## wp1716

## Source details

**Canonical URL:** [wp1716](https://www.imf.org/-/media/files/publications/wp/wp1716.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/wp1716.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/wp1716.pdf.json)

---

### Motivation and research question
- Investigates how changes in U.S. interest rates affect risk taking by financial institutions in the global syndicated term loan market.
- Focuses on whether low U.S. interest rates encourage lenders to supply riskier credit to non-U.S. borrowers, including borrowers in emerging market economies (EMEs), and whether responses differ between bank and nonbank lenders.
- Distinguishes responses to changes in short-term and long-term U.S. interest rates to provide evidence of a cross-border spillover effect of monetary policy associated with the global lending channel.

### Rationale for using the global syndicated term loan market
- Market comparable in size to the global corporate bond market and provides trillions of U.S. dollars a year to mostly nonfinancial corporations.
- Vast majority of syndicated term loans are made to speculative-grade and obscure, non-rated corporations, implying large positions in risky leveraged assets by lenders.
- Significant portion of non-U.S. borrowers are from EMEs, enabling identification of cross-border spillover effects.
- Banks originate the majority of loans but sell them quickly to nonbank lenders, so loan risk often reflects behavior of the broad financial system rather than banks alone.
- Loan spreads are fixed at origination and used as an ex-ante measure of credit risk.

### Theoretical background and measurement strategy
- Mixed theoretical predictions:
  - Risk-shifting models with asymmetric information and limited liability predict higher interest rates can exacerbate agency problems and encourage bank risk taking.
  - Portfolio allocation / “search for yield” predicts a fall in interest rates on safe assets induces shifts toward riskier assets for institutions with fixed-rate obligations.
- Net effect depends on:
  - Nature of interest rate changes—short-term versus long-term.
  - Type of financial institution—banks versus nonbanks.
- Measurement advantages:
  - Uses ex-ante measure of credit risk proxied by loan spreads fixed at origination.
  - Loan spreads alleviate endogeneity concerns and focus on ex-ante investor risk perceptions.
  - Evidence: loan spreads are strongly correlated with probabilities of borrower default over a one-year horizon (a Basel II-consistent risk parameter) reported by lenders for a subsample.

### Data, market features, and identification
- Data source: Thomson Reuters Loan Pricing Corporation (LPC)’s DealScan, covering 1995–2013, with detailed information on individual syndicated loans, borrower characteristics, and lending syndicate composition.
- Scope: term loans denominated in U.S. dollars, indexed to the U.S. dollar LIBOR, usually maturity of 5-7 years; credit lines excluded.
- Lender and borrower identification:
  - Ultimate counterparty (parent organization) aggregation used; lender types classified via two-digit SIC codes into banks and nonbank financial institutions.
  - Borrower industries classified at one-digit SIC codes.
- Market composition and dynamics:
  - Originations of term loans reached to about $2 trillion in 2013, of which roughly $700 billion were made to U.S. borrowers and $1.3 trillion to non-U.S. borrowers.
  - Syndicated lines of credit added another $2 trillion to originations in the overall syndicated loan markets.
  - At origination banks and bank holding companies (BHCs) accounted for about 80 percent of originations volume globally.
  - Median of banks’ ownership shares of syndicated loans in the U.S. market declines from about 80 percent to less than 20 percent in just a few weeks after origination.
- Identification strategy:
  - Restrict sample to loans originated to non-U.S. borrowers to treat U.S. monetary conditions as exogenous to other countries.
  - Include controls for global factors (risk appetite, uncertainty) and fixed effects: lender, borrower, lender country-group-year, lender type-year, borrower country-group-year, borrower industry-year, and syndicate characteristics.

### Empirical model (syndicate regressions) — specification highlights
- Benchmark semi-log syndicate model:
  - log(spreadj,l,b,t) = αl + αb + βF RFt + βT RTt + Xt γ + θl,c,y + θl,i,y + θb,c,y + θb,i,y + φj,t + εj,l,b,t
- Key variables:
  - spreadj,l,b,t: all-in-drawn spread for loan j by lender l to borrower b at time t.
  - RFt: federal funds rate.
  - RTt: 10-year U.S. Treasury (zero-coupon) interest rate.
  - Xt includes Merrill Lynch low-grade corporate bond spread, VIX-derived variance risk premium, European sovereign crisis spread (Italian minus German sovereign yields), U.S.-news based uncertainty index, and Michigan expected 12-month inflation.
- Estimation approach:
  - Regressions run quarterly.
  - Pre-crisis and post-crisis regressions estimated separately.
  - Post-crisis: federal funds rate excluded (zero lower bound beginning 2008:Q4); include U.S. Treasury rate only.
  - Full-sample models include interactions with Post2008 dummy and crisis controls.

### Main empirical findings — syndicated regressions (loan-level)
- Pre-crisis period:
  - Significantly negative relationship between short-term interest rates (federal funds rate) and ex-ante risk taking; movements in long-term rate statistically insignificant.
- Post-crisis period:
  - Significantly negative relationship between long-term interest rate (10-year Treasury) and risk taking.
- Marginal economic effects (using equation 3 and neighborhood of average spreads):
  - Pre-crisis: one-standard deviation decrease in the federal funds rate of 1.82 percentage point leads to an increase in loan spread by 11 basis points.
  - Post-crisis: one-standard deviation decrease in the 10-year Treasury rate of 0.71 percentage point raises the loan spread by 28 basis points.
- Other variable influences:
  - Variance risk premium: statistically insignificant across specifications.
  - U.S.-news uncertainty index: negative and statistically significant in pre-crisis (elevated uncertainty moderated lenders’ appetite for riskier loans).
  - Low-grade bond spreads: positive and consistently statistically significant (compression of loan spreads associated with broad improvements in investors’ appetite for term loans).
  - European sovereign yield spread: negative in the post-crisis column covering the European sovereign debt crisis period.

### Heterogeneity by lender type and geography
- Banks vs nonbanks:
  - Banks increased risk taking in response to both a drop in short-term rates before the crisis and a decline in long-term rates after the crisis.
  - Nonbanks increased risk taking only when long-term rates decreased in the post-crisis period.
  - Interpretation: banks’ risk taking driven by both cost-of-fund and return-on-safer-asset concerns; nonbank institutions’ risk taking motivated by return-on-safer-asset or search-for-yield incentives.
- Lender nationality and borrower region:
  - Loans to EME borrowers and non-U.S. borrowers produce results close to full sample, supporting cross-border spillover evidence.
  - For U.S. lenders to non-U.S. borrowers, the magnitude of the 10-year Treasury coefficient in the post-crisis period is much larger in absolute terms.

### Portfolio-level results (summary)
- Portfolio regressions confirm differential effects of short-term and long-term interest rates on ex-ante risk taking in pre- and post-crisis periods.
- Full-sample portfolio model: interaction of 10-year Treasury rate and the post-crisis dummy is negative and statistically significant.
- Table highlights:
  - Post-2008 coefficient: 1.327 ∗∗∗ (Table 7, portfolios of all lenders to non-U.S. borrowers).
  - Bank × Fed. funds rate, pct: Pre-crisis −0.029 ∗∗ ; Post-crisis −0.036 ∗∗∗ (Table 8).
  - Nonbank × Fed. funds rate, pct: 0.000 (pre-crisis); −0.016 (post-crisis) (Table 8).

### Robustness checks
- Immediate counterparty basis estimation: results weaker but broadly consistent with ultimate counterparty basis.
- Subsample analyses:
  - Omitting investment-grade borrowers (restricting to nonrated and speculative borrowers) produces statistically stronger results.
  - Restricting to most active lenders (lenders that lend in the market in every quarter) produces stronger results.
- Additional checks in Appendix largely support main findings.

### Caveats and limitations
- Do not observe lenders’ entire portfolios; lenders may hedge syndicated loan exposures in other markets—additional risk taking in the syndicated loan market may be offset by hedges.
- Post-crisis long-term rate effects may partly reflect expansionary monetary policies in other large advanced economies; interpret results as international spillovers of low interest rates in general rather than U.S. monetary policy exclusively.
- Analysis limited to the global syndicated loan market; cannot fully resolve whether additional risk taking induced by accommodative monetary policy is excessive—requires comprehensive welfare analysis.

### Policy implications and unanswered questions
- Evidence of cross-border spillovers: low interest rates in one country encourage riskier lending to other countries, potentially undermining domestic central bank efforts to affect risk taking and credit cycles.
- Central banks may need coordination of multiple policy tools, such as monetary and macro-prudential policies, to ensure both economic and financial stability.
- Open questions:
  - Whether additional risk taking induced by expansionary monetary policy is welfare-improving or excessive.
  - The extent to which hedging transfers risk to other financial intermediaries and aggregate systemic consequences.

### Key statistics and descriptive evidence (selected, preserved exactly)
- Originations of term loans reached to about $2 trillion in 2013, of which roughly $700 billion were made to U.S. borrowers and $1.3 trillion to non-U.S. borrowers.
- Loans extended to nonrated borrowers dominated originations; risky loans—with contractual loan spread over 125-150 points—account for the bulk of loan originations.
- Average loan spread for EME borrowers: 1.67 percentage points.
- Average loan spread for non-U.S. borrowers: 1.86 percentage points.
- Average federal funds rate: 3.0 percent; standard deviation: 2.3 percent.
- 10-year Treasury rate standard deviation: 1.4 percent.
- Average spread during pre-crisis period: about 100 basis points.
- Average spread during post-crisis period: about 260 basis points.
- Table 1 (sample 2010:Q1-2013:Q4; over 700 term loans):
  - Loan spread, pct: Mean 3.101; Std. dev. 1.633; 10th p. 1.500; 50th p. 2.750; 90th p. 5.000
  - Probability of default, pct: Mean 2.317; Std. dev. 7.110; 10th p. 0.170; 50th p. 0.800; 90th p. 4.820
  - 10-year Treas. rate, pct: Mean 2.346; Std. dev. 0.581; 10th p. 1.710; 50th p. 2.050; 90th p. 3.200
  - Variance risk premium, pct sq.: Mean 14.439; Std. dev. 6.218; 10th p. 7.365; 50th p. 14.034; 90th p. 25.099
- Table 3 (sample 1995:Q1-2013:Q4; loans to non-U.S. borrowers):
  - Loan spread, pct — Non-U.S. borrowers: Mean 1.860; Std. dev. 1.529; 10th p. 0.425; 50th p. 1.400; 90th p. 3.875
  - Loan spread, pct — EME borrowers: Mean 1.667; Std. dev. 1.352; 10th p. 0.425; 50th p. 1.250; 90th p. 3.500
  - Fed. funds rate, pct: Mean 3.010; Std. dev. 2.336; 10th p. 0.119; 50th p. 3.056; 90th p. 5.723
- Selected regression coefficients (syndicate regressions, summarized):
  - Fed. funds rate, pct: Pre-crisis −0.062 ∗∗ ; Post-crisis −0.075 ∗∗∗ (Table 4)
  - 10-year Treas. rate, pct: Post-crisis −0.150 ∗∗∗ (Table 4)
  - Fed. funds rate, pct (U.S. lenders to non-U.S. borrowers): Pre-crisis −0.065 ∗∗ ; Post-crisis −0.049 ∗∗ (Table 5)
  - 10-year Treas. rate (U.S. lenders): Post-crisis −0.261 ∗∗∗ (Table 5)
  - Fed. funds rate, pct (non-U.S. lenders to EME borrowers): Pre-crisis −0.068 ∗∗ ; Post-crisis −0.079 ∗∗∗ (Table 6)

*Source: IMF Working Paper — "1. Introduction", "3.1 Global syndicated loan market", "4.1 Syndicate regressions", "5.3 Robustness checks and caveats" (wp1716).*

### 1. Introduction ........................................................................................................

### 1. Introduction

### Motivation and research question
- Investigates how changes in U.S. interest rates affect risk taking by financial institutions in the global syndicated term loan market.
- Focuses on whether low U.S. interest rates encourage lenders to supply riskier credit to non-U.S. borrowers, including borrowers in emerging market economies (EMEs), and whether responses differ between bank and nonbank lenders.
- Distinguishes responses to changes in short-term and long-term U.S. interest rates to provide evidence of a cross-border spillover effect of monetary policy associated with the global lending channel.

### Rationale for using the global syndicated term loan market
- Market comparable in size to the global corporate bond market and provides trillions of U.S. dollars a year to mostly nonfinancial corporations.
- Vast majority of syndicated term loans are made to speculative-grade and obscure, non-rated corporations, implying large positions in risky leveraged assets by lenders.
- Significant portion of non-U.S. borrowers are from EMEs, enabling identification of cross-border spillover effects.
- Banks originate the majority of loans but sell them quickly to nonbank lenders, so loan risk often reflects behavior of the broad financial system rather than banks alone.
- Loan spreads are fixed at origination and used as an ex-ante measure of credit risk.

### Theoretical background
- Mixed predictions on relationship between interest rates and financial-sector risk taking:
  - Risk-shifting models with asymmetric information and limited liability predict higher interest rates can exacerbate agency problems and encourage bank risk taking (references: Dell’Ariccia, Laeven, and Marquez (2014); Acharya and Viswanathan (2011)).
  - Portfolio allocation theory predicts a fall in interest rates on safe assets induces shifts toward riskier assets ("search for yield" for institutions with fixed-rate obligations; Rajan (2005)).
- Net effect depends on:
  - Nature of interest rate changes—short-term versus long-term.
  - Type of financial institution—banks versus nonbanks.

### Measurement strategy and advantages of ex-ante loan spreads
- Uses ex-ante measure of credit risk proxied by loan spreads fixed at origination.
- Advantages:
  - Helps alleviate endogeneity concerns of interest rate changes.
  - Focuses on ex-ante investor risk perceptions at loan origination rather than ex-post loan performance.
- Evidence provided that loan spreads are strongly correlated with probabilities of borrower default over a one-year horizon (a Basel II-consistent risk parameter) reported by lenders for a subsample of loans.
- Typical loan maturities of several years imply loan spreads capture credit risk smoothed through the cycle.

### Identification challenges and strategies
- Two main identification challenges:
  - Endogeneity of interest rates to economic conditions.
  - Changes in credit risk may reflect both demand (pull) and supply (push) factors.
- Strategies to address these challenges:
  - Restrict sample to loans originated to non-U.S. borrowers to separate U.S. interest rate and credit cycles and treat U.S. monetary conditions as exogenous to other countries.
  - Include variables capturing economic uncertainty and risk appetite to control for global factors.
  - Run loan-level analysis with controls for both credit supply and demand:
    - Lender country- and lender type-year effects to control for lender-specific credit supply cycles.
    - Borrower-type year and borrower-region year fixed effects to control for borrower-specific credit demand cycles.

### Main empirical findings
- Robust evidence that lower U.S. interest rates encourage lenders to extend riskier loans in the global syndicated loan market, with different patterns before and after the global financial crisis:
  - Pre-crisis period:
    - A decline in U.S. short-term interest rates encouraged lenders of various types and nationalities to originate riskier loans to non-U.S. borrowers.
    - After accounting for this effect, movements in long-term interest rates did not have statistically significant effect on risk taking.
    - Given the Federal Reserve explicitly targeted the federal funds rate during this period, results present evidence of a risk taking channel of monetary policy via short-term rates.
    - Footnote: "The Fed was very successful in achieving the target, with the spread between the effective federal rate and its target averaged just a couple of basis points with a low variance."
  - Post-crisis period:
    - Lenders of various types and nationalities tended to originate riskier loans to non-U.S. borrowers in response to a decline in long-term interest rates.
    - Short-term rates were close to zero lower bound during this period, so results are implicitly conditional on short-term rates being low.
    - The U.S. 10-year Treasury rate acted as a de facto target for longer-term rates; effects on risk taking reflect return-on-safer-assets and search-for-yield motives.
- Global factors (e.g., European sovereign spread and economic uncertainty index) played a limited role in explaining ex-ante credit risk of the syndicated loans; effects of interest rates on risk taking remain robust after controlling for these factors.
- Portfolio-level analysis (lender loan portfolios) reinforces loan-level findings.
- Differential responses by lender type:
  - Banks increased risk taking in response to both a drop in short-term rates before the crisis and a decline in long-term rates after the crisis.
  - Nonbanks increased risk taking only when long-term rates decreased in the post-crisis period.

### Contributions to the literature
- Demonstrates international spillovers of monetary policy through the quality (rather than quantity) of cross-border credit flows: low interest rates in one country encourage riskier lending to other countries, posing challenges for central banks, particularly in EMEs.
- Distinguishes roles of short-term and long-term interest rates in the risk taking channel of monetary policy and compares transmission channels before and after the crisis.
- Covers the broad financial system including both banks and nonbanks, highlighting the importance of nonbank institutions given their rapid expansion in the post-crisis period.

### Data and paper organization (overview)
- Data source: Thomson Reuters Loan Pricing Corporation (LPC)’s DealScan, covering 1995–2013, with detailed information on individual syndicated loans, borrower characteristics, and lending syndicate composition.
- Remainder of paper:
  - Section 2: literature review.
  - Section 3: data description and demonstration that loan spreads are reasonable proxies for ex-ante credit risk.
  - Section 4: empirical methodology.
  - Section 5: empirical results and robustness tests.
  - Section 6: conclusions and implications for financial stability.

*Source: IMF Working Paper — "1. Introduction" (wp1716).*

### 3.1  Global syndicated loan market

### 3.1  Global syndicated loan market

### Definition and key features
- A syndicated loan is extended to a borrower by multiple lenders forming a syndicate and is administered by an agent, typically a bank.
- The loan’s interest rate is floating: it equals a contractual loan spread over a benchmark interest rate; the spread is determined at origination, reflects lenders’ judgments about the loan’s credit risk, and stays constant over the loan’s duration unless covenants are violated.
- For a given loan, its interest rate changes as the benchmark interest rate moves, minimizing lenders’ interest rate risk.

### Scope and dataset focus
- Analysis focuses on term loans that are:
  - denominated in U.S. dollars,
  - indexed to the U.S. dollar LIBOR,
  - originated in the global syndicated loan market.
- These loans:
  - are dispersed at origination,
  - usually have a maturity of 5-7 years,
  - are more similar to corporate bonds than credit lines.
- Credit lines are excluded for two stated reasons:
  - lenders of credit lines are predominantly banks,
  - credit lines have complex pricing and drawdowns that are endogenous to the credit cycle and not available in the dataset.

### Market size and composition (origination)
- Originations of syndicated term loans are comparable in size to issuance of nonfinancial corporate bonds.
- Historical dynamics:
  - New loan originations rose rapidly through 2007, collapsed during the crisis, and recovered substantially from the 2009 trough.
  - Originations of term loans reached to about $2 trillion in 2013, of which roughly $700 billion were made to U.S. borrowers and $1.3 trillion to non-U.S. borrowers.
- Syndicated lines of credit (mostly to borrowers in advanced economies) added another $2 trillion to originations in the overall syndicated loan markets.

### Identification of lenders and borrowers; classification
- Lending analyzed on an ultimate counterparty basis:
  - lenders aggregated up to parent organizations—ultimate lenders;
  - locations of parent organizations—ultimate borrowers—used to determine borrower locations.
- Rationale: risk management and borrower funding decisions are carried out at parent-company level; parent companies tend to guarantee subsidiary debt.
- Lender types:
  - Classified by two-digit SIC codes in DealScan.
  - Two types considered: banks and nonbank financial institutions (including finance companies; investment banks; investment funds; structured products; and insurance companies).
  - Nonbank financial lenders have modest ownership shares at origination; grouped together for economic meaningfulness.
- Borrower classification:
  - One-digit SIC codes for primary industries of borrowers (with limited fallbacks to subsidiary identifiers if missing).

### Risk profile and pricing of syndicated term loans
- Risk concentration:
  - Vast majority of syndicated term loans are extended to risky, opaque borrowers.
  - Loans to nonrated borrowers dominated originations; investment-grade and speculative-grade borrowers accounted for only a moderate fraction.
- Pricing:
  - Loan spreads on loans to nonrated borrowers are typically much higher than those on loans to investment-grade borrowers, but somewhat lower than those on loans to speculative-grade borrowers.
  - Risky loans—with contractual loan spread over 125-150 points—account for the bulk of loan originations.
  - Loans extended to nonrated and speculative-grade borrowers tend to be leveraged.
- Geographic composition:
  - Loans extended to borrowers from EMEs accounted for a sizable share at origination.
- Lender-type pricing differences:
  - Nonbank lenders tend to originate riskier loans than banks; contractual loan spreads for new loans made by nonbanks are significantly wider than spreads for loans made by banks.

### Market share and secondary distribution dynamics
- At origination:
  - Banks and bank holding companies (BHCs) accounted for about 80 percent of originations volume globally; the rest originated by nonbank lenders.
- After origination / ownership in seasoned loans:
  - Banks often originate syndicated loans with intent to distribute to other lenders (pre-committed buyers or likely buyers shortly after origination).
  - In the U.S. market, banks and BHCs own only over a third of all term loans in seasoned holdings; nonbank lenders (investment funds, structured finance products, portfolio managers, finance companies, and others) account for the rest.
  - Median of banks’ ownership shares of syndicated loans in the U.S. market declines from about 80 percent to less than 20 percent in just a few weeks after origination.

### Implications for analysis and identification
- Panel structure and identification:
  - Analysis uses the panel structure to assume lenders take U.S. short-term and long-term interest rates as given and make portfolio decisions in response to changes in U.S. dollar-denominated cost of funds and returns on safe assets.
  - Identification is strongest when restricting the panel to non-U.S. lenders extending loans to non-U.S. borrowers (U.S. interest rate changes likelier exogenous to loan credit risk).
  - Panel fixed effects help control for unobserved factors affecting lenders and borrowers.
- Measurement choice:
  - Loan spreads are used as proxies for ex-ante credit risk; rationale and empirical validation of spreads as reliable proxies are presented later (see section 3.3).
  - Acknowledged caveat: changes in interest rates may also cause spreads to change for given credit risk levels; regressions include spreads of U.S. low-grade corporate bonds over the U.S. Treasury rate to control for general movements in risk aversion.

_Italic: Source: wp1716 - 3.1  Global syndicated loan market (IMF PDF chapter)._

### 4.1  Syndicate regressions

### 4.1  Syndicate regressions

### Model specification and key variables
- Benchmark syndicate model (semi-log):
  - log(spreadj,l,b,t) = αl + αb + βF RFt + βT RTt + Xt γ (high-frequency push factors) + θl,c,y + θl,i,y (low-frequency push factors) + θb,c,y + θb,i,y (low-frequency pull factors) + φj,t + εj,l,b,t
- spreadj,l,b,t: all-in-drawn spread for loan (syndicate) j made/reported by lender l to borrower b at time t.
- RFt: federal funds rate.
- RTt: 10-year U.S. Treasury (zero-coupon) interest rate.
- Xt (controls for risk appetite and macroeconomic factors) includes:
  - Merrill Lynch low-grade corporate bond spread (measures risk aversion and credit risk compensation).
  - VIX-derived variance risk premium.
  - European sovereign crisis spread (Italian minus German sovereign yields).
  - Economic uncertainty: U.S.-news based uncertainty index of Baker, Bloom, and Davis (2015).
  - Expected inflation: Michigan consumer survey-based expected 12-month inflation.
- Fixed effects:
  - Lender l fixed effect αl and borrower b fixed effect αb.
  - Lender country-group-year θl,c,y and lender type-year θl,i,y (lender countries assigned to either 3 groups—the United States, non-U.S. advanced economies, and EMEs—or 2 groups; lender types: banks, nonbank financial lenders, nonfinancial lenders).
  - Borrower country-group-year θb,c,y and borrower industry-year θb,i,y (borrower countries assigned to 2 groups: non-U.S. advanced economies and EMEs; borrower industry SIC codes at one digit level).
  - Syndicate characteristics φj,t (e.g., number of lenders in a syndicate).
- Errors εj,l,b,t: white noise; clustered by time.
- Regression frequency: quarterly (loan data available monthly, but quarterly chosen for economic volumes and syndication timing considerations).

### Interpretation and estimation strategy
- βF and βT capture sensitivity of risk taking to changes in U.S. interest rates.
- Because model is semi-log, marginal effect for loan j around a level of loan spreads:
  - ∆(spreadj,...) = ̂spreadj,... βi ∆Ri, i = F, T.
- Estimation approach:
  - Pre-crisis and post-crisis regressions estimated separately due to structural break in interest rate series.
  - Pre-crisis: include both short-term (RF) and long-term (RT) interest rates.
  - Post-crisis: federal funds rate excluded (zero lower bound beginning 2008:Q4); include U.S. Treasury rate only.
  - Extended full-sample model includes RF, RT, interactions with Post2008 dummy, Post2008, Crisist, and other controls.

### Main findings — syndicated regressions (loan-level)
- Pre-crisis period:
  - Significantly negative relationship between short-term interest rates (federal funds rate) and ex-ante risk taking.
  - Movements in long-term rate (10-year Treasury) statistically insignificant in this period.
- Post-crisis period:
  - Significantly negative relationship between long-term interest rate (10-year Treasury) and risk taking.
- Full-sample specification (including crisis quarters):
  - Federal funds rate effect in pre-crisis remains similar; coefficient for long-term rate not statistically significant (likely due to crisis quarters).
- Marginal economic effects (calculated using equation 3 and neighborhood of average spreads):
  - Pre-crisis: one-standard deviation decrease in the federal funds rate of 1.82 percentage point leads to an increase in loan spread by 11 basis points.
  - Post-crisis: one-standard deviation decrease in the 10-year Treasury rate of 0.71 percentage point raises the loan spread by 28 basis points.
  - These imply the economic effect of the long-term rate on risk taking in the post-crisis period appears to be much stronger than that of the short-term rate in the pre-crisis period, despite the smaller standard deviation of the longer-term interest rate.
- Other explanatory variables:
  - Variance risk premium: statistically insignificant across specifications.
  - U.S.-news uncertainty index: negative coefficient and statistically significant in the pre-crisis period (elevated uncertainty moderated lenders’ appetite for riskier loans).
  - Low-grade bond spreads: positive and consistently statistically significant (compression of loan spreads associated with broad improvements in investors’ appetite for term loans).
  - European sovereign yield spread: negative in the post-crisis column covering the European sovereign debt crisis period.

### Sample cuts and robustness
- Loans to non-U.S. borrowers by U.S. lenders:
  - Regression results similar to broad sample.
  - Magnitude of estimated coefficient for 10-year Treasury rate in the post-crisis period is much larger in absolute terms for the U.S. lender sample, suggesting stronger economic effect for U.S. lenders.
- Loans to EME borrowers by non-U.S. lenders:
  - Results close to those for full non-U.S. borrower sample, supporting robustness and cross-border spillover evidence.
- Aggregate summary statistics (reported for non-U.S. borrowers and EME borrowers):
  - Average loan spread for EME borrowers: 1.67 percentage points.
  - Average loan spread for non-U.S. borrowers (includes non-U.S. advanced economies and EMEs): 1.86 percentage points.
  - Average federal funds rate: 3.0 percent; standard deviation: 2.3 percent.
  - 10-year Treasury rate standard deviation: 1.4 percent.
  - Average spread during pre-crisis period: about 100 basis points.
  - Average spread during post-crisis period: about 260 basis points.

### Key implications from syndicated regressions
- Evidence that lower interest rates in the U.S. induce non-U.S. lenders to extend loans to riskier borrowers in the global syndicated term loan market.
- The dominant channel shifted from short-term (pre-crisis) to long-term rates (post-crisis), with stronger economic impact per standard deviation of rate movements in the post-crisis period.

### Connections to portfolio-level results and lender types (summary)
- Portfolio regressions (overview):
  - Confirm differential effects of short-term and long-term interest rates on ex-ante risk taking in pre- and post-crisis periods.
  - Full-sample portfolio model shows the interaction of 10-year Treasury rate and the post-crisis dummy is negative and statistically significant.
- Bank vs nonbank lenders (portfolio level):
  - Banks:
    - Adjust risk portfolios in response to changes in the federal funds rate before the crisis and also to the 10-year Treasury rate after the crisis.
  - Nonbank lenders:
    - Respond only to changes in the 10-year Treasury rate in the post-crisis period.
  - Interpretation: banks’ risk taking driven by both cost-of-fund and return-on-safer-asset concerns; nonbank institutions’ risk taking motivated by return-on-safer-asset or search-for-yield incentives.
  - Policy relevance: rising share of nonbank participants in the global syndicated term loan market may present different financial stability implications and policy challenges.

*Source: wp1716 - 4.1  Syndicate regressions*

### 5.3  Robustness checks and caveats

### 5.3  Robustness checks and caveats

### Robustness checks
- Reestimated regression models 2 to 6 on an immediate counterparty basis; results on an immediate basis are weaker but still consistent with ultimate counterparty results.
- Estimated models on two additional subsamples:
  - Loans extended to nonrated and speculative borrowers (i.e. excluding investment-grade borrowers).
  - Loans originated by the most active lenders (lenders that lend in the market in every quarter).
- Omitting loans to investment-grade borrowers and restricting to loans from the most active lenders produce statistically stronger results.
- Appendix provides further robustness checks.

### Caveats and limitations
- Do not observe lenders’ entire portfolios, which may include a variety of instruments; lenders may hedge exposures to the syndicated term loan market in other financial markets.
  - Additional risk taking in the syndicated loan market as a result of low interest rates may not necessarily represent a general increase in the riskiness of the financial sector if hedges offset exposures.
- Results on the risk taking channel through long-term interest rates in the post-crisis period may be partly due to expansionary monetary policies adopted in other large advanced economies, not just in the United States.
  - These results may be better interpreted as the international spillovers of risk taking associated with low interest rates in general rather than U.S. monetary policy per se.

### Key findings (summary from Conclusions)
- Evidence of a negative relationship between interest rates and ex-ante risk taking in the global syndicated term loan market.
- Differentiation of risk taking channels before and after the global financial crisis:
  - Pre-crisis: as U.S. short-term interest rate declined, lenders of various types and nationalities tended to originate riskier loans to both U.S. and non-U.S. borrowers.
  - Post-crisis: lenders of various types and nationalities tended to originate riskier loans in response to declines in long-term interest rates.
- Results are particularly stronger when restricting the sample to non-U.S. lenders and borrowers from EMEs.
- Evidence of cross-border spillovers: low interest rates in one country encouraging risk taking in other countries.
  - Cross-border spillovers can undermine domestic central bank efforts to affect risk taking and credit cycles (e.g., a domestic central bank tightening may be offset by foreign central banks loosening).
- Policy coordination implications:
  - Central banks may need to rely on coordination of multiple policy tools, such as monetary and macro-prudential policies, to ensure both economic and financial stability (see Rey (2015) and Obstfeld (2015) for further policy discussions related to the global financial cycle).

### Policy implications and unanswered questions
- Increased risk taking by lenders is one channel through which accommodative monetary policy spurs economic activity; greater risk taking may also pose risks to financial stability domestically and globally.
- The paper’s evidence is limited to the global syndicated loan market; lenders may have hedged potential losses from syndicated loan defaults, with hedges potentially transferring losses to other financial intermediaries.
  - In aggregate, the financial system could be strained as a result of additional risk taking.
- The analysis does not resolve whether additional risk taking induced by expansionary monetary policy is excessive; addressing this requires a comprehensive welfare analysis.
- Assessing the trade-off between monetary-policy-induced risk taking that supports economic activity and the risks to financial stability is a high priority for the Federal Reserve and central banks globally.

*Source: wp1716 - 5.3  Robustness checks and caveats*

### References

### References

### Key empirical findings from figures and captions
- Issuance and originations:
  - Originations of syndicated term loans (U.S. dollar-denominated, LIBOR-indexed, global market) are comparable in size to issuance of nonfinancial corporate bonds. (Figure 1 (a))
  - The majority of syndicated term loans are made to speculative-grade and unrated debt borrowers; these loans tend to be leveraged. (Figure 1 (b))
  - Originations of syndicated term loans made to non-U.S. borrowers are used to estimate regression models 2 to 6. (Figure 1 (c))
- Loan pricing and risk:
  - Loan spreads fixed at origination are highly positively correlated with through-the-cycle probabilities of borrower default over a one year horizon. (Figure 4)
- Lender ownership:
  - Banks account for 80 percent of loan ownership at origination. (Figure 3 (a))
  - Median bank share in syndicated loans falls from 90 percent at origination to 20 percent within a month, indicating rapid sell-off to nonbank financial intermediaries. (Figure 3 (b))
- Interest rate environment:
  - Panel (a) of Figure 2 suggests splitting the sample into pre- and post-crisis periods because the federal funds rate reached the zero lower bound in late 2008.

### Descriptive statistics and key regression evidence
- Table 1: Descriptive statistics for loan pricing regressions (sample 2010:Q1-2013:Q4; over 700 term loans)
  - Loan spread, pct: Mean 3.101; Std. dev. 1.633; 10th p. 1.500; 50th p. 2.750; 90th p. 5.000
  - Probability of default, pct: Mean 2.317; Std. dev. 7.110; 10th p. 0.170; 50th p. 0.800; 90th p. 4.820
  - 10-year Treas. rate, pct: Mean 2.346; Std. dev. 0.581; 10th p. 1.710; 50th p. 2.050; 90th p. 3.200
  - Variance risk premium, pct sq.: Mean 14.439; Std. dev. 6.218; 10th p. 7.365; 50th p. 14.034; 90th p. 25.099
- Table 2: Loan spreads as proxies for ex-ante credit risk (based on 709 observations)
  - log(PD, pct): coefficient 0.231 ∗∗∗ (column 1), 0.206 ∗∗∗ (column 2), 0.190 ∗∗∗ (column 3), 0.189 ∗∗∗ (column 4)
  - log(loan amount, $ bill.): −0.030 ∗∗ (column 2), −0.036 ∗∗∗ (column 3), −0.040 ∗∗∗ (column 4)
  - log(duration, years): 0.166 ∗∗∗ (column 1), 0.128 ∗∗∗ (column 2), 0.157 ∗∗∗ (column 3)
  - LBO or takeover loan purpose: 0.223 ∗∗∗ (column 1), 0.151 ∗∗ (column 2), 0.152 ∗∗ (column 3)
  - Public company: −0.195 ∗∗∗ (column 1), −0.178 ∗∗∗ (column 2), −0.175 ∗∗∗ (column 3)
  - R-sq. adj.: 0.40 (col 1), 0.50 (col 2), 0.56 (col 3), 0.57 (col 4); RMSE: 0.37, 0.34, 0.32, 0.31 respectively
  - Note: Dependent variable is log(loan spread, pct); inclusion of loan characteristics improves goodness of fit but log(PD, pct) remains strongly explanatory.
- Table 3: Descriptive statistics for loan portfolio spread regressions (sample 1995:Q1-2013:Q4; 9,980 loans to non-U.S. borrowers; 7,589 to EME borrowers)
  - Loan spread, pct — Non-U.S. borrowers: Mean 1.860; Std. dev. 1.529; 10th p. 0.425; 50th p. 1.400; 90th p. 3.875
  - Loan spread, pct — EME borrowers: Mean 1.667; Std. dev. 1.352; 10th p. 0.425; 50th p. 1.250; 90th p. 3.500
  - Fed. funds rate, pct: Mean 3.010; Std. dev. 2.336; 10th p. 0.119; 50th p. 3.056; 90th p. 5.723
- Syndicate regressions (Loans to non-U.S. borrowers; Tables 4–6; multiple samples and subsamples)
  - Table 4 (all lenders to non-U.S. borrowers):
    - Fed. funds rate, pct: Pre-crisis −0.062 ∗∗ ; Post-crisis −0.075 ∗∗∗
    - 10-year Treas. rate, pct: Post-crisis −0.150 ∗∗∗ (column 2)
    - Low-grade bond spread, pct: 0.099 ∗∗ (pre-crisis), 0.127 ∗∗∗ (post-crisis)
    - News-based uncertainty index: −0.003 ∗∗∗ (pre-crisis)
  - Table 5 (U.S. lenders to non-U.S. borrowers):
    - Fed. funds rate, pct: Pre-crisis −0.065 ∗∗ ; Post-crisis −0.049 ∗∗
    - 10-year Treas. rate (z.c.), pct: Post-crisis −0.261 ∗∗∗
    - Low-grade bond spread, pct: 0.189 ∗∗∗ (pre-crisis); 0.070 ∗∗∗ (full)
    - News-based uncertainty index: −0.004 ∗∗∗ (pre-crisis)
  - Table 6 (non-U.S. lenders to EME borrowers):
    - Fed. funds rate, pct: Pre-crisis −0.068 ∗∗ ; Post-crisis −0.079 ∗∗∗
    - Variance risk premium, pct sq.: 0.005 ∗∗∗ (pre-crisis)
    - Low-grade bond spread, pct: 0.116 ∗∗∗ (pre-crisis); 0.129 ∗∗∗ (post-crisis)
    - Dummy for crisis quarters: −0.183 ∗∗∗
- Portfolio regressions (Tables 7–8)
  - Table 7 (portfolios of loans by all lenders to non-U.S. borrowers):
    - Fed. funds rate, pct: Pre-crisis −0.028 ∗ ; Post-crisis −0.036 ∗∗∗
    - 10-year Treas. rate, pct: Post-crisis −0.219 ∗∗
    - Variance risk premium, pct sq.: 0.002 ∗ (full)
    - News-based uncertainty index: −0.002 ∗∗∗ (full)
    - Post-2008 coefficient: 1.327 ∗∗∗
  - Table 8 (bank vs. nonbank financial lenders):
    - Bank × Fed. funds rate, pct: Pre-crisis −0.029 ∗∗ ; Post-crisis −0.036 ∗∗∗
    - Nonbank × Fed. funds rate, pct: 0.000 (pre-crisis); −0.016 (post-crisis)
    - Bank × 10-year Treas. rate, pct: Pre-crisis 0.044 ∗ ; Post-crisis −0.224 ∗∗
    - Post-2008 coefficient: 1.362 ∗∗∗
  - Notes: Regression specifications include extensive fixed effects (lender type-year, lender country-year, borrower country-year, borrower industry-year where relevant). Errors clustered by quarters.

### Robustness checks (Appendix)
- Counterparty consolidation:
  - Reestimating regression models 2 to 6 on an immediate counterparty basis (no consolidation under parent organizations) yields:
    - Syndicate regressions (Table A1) results very similar to ultimate counterparty basis.
    - Portfolio regressions (Table A2) are weaker than on an ultimate counterparty basis—differences in statistical significance and explanatory power likely because risk management decisions tend to be made at a parent level.
- Sample composition:
  - Results do not necessarily apply to syndicated term loans made to investment-grade debt borrowers; omitting loans to investment-grade borrowers gives statistically stronger results (a falsification test).
  - Results are stronger for most active lenders in the global syndicated market; these lenders are large, sophisticated institutions that lend every quarter, but their number in the sample is modest.
- Selected robustness coefficients (immediate counterparty basis):
  - Table A1 (all lenders to non-U.S. borrowers, immediate basis):
    - Fed. funds rate, pct: Pre-crisis −0.069 ∗∗∗ ; Post-crisis −0.082 ∗∗∗
  - Table A2 (portfolio regressions, immediate basis):
    - Fed. funds rate, pct: Pre-crisis −0.014; Post-crisis −0.030 ∗∗∗
    - 10-year Treas. rate, pct: Pre-crisis 0.065 ∗∗∗ ; Post-crisis −0.215 ∗∗

*Italic: Source — wp1716 - References (PDF chapter/section).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/wp1716.pdf_