FHLB members, regardless of their top holder size. Moreover, passive FHLB members held a higher proportion of their assets in mortgagerelated assets than did entities without advances, regardless of their top holder size. The timeseries patterns for the portfolio shares of other real estate lending were quite different across the three top holder size groups: Over the time period, small top holder FHLB members increased their other real estate portfolio share modestly, medium top holder FHLB members increased their other real estate portfolio share rapidly, and large top holder FHLB members, who had the smallest such portfolio share at the beginning of the period, also had a modest increase in their other real estate portfolio. Small top holder FHLB members with advances held similar proportions of other real estate loans in each quarter regardless of whether they were active or passive members. This similarity in the timeseries data across active and passive FHLB members for other real estate loan portfolio shares was also apparent for medium top holder FHLB members. In contrast, large top holder active members held substantially higher proportions of other real estate loans in their portfolios than was held by large top holder passive members, especially after the first few years of the sample.

Figure 7c: Portfolio Composition: Other Assets Ratios of Domestic Deposits (Percent Total Assets) and FHLB Advances (Percent Total Assets) By FHLB Advance User Status and Bank Top Holde Size Category (Small, Medium and Large): 1994 2006. This is a sixpanel chart showing how the portfolio composition of bank top holders has changed over the time period 1994 to 2006. The first column is for small top holders, the middle column is for medium top holders and the final column is for large top holders. The first row tracks commercial and industrial loans as a percentage of total assets and the second row tracks securities (excluding domestic mortgagebacked securities) as a percentage of total assets. With the exception of large top holders without advances, all sizes of top holders and all FHLB usage groups increased their holdings of commerical and industrial loans until around the turn of the century, when their holdings began to decrease. Large top holders without advances were generally decreasing their holdings of commercial and industrial loans over the entire time period. The curves indicate that small and mediumsized top holder FHLB members with advances did not have as dramatic of a runoff in their commercial and industrial portfolio after the turn of the century as did entities without advances. For the largest top holders, active FHLB advance users appear to have been able to mitigate the runoff in the commercial and industrial lending portfolio whereas passive FHLB advance users had a steep decline in this lending activity. Nevertheless, the passive FHLB advance users among the largest top holders held about the same proportion of their portfolio in commercial and industrial loans as did small and mediumsized top holders by the end of the sample period. Small and mediumsized top holders that did not rely on FHLB advances tended to hold a higher proportion of their portfolio in securities excluding mortgagebacked assets than did comparablysized users of FHLB advances. In contrast, the largest top holders without FHLB advances held similar proportions of securities excluding mortgagebacked assets as did comparablysized FHLB advance users. For smalland mediumsized top holders, securities as a percentage of total assets has generally been falling over the time period across FHLB advance usage groups, especially with a noticable drop around 2001. This pattern does not hold for large top holders, where active users have seen their securities excluding mortgagebacked assets as a percentage of total assets fall, passive users have generally seen that percentage rise and large top holder with no advances saw this percentage generally increase until about 2004, when it began a steep drop. Large top holders across FHLB advance usage groups also hold less securities as a percentage of total assets than small or medium top holders (usually around 6 to 10 percent versus the 15 to 25 percent commonly seen in the small and medium top holder groups).

Figure 8a: Commercial Bank Portfolio Responses to FHLB Advance Shocks Mortgage Lending (MORT) Relative to Commercial and Industrial (C&I) and Other Real Estate Lending (OREL) By Bank Top Holder Size Category (Small and Large) Data for 1997:Q3 2000:Q4. This is a four panel chart arranged twobytwo. The top row is for small top holders with advances and the bottom row is for large top holders with advances. Each panel has two curves. The left panels have curves for the mortgage and C&I responses to FHLB advance shocks, while the right panels have curves for the mortgage and OREL response to FHLB advance shocks. The mortgage curves on each panel in the same row are identical. Each curve also is surrounded by a shaded 90% confidence interval. The horizontal axis represents the number of quarters since the shock and runs from 0 to 30. The vertical axis represents percentage change from the base in basis points. All response curves start at zero change from the base in the first quarter. In the top left panel, in the first period after the shock, both MORT and C&I response curves become positive, reaching their peak at around the fifth quarter, then both converging back towards zero. The MORT curve is always significantly positive and the C&I curve becomes significantly positive around week 8 and remains so for the remainder of the panel. Both the MORT and C&I curves are always within the confidence interval for the other curve. In the top right panel, both MORT and OREL response curves become positive before converging back towards zero. MORT is always significantly different from zero after the first quarter, while OREL becomes significantly different from zero in the seventh quarter and remains so for the rest of the time period. MORT is always within the confidence interval for OREL, while OREL remains outside the confidence interval for MORT for the first few periods. In the bottom left panel, the MORT response curve becomes positive before converging back towards zero. The C&I response curve initially becomes negative, but becomes positive in the fourth quarter and eventually converges towards zero. At no time is the C&I curve statistically distinguishable from zero. The MORT response curve is indistinguishable from zero at the beginning and end of the panel, but from the sixth to sixteenth quarter, it is. Both response curves are always within the confidence intervals of the other curve. In the bottom right panel, both the MORT and OREL response curves become positive and eventually converge towards zero. Both are statistically different from zero for around ten quarters; MORT beginning around quarter six and OREL beginning around quarter three. Both response curves are always within the confidence intervals of the other curve.

Figure 8b: Commercial Bank Portfolio Responses to FHLB Advance Shocks Mortgage Lending (MORT) Relative to Commercial and Industrial (C&I) and Other Real Estate Lending (OREL) By Bank Top Holder Size Category (Small and Large) Data for 2001:Q1 2006:Q3. This is a four panel chart arranged twobytwo. The top row is for small top holders with advances and the bottom row is for large top holders with advances. Each panel has two curves. The left panels have curves for the mortgage and C&I responses to FHLB advance shocks, while the right panels have curves for the mortgage and OREL response to FHLB advance shocks. The mortgage curves on each panel in the same row are identical. Each curve also is surrounded by a shaded 90% confidence interval. The horizontal axis represents the number of quarters since the shock and runs from 0 to 30. The vertical axis represents percentage change from the base in basis points. All response curves start at zero in the first quarter. Both the MORT and C&I response curves in the upper left panel are positive after the first quarter and eventually converge back towards zero. MORT is briefly significantly different from zero during at the beginning of the time period. C&I is never significantly different from zero. Both response curves are always within the confidence intervals of the other curve. In the upper right panel, the MORT response curve becomes positive after the first quarter (and is significant for one quarter) and then converges back towards zero. The OREL response curve becomes negative after the first quarter (and is significant for two quarters) before converging back towards zero. In the lower left panel, both MORT and C&I response curves are negative after the first quarter before converging back towards zero. Both the MORT and C&I response curves are statistically indistinguishable from zero for the entire period. Both response curves are always within the confidence intervals of the other curve. The lower right panel appears very similar to the lower left; both MORT and OREL response curves are negative after the first quarter before converging back towards zero. Both the MORT and OREL response curves are statistically indistinguishable from zero for the entire period. Both response curves are always within the confidence intervals of the other curve.

Figure 9a: Commerical Bank FHLB Advance Responses to Loan Shocks Mortgage Lending (MORT) Relative to Commercial and Industrial (C&I) and Other Real Estate Lending (OREL) By Top Holder Size Category (Small, Medium, and Large) Data for 1997:Q3 2000:Q4. This is a six panel chart arranged with two columns and three rows. The top row is for small top holders with advances, the middle is for medium top holders with advances and the bottom row is for large top holders with advances. Each panel has two curves. The left panels have curves for the FHLB advance responses to MORT and C&I shocks, while the right panels have curves for the FHLB advance responses to MORT and OREL shocks. The mortgage curves on each panel in the same row are identical. Each curve also is surrounded by a shaded 90% confidence interval. The horizontal axis represents the number of quarters since the shock and runs from 0 to 30. The vertical axis represents percentage change from the base in basis points. In the upper left panel, the MORT and C&I curves are both significantly positive. The FHLB advances response to a MORT shock is also signficantly greater than to advances response to a C&I shock for the first several quarters. Both curves converge to zero. The upper right panel looks similar to the upper left panel, but with OREL shocks instead of C&I shocks. The MORT and OREL curves are both significantly positive. The FHLB advances response to a MORT shock is also signficantly greater than to advances response to a OREL shock for the first several quarters. Both curves converge to zero. The panels in the middle row (for medium top holders with advances) have the same features that were described for the panels in the upper row (for small top holders with advances), though the magnitude of the shocks has generally increased, especially for MORT. The panels in the lower row (for large top holders with advances) also have the same features of the higher rows, with still higher magnitudes for MORT (about twice as large as those for small top holders with advances).

Figure 9b: Commerical Bank FHLB Advance Responses to Loan Shocks Mortgage Lending (MORT) Relative to Commercial and Industrial (C&I) and Other Real Estate Lending (OREL) By Top Holder Size Category (Small, Medium, and Large) Data for 2001:Q1 2006:Q3. This is a six panel chart arranged with two columns and three rows. The top row is for small top holders with advances, the middle is for medium top holders with advances and the bottom row is for large top holders with advances. Each panel has two curves. The left panels have curves for the FHLB advance responses to MORT and C&I shocks, while the right panels have curves for the FHLB advance responses to MORT and OREL shocks. The mortgage curves on each panel in the same row are identical. Each curve also is surrounded by a shaded 90% confidence interval. The horizontal axis represents the number of quarters since the shock and runs from 0 to 30. The vertical axis represents percentage change from the base in basis points. In the upper left panel, the MORT and C&I curves are both significantly positive. The FHLB advances response to a MORT shock is also signficantly greater than to advances response to a C&I shock for almost all 30 quarters. Both curves converge to zero. In the upper right panel, both the MORT and OREL curves are significantly positive for all 30 quarters. Near the beginning of the panel, the MORT curve is briefly outside of the OREL confidence interval, but everywhere else, the OREL and MORT curves are within the confidence interval of the other curve. The middle left panel (for medium top holders with advances) resembles the top left panel (for small top holders with advances). The MORT and C&I curves are both significantly positive. The FHLB advances response to a MORT shock is also signficantly greater than to advances response to a C&I shock for about the first 10 quarters. Both curves converge to zero. In the middle right panel, both the MORT and OREL curves are significantly positive for the entire 30 quarters and converge towards zero. The confidence intervals for the two curves overlap throughout the entire panel. Both of the lower panels resemble the middle right panel, but with much wider confidence intervals and larger magnitudes. MORT, C&I and OREL curves are all significantly positive. The C&I curve remains significantly positive until near the end of the panel, the MORT curve remains significant until the eighth quarter and the OREL curve remains significant until the 13th quarter. The confidence intervals for MORT and C&I in the left panel and MORT and OREL in the right panel overlap throughout the panels and often encompass the other curve.

Figure 10a: Small Bank Portfolio Responses to Federal Funds Shocks Banks with FHLB Advances versus those without Advances Mortgage Loans, Commercial & Industrial Loans, and Other Real Estate Loans By Time Period (1997:Q3 2000:Q4 and 2001:Q1 2006:Q3). This is a six panel chart arranged with two columns and three rows. The top row looks at mortgage responses, the middle row looks and C&I responses and the bottom row looks at other real estate loans (OREL) responses. The panels on the left use data from the time period 1997:Q3 2000:Q4. The panels on the right use data from the time period 2001:Q1 2006:Q3. Each panel has two curves: one for small top holders with advances and the other for small top holders without advances. Each curve is surrounded by a shaded 90% confidence interval. The horizontal axis represents the number of quarters since the shock and runs from 0 to 30. The vertical axis represents percentage change from the base in basis points. In the upper left panel, the response curves for both small with advances and small without advances are initially significantly positive, become more so, and then converge towards zero, becoming for a while significantly negative in the process. Initially the confidence intervals for the curves overlap, but the curves do become significantly different later. When the confidence intervals for the curves no longer overlap, the curve for small top holders with advances is more positive than the other.In the upper right panel, the initial responses to the the shock are positive for both small top holders with and without advances. The curve for small top holders with advances initially increases before decreasing and becoming negative, ultimately converging towards zero. The curve is initially significantly above zero and shortly after the curve becomes negative, it becomes significantly less than zero. Small top holders without advances start with an intially positive shock which immediately decreases, becoming negative and ultimately converging towards zero. The curve is initially statistically indistinguishable from zero, but later becomes negative to a statistically significant degree. Initially the two curves have overlapping confidence intervals, but for several quarters in the middle, and again at the end of the panel, the curves are statistically different from each other. The curve for small with advances is initially more positive than the curve for small without advances, including when the curves are first statistically distinguishable, however at the end of he panel when the curves are converging towards zero, the curve for small without advances is slightly more positive than the other curve to a statistically significant degree. In the middle left panel, both curves (for small with and without advances) start positive and immediately decline, becoming negative before converging towards zero. Both curves are initially statistically positive and later both become statistically negative. The confidence intervals for the two curves initially overlap, but quickly the curves do become statistically distinguishable. The response curve for small with advances is always more positive than the curve for small without advances. In the middle right panel, both curves start positive. The small with advances curve is increasing for the first quarter, but then begins to fall until it is negative and eventually converges towards zero. The small without advances curve begins falling immediately until it too is negative and eventually converges towards zero. Both curves are ultimately significantly negative. The two curves are generally close to each other and almost always are within the confidence interval of the other curve (the small with advances curve briefly is above and outside of the confidence interval for the small without advances curve). In the lower left panel, both curves are initially significantly positive. Both curves quickly fall (the small without advances curve slightly rises for one quarter first) and become significantly negative before converging towards zero. For the entire panel, the small with advances curve is above the small without advances curve. After the first few quarters, the confidence intervals for the curves no longer overlap. In the lower right panel, the small with advances curve begins positive and falls to become significantly negative before converging towards zero. The small with advances curve begins negative (though not significantly so) and falls further before converging towards zero. At all times, the confidence intervals for the two curves are intertwined.

Figure 10b: Medium Bank Portfolio Responses to Federal Funds Shocks Banks with FHLB Advances versus those without Advances Mortgage Loans, Commercial & Industrial Loans, and Other Real Estate Loans By Time Period (1997:Q3 2000:Q4 and 2001:Q1 2006:Q3). This is a six panel chart arranged with two columns and three rows. The top row looks at mortgage responses, the middle row looks and C&I responses and the bottom row looks at other real estate loans (OREL) responses. The panels on the left use data from the time period 1997:Q3 2000:Q4. The panels on the right use data from the time period 2001:Q1 2006:Q3. Each panel has two curves: one for medium top holders with advances and the other for medium top holders without advances. Each curve is surrounded by a shaded 90% confidence interval. The horizontal axis represents the number of quarters since the shock and runs from 0 to 30. The vertical axis represents percentage change from the base in basis points. In the upper left panel, both curves start significantly positive and increase for several quarters, when they begin to fall, eventually becoming significantly negative before converging back towards zero. Initially, the confidence intervals for the two curves overlap, however, about when the curves are approaching their nadir before converging back towards zero, the overlap between the confidence intervals ends. The curve for medium with advances is less negative. In the upper right panel, the medium with advances curve starts at zero and increases for a quarter before falling, ultimately becoming significantly negative before converging back towards zero. The medium without advances curve begins significantly positive and also increases for one quarter before falling, ultimately becoming significantly negative before converging back towards zero. For almost the entire panel, the confidence intervals for the two curves overlap. Near the very end, the confidence interval for the medium without advances curve becomes just fully above the confidence interval of the other curve. In the middle left panel, both curves are insignificantly negative and fall for a short period before converging towards zero. They both quickly become significantly negative. Initially their confidence intervals overlap, but the confidence interval for the curve for medium top holders with advances is significantly above the other confidence interval within ten quarters of the beginning of the panel. In the middle right panel, both curves begin negative and rise before falling again, becoming significantly negative, then converging towards zero. The medium with advances curve is above the medium without advances curve for the entire panel and briefly becomes positive for a few quarters at the beginning of the panel. For the entire panel, the confidence intervals of the two curves overlap. In the lower left panel, the medium with advances curve begins sigificantly positive and quickly falls to be significantly negative before converging to zero. The medium without advance curve begins insignificantly negative and falls further, becoming significantly negative, before converging towards zero. For the entire panel, the medium with advances curve is above the medium without advances curve. At all points, the confidence intervals for the two curves do not overlap. In the lower right panel, both curves begin positive (the medium with advances curve significantly so) and quickly fall to become significantly negative before converging back to zero. At all points in the panel, the confidence intervals for the two curves overlap.

Figure 10c: Large Bank Portfolio Responses to Federal Funds Shocks Banks with FHLB Advances versus those without Advances Mortgage Loans, Commercial & Industrial Loans, and Other Real Estate Loans By Time Period (1997:Q3 2000:Q4 and 2001:Q1 2006:Q3). This is a six panel chart arranged with two columns and three rows. The top row looks at mortgage responses, the middle row looks and C&I responses and the bottom row looks at other real estate loans (OREL) responses. The panels on the left use data from the time period 1997:Q3 2000:Q4. The panels on the right use data from the time period 2001:Q1 2006:Q3. Each panel has two curves: one for large top holders with advances and the other for large top holders without advances. Each curve is surrounded by a shaded 90% confidence interval. The horizontal axis represents the number of quarters since the shock and runs from 0 to 30. The vertical axis represents percentage change from the base in basis points. The fact that the confidence intervals for the large with advances curves in all panels are much larger stands out when comparing this figure to figures 10A and 10B. Furthermore, unlike other curves in figures 10A and 10B, the confidence intervals for the large without advances curve get wider towards the right of the panels (as more quarters pass since the shock.) Thus the confidence intervals for this curve in all the panels in figure 10C are certainly not converging to zero. In the upper left panel, the curve for large without advances starts significantly positive, increases for one quarter and then converges towards zero. The width of the confidence interval quickly increases until the curve no longer is significantly positive. The curve for large with advances initially starts out negative, becomes positive in the second quarter, then falls, soon becoming negative again before converging towards zero. The first several quarters of this panel are the only points in the entire Figure 10C where the confidence intervals for the curves do not overlap. In the upper right panel, the large with advances curve begins as insignificantly negative and falls, becoming significantly negative before converging towards zero. The large without advances curve begins positive, falls to become negative, then converges towards zero. The large with advances curve is always statistically indistinguishable from zero. The confidence intervals for both curves always at least overlap and in fact in almost every quarter, the large without advance confidence interval completely envelops the confidence interval for the large with advances curve. In the middle left panel, the large with advances curve is always significantly negative. It falls for the first quarter, then begins converging towards zero. The large without advances curve is always negative, though also always insignificantly so. It too falls for the first quarter, then begins convergins towards zero. In every quarter the confidence interval for the large without advances curve completely envelops the confidence interval for the large with advances curve. In the middle right panel, both the large with advances and large without advances curves begin positive, fall for several quarters, becoming negative, before converging towards zero. Near the beginning, the large without advances curve is briefly significantly positive and near the middle of the panel, both curves are significantly negative. The confidence intervals for both curves always at least overlap and in fact in almost every quarter, the large without advance confidence interval completely envelops the confidence interval for the large with advances curve. In the lower left panel, both the large with advances and large without advances curve start out negative. Large with advances is significantly negative, becomes increasingly so over the next several quarters, then converges to zero. The confidence interval for the large without advances curve always easily encompasses the large with advances confidence interval. The large without advances curve is always insignificantly negative and eventually converges back towards zero. In the lower right panel, both the large with advances curve and large without advances curve start out insignificantly positive. Both curves fall, become negative and eventually converge back towards zero. Although the large without advances curve is always above the large with advances curve, in every quarter but the first the large without advances confidence interval encompasses the large with advances confidence interval. The large with advances curve eventually becomes significantly negative, but the large without advances curve does not.

Figure 11a: Small Bank Portfolio Responses to GDP Shocks Banks with FHLB Advances versus those without Advances Mortgage Loans, Commercial & Industrial Loans, and Other Real Estate Loans By Time Period (1997:Q3 2000:Q4 and 2001:Q1 2006:Q3). This is a six panel chart arranged with two columns and three rows. The top row looks at mortgage responses, the middle row looks and C&I responses and the bottom row looks at other real estate loans (OREL) responses. The panels on the left use data from the time period 1997:Q3 2000:Q4. The panels on the right use data from the time period 2001:Q1 2006:Q3. Each panel has two curves: one for large top holders with advances and the other for large top holders without advances. Each curve is surrounded by a shaded 90% confidence interval. The horizontal axis represents the number of quarters since the shock and runs from 0 to 30. The vertical axis represents percentage change from the base in basis points. In the upper left panel, both the small with advances and small without advances curve begin as significantly negative, fall further for about five quarters, then converge towards zero. At all points, both curves are significantly below zero, the small without advances curve is below the small with advances curve, and the confidence intervals for the two curves do not overlap. In the upper right panel, the small with advances curve begins insignificantly negative, falls further (becoming significantly negative) then eventually converges towards zero. The small without advances curve begins insigificantly positive, falls to become significantly negative, then converges towards zero. As the rate of decrease for the small without advances curve (and its confidence interval) is higher than the small with advances curve, for several quarters the small without advances curve and confidence interval is below the small with advances curve and confidence interval with no overlap. Apart from this brief period, the two confidence intervals overlap. In the middle left panel, the small with advances curve begins as insiginificantly negative, falls for about five quarters (becoming significantly negative) then converges towards zero. The small without advances curve begins as significantly negative, also falls for about five quarters, then converges towards zero. The small without advances curve is always below the small with advances curve, and except for a few quarters at the beginning of the panel, the confidence intervals for the two curves do not overlap. In the middle right panel, both curves start insignificantly positive, fall to become significantly negative, then converge towards zero. The rate of fall for the small without advances curve is greater, and for several quarters the confidence interval for the small without advances curve is below the confidence interval for the small with advances curve. In the lower left panel, both the small with advances and small without advances curve begin as signficantly negative, fall further for about six quarters, then converge towards zero. At all points, both curves are significantly below zero, the small without advances curve is below the small with advances curve, and the confidence intervals for the two curves do not overlap. In the lower right panel, both curves start significantly negative, fall further for about six quarters, then converge towards zero. The rate of fall for the small without advances curve is greater, and for several quarters the confidence interval for the small without advances curve is below the confidence interval for the small with advances curve.

Figure 11b: Medium Bank Portfolio Responses to GDP Shocks Banks with FHLB Advances versus those without Advances Mortgage Loans, Commercial & Industrial Loans, and Other Real Estate Loans By Time Period (1997:Q3 2000:Q4 and 2001:Q1 2006:Q3). This is a six panel chart arranged with two columns and three rows. The top row looks at mortgage responses, the middle row looks and C&I responses and the bottom row looks at other real estate loans (OREL) responses. The panels on the left use data from the time period 1997:Q3 2000:Q4. The panels on the right use data from the time period 2001:Q1 2006:Q3. Each panel has two curves: one for large top holders with advances and the other for large top holders without advances. Each curve is surrounded by a shaded 90% confidence interval. The horizontal axis represents the number of quarters since the shock and runs from 0 to 30. The vertical axis represents percentage change from the base in basis points. In the top left panel, both the medium with advances and medium without advances curves begin significantly negative and continue to fall before converging towards zero. Both curves are significantly below zero for the entire panel. The medium with advances curve (and its confidence interval) falls faster than the medium without advances curve and for a short time the confidence interval for the medium with advances curve is completely below the confidence interval for the medium without advances curve. The medium with advances curve also converges to zero faster than the medium without advances curve, and for about the right half of the panel, the confidence interval for the medium with advances curve is completely above the confidence interval for the medium without advances curve. In the top right panel, the medium with advances curve begins insignificantly positive and quickly falls to become significantly negative before converging towards zero. The medium without advances curve begins insignificantly negative, falls further to become significantly negative, then converges towards zero. At all points, the confidence intervals for the two curves overlap. In the middle left panel, both curves start significantly negative, fall for around five quarters, then begin to converge towards zero. The medium with advances curve converges to zero faster than the medium without advances curve and though the confidence intervals for the two curves overlap at the left side of the panel, soon after the curves begin to head to zero, the confidence interval for the medium with advances curve is completely above the confidence interval for the medium without advances curve. In the middle right panel, the medium without advances curve begins insignificantly positive and quickly falls to become significantly negative before converging towards zero. The medium with advances curve begins significantly negative, falls further, then converges towards zero. At all points, the confidence intervals for the two curves overlap. In the lower left panel, both curves start significantly negative, fall for around six quarters, then begin to converge towards zero. The medium with advances curve is always above the medium without advances curve. It also converges to zero faster than the medium without advances curve and though the confidence intervals for the two curves overlap at the left side of the panel, soon after the curves begin to head to zero, the confidence interval for the medium with advances curve is completely above the confidence interval for the medium without advances curve. In the lower right panel, both curves begin significantly negative, and fall further before converging to zero. At all points in the panel, the curves are significantly negative and have overlapping confidence intervals.

Figure 11c: Large Bank Portfolio Responses to GDP Shocks Banks with FHLB Advances versus those without Advances Mortgage Loans, Commercial & Industrial Loans, and Other Real Estate Loans By Time Period (1997:Q3 2000:Q4 and 2001:Q1 2006:Q3). This is a six panel chart arranged with two columns and three rows. The top row looks at mortgage responses, the middle row looks and C&I responses and the bottom row looks at other real estate loans (OREL) responses. The panels on the left use data from the time period 1997:Q3 2000:Q4. The panels on the right use data from the time period 2001:Q1 2006:Q3. Each panel has two curves: one for large top holders with advances and the other for large top holders without advances. Each curve is surrounded by a shaded 90% confidence interval. The horizontal axis represents the number of quarters since the shock and runs from 0 to 30. The vertical axis represents percentage change from the base in basis points. The fact that the confidence intervals for the large with advances curves in all panels are much larger stands out when comparing this figure to figures 11A and 11B. Furthermore, unlike other curves in figures 11A and 11B, the confidence intervals for the large without advances curve get wider towards the right of the panels (as more quarters pass since the shock.) Thus the confidence intervals for this curve in all the panels in figure 11C are certainly not converging to zero. In the upper left panel, the large with advances curve begins significantly negative, falls for several quarters, then converges towards zero, remaining statistically negative the entire time. The large without advances curve begins insignificantly negative, rises to become positive, then converges towards zero. At all times in the panel, the confidence interval for the large without advances curve encompasses the confidence interval for the large with advances curve. In the upper right panel, the large with advances curve begins as insignificantly positive, then falls to become significantly negative before converging towards zero. The large without advances curve begins as positive, falls to become negative, then converges towards zero. It is always insignificant and its confidence interval always encompasses the confidence interval for the large with advances curve. In the middle left panel, the large with advances curve begins significantly negative, falls further for several quarters, then converges towards zero, remaining significantly negative the entire time. The large without advances curve begins as insignificantly negative, falls further, becoming significantly negative for about 10 quarters, then converges towards zero. For the first several quarters, the confidence interval for the large without advances curve overlaps but does not encompass the confidence interval for the large with advances curve. This is the only point in all of figure 11C (including the remainder of this panel) where the confidence interval for the large without advances curve does not completely encompass the confidence interval for the large with advances curve. In the middle right panel, the large with advances curve begins significantly negative, falls further for several quarters, then converges towards zero. The large without advances curve begins as insignificantly negative, falls further, becoming significantly negative for about 15 quarters, then converges towards zero. Throughout the entire panel, the confidence interval for the large without advances curve encompasses the confidence interval for the large with advances curve. In the lower left panel, the large with advances curve begins significantly negative, falls further for several quarters, then converges towards zero. The large without advances curve, always insignificant, begins negative, falls further then converges towards zero. Throughout the entire panel, the confidence interval for the large without advances curve encompasses the confidence interval for the large with advances curve. In the lower right panel, the large with advances curve begins significantly negative, falls further for about 15 quarters, then converges towards zero. The large without advances curve, always insignificant, begins negative, falls further for several quarters then converges towards zero. Throughout the entire panel, the confidence interval for the large without advances curve encompasses the confidence interval for the large with advances curve.


Table 1: FHLB Commercial Bank and Thrift Members
(2006:Q3)
Entity Type Number of Entities Number of Borrowers Advances Outstanding (Billions) % of FHLB Advances % of FHLB Capital Stock % of Advances to Borrower Assets
Asset Size (1) (2) (3) (4) (5) (6)
Commercial Banks: Less than 100 million 2,260 1,457 5.9 0.9 1.2 6.7
Commercial Banks: 100 million to 1 billion 3,239 2,594 52.0 8.1 9.7 6.7
Commercial Banks: Greater than 1 billion 441 393 227.9 35.4 33.3 6.0
Commercial Banks: Subtotal 5,940 4,444 285.8 44.4 44.1 --
Thrifts: Less than 100 million 380 232 1.5 0.2 0.4 11.3
Thrifts: 100 million to 1 billion 734 622 30.1 4.7 4.7 13.9
Thrifts: Greater than 1 billion 152 142 296.5 46.0 39.6 18.4
Thrifts: Subtotal 1,266 996 328.2 51.0 44.7 --
Source: Federal Housing Finance Board (FHFB).


Table 2: Commercial Bank Top Holder Members of the FHLB System
(2006:Q3)
Bank Top Holder Size Category Number of Top Holders Number of Borrowers Borrowers (Percent of Top Holders) Advances Outstanding (Billions) Percent of FHLB Advances Percent of Advances to Borrower Assets
column number (1) (2) (3) (4) (5) (6)
Small 2166 1419 65.5 $6.4 1.0 6.6
Medium 2584 2133 82.5 $56.1 8.7 7.1
Large 297 274 92.3 $304.5 47.3 7.1
Total 5047 3826 75.8 $367.0 57.0 --
Memo: 10 Largest Top Holder Members 10 10 100.0 $134.8 20.9 6.9
Note: Small top holders are those with assets at or below the 50th percentile of the distribution of total assets. Medium top holders are those with assets between the 50th and 95th percentiles of the distribution of total assets. Large top holders are those with assets at or above the 95th percentile of the distribution of total assets.


Table 3: Liability Structure of Commercial Banks by Top Holder Size Category
Year-end 2001 and Year-end 2005
TOP HOLDER SIZE Balance Sheet Items Year-end 2001, Top Holders With Advances (Billions of Dollars) Year-end 2001, Top Holders With Advances (Percent of Total Assets) Year-end 2001, Top Holders Without Advances (Billions of Dollars) Year-end 2001, Top Holders Without Advances (Percent of Total Assets) Year-end 2005, Top Holders With Advances (Billions of Dollars) Year-end 2005, Top Holders With Advances (Percent of Total Assets) Year-end 2005, Top Holders With Advances (Billions of Dollars) Year-end 2005, Top Holders With Advances (Percent of Total Assets)
Small: Total Liabilities: Core Deposits 41 70 69 74 62 67 65 72
Small: Total Liabilities: Foreign Deposits 0 0 0 0 0 0 0 0
Small: Total Liabilities: Subordinated Debt 0 0 0 0 0 0 0 0
Small: Total Liabilities: Large Time Deposits 7 13 12 13 13 14 13 14
Small: Total Liabilities: Other Managed Liabilities 4 7 1 1 7 8 1 1
Small: Total Liabilities: FHLB Advances 4 6 0 0 6 6 0 0
Small: Total Liabilities: Other 0 1 1 1 1 1 1 1
Small: Total Equity Capital 6 10 11 12 9 10 12 13
Small: Total Assets 59 100 93 100 92 100 91 100
Medium: Total Liabilities: Core Deposits 342 68 152 72 490 65 134 69
Medium: Total Liabilities: Foreign Deposits 1 0 1 0 1 0 0 0
Medium: Total Liabilities: Subordinated Debt 0 0 0 0 0 0 0 0
Medium: Total Liabilities: Large Time Deposits 68 13 30 14 121 16 32 17
Medium: Total Liabilities: Other Managed Liabilities 43 9 4 2 71 9 5 2
Medium: Total Liabilities: FHLB Advances 32 6 0 0 53 7 0 0
Medium: Total Liabilities: Other 4 1 2 1 5 1 2 1
Medium: Total Equity Capital 46 9 22 11 71 9 22 11
Medium: Total Assets 505 100 210 100 759 100 195 100
Large: Total Liabilities: Core Deposits 2058 54 459 27 2511 45 910 42
Large: Total Liabilities: Foreign Deposits 20 1 72 4 66 1 83 4
Large: Total Liabilities: Subordinated Debt 63 2 30 2 80 1 41 2
Large: Total Liabilities: Large Time Deposits 273 7 148 9 524 9 175 8
Large: Total Liabilities: Other Managed Liabilities 691 18 285 17 941 17 274 13
Large: Total Liabilities: FHLB Advances 158 4 0 0 198 4 0 0
Large: Total Liabilities: Other 190 5 210 12 266 5 225 10
Large: Total Equity Capital 344 9 142 8 533 10 229 10
Large: Total Assets 3830 100 1691 100 5527 100 2182 100
Memo: Number of Firms                
Small 1087 1087 2187 2187 1374 1374 1771 1771
Medium 1910 1910 1037 1037 2118 2118 710 710
Large 262 262 54 54 271 271 36 36
Total 3259 3259 3278 3278 3763 3763 2517 2517
Source: Bank Call Reports and Bank Holding Company Consolidated Reports.


References

Ambrose, B.W., M. LaCour-Little, and A.B. Sanders, 2004, "The Effect of Conforming Loan Status on Mortgage Yield Spreads: A Loan Level Analysis," Journal of Real Estate Economics, 32, pp. 541-569.

Arellano, M. and O. Bover, 1995, "Another Look at the Instrumental Variable Estimation of Error Component Models," Journal of Econometrics, 68, pp. 29-51.

Baker-Botts L.L.P. 2003. Report to the Board of Directors of the Federal Home Loan Mortgage Corporation: Internal Investigation of Certain Accounting Matters, December 10, 2002 - July 21, 2003. Available at: www.freddiemac.com/news/board_report.

Bennett, R. L., M. D. Vaughan, and T. J. Yeager, 2005. "Should the FDIC Worry about the FHLB? The Impact of Federal Home Loan Bank Advances on the Bank Insurance Fund," FDIC Center for Financial Research working paper 2005-10.

Berger, A. N., 2003, "The Economic Effects of Technological Progress: Evidence from the Banking Industry," Journal of Money, Credit, and Banking, 35, pp. 141-176.

Bernanke, B.S. and A.S. Blinder, 1992, "The Federal Funds Rate and the Channels of Monetary Transmission," American Economic Review, September, pp. 901-921.

Den Haan, W.D., S. Sumner, and G. Yamashiro, 2004, "Bank Loan Components and the Time-Varying Effects of Monetary Policy Shocks," Centre for Economic Policy Research discussion paper 4724.

Eisenbeis, R.A., W.S. Frame, and L.D. Wall, 2006, "An Analysis of the Systemic Risks Posed by Fannie Mae and Freddie Mac and An Evaluation of the Policy Options for Reducing Those Risks." Federal Reserve Bank of Atlanta working paper 2006-2.

Federal Home Loan Banks' Office of Finance, 2004, "Federal Home Loan Banks: Quarterly Financial Report for the Six Months Ended June 30, 2004." Available at: http://www.fhlb-of.com/specialinterest/finreportframe.html.

Flannery, M. J. and W.S. Frame, 2006, "The Federal Home Loan Bank System: The Other Housing GSE," Federal Reserve Bank of Atlanta, Economic Review, 91, third quarter, pp. 33-54.

Fortune, P., 1976, "The Effect of FHLB Bond Operations on Savings Inflows at Savings and Loan Associations: Comment," Journal of Finance, 31, pp. 963-972

Frame, W. S., 2003, "Federal Home Loan Bank Mortgage Purchases: Implications for Mortgage Markets," Federal Reserve Bank of Atlanta, Economic Review, 88, third quarter, pp. 17-31.

Frame, W.S. and L.J. White, 2004. "Regulating Housing GSEs: Thoughts on Institutional Structure and Authorities," Federal Reserve Bank of Atlanta Economic Review, 89, second quarter, pp. 87-102.

Frame, W.S. and L.J. White, 2005. "Fussing and Fuming over Fannie and Freddie: How Much Smoke, How Much Fire?" Journal of Economic Perspectives, 19, pp. 159-184.

Frame, W. S., A. Srinivasan, and L. Woosley, 2001. "The Effect of Credit Scoring on Small Business Lending," Journal of Money, Credit, and Banking, 33, pp. 813-825.

Goldfield, S. M., D. M. Jaffee, and R. E. Quandt, 1980. "A Model of FHLBB Advances: Rationing or Market Clearing?" Review of Economics and Statistics, 62, pp. 339-347.

Hancock, D. and J.A. Wilcox, 1995, "Bank Capital Shocks: Dynamic Effects on Securities, Loans, and Capital," Journal of Banking and Finance, 19, pp. 661-677.

Hancock, D., A. Lehnert, S.W. Passmore, and S. Sherlund, 2005, "An Analysis of the Potential Competitive Impacts of Basel II Capital Standards on U.S. Mortgage Rates and Mortgage Securitization," Basel II White Paper, Board of Governors of the Federal Reserve System, Washington, DC, April.

Heuson, A., S.W. Passmore and R. Sparks, 2001, "Credit Scoring and Mortgage Securitization: Implications for Mortgage Rates and Credit Availability," Journal of Real Estate Finance and Economics, 23, November, pp. 337-363.

Jackson, B., 2004, "New Collateral Types? Not all FHLBs on Board," American Banker Community Banking Supplement, December 14, p. 4.

Kashyap, A.K., and J.C. Stein, 2000, "What do a Million Observations on Banks Say about the Transmission of Monetary Policy," American Economic Review, 90, pp. 407-428.

Kwon, J.K. and R.M. Thornton, 1971, "An Evaluation of the Competitive Effect of FHLB Open Market Operations on Savings Inflows at Savings and Loan Associations," Journal of Finance, 26, pp. 699-712.

Love, I. 2001, "Estimating Panel-data Autoregressions: Package of Programs for Stata," Columbia University, Mimeo.

Love, I. and L. Zicchino, 2006, "Financial Development and Dynamic Behavior: Evidence from Panel VAR," Quarterly Review of Economics and Finance, 46, pp. 190-210.

McCool, T.J., 2005, "Federal Home Loan Bank System: An Overview of Changes and Current Issues Affecting the System," Testimony before the Committee on Banking Housing, and Urban Affairs, U.S. Senate, April 13.

McKenzie, J., 2002, "A Reconsideration of the Jumbo/Non-jumbo Mortgage Rate Differential," Journal of Real Estate Finance and Economics, 25, pp. 197-213.

Ostas, J.R., 1981, "The Federal Home Loan Bank System: Cause or Cure for Disintermediation?" Journal of Monetary Economics, 8, pp. 231-246

Passmore, S.W., S. Sherlund, and G. Burgess, 2005, "The Effect of Housing Government-Sponsored Enterprises on Mortgage Rates," Real Estate Economics, 33, pp. 427-463.

Paul, Weiss, Rifkin, Wharton, and Garrison LLP, 2006. "A Report to the Special Review Committee of the Board of Directors of Fannie Mae." Available at: http://download.fanniemae.com/report.pdf.

Silber, W. L., 1973. "A Model of the Federal Home Loan Bank System and Federal National Mortgage Association Behavior," Review of Economics and Statistics, 55, pp. 308-320.

Stiglitz, J. E. and A. Weiss, 1981. "Credit Rationing in Markets with Imperfect Information," American Economic Review, 71, pp. 393-410.

Thomson, J.B., 2002. "Commercial Banks' Borrowing from the Federal Home Loan Banks," Federal Reserve Bank of Cleveland, Economic Commentary, July.

Tuccillo, J.A., F.E. Flick, and M.R. Ranville, 2005. "The Impact of Advances on Federal Home Loan Bank Portfolio Lending: A Statistical Analysis," Working Paper, Available at: www.fhlbanks.com/html/council_news.html.

U.S. Office of Federal Housing Enterprise Oversight, 2006. Report of the Special Examination of Fannie Mae. Available at: http://www.ofheo.gov/media/pdf/FNMSPECIALEXAM.PDF

U.S. Office of Federal Housing Enterprise Oversight, 2004. Report of Findings to Date: Special Examination of Fannie Mae. Available at: www.ofheo.gov/media/pdf/

FNMfindingstodate17sept04.pdf.

U.S. Office of Federal Housing Enterprise Oversight. 2003. Report of the Special Examination of Freddie Mac. Available at: http://www.ofheo.gov/media/pdf/specialreport122003.pdf.

Van Horne, J.C., 1973, "The Effect of FHLB Bond Operations on Savings Inflows at Savings and Loan Associations: Comment," Journal of Finance, 28, pp. 194-97.

Wall, L.D., R.A. Eisenbeis, and W.S. Frame, 2005. "Resolving Large Financial Intermediaries: Banks versus Housing Enterprises," Journal of Financial Stability, 1, pp. 386-425.


Footnotes

1. Two other GSEs serve agriculture: the Farm Credit System and the Federal Agricultural Mortgage Corporation (Farmer Mac). The Student Loan Marketing Association (Sallie Mae) is also a GSE serving education, although it is in the process of privatization under the name SLM Corp. Return to Text
2. Housing GSEs have attracted a considerable amount of attention in recent years from both the media and policymakers. Much of the media attention has centered on the accounting scandals at Fannie Mae and Freddie Mac, which resulted in the dismissal of senior executives at each institution. For discussions of the accounting issues, see: (1) Baker-Botts LLP (2003) and U.S. Office of Federal Housing Enterprise Oversight (2003) for Freddie Mac, and (2) U.S. Office of Federal Housing Enterprise Oversight (2004, 2006) and Paul, Weiss, Rifkin, Wharton, and Garrison LLP (2006) for Fannie Mae. For an overview of the relevant public policy issues, see: Frame and White (2004, 2005), Wall, Eisenbeis, and Frame (2005), and Eisenbeis, Frame, and Wall (2006). Return to Text
3. See McKenzie (2002) for a review of this literature. Ambrose, LaCour-Little, and Sanders (2004) and Passmore, Sherlund, and Burgess (2005) provide recent contributions. Return to Text
4. Two studies have shown that FHLB members tend to hold more mortgage-related asset holdings, but neither study was able to credibly establish that these higher mortgage holdings are a consequence of FHLB membership or FHLB activities. Neither Thomson (2002) nor Tuccillo, Flick, and Ranville (2005) can ascribe a causal relationship since the reverse relation is unaccounted for (i.e., that more active mortgage lenders are those most likely to join the FHLB System). These papers are also hampered by the fact that the other explanatory variables in the empirical models (the other portions of bank portfolios) are treated as exogenous, when they are endogenously determined in practice. Return to Text
5. See Federal Home Loan Bank Mission, 12 C.F.R. § 940 (2006), and "Mission of the Banks," 65 Fed. Reg. 25, 278 (May 1, 2000). Other FHLB activities include (1) acquiring member assets (e.g., mortgages), (2) stand-by letters of credit, (3) intermediary derivative contracts, and (4) debt or equity investments (that primarily benefit households below 80 percent of area median income). Return to Text
6. Advances are the historical channel by which the FHLBs served their public mission and these collateralized loans still comprise about 62 percent of the FHLB System's consolidated balance sheet. Commercial banks, although only eligible for FHLB membership since the passage of the Financial Institutions Recovery and Reform Act of 1989, now account for 73 percent of FHLB System membership. Return to Text
7. The 12 FHLBs are located in Atlanta, Boston, Chicago, Cincinnati, Dallas, Des Moines, Indianapolis, New York, Pittsburgh, San Francisco, Seattle, and Topeka. The Office of Finance is located in Reston, Virginia. Return to Text
8. Some financial institutions do maintain charters in multiple FHLB districts, which allow them to be members of more than one FHLB. This creates a degree of inter-FHLB competition. Return to Text
9. The regulator of the Federal Home of Loan Banks recently authorized the Federal Home Loan Bank of Chicago to issue subordinated debt that would not be a joint liability of all 12 FHLBs. This is the first instance of debt that was not a joint liability being authorized. Return to Text
10. Data for the FHLB System as of year-end 2006 is available from Federal Home Loan Banks' Office of Finance at http://www.fhlb-of.com/specialinterest/financialframe.html. Return to Text
11. See Frame (2003) for a detailed discussion of the FHLB mortgage programs. Return to Text
12. Retained earnings account for only six percent of the FHLB System's total equity capital. History provides an explanation: Congress previously took the FHLBs' retained earnings to help pay for the thrift bailout and thereafter the institutions began to pay out almost all earnings as dividends. The Financial Modernization Act of 1999 clarified that a particular class of FHLB shareholders would legally own the institution's retained earnings (as well as surplus, undivided earnings, and equity reserves) going forward. Return to Text
13. See 12 U.S.C. § 1430 (a)(2)(A). Return to Text
14. See 12 U.S.C. 1430(a)(3) for a complete list of eligible collateral. Federal Agency securities are generally synonymous with debt and mortgage-backed securities issued by government sponsored enterprises. Return to Text
15. In particular, the FHLB maintains a claim senior to depositors, unsecured and secured creditors, and the claims of any receiver, conservator, or trustee. The only excepted claims are those entitled priority under otherwise applicable law or where a secured party has perfected a security interest in specific assets. When resolving an insolvent depository institution, the FDIC has made it a practice to simply make FHLB creditors whole straightaway, including prepayment penalties associated with advances. See Bennett, Vaughan, and Yeager (2005) for a description of how FHLB advances may increase the probability of bank default and raise the FDIC's expected losses given default. Return to Text
16. No FHLB has ever suffered a loss on an advance. Return to Text
17. This view is consistent with that articulated in early studies of the FHLB System by Silber (1973) and Goldfield, Jaffee, and Quandt (1980). Related studies examined whether FHLB debt issuance may also actually lead to some disintermediation (crowding-out): Kwon and Thornton (1971), Van Horne (1973), Fortune (1976), and Ostas (1981). Return to Text
18. The statute defines "community financial institutions" as banks, thrifts, or credit unions with total assets of less than $500 million. (This cutoff is annually adjusted for inflation and stood at $587 million for 2006). Unlike the situation before 1999, community financial institutions may join an FHLB even if they do not hold 10 percent of their assets in residential mortgage-related assets. In practice, however, removing this restriction was not very important because most small institutions (and many large ones) maintain at least 10 percent of their assets in residential mortgage-related assets. In addition, community financial institutions may pledge small business, small farm, and small agribusiness loans as collateral for their FHLB advances, rather than being limited to mortgage assets. In practice, however, few FHLBs have actually accepted those alternative forms of collateral (Jackson, 2004).Return to Text
19. Focusing on the portfolio decision in the absence of capital requirements, a risk-neutral mortgage originator will offer a mortgage if qr+(1-q)r_{d }= r_{f} where r is the mortgage rate received by the lender if the borrower does not default, r_{d } is the expected return to the lender if the borrower does default, and r_{f} is the expected return on an alternative investment. Rewriting this expression in terms of an equality and solving for r, it is easily demonstrated that the inverse supply function for mortgages is decreasing in q and r_{d }, but increasing in r_{f}. See Heuson, Passmore, and Sparks (2001, p. 340). Return to Text
20. The purple line incorporates the market's credit risk-sensitive capital requirement. This marginal cost curve with respect to credit quality implicitly assumes that other marginal costs for loan financing do not vary with respect to credit quality. Thus, the curvature simply reflects the effective cost of capital to back the credit risk (or an equivalent credit guarantee). Return to Text
21. The blue line at q_{2 } is determined by the originator's comparison of the marginal profit derived from holding the loan to the price offered by the securitizer for selling the mortgage. Return to Text
22. In contrast, in an adverse selection model (such as proposed by Stiglitz and Weiss, 1981), when mortgage rates rise lower risk borrowers drop out of the pool of potential borrowers. This type of adverse selection model assumes that borrowers with higher default risks have higher expected returns from their investment projects (in this case, the project is a home purchase). In our model, however, the benefits associated with homeownership are not related to a household's default probability. In this case, rising mortgage rates simply raise the cost of homeownership without any offsetting effects. Return to Text
23. As noted earlier, the model presented here is a stylized version of Heuson, Passmore, and Sparks (2001). More generally, the underwriting standards of market participants - depositories and securitizers alike - may change as mortgage rates change (i.e., the black vertical dashed lines may move to the left or the right). Return to Text
24. Alternatively, these loans can be viewed as being sold into a market for standardized loans or loan participations among banks. Return to Text
25. When two entities merge, the Federal Housing Finance Board (FHFB) information does not add the FHLB advances outstanding for the predecessor and the successor. Rather, the successor entity has its own FHLB advances as of the date of the merger and any additional advances extended after the merger date. Quarterly Call Report data on FHLB advances is only available 2001:Q1 and beyond. These data pool the advances of the predecessor and successor. Using Call Report data rather than FHFB data for FHLB advances for the 2001:Q1-2006:Q3 period did not materially or qualitatively affect the empirical results presented below. Return to Text
26. Data for the remaining year-ends between these dates had similar histogram patterns for each top holder size group. Return to Text
27. Each bank (1) had positive net loans and leases and positive equity capital, (2) was headquartered in one of the fifty U.S. states, and (3) indicated that its primary activity was commercial banking. Return to Text
28. Using generally accepted accounting principals (GAAP), securities that are "held-to-maturity" are included at their amortized cost, but securities "available-for-sale" are included at their fair value. Return to Text
29. Home equity lines are typically secured by a junior lien and usually are accessible by check or credit card. The reported value on the Call Report is the amount outstanding as of the report date, not the total amount that the customer is authorized to borrow under such arrangements. Return to Text
30. MORT = RCON1797+RCON5367+RCON5368. (RCON is the Call Report mnemonic for domestic balance sheet and income information for banks. RCFD is the Call Report mnemonic for both domestic and foreign information for banks.) Return to Text
31. OREL = RCON1415+RCON1420+RCON1460+RCON1480. Return to Text
32. SEC = RCON1754+RCON1773. Return to Text
33. C&I = RCON1766. For banks with less than $300 million in assets, this item is only reported on a consolidated basis (i.e., commercial and industrial loans = RCFD1766). Return to Text
34. DEP = RCON2702. Return to Text
35. Entities without advances include FHLB members with no advances and non-members. Return to Text
36. Aggregate data for each portfolio share were constructed by summing the balance sheet component across entities in the subgroup and dividing by the sum of total assets across entities in the relevant subgroup. This procedure effectively weights the top holders' portfolio shares by total assets. Return to Text
37. See Kashyap and Stein (2000). Return to Text
38. Year-end data are presented for 2001 and 2005. Data for the remaining year-ends between these dates provided similar patterns in the liability structures for each top holder size group. Return to Text
39. These yields are interpolated by the U.S. Department of the Treasury from the daily yield curve based mainly on quarter-end "on-the-run" Treasury securities. Return to Text
40. See Love (2001) and Love and Zicchino (2002). Return to Text
41. Standard errors for the impulse-response functions reported below are obtained using Monte Carlo simulation. Random draws of errors are used together with the estimated coefficients and their variance-covariance matrix to re-compute impulse-responses. This procedure is repeated 1000 times. Then, the 5th and 95th percentiles of the resulting distribution are used as a confidence interval for each element of an impulse-response. See Love (2001).Return to Text
42. Our approach to VAR modeling is similar to the "semi-structural" approach used in Bernanke and Mihov (1998). Using their framework, in our context the bank's "policy variables" are the changes in loans extended for each loan type; advance demand is simply a result of these policy decisions. Since the bank is a economic unit in a country (whereas the central bank is the decision maker in Bernanke and Mihov), macroeconomic variables are not contemporaneously determined by bank polices, but rather vice versa. Return to Text
43. Other orderings of the variables did not materially affect the portfolio responses to macroeconomic shocks reported below. For example, we considered the following four alternative orderings: (1) FFR, YIELD, GDP, MORT, ADV, C&I, OREL, DEP, SEC; (2) FFR, YIELD, GDP, DEP, MORT, C&I, OREL, SEC, ADV; (3) FFR, YIELD, GDP, DEP, MORT, ADV, C&I, OREL, SEC; and (4) FFR, YIELD, GDP, C&I, OREL, ADV, MORT, DEP, SEC.Return to Text
44. Impulse-response functions were calibrated to the average one-standard deviation percentage change in the three loan categories, namely mortgages, C&I loans and other real estate loans, across the two entity size categories. Return to Text
45. All shocks have been standardized by using the average shock across loan types and top holder groups (with and without advances). Return to Text
46. Reported impulse-response functions for the large top holder groups are derived using a first-order VAR with an identical variable ordering that was used for third-order VAR models reported for smaller top holders. Return to Text
47. Reported impulse-response functions for the large top holder groups are derived using a first-order VAR with an identical variable ordering that was used for the third-order VAR models reported for smaller top holders. Return to Text

♣ This version is optimized for use by screen readers. Descriptions for all mathematical expressions are provided in LaTex format. A printable pdf version is available. Return to Text
Home | Economic Research and Data | Accessibility | Contact us
Last update: August 2, 2007