A. The Database

The FAs provide the most comprehensive overview of financial transactions in the United States. The database contains information on 25 sectors of the economy (e.g., households, depository institutions, money market mutual funds) and on the flows and levels of their assets and liabilities, broken down into 22 different financial instruments (e.g., home mortgages, corporate equities). The data are available on a quarterly frequency, in most cases going as far back as 1952.

As a whole, the FAs paint a picture of who owes what. Unfortunately, however, the FAs do not contain information on the financial interconnections between counterparties. In other words, the FAs do not provide information on the “whom-to-whom lending.”³ An example helps clarify what is contained or not in the FAs: regarding commercial paper, Table L.208 shows that, as of 2014Q4, of the $930 billion outstanding commercial paper, $182 billion had been issued by nonfinancial corporations and $65 billion by asset-backed security (ABS) issuers. The table also contains information on the holders: for instance, $36 billion worth of commercial paper was held by pension funds and $334 billion by money market mutual funds. However, the table does not specify how much of the commercial paper held by pension funds was issued by the nonfinancial corporations, by the ABS issuers, or by some other issuer. That is, the FAs do not specify the connection between pension funds and ABS issuers via commercial paper. In the next subsection, we outline a method to construct such interconnections with the data at hand.

B. Estimating Financial Interconnections

We are interested in measuring financial linkages across different sectors of the economy. These linkages are established via specific financial instruments, with more than one instrument being used most of the time. For example, banks and households are linked via mortgages and consumer loans. At the same time, banks are linked to money market mutual funds via commercial paper. It is through such linkages that credit and funding risks among different sectors arise, facilitating the transmission and potential amplification of shocks to a specific asset class or sector. Often these linkages are an under-used source of information and measuring them is paramount to monitoring risks to the financial system and the economy as a whole.

To get a measure of financial interconnections, we first obtain assets and liabilities of each sector for each debt instrument in the FAs. We focus on debt instruments because these contracts carry the features that are more likely to enable the transmission and amplification of shocks throughout the system.⁴ We then assume that the whom-to-whom lending (for each specific instrument) is proportional to each sector’s asset holdings of that instrument.⁵ A step-by-step description and example of this procedure is helpful:

1. For each instrument, we obtain the assets and liabilities held by each sector from Tables L.204 to L.226. For example, Table L.210 (partially reproduced in Figure 1) indicates that, in 2010, there was a total of $7,598 billion in agency- and GSE-backed securities issued by budget agencies ($24.2 billion), government-sponsored enterprises ($6,435 billion), and agency- and GSE-backed mortgage pools ($1,140 billion). The same table shows that private pension funds held $184 billion out of the total amount. But note that the table does not specify who issued the securites the pension funds held—budget agencies, government-sponsored enterprises, or agency- and GSE-backed mortgage pools.

U.S. Flow of Funds: Agency- and GSE-Backed Securities

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Source: Financial Accounts of the United States.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

U.S. Flow of Funds: Agency- and GSE-Backed Securities

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Source: Financial Accounts of the United States.

• Download Figure
• Download figure as PowerPoint slide

U.S. Flow of Funds: Agency- and GSE-Backed Securities

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Source: Financial Accounts of the United States.

• Download Figure
• Download figure as PowerPoint slide

2. With the information on asset holdings by each sector, we calculate the share of assets held by each sector. For instance, private pension funds held 2.4 percent of the total amount of agency- and GSE-backed securities outstanding in 2009 (that is, $184 billion out of $7,598 billion).

Finally, we allocate each sector’s liabilities obtained in step 1 uniformly according to the asset shares computed in step 2. For example, private pension funds held 2.4 percent of the agency- and GSE-backed securities issued by budget agencies (i.e., 2.4 percent of $24.2 billion = $0.6 billion), by government-sponsored enterprises (i.e., 2.4 percent of $6,435 billion = $156 billion), and by agency- and GSE-backed mortgage pools (i.e., 2.4 percent of $1,140 billion = $28 billion). Applying this procedure to each instrument we are interested in yields 22 three-dimensional matrices of size 25-by-25-by-252 (22 being the number of instruments; 25 being the number of sectors, and 252 being the number of quarters from 1952Q1 to 2014Q4). We then consolidate these matrices to arrive at a more general breakdown of sectors and instruments. In particular, we consolidate the Financial Accounts sectors and create the following groupings for the analysis:

• Households (HSH): Households and nonprofit organizations;

• Nonfinancial business (NFB): Nonfinancial corporate and nonfinancial non-corporate businesses;

• Banks (BNK): U.S.-chartered depository institutions, foreign banking offices in the United States, banks in U.S.-affiliated areas, credit unions;

• Nonbanks (NoBNK): Property-casualty insurance, life insurance companies, private pension funds, state and local government retirement funds, federal governemnt retirement funds, money market mutual funds, mutual funds, closed-end funds, ETFs, government-sponsored enterprises, agency- and GSE-backed mortgage pools, ABS issuers, finance companies, REITs, broker-dealers, funding corporations, holding companies;⁶

• Rest of the World (RoW).

Regarding instruments, recall that we are interested in studying the pattern of different types of credit flows over the business cycle. Hence, we make a distinction between those instruments that are equity or equity-like (such as corporate equities and mutual fund shares), and those that are debt and debt-like (i.e., those instrument that embed a de jure or de facto promise to repay at par). Thus we group the FAs’ instruments into the following categories:

• Equity-like: corporate equities and mutual fund shares;

• Debt and debt-like:

  • Short-term (and non-insured): Shares of money market funds, fed funds and repo agreements, open market paper (mostly commercial paper);

  • Long-term: Checkable deposits, time and savings deposits, Treasuries, GSE-backed securities, municipal decurities and loans, corporate and foreign bonds, bank loans not classified, other loans, total mortgages (home, multifamily, commercial, and farm), consumer credit, trade credit, and security credit.

Before proceeding, some comments about our procedure are in order. First, we compute these matrices using the ‘level’ of credit tables from the FAs and derive real credit series by deflating in the same manner as GDP (i.e., using the GDP deflator).

Second, by assuming that the whom-to-whom linkages are proportional to sectors’ holdings of each instrument, we are abstracting from specific concentrations and/or risk profiles that may be present in the data. That could be the case, for instance, if there is relationship lending at the sector level or a sectoral preference by specific lenders for a specific type of assets issued by a specific type of borrower. For instance, in the previous example, this may be the case if pension funds preferred to hold agency- and GSE-backed securities issued only by budget agencies. Similarly, our approach treats all debt from a given sector as equally risky and may be ignoring preferences or institutional constraints that create demand for specific asset classes. For example, pension funds may be required to hold only investment-grade assets. This may generate inaccuracies especially when classifications of debt securities have subcomponents with very different characteristics, such as municipal securities and loans. Unfortunately, the lack of comprehensive information on such investor preferences and granular data on risk profiles prevents the construction of a more accurate database. That said, the distribution we propose is anchored in the data, minimizes the number of required assumptions, and is the one followed by other authors (cf., Bhatia and Bayoumi, 2012; Gallin, 2013).⁷

Third, our measures capture gross and direct linkages. We do not attempt to net out the amount of lending between two sectors that is intermediated via a third sector. That is, in measuring credit from banks to households, we estimate direct loans that banks make to households, as opposed to trying to estimate the amount of credit that banks give to households via nonbanks (see Gallin, 2013, for an analysis along these lines). Therefore, our approach is appropriate for the purpose of documenting differences in bank and nonbank credit patterns throughout the macroeconomic cycle and around systemic events.

Fourth, since we are interested in the macroeconomic role of credit flows, some level of aggregation of nonbank entities is needed in Section III. However, grouping financial institutions with different business models and risk characteristics may also prevent the identification of different behavioral patterns among types of entities. To address this concern, we use a more disaggregated definition of nonbanks to examine the role of financial interconnections around systemic crises in Section IV.

A. Business and Credit Cycles

As argued, understanding how credit and output co-move through the business cycle is necessary to design effective macroeconomic and macroprudential policies that appropriately balance growth and financial stability objectives. With this goal in mind, we follow Claessens et al. (2012) and apply Harding and Pagan (2002)’s algorithm to our various credit series.⁸ In this manner, we identify the phases that each series has undergone in our 63-year sample. That is, we determine when a series reached a peak or trough, and when it was in a boom or bust phase. After that, we assess the length and depth of credit cycles compared to the business cycle. We also measure the ratio of synchronicity between various series. In other words, we measure how frequently both series are in the same phase of the cycle (e.g., both booming, both at a peak, etc.).⁹

The main patterns detected by this exercise are as follows (Table 1):

• The cycles of bank and nonbank credit to the private sector (households and businesses) are longer than the output cycle.¹⁰ However, these cycles have similar amplitudes to the GDP cycle (except for nonbank credit to households), as evidenced by the coefficient of variation.

• Bank-to-bank credit and bank-to-nonbank credit tend to exhibit shorter cycles. And although shorter, these cycles tend to be relatively deep.

• Bank credit to households and bank credit to nonfinancial businesses have the shallowest cycles, but are the most similar in amplitude to GDP.

Table 1.

Credit Cycle Length and Depth during 1952Q1 – 2014Q4

Note: Coefficient of variation is defined as the standard deviation divided by the mean of a series.Source: Authors’ calculations

Table 1.

Credit Cycle Length and Depth during 1952Q1 – 2014Q4

	Average cycle duration (in years)	Coefficient of variation	Share of GDP in 2014
GDP	6.3	0.5	…
Bank credit to households	10.5	0.6	25.6
Bank credit to nonfinancial businesses	10.5	0.6	21.6
Bank credit to banks	3.2	1.6	0.4
Bank credit to nonbanks	6.3	1.0	17.1
Nonbank credit to households	21.0	1.1	42.3
Nonbank credit to nonfinancial businesses	7.9	0.7	25.8
Nonbank credit to banks	6.3	1.1	8.2
Nonbank credit to nonbanks	10.5	1.2	44.5

Note: Coefficient of variation is defined as the standard deviation divided by the mean of a series.Source: Authors’ calculations

View Table

Table 1.

Credit Cycle Length and Depth during 1952Q1 – 2014Q4

	Average cycle duration (in years)	Coefficient of variation	Share of GDP in 2014
GDP	6.3	0.5	…
Bank credit to households	10.5	0.6	25.6
Bank credit to nonfinancial businesses	10.5	0.6	21.6
Bank credit to banks	3.2	1.6	0.4
Bank credit to nonbanks	6.3	1.0	17.1
Nonbank credit to households	21.0	1.1	42.3
Nonbank credit to nonfinancial businesses	7.9	0.7	25.8
Nonbank credit to banks	6.3	1.1	8.2
Nonbank credit to nonbanks	10.5	1.2	44.5

Note: Coefficient of variation is defined as the standard deviation divided by the mean of a series.Source: Authors’ calculations

In terms of the cyclical dynamics, we find that (Table 2):

• Nonbank credit tends to be more procyclical than bank credit. This can be seen by comparing the synchronicity ratios in segment a of Table 2 to those in segment b. For instance, nonbank credit to businesses is in the same phase as output 79 percent of the time, whereas bank credit to businesses coincides with output 72 percent of the time.

• Credit within the financial system is less procyclical than credit to the nonfinancial sector. For example, bank-to-bank credit has a coefficient of 0.43 compared to 0.71 for bank credit to households and 0.72 for bank credit to businesses (cf., segment a). Similarly, nonbank credit to banks (0.68) and to nonbanks (0.75) is also less procyclical than nonbank credit to households (0.78) and to businesses (0.79), (cf., segment b).

• Credit outflows to the rest of the world tends to be less procyclical that credit to the private sector, with nonbank credit being substantially more synchronized with the business cycle (0.69, at the bottom of segment b) than bank credit (0.50, at the bottom of segment a).

Table 2.

Synchronization Ratios of Output and Credit Cycles

Note: Synchronization ratio is the ratio of number of observations where GDP and credit are in the same phase divided by total observations. Output-upward (output-downward) ratio is the ratio of observations where GDP and credit are both booming (slumping) divided by number of periods when GDP was booming (slumping). Source: Authors’ calculations

Table 2.

Synchronization Ratios of Output and Credit Cycles

		Synchronization ratio						Output-upward synchronization ratio	Output-downward synchronization ratio
		Full sample		1952-1979		1980-2014
Output cycle and bank credit cycle
	Bank credit to households	0.71		0.84		0.61		0.77	0.60
	Bank credit to nonfinancial businesses	0.72		0.79		0.66		0.82	0.35
	Bank credit to banks	0.43		0.46		0.42		0.47	0.35
	Bank credit to nonbanks	0.70		0.74		0.73		0.80	0.30
	Bank credit to ROW	0.50		0.56		0.46		0.54	0.45
Output cycle and nonbank credit cycle
	Nonbank credit to households	0.78		0.85		0.86		0.89	0.40
	Nonbank credit to nonfinancial businesses	0.79		0.81		0.76		0.90	0.30
	Nonbank credit to banks	0.68		0.76		0.62		0.78	0.30
	Nonbank credit to nonbanks	0.75		0.73		0.76		0.87	0.20
	Nonbank credit to ROW	0.69		0.70		0.69		0.78	0.30
Bank credit cycle and nonbank credit cycle
	Credit to households	0.76		0.90		0.57		…	…
	Credit to nonfinancial businesses	0.81		0.93		0.72		…	…
	Credit to banks	0.51		0.63		0.42		…	…
	Credit to nonbanks	0.75		0.72		0.75		…	…
Output cycle and ROW credit cycle
	ROW credit to households	…		…		…		…	…
	ROW credit to nonfinancial businesses	0.73		0.65		0.80		0.84	0.30
	ROW credit to banks	0.61		0.59		0.64		0.70	0.20
	ROW credit to nonbanks	0.66		0.61		0.83		0.73	0.55
	ROW credit to ROW	0.63		0.64		0.64		0.73	0.25

Note: Synchronization ratio is the ratio of number of observations where GDP and credit are in the same phase divided by total observations. Output-upward (output-downward) ratio is the ratio of observations where GDP and credit are both booming (slumping) divided by number of periods when GDP was booming (slumping). Source: Authors’ calculations

View Table

Table 2.

Synchronization Ratios of Output and Credit Cycles

		Synchronization ratio						Output-upward synchronization ratio	Output-downward synchronization ratio
		Full sample		1952-1979		1980-2014
Output cycle and bank credit cycle
	Bank credit to households	0.71	View Expanded	0.84	View Expanded	0.61	View Expanded	0.77	0.60
	Bank credit to nonfinancial businesses	0.72		0.79		0.66		0.82	0.35
	Bank credit to banks	0.43		0.46		0.42		0.47	0.35
	Bank credit to nonbanks	0.70		0.74		0.73		0.80	0.30
	Bank credit to ROW	0.50		0.56		0.46		0.54	0.45
Output cycle and nonbank credit cycle
	Nonbank credit to households	0.78	View Expanded	0.85		0.86		0.89	0.40
	Nonbank credit to nonfinancial businesses	0.79		0.81		0.76		0.90	0.30
	Nonbank credit to banks	0.68		0.76		0.62		0.78	0.30
	Nonbank credit to nonbanks	0.75		0.73		0.76		0.87	0.20
	Nonbank credit to ROW	0.69		0.70		0.69		0.78	0.30
Bank credit cycle and nonbank credit cycle
	Credit to households	0.76		0.90	View Expanded	0.57	View Expanded	…	…
	Credit to nonfinancial businesses	0.81		0.93		0.72		…	…
	Credit to banks	0.51		0.63		0.42		…	…
	Credit to nonbanks	0.75		0.72		0.75		…	…
Output cycle and ROW credit cycle
	ROW credit to households	…		…		…		…	…
	ROW credit to nonfinancial businesses	0.73		0.65	View Expanded	0.80	View Expanded	0.84	0.30
	ROW credit to banks	0.61		0.59		0.64		0.70	0.20
	ROW credit to nonbanks	0.66		0.61		0.83		0.73	0.55
	ROW credit to ROW	0.63		0.64		0.64		0.73	0.25

Note: Synchronization ratio is the ratio of number of observations where GDP and credit are in the same phase divided by total observations. Output-upward (output-downward) ratio is the ratio of observations where GDP and credit are both booming (slumping) divided by number of periods when GDP was booming (slumping). Source: Authors’ calculations

One interpretation for the longer cycles of credit to the private sector is that some lending linkages take longer to build and to unwind than some real variables (e.g., jobs can be created and destroyed very quickly in the U.S.). For banks, this may be a consequence of “relationship lending” whereas for nonbanks the longer cycle could reflect the slow-moving nature of real estate cycles that underlies a large fraction of nonbank lending. More generally, the same mechanisms that explain the close interaction between financial conditions and macroeconomic developments could also explain the longer duration of credit cycles.¹¹ For example, financial accelerator models demonstrate the importance of feedback effects across financial markets. In other models with credit frictions, financial cycles occur not only because financial variables reflect developments in the real economy but also because they act as propagation mechanisms themselves. Yet other models focus on the special role played by banks in financial intermediation and emphasize the bank lending channel, where a negative liquidity or interest rate shock limits banks’ access to funding and forces them to adjust their lending with consequences for asset prices. A similar argument can be applied to financial intermediaries more broadly: leverage and balance sheet health affects intermediaries’ ability to lend and has implications for asset prices, including through fire sales and contagion across institutions as well as asset classes.

Next, we explore whether the findings above apply across subperiods. We split our sample into two subsamples (from 1952 to 1979 and from 1980 to 2014) to account for the considerable transformation undergone by the financial system after the 1970s, when new financial instruments, such as money market mutual funds, commercial paper, repos, and mortgage securitization took off and redefined financial activity. Our findings are detailed below.

• Bank credit to the private sector has become markedly less procyclical (segments c and d). During 1980-2014, bank credit to households was in the same cyclical phase as output 61 percent of the time, compared to 84 percent in the previous three decades. Bank credit to businesses has also become less procyclical (with a coefficient going from 0.79 to 0.66). Nonbank credit to the private sector has changed less.

• Bank and nonbank credit cycles have generally become less synchronized, as revealed by comparing the higher synchronization ratios in segment e to those of segment f (of course, this result is another manifestation of the growing divergence in procyclicality mentioned above). An exception to this finding is bank credit to nonbanks and nonbank credit to nonbanks, which have become slightly more synchronized (bottom of segments e and f).

• External credit has become markedly more procyclical, especially lending to businesses and to nonbanks, which increased their coefficients from 0.65 to 0.80 and from 0.61 to 0.83, respectively (segments g and h). Interestingly too, the amount of financing that the rest of the world intermediates through the U.S. financial system has maintained a constant level of synchronicity with the U.S. business cycle (cf., coefficient of 0.64).

We also explore if it matters whether output is in an upswing or downturn for the cyclicality of credit. That is, conditioning on the phase of the business cycle, we compute how frequently each credit cycle is in the same phase as output. The difference in coefficients is remarkable, with the synchronicity ratios being much higher during output upswings than downturns (last two columns of Table 2). This is due to the fact that GDP and credit series are more often in expansion phases than in contraction phases. That is, expansions tend to be protracted, whereas contractions are sudden and sharper.

Additionally, we investigate the degree to which bank credit and nonbank credit are simultaneously synchronized with the business cycle (Table 3). Although synchronization ratios are broadly lower with respect to their bank and nonbank counterparts, they exhibit similar properties. As with both bank credit and nonbank credit, joint synchronization between both forms of credit to nonfinancial businesses and output in the full sample is the most procyclical (0.69) while external credit and credit to banks are the least procyclical (0.40 and 0.35, respectively). Similarities to bank and nonbank cyclical patterns extend across subsamples, as well as in upswings and downturns.

Table 3.

Synchronization Ratios of Output, Bank Credit, and Nonbank Credit Cycles

Note: Synchronization ratio is the ratio of number of observations where GDP and both types of credit are in the same phase divided by total observations. Output-upward (output-downward) ratio is the ratio of observations where GDP and both types of credit are both booming (slumping) divided by number of periods when GDP was booming (slumping). Source: Authors’ calculations

Table 3.

Synchronization Ratios of Output, Bank Credit, and Nonbank Credit Cycles

	Synchronization ratio			Output-upward	Output-downward
	Full sample	1952-1979	1980-2014	synchronization ratio	synchronization ratio
Output cycle, bank credit cycle, and nonbank credit cycle
Credit to households	0.65	0.81	0.55	0.73	0.40
Credit to nonfinancial businesses	0.69	0.78	0.61	0.79	0.30
Credit to banks	0.35	0.45	0.29	0.41	0.10
Credit to nonbanks	0.61	0.61	0.63	0.72	0.10
Credit to ROW	0.40	0.44	0.39	0.46	0.20

Note: Synchronization ratio is the ratio of number of observations where GDP and both types of credit are in the same phase divided by total observations. Output-upward (output-downward) ratio is the ratio of observations where GDP and both types of credit are both booming (slumping) divided by number of periods when GDP was booming (slumping). Source: Authors’ calculations

View Table

Table 3.

Synchronization Ratios of Output, Bank Credit, and Nonbank Credit Cycles

	Synchronization ratio			Output-upward	Output-downward
	Full sample	1952-1979	1980-2014	synchronization ratio	synchronization ratio
Output cycle, bank credit cycle, and nonbank credit cycle
Credit to households	0.65	0.81	0.55	0.73	0.40
Credit to nonfinancial businesses	0.69	0.78	0.61	0.79	0.30
Credit to banks	0.35	0.45	0.29	0.41	0.10
Credit to nonbanks	0.61	0.61	0.63	0.72	0.10
Credit to ROW	0.40	0.44	0.39	0.46	0.20

Note: Synchronization ratio is the ratio of number of observations where GDP and both types of credit are in the same phase divided by total observations. Output-upward (output-downward) ratio is the ratio of observations where GDP and both types of credit are both booming (slumping) divided by number of periods when GDP was booming (slumping). Source: Authors’ calculations

B. Interest Rates and Credit

The divergent cyclical behavior across types of credit identified above could translate into a different response of bank and nonbank lending to unexpected monetary policy changes. In this subsection, we perform some VAR analyses that seem to confirm this hypothesis. In particular, we estimate a series of small VAR specifications using real GDP, the fed funds rate, and the various real credit measures. We use the first difference of the logarithms of all variables except the fed funds rate. In our baseline model, we use one lag and focus on the period 1980Q1 to 2014Q4 given that it is after 1980 when the nonbank sector developed in earnest and also when the Federal Reserve began to explicitly target the fed funds as a policy variable. As robustness checks, we consider other subsample periods and two-lag specifications.

To study the impact of a monetary contraction on the credit measures, we assume an exogenous unanticipated shock of 100 basis points to the fed funds rate. In doing so, we follow the literature spawned by Bernanke and Blinder (1992) and use the fed funds rate as a relevant indicator of monetary policy (see, for instance, Den Haan and Sterk, 2011, who conduct a similar exercise to ours).

Figures 2 and 3 show the impulse response functions corresponding to the estimated VAR for the period 1980Q1 to 2014Q4. We find that while the growth rate of bank credit to households slows down in response to an interest rate shock, growth of nonbank credit to households actually accelerates (with significant coefficients for both bank and nonbank credit measures). Note, however, that the impact on credit to the business sector is not significant. Moreover, the flow of credit from nonbanks to other financial entities (bank and nonbank) also picks up and takes a long time to return to pre-shock growth rates (bottom panels of Figure 3).

Interest Rate Shocks and Bank and Nonbank Credit to Private Sector: 1980Q1–2014Q4

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Source: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Interest Rate Shocks and Bank and Nonbank Credit to Private Sector: 1980Q1–2014Q4

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Source: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

Interest Rate Shocks and Bank and Nonbank Credit to Private Sector: 1980Q1–2014Q4

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Source: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

Interest Rate Shocks and Intra-Financial System Credit: 1980Q1–2014Q4

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Source: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Interest Rate Shocks and Intra-Financial System Credit: 1980Q1–2014Q4

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Source: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

Interest Rate Shocks and Intra-Financial System Credit: 1980Q1–2014Q4

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Source: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

To assess the robustness of our results, we consider various estimations over different subsample periods and also by increasing the number of lags to two (the impulse-response functions of such tests are reported in Appendix I). We find that the direction of the (medium-term) response of each type of credit to an interest rate shock remains the same across these changes in the estimations but that the coefficients are most often not significant. Therefore, we treat these results with caution and view them as a prompt for further and more detailed research using more disaggregated data on each of the different types of credit (e.g., ABS issuers, GSEs, etc.). The most relevant lesson of this exercise, in our view, is that there seems to be growing evidence of different reactions of bank and nonbank credit to changes in monetary policy.

Our results, however, are also consistent with the findings of Den Haan and Sterk (2011) and Nelson et al. (2015). Specifically, the former find that following a monetary tightening bank mortgages decline but that nonbank mortgages increase. Nelson et al. (2015), in turn, find that a contractionary monetary policy shock has a persistent negative impact on the asset growth of commercial banks but increases the asset growth of shadow banks.

The varying reaction of banks and nonbanks to policy rate shocks can be interpreted as symptomatic of the malleability of the complex U.S. financial system. Indeed, a highly interconnected system means that it is easier for financial entities to circumvent a tightening of policy measures. Faced with a tighter liquidity environment, banks may rely on financial innovation (e.g., securitization) to convert illiquid assets into liquid funds which could, in turn, lead to a migration of credit to nonbanks. Such is the dynamic found in Loutskina (2011), who finds evidence that securitization provides banks with an effective channel to transform illiquid assets into liquid ones and thus (by making the liquidity shock less binding) can potentially offset the impact of the Fed’s policies to contract the loan supply.

C. Credit in Recessions and Recoveries: A Brief Historical Retrospective

In this subsection, we examine the typical credit dynamics around recessions as well as expansions. We use recession dates as determined by the NBER. Given the structural changes the U.S. economy has undergone over the sample period, one may argue that the average or median dynamics are not representative. Hence, we provide the change experienced in each of the credit series prior to each business cycle peak and after each business cycle trough in Tables 4 and 5. In Appendix II, we provide a narrative of the evolution of credit measures around each of the ten post-war recessions. To guide the discussion of the typical dynamics, we plot median quarterly growth in annualized terms across time for each type of credit in Figures 4 and 5.

Table 4.

Credit Growth before Business Cycle Peaks

(Annualized growth; 6 quarters pre-peak)

Source: Authors’ calculations.

Table 4.

Credit Growth before Business Cycle Peaks

(Annualized growth; 6 quarters pre-peak)

		1953Q3	1957Q3	1960Q2	1969Q4	1973Q4	1980Q1	1981Q3	1990Q3	2001Q1	2007Q4
Bank credit to:
	Households	16.4	7.5	11.4	3.0	8.2	1.2	−4.3	−0.4	5.3	3.0
	Nonfinancial Businesses	3.1	5.3	8.0	6.3	11.3	3.4	0.0	−2.6	7.8	9.5
	Banks	−4.5	14.2	−15.2	−5.7	−7.2	2.9	−9.2	−23.2	22.1	23.6
	Nonbanks	15.5	−3.8	14.8	13.0	19.0	4.6	4.6	0.0	2.7	0.6
Nonbank credit to:
	Households	16.8	7.4	9.1	1.4	2.1	15.2	2.9	12.1	6.2	6.7
	Nonfinancial Businesses	9.8	4.3	3.3	3.3	5.8	4.0	1.9	7.0	4.8	5.2
	Banks	1.2	3.1	3.4	5.1	12.5	19.5	21.2	−5.8	6.9	8.0
	Nonbanks	35.7	11.6	24.9	5.0	15.0	13.1	19.6	8.0	12.7	8.1

Source: Authors’ calculations.

View Table

Table 4.

Credit Growth before Business Cycle Peaks

(Annualized growth; 6 quarters pre-peak)

		1953Q3	1957Q3	1960Q2	1969Q4	1973Q4	1980Q1	1981Q3	1990Q3	2001Q1	2007Q4
Bank credit to:
	Households	16.4	7.5	11.4	3.0	8.2	1.2	−4.3	−0.4	5.3	3.0
	Nonfinancial Businesses	3.1	5.3	8.0	6.3	11.3	3.4	0.0	−2.6	7.8	9.5
	Banks	−4.5	14.2	−15.2	−5.7	−7.2	2.9	−9.2	−23.2	22.1	23.6
	Nonbanks	15.5	−3.8	14.8	13.0	19.0	4.6	4.6	0.0	2.7	0.6
Nonbank credit to:
	Households	16.8	7.4	9.1	1.4	2.1	15.2	2.9	12.1	6.2	6.7
	Nonfinancial Businesses	9.8	4.3	3.3	3.3	5.8	4.0	1.9	7.0	4.8	5.2
	Banks	1.2	3.1	3.4	5.1	12.5	19.5	21.2	−5.8	6.9	8.0
	Nonbanks	35.7	11.6	24.9	5.0	15.0	13.1	19.6	8.0	12.7	8.1

Source: Authors’ calculations.

Table 5.

Credit Growth after Business Cycle Troughs

(Annualized growth; 6 quarters post-trough)

Source: Authors’ calculations.

Table 5.

Credit Growth after Business Cycle Troughs

(Annualized growth; 6 quarters post-trough)

		1954Q2	1958Q2	1961Q1	1970Q4	1975Q1	1980Q3	1982Q4	1991Q1	2001Q4	2009Q2
Bank credit to:
	Households	15.1	11.4	9.4	9.8	6.6	−4.9	8.7	−3.5	10.9	−0.4
	Nonfinancial Businesses	8.7	8.2	10.2	10.7	−1.4	3.3	11.8	−7.8	−1.4	−7.3
	Banks	−16.0	−25.8	6.5	2.7	22.5	−9.2	10.1	−3.4	65.4	37.5
	Nonbanks	8.6	17.8	21.8	15.5	0.0	5.5	9.0	5.3	14.4	0.0
Nonbank credit to:
	Households	17.9	8.0	4.9	5.7	4.1	5.3	11.7	10.7	10.3	−7.3
	Nonfinancial Businesses	5.5	3.9	7.4	4.6	−0.2	2.6	7.0	1.2	1.6	2.6
	Banks	2.2	5.4	10.9	9.2	7.4	23.3	5.7	−7.4	6.3	9.3
	Nonbanks	18.8	20.3	4.9	20.0	14.0	21.1	22.5	10.1	5.2	−10.5

Source: Authors’ calculations.

View Table

Table 5.

Credit Growth after Business Cycle Troughs

(Annualized growth; 6 quarters post-trough)

		1954Q2	1958Q2	1961Q1	1970Q4	1975Q1	1980Q3	1982Q4	1991Q1	2001Q4	2009Q2
Bank credit to:
	Households	15.1	11.4	9.4	9.8	6.6	−4.9	8.7	−3.5	10.9	−0.4
	Nonfinancial Businesses	8.7	8.2	10.2	10.7	−1.4	3.3	11.8	−7.8	−1.4	−7.3
	Banks	−16.0	−25.8	6.5	2.7	22.5	−9.2	10.1	−3.4	65.4	37.5
	Nonbanks	8.6	17.8	21.8	15.5	0.0	5.5	9.0	5.3	14.4	0.0
Nonbank credit to:
	Households	17.9	8.0	4.9	5.7	4.1	5.3	11.7	10.7	10.3	−7.3
	Nonfinancial Businesses	5.5	3.9	7.4	4.6	−0.2	2.6	7.0	1.2	1.6	2.6
	Banks	2.2	5.4	10.9	9.2	7.4	23.3	5.7	−7.4	6.3	9.3
	Nonbanks	18.8	20.3	4.9	20.0	14.0	21.1	22.5	10.1	5.2	−10.5

Source: Authors’ calculations.

Credit Growth Around Business Cycle Peaks

(Median ± 25^th percentile; Quarterly annualized growth)

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Note: t corresponds to beginning of recession.Source: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Credit Growth Around Business Cycle Peaks

(Median ± 25^th percentile; Quarterly annualized growth)

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Note: t corresponds to beginning of recession.Source: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

Credit Growth Around Business Cycle Peaks

(Median ± 25^th percentile; Quarterly annualized growth)

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Note: t corresponds to beginning of recession.Source: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

Credit Growth Around Business Cycle Troughs

(Median ± 25^th percentile; Quarterly annualized growth)

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Note: t corresponds to end of recession.Source: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Credit Growth Around Business Cycle Troughs

(Median ± 25^th percentile; Quarterly annualized growth)

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Note: t corresponds to end of recession.Source: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

Credit Growth Around Business Cycle Troughs

(Median ± 25^th percentile; Quarterly annualized growth)

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Note: t corresponds to end of recession.Source: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

We note that the growth of credit to the private sector (households and businesses) begins to slow down prior to the onset of the recession, and does not pick up again before four quarters into the recession. Bank credit to households tends to have a wider range of negative growth in recessions, whereas nonbank credit to households may recover as early as two quarters post-peak. Intra-financial system credit (i.e., bank and nonbank credit to banks and nonbanks), on the other hand, exhibits more volatility before and after recessions, and it is hard to discern a clear pattern. Note also, that the scale of these variations is quite significant.

We also construct parallel event studies for credit behavior around business cycle troughs. Here, we find that credit to households tends to recover quickly, typically within two to three quarters. Credit recoveries tend to vary widely for credit to households, while credit to businesses is more stable and recovers after three quarters. As before, interbank and inter-nonbank credit is volatile after troughs, recovering after about four quarters.

A. The 1980s: Financial Deregulation, a Banking Crisis, and New Interconnections

Financial regulation underwent fundamental changes in the 1970s and 1980s. Money market mutual funds were legalized in the mid-1970s and, contrary to banks and thrifts, were not subject to deposit rate ceilings. Such exemption soon became a major competitive threat for depository institutions, who then lobbied Congress succesfully to phase out the cap on interest rates and broaden the range of acceptable investments by banks.¹³ Coetaneously, momentous legislative changes were enacted to spur the growth of GSEs and their sponsored vehicles.¹⁴ Notably, GSEs were allowed to securitize assets, granted hefty credit lines with the Treasury, given exemptions from certain state and local taxes, and held to much lower capital requirements than banks and thrifts. And beyond the direct advantage these measures afforded, GSEs began to benefit from the ‘quasi-sovereign’ status granted by markets who funded them at rates close to that of Treasuries. Unsurprisingly, GSEs progressively became one of the dominant players in the housing market.

Faced with mounting competition from nonbanks, banks quickly took advantage of the laxer regulatory environment to switch from traditional home mortgages into more profitable but riskier projects, such as commercial real estate and high-yielding leveraged buyouts, oil extraction, and off-balance sheet products. Thus, we estimate that in less than a decade, banks reshuffled their portfolios boosting credit to businesses in an amount equivalent to 7 percentage points of GDP and cutting down loans to households although by a smaller magnitude (4 percentage points of output; Figure 6a). Banks’ portfolio recomposition turned out to be much riskier than anticipated, as the inherent riskiness of the new activities was exacerbated by regulations that curtailed banks’ ability to invest out-of-state. Thus, a hefty measure of geographical concentration risk was added to banks’ investments, also fostering overvaluations in several regional property markets and business activities.

Credit to the Nonfinancial Private Sector

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Sources: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Credit to the Nonfinancial Private Sector

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Sources: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

Credit to the Nonfinancial Private Sector

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Sources: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

Risks materialized in two waves. First, as the regional real estate bubbles sequentially burst—moving from Texas to the Southeast and Mid-Atlantic to California and Arizona—bank and thrift failures skyrocketted. A second round of defaults erupted as risky corporate investments began to grossly underperform. The final tally was close to 3,000 failed depository institutions worth nearly a trillion dollars, with 70 percent of the losses coming from savings and loans thrifts, thus giving name to the episode.

But as bank lending to households began to drop in 1980, nonbanks stepped in to provide increasing amounts of credit over a three-decade climb that culminated with the 2007 Great Financial Crisis (Figure 6b); nonbank credit to households reached 60 percent of GDP by end-2007. Large part of this growth was driven by the GSE complex which, as argued, had been empowered by legislation to expand credit to households vastly. Within a decade, GSEs doubled their mortgage portfolios while agency- and GSE-backed mortgage pools swelled almost ten-fold. That is, by 1990, the market volume of GSE securities surpassed $1.1 trillion, or sixteen percent of GDP, compared to less than $190 billion at end-1979.¹⁵

The vigorous expansion of nonbank credit to households was accompanied by a structural change that redefined the U.S. financial system in a way that became consequential thirty years later. This fundamental transformation was the creation of new and growing interconnections among all types of financial entities, but most notably among nonbanks (Figure 7a). During the 1980s, estimated credit flows among nonbanks ballooned to reach a size equivalent to almost one-third of GDP by end-1990. Remarkably, however, interconnections between banks and nonbanks grew, but remained stable at around 14 percent of GDP until the 2000s (this can be seen by adding the blue and black lines in Figure 7a).

Financial Interconnections

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Sources: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Financial Interconnections

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Sources: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

Financial Interconnections

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Sources: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

The unrelentless growth of the new financial interconnections begs a further inquiry into their implications for financial stability. Is a more interconnected system desirable since it helps diversify risks—as it was often argued before 2007—or do new linkages mask vulnerabilities that surpass potential benefits? In the next subsection we argue that the key financial interconnections that proved fatal in the 2007–08 crisis had a gestation period of almost thirty years and that they could have been ‘macroeconomicaly’ visible through analyses like ours.

B. From the 1990s to 2007: More Interconnections and a Full-Blown Financial Crisis

To investigate further how the interconnections that swelled over the three decades preceeding the GFC bear on financial stability, we take a more granular look at the different types of nonbank entities. We split the nonbank sector into three subgroups: long-term funders, short-term funders, and intermadiate funders.¹⁶

• Long-term funders are financial intermediaries with typically long-term investment horizons. This category includes private and public pension funds, casualty and life insurance companies, mutual funds, ETFs and closed-end funds.

• Short-term funders, on the other hand, are intermediaries with genuinely short investment horizons and whose liabilities are redeemable on demand. This category is comprised of money market mutual funds.¹⁷

• Intermediate funders include the rest of nonbank entities that provide credit to nonfinancial borrowers, and whose investment strategies vary at times between shorter-or longer-term. In this category we include the GSEs, agency- and GSE-backed mortgage pools, ABS issuers, finance companies, REITs, broker-dealers, holding companies, and funding corporations.

With this classification at hand, we examine the interconnections among nonbanks. That is, we measure credit flows between each pair of entity types (e.g., credit from long-term funders to short-term funders, from short-term funders to intermediate funders, and so on). We find that intermediate funders consistently received more than 90 percent of all the credit in the nonbank sector from 1980 onwards (Figure 7b). Interestingly, credit amongst intermediate funders doubled in the 1980s from 20 percent to 40 percent of all nonbank credit, and remained at that level until the 2007 crisis. Since then it has retrenched to account for just around one-fourth of nonbank credit.

The next natural question is whether it was possible to quantify somewhat the risks embedded in these interconnections. Admitting that the available data on financial linkages are deficient, we attempt a rough estimation of the credit and funding exposures faced by each subsector (Figure 8). We measure credit risk as the ratio of credit between two types of entities over the total financial assets of the lender.¹⁸ Funding risk is approximated as the ratio of credit between two types of entities over the total liabilities of the borrower (cf., FSB 2012).

Credit and Funding Risk Measures

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Credit and Funding Risk Measures

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

• Download Figure
• Download figure as PowerPoint slide

Credit and Funding Risk Measures

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

• Download Figure
• Download figure as PowerPoint slide

Given that intermediate funders were at the epicenter of the 2007 crisis (along with short-term funders, of course), we focus first on this group to assess vulnerabilities in the financial system. Our measure of credit risk reveals the secular accumulation of credit exposures to intermediate funders by all subsectors of the financial system, including banks, between 1980 and 2006 (Figure 9). Such growing exposure to a rather concentrated type of credit risk (i.e., the household sector as revealed by Figure 6b) should have raised concerns and thus elicited a closer monitoring, especially at a time when the financial system was becoming less regulated.¹⁹

Evolution of Credit Exposures to Intermediate Funders

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Sources: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Evolution of Credit Exposures to Intermediate Funders

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Sources: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

Evolution of Credit Exposures to Intermediate Funders

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Sources: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

Moreover, these credit vulnerabilities were compounded by the unstable nature of a non-negligible fraction of the funding received by nonbanks. By focusing on specific instruments that are either runnable—such as money market fund shares—or subject to high roll-over risk—such as commercial paper and repo contracts—we can derive a lower-bound estimate of the fraction of interconnections that are short-term and therefore more vulnerable to runs (Figure 10).²⁰ We find that, on the eve of the GFC, around 36 percent of the funding among nonbanks was of short-term nature. And, as was soon to be shown, this vulnerability crystallized into widespread financial distress when nonbanks began to quickly withdraw funding from each other in response to escalating credit risk concerns. For intermediate funders in particular, the share of short-term funding from nonbanks amounted to 15 percent of all their liabilities. Note, however, that by 2007Q4, short-term funding among nonbanks was not at its historical high, although it still remained high enough to be a large vulnerability were it to dissipate suddenly and needed to be replaced quickly, as was soon to be the case. Moreover, it was the combination of still relatively high short-term funding and large credit risks that proved problematic.

Short-Term Funding Among Nonbanks

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Sources: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Short-Term Funding Among Nonbanks

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Sources: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

Short-Term Funding Among Nonbanks

Citation: IMF Working Papers 2015, 143; 10.5089/9781513516448.001.A001

Sources: Authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

As we now know, these elements—high and lightly supervised interconnections among financial entities (notably via securitizations), the build-up of credit risks, and large maturity transformation (on the back of the growing commercial paper and repo markets²¹)—set the stage for the crisis dynamics that unraveled in 2007–08. After the fact, however, and from a macro-financial perspective one could argue that a framework to monitor these financial flows and measure their attendant risks, as proposed here, could be useful going forward.