## 1.  Sources of Financing for the U.S. Current Account Deficit

## Source details

**Canonical URL:** [1.  Sources of Financing for the U.S. Current Account Deficit](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2007/_wp07177.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2007/_wp07177.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2007/_wp07177.pdf.json)

---

### Financial integration and home bias
- Since 1994, cross-border capital flows increased by more than 500 percent.
- Observed home bias (share of foreign assets in portfolios) diminished in most developed countries and for most classes of investors.
- Contributing factors listed:
  - rapid development of international trade;
  - some reduction in the use of capital controls;
  - development of sophisticated financial instruments for handling currency and other risks;
  - improved marketing of financial products constructed from foreign assets.
- Implication: Greater international financial integration raises the prospect of easier financing for fiscal or current account deficits and increases sensitivity of capital flows to relative conditions (including relative interest rates).

### Focus and approach of the paper
- Focus: bond flows identified as the dominant source of private sector funding for the US current account deficit.
  - Statistic: net bond (including Treasuries, agencies, and corporates) purchases by foreigners accounted for more than 88% of portfolio inflows in the 12 months through October 2006, and more than 100% of the current account deficit during that period.
- Structure:
  - literature review;
  - two-country theoretical model (based largely on Blanchard, Giavazzi, and Sa (2005));
  - empirical estimation of impact on bond purchases of expectations for probable returns on an investment in foreign bonds;
  - examination of successive time spans to assess changes over time;
  - panel VAR and panel instrumental variables approaches to exploit cross-sectional structure and correct for endogeneity (notably spread differentials).

### Literature survey — equity and fixed income flows (summarized)
- Home bias explanations: information asymmetries, incomplete hedging markets, sovereign risk, institutional constraints (capital controls, legal/regulatory restrictions, capital risk-weighting).
- Fixed income literature highlights:
  - Greenspan (2005): reduction in home bias increased sensitivity of inflows to interest differentials and expectations.
  - Bernanke (2005): "global savings glut" lowered long-term real interest rates and increased US current account deficit.
  - Dooley, Folkerts-Landau, and Garber (2003): foreign official bond purchases supporting the dollar.
  - Frey and Moec (2005): foreign central bank buying might have lowered long-term US Treasury yields by up to 115 bps in 2004.
  - Warnock and Warnock (2005): foreign inflows estimated to reduce US long-term bond yields by 150 basis points at study time.
  - Lane and Milesi-Ferretti (2005): no evidence of positive correlation between lagged returns and bond inflows (but do find it for equities).

### Theoretical model (two-country, bond-focused) — key definitions and implications
- Based on Blanchard, Giavazzi, and Sa (2005). Key definitions preserved:
  - US national wealth W_U = X_U - F, where X_U is value of US asset market and F is net debt of the US.
  - Foreign country V: W_V/ε = X_V/ε + F, where ε is exchange rate (units of V currency per dollar).
  - Share of foreign national wealth held in US assets η(R,t) = 1 - σ_V(R,t), increasing in R and t; t denotes degree of home bias of country V (t = 0 implies holdings only V assets).
- Functional form used:
  - η(R,t) = (1 + e^{-1}e^{-Rt})^{-1}
- Derived relationships and equations preserved verbatim in source (including Φ definition and Equations 1–4).
- Model implications:
  - Bond flows are a positive function of GDP growth, of Ṙ, and of ṫ (reductions in home bias).
  - Higher R generally leads to higher bond flows except in specific extreme cases detailed in the source.
  - A reduction in home bias increases sensitivity of bond flows to the uncovered interest parity ratio.
  - Short-run exchange rate effect of increased t: dollar appreciates as outflows to the US increase.
  - Long-run exchange rate effect of increased t: dollar weaker than otherwise because larger F requires a weaker dollar to generate trade surplus to service debt.

### Data sources, custodial adjustment, and panel construction
- Primary flow data: US Treasury International Capital Flows dataset, period 1994–2006.
- Custodial bias: TIC records country of initial purchase, not final beneficial owner; leads to overstatement of flows to financial centers (UK, Cayman Islands, Singapore).
- Adjustment procedure:
  - Match TIC flows to custodial data on stocks of foreign holdings via interpolation so cumulative flows match differences between initial and terminal stocks in custodial intervals.
  - Pre-adjust monthly holdings for transactions, price changes, and exchange rate effects using:
    - A_{i,t} = r_t A_{i,t-1} + NP_{i,t} + (GP_{i,t} - GS_{i,t})(1 - T_i)
  - Transaction cost assumptions:
    - 5 basis points on U.S. Treasury debt;
    - 10 basis points on U.S. agency debt;
    - 25 basis points on U.S. corporate debt.
- Yields and exchange rates: Bloomberg. Macroeconomic data: Fund’s International Financial Statistics database.
- Interest rate choice: long-term interest rates (typically ten-year bonds); exception: China uses short-term rates.
- Panel construction:
  - Period 1: Jan 1995 – Dec 2001 (12-country × 84-month panel = 1008 observations).
  - Period 2: Jan 2002 – Apr 2006 (12 × 42 = 504 observations — adjusted data availability).
  - Twelve counterpart economies: Australia, Brazil, Canada, China, Euroland, Japan, Korea, New Zealand, Norway, Sweden, Switzerland, United Kingdom.

### Summary statistics (as reported)
- Table 1: Period 1 (Jan 1995 – Dec 2001)
  - Bond Flows (% of GDP): Mean 0.43, Min -16.23, Max 16.17, StdDev 2.93, ρ .203, No Obs 1008
  - GDP Growth (% m/m saar): Mean 4.23, Min -75.99, Max 196.08, StdDev 18.58, ρ -.240, No Obs 956
  - Expected Dlr App (% annual): Mean -2.01, Min -20.81, Max 20.94, StdDev 6.09, ρ .867, No Obs 1008
  - ST Interest Rate (i): Mean 7.05, Min 0.00, Max 85.43, StdDev 4.39, ρ .963, No Obs 891
  - LT Interest Rate (l): Mean 6.03, Min 0.78, Max 29.74, StdDev 3.26, ρ .992, No Obs 794
  - ST Spread (i_US - i_F): Mean -1.81, Min -79.43, Max 36.50, StdDev 8.38, ρ .963, No Obs 891
  - LT Spread (l_US - l_F): Mean -0.22, Min -25.15, Max 6.50, StdDev 3.26, ρ .990, No Obs 860
- Table 2: Period 2 (Jan 2002 – April 2006)
  - Bond Flows (% of GDP): Mean 1.62, Min -82.51, Max 42.53, StdDev 6.67, ρ .007, No Obs 648
  - GDP Growth (% m/m saar): Mean 5.22, Min -66.30, Max 373.90, StdDev 24.39, ρ -.233, No Obs 633
  - Expected Dlr Aprec (% annual): Mean -0.92, Min -23.75, Max 37.53, StdDev 5.21, ρ .806, No Obs 624
  - ST Interest Rate (i): Mean 4.50, Min 0.00, Max 26.32, StdDev 4.78, ρ .995, No Obs 648
  - LT Interest Rate (l): Mean 5.72, Min 0.54, Max 37.24, StdDev 5.26, ρ .987, No Obs 606
  - ST Spread (i_US - i_F): Mean -2.32, Min -25.07, Max 5.25, StdDev 4.96, ρ .995, No Obs 648
  - LT Spread (l_US - l_F): Mean -1.31, Min -33.65, Max 4.00, StdDev 5.13, ρ .988, No Obs 648

### Empirical specification, identification, and estimation methods
- Core linearized empirical model (linearization of Equation 2) and final employed equation preserved as:
  - it it it it R Y Y R Y BF ε β β βα ++ ⎟⎟⎠⎞⎜⎜⎝⎛ ++ = ⎟⎠⎞⎜⎝⎛ & 321
  - Equation 5 (as in source): it it itit t tiUS it Y Y eeErr Y BF ε β β βα + ⎟⎟⎠⎞⎜⎜⎝⎛ +−+−+= ⎟⎠⎞⎜⎝⎛ + & 3121 )1/)(()( 
- Exchange rate expectations:
  - Forward prices unsuitable due to covered interest parity.
  - Adaptive expectations coefficients not significantly different from zero.
  - Adopted proxy: averages of private sector forecasts (Consensus Forecasts) where available.
  - Correlation coefficients between interest spread and Consensus expectations:
    - Period 1 (1995-2001): -.1431
    - Period 2 (2002-2006): -.3470
    - Whole Sample (1995-2006): -.2479
- Identification and estimation:
  - Endogeneity concern: spread likely correlated with ε_it because higher bond inflows should lower spreads.
  - Mitigation methods:
    - heteroskedasticity-adjusted two-stage least squares (2SLS) using spread lagged one period as instrument;
    - panel vector autoregressions (panel VAR) to capture lagged impacts.
  - Panel estimators: pooled, fixed effects, and “partial fixed effects” 2SLS (country dummies included only where fixed effects significant, with Canada as implicit omitted intercept).

### Key empirical results — 2SLS, fixed effects, partial fixed effects
- Pooled 2SLS (dependent variable: bond flows as percent of own-country GDP; T-statistics in parentheses)
  - Period 1:
    - Interest rate spread (%) .161 (2.44)
    - Growth (% ann) .006 (0.27)
    - Expected Dollar Appreciation (% annual) .011 (1.88)
    - Additional coefficient .030 (1.68) (preserved as in source)
  - Period 2:
    - Interest rate spread (%) .028 (0.23)
    - Growth (% ann) .096 (3.87)
    - Expected Dollar Appreciation (% annual) .049 (2.14)
    - Additional coefficient .009 (0.21)
  - Period 2 regression: 624 obs; R-squared = .0923; sigma = 6.515
  - Interpretation: significant impact of interest rate spread on bond inflows in Period 2 (pooled), not in Period 1; own-country GDP growth significant and larger in Period 2; exchange rate expectations marginal in Period 1 and weaken in Period 2.
- Fixed effects 2SLS (Table 5)
  - Period 1:
    - Interest rate spread (%) .143 (2.19)
    - Growth (% ann) -.025 (-0.41)
    - Expected Dollar Appreciation .009 (1.56)
    - Additional .027 (1.34)
  - Period 2:
    - Interest rate spread (%) -.013 (-0.10)
    - Growth (% ann) .125 (1.02)
    - Expected Dollar Appreciation .042 (1.90)
    - Additional .003 (0.04)
  - Period 2 regression: 624 obs; R-squared = .1217; sigma = 6.409
  - Note: Period 2 coefficient magnitude on spread rises but t-statistic declines from pooled to fixed effects, likely reflecting multicollinearity with country dummies.
- “Partial fixed effects” 2SLS (Table 6 — country dummies for Canada, China, Korea, Switzerland)
  - Period 2:
    - Lag Bond Flows -.006 (-0.05)
    - Interest rate spread (%) .066 (2.72)
    - Growth (% ann) .043 (2.04)
    - Expected Dollar Appreciation (% ann) .017 (0.42)
  - Chi-squared test that remaining country intercepts are zero: Prob > chi2 = 0.7346 (null not rejected).
  - Interpretation: supports theory that interest rates and own-country growth positively influence bond flows and that average home bias declined from Period 1 to Period 2.

### Panel VAR results and impulse responses
- Panel VARs (two lags) with four endogenous variables: bond flows, GDP growth, interest rate spread, exchange rate expectations; fixed effects removed via Helmert forward mean-differencing; Monte Carlo for confidence intervals.
- Table 7 (Period 2 VAR residual correlations and standard deviations)
  - Correlation w/ Bond Flows:
    - GDP Growth .1797
    - Spread .0115
    - Exp Dollar App .0679
  - Correlation w/ Spread:
    - GDP Growth -.0184
    - Bond Flows .0115
    - Exp Dollar App .3176
  - Correlation w/ Exp Dlr App:
    - GDP Growth .0544
    - Bond Flows .0679
    - Spread .3176
  - Standard deviations:
    - GDP Growth 22.7007
    - Bond Flows 6.7722
    - Spread 0.6826
    - Exp Dollar App 3.2630
- Main VAR findings:
  - Dramatic increase in responsiveness of bond flows to interest rate changes between Periods 1 and 2.
    - Period 1 response to a spread shock: not positive.
    - Period 2 response: positive, statistically significant over the long run, and larger than in 2SLS regressions; response appears persistent.
  - Response to exchange rate appreciation shock: significantly positive in Period 1 (VAR), vanished in Period 2.
- Scaling implication preserved:
  - Based on world output ~ $44 trillion in 2005, of which about $12 trillion produced in the United States, a 1% reduction in the interest rate spread between the United States and the rest of the world would reduce bond inflows to the US by about $53 billion over the course of a year.

### Conclusions, implications, and quantitative summary
- Main conclusions:
  - Reduction in home bias (increase in t) between Period 1 and Period 2 increased the elasticity of substitution between domestic and foreign bonds.
  - This made bond flows more responsive to interest rate differentials over time (notably in Period 2).
  - Own-country GDP growth increases demand for foreign bonds (even when scaled as percent of own-country GDP).
  - Exchange rate expectations (as measured) have limited explanatory power; investors may treat exchange rates as a random walk with no drift.
  - Increased financial integration does not lessen the eventual degree of exchange rate and current account adjustment needed to rectify imbalances, but can delay adjustment.
- Quantitative highlights:
  - Periods analyzed: 1995-2001 (Period 1) and 2002-2005 (Period 2).
  - Empirical pooled non-IV coefficients (from Conclusion section):
    - Period 1 pooled: Lagged Bond Flows .160 (4.65); Interest rate spread (%) .007 (0.22); Own-country growth .011 (1.70); Expected dollar appreciation .029 (1.62)
    - Period 2 pooled: Lagged Bond Flows .028 (0.69); Interest rate spread (%) .098 (1.80); Own-country growth .049 (4.42); Expected dollar appreciation .009 (0.17)
  - Fixed effects non-IV (from Conclusion section):
    - Period 1 fixed effects: Lagged Bond Flows .142 (4.09); Interest rate spread (%) -.015 (-0.22); Own-country Growth .009 (1.36); Expected Dollar Appreciation .027 (1.29); F-Test Prob=.32
    - Period 2 fixed effects: Lagged Bond Flows -.009 (-0.19); Interest rate spread (%) .024 (0.14); Own-country Growth .044 (3.69); Expected Dollar Appreciation .034 (0.44); F-Test Prob=.24
  - Scaling result: a 1% reduction in the interest rate spread against the US implies an estimated reduction in annual US bond inflows of about $53 billion.  Total bond inflows amounted to over $800 billion in 2005 (as cited in the source).

*Source: _wp07177 — IMF Working Paper: "1. Sources of Financing for the U.S. Current Account Deficit" (Introduction, Empirical Results, and Conclusion).*

### 1.  Sources of Financing for the U.S. Current Account Deficit..........................................4

### 1.  Sources of Financing for the U.S. Current Account Deficit

### Major sections in this content unit
- 1.  Sources of Financing for the U.S. Current Account Deficit..........................................4
- 2.  Japan – Holdings of Long-Term U.S. Bond Debt.........................................................9
- 3. Impulse Response Functions........................................................................................18

### Tables included
- Table 1.  Summary Statistics for Period 1 Data (1/95-12/01).....................................................11
- Table 2. Summary Statistics for Period 2 Data (1/02-4/06).......................................................11
- Table 3.  Correlations Between Interest Spreads and Exchange Rate Expectations...................12
- Table 4. Pooled Two-Stage Least Squares Regressions............................................................14
- Table 5. Fixed Effects Two-Stage Least Squares Regressions..................................................15
- Table 6. “Partial Fixed Effects” 2SLS Regressions...................................................................16
- Table 7. Correlation Coefficients and Std Dev of Residuals from Pd 2 VAR...........................17

### Appendix material
- Appendix: Panel Estimates Without Instrumental Variables....................................................................20

*Source: _wp07177 - 1.  Sources of Financing for the U.S. Current Account Deficit..........................................4*

### Introduction

### Introduction

### Financial integration and home bias
- Since 1994, international capital markets have grown severalfold, with an increase in cross-border capital flows of more than 500 percent.
- Over the same period, the incidence of observed home bias, as indicated by the share of foreign assets in portfolios, has diminished in most developed countries and for most classes of investors.
- Factors contributing to greater financial integration cited:
  - rapid development of international trade;
  - some reduction in the use of capital controls;
  - development of sophisticated financial instruments for handling currency and other risks;
  - improved marketing of financial products constructed from foreign assets.
- Greater international financial integration raises the prospect of easier financing for fiscal or current account deficits and increases sensitivity of capital flows to relative conditions (including relative interest rates).

### Focus and approach of the paper
- The paper focuses on bond flows, identified as the dominant source of private sector funding for the US current account deficit.
  - Example statistic: net bond (including Treasuries, agencies, and corporates) purchases by foreigners accounted for more than 88% of portfolio inflows in the 12 months through October 2006, and more than 100% of the current account deficit during that period.
- Structure:
  - brief literature review;
  - presentation of a simple two-country theoretical model (based largely on Blanchard, Giavazzi, and Sa (2005));
  - empirical estimation of the impact on bond purchases of expectations for probable returns on an investment in foreign bonds;
  - examination of successive time spans to assess changes over time;
  - use of panel VAR and panel instrumental variables approaches to exploit cross-sectional structure and correct for endogeneity (notably spread differentials).

### Literature survey — home bias, equity, and fixed income flows
- Home bias: longstanding literature since the early 1970s noting strong investor preference for domestic over foreign assets; common explanations include information asymmetries, incomplete hedging markets, sovereign risk, institutional constraints (capital controls, legal/regulatory restrictions, capital risk-weighting).
- Determinants of equity flows:
  - Portes and Rey (2005): gravity model for cross-border equity flows; geographical distance and proxies for information asymmetries explain substantial variation.
  - Edison and Warnock (2006): capital account liberalization has strong short-term impact on equity inflows to emerging markets and a more attenuated long-term effect.
- Fixed income flows and bond yields:
  - Greenspan (2005): secular reduction in home bias enabled large sustained deficits but increased sensitivity of inflows to interest differentials and expectations.
  - Bernanke (2005): "global savings glut" lowering long-term real interest rates and increasing the US current account deficit.
  - Dooley, Folkerts-Landau, and Garber (2003): tacit “Bretton Woods II” arrangement with foreign official bond purchases supporting the dollar.
  - Frey and Moec (2005): using TIC data on official purchases, long-term US Treasury yields would have been up to 115 bps higher in 2004 without foreign central bank buying.
  - Warnock and Warnock (2005): estimate total impact from foreign inflows on US long-term bond yields of 150 basis points at the time of the study.
  - Lane and Milesi-Ferretti (2005): do not find evidence of positive correlation between lagged returns and bond inflows (but do for equities).

### Theoretical model (two-country, bond-focused)
- Based on Blanchard, Giavazzi, and Sa (2005). Key elements and definitions preserved:
  - US national wealth W_U = X_U - F, where X_U is value of US asset market and F is net debt of the US.
  - Foreign country V: W_V/ε = X_V/ε + F, where ε is exchange rate (units of V currency per dollar).
  - Share of W_U in domestic assets σ_U(R,s) with R as uncovered interest parity ratio and s parameter for US home bias; in the simplified bond-focused model s = 0 and σ_U(R,s) = 1 (US investors hold only US assets).
  - For country V, share of national wealth held in US assets η(R,t) = 1 - σ_V(R,t), η(.) is increasing in R and t; t denotes degree of home bias of country V (t = 0 implies holdings only V assets).
- Functional form used for η(R,t):
  - η(R,t) = (1 + e^{-1}e^{-Rt})^{-1} (as given in the source text).
- Derived relationships (notations preserved):
  - Φ = η/(1-η) = e^{-1}e^{Rt} - e^{-1}
  - Equation 1 (as presented in the source): }()V Rt V Rt XtRRteeXeeeF ⎟⎠⎞⎜⎝⎛++−= •• − • −− • 111 ε
  - Equation 2 (scaled by X_V to proxy by GDP): }V V Rt V V Rt V X X etRRtee X X ee X F • − •• − • − • − ⎟⎠⎞⎜⎝⎛ ++ = 111 ε
  - Equation 3: derivative with respect to R (presented verbatim in the source): }⎟⎟⎟⎠⎞⎜⎜⎜⎝⎛ +++= ••• • − • tRttR X X tee dR X F d t V V Rt V 2 1 ε
  - Equation 4: derivative of Equation 3 with respect to t (presented verbatim in the source).
- Model implications preserved:
  - Bond flows are a positive function of GDP growth, of Ṙ, and of ṫ (reductions in home bias).
  - Higher R generally leads to higher bond flows except in specific extreme cases described (inequality condition preserved as text).
  - A reduction in home bias increases sensitivity of bond flows to the uncovered interest parity ratio.
  - Short-run vs long-run exchange rate effect of increased t: short run — dollar appreciates as outflows to the US increase; long run — dollar weaker than otherwise because larger F requires a weaker dollar to generate trade surplus to service debt.

### Data sources and adjustments
- Primary capital flows data: US Treasury International Capital Flows dataset, period 1994-2006, bilateral flows of bonds and equities between the United States and other jurisdictions.
- Custodial bias issue: TIC records country of initial purchase, not necessarily final beneficial owner; custodial bias leads to overstatement of flows to/from financial centers (UK, Cayman Islands, Singapore).
- Adjustment procedure:
  - Match TIC-based flows to less frequent custodial data on stocks of foreign holdings of US assets via interpolation so cumulative flows match differences between initial and terminal stocks in custodial intervals.
  - Before custodial adjustment, adjust monthly holdings for transactions, price changes, and exchange rate effects using formula (as in Warnock (2005)):
    - A_{i,t} = r_t A_{i,t-1} + NP_{i,t} + (GP_{i,t} - GS_{i,t})(1 - T_i)
  - Transaction cost assumptions:
    - 5 basis points on U.S. Treasury debt;
    - 10 basis points on U.S. agency debt;
    - 25 basis points on U.S. corporate debt.
- Data on yields and exchange rates: Bloomberg. Macroeconomic data: Fund’s International Financial Statistics database.
- Interest rate choice: long-term interest rates (typically ten-year bonds); exception: China uses short-term rates (10-year bond yields unavailable).
- Panel construction:
  - Period division: Period 1 Jan 1995 – Dec 2001 (12-country × 84-month panel = 1008 observations).
  - Period 2 Jan 2002 – Apr 2006 (12 × 42 = 504 observations — adjusted data availability).
  - Twelve counterpart economies: Australia, Brazil, Canada, China, Euroland, Japan, Korea, New Zealand, Norway, Sweden, Switzerland, United Kingdom.
- Summary statistics (preserved exactly from tables):
  - Table 1: Period 1 (Jan 1995 – Dec 2001)
    - Bond Flows (% of GDP): Mean 0.43, Min -16.23, Max 16.17, StdDev 2.93, ρ .203, No Obs 1008
    - GDP Growth (% m/m saar): Mean 4.23, Min -75.99, Max 196.08, StdDev 18.58, ρ -.240, No Obs 956
    - Expected Dlr App (% annual): Mean -2.01, Min -20.81, Max 20.94, StdDev 6.09, ρ .867, No Obs 1008
    - ST Interest Rate (i): Mean 7.05, Min 0.00, Max 85.43, StdDev 4.39, ρ .963, No Obs 891
    - LT Interest Rate (l): Mean 6.03, Min 0.78, Max 29.74, StdDev 3.26, ρ .992, No Obs 794
    - ST Spread (i_US - i_F): Mean -1.81, Min -79.43, Max 36.50, StdDev 8.38, ρ .963, No Obs 891
    - LT Spread (l_US - l_F): Mean -0.22, Min -25.15, Max 6.50, StdDev 3.26, ρ .990, No Obs 860
  - Table 2: Period 2 (Jan 2002 – April 2006)
    - Bond Flows (% of GDP): Mean 1.62, Min -82.51, Max 42.53, StdDev 6.67, ρ .007, No Obs 648
    - GDP Growth (% m/m saar): Mean 5.22, Min -66.30, Max 373.90, StdDev 24.39, ρ -.233, No Obs 633
    - Expected Dlr Aprec (% annual): Mean -0.92, Min -23.75, Max 37.53, StdDev 5.21, ρ .806, No Obs 624
    - ST Interest Rate (i): Mean 4.50, Min 0.00, Max 26.32, StdDev 4.78, ρ .995, No Obs 648
    - LT Interest Rate (l): Mean 5.72, Min 0.54, Max 37.24, StdDev 5.26, ρ .987, No Obs 606
    - ST Spread (i_US - i_F): Mean -2.32, Min -25.07, Max 5.25, StdDev 4.96, ρ .995, No Obs 648
    - LT Spread (l_US - l_F): Mean -1.31, Min -33.65, Max 4.00, StdDev 5.13, ρ .988, No Obs 648

### Empirical specification and measurement choices
- Linearized empirical model (core linear model as used):
  - it it it it R Y Y R Y BF ε β β βα ++ ⎟⎟⎠⎞⎜⎜⎝⎛ ++ = ⎟⎠⎞⎜⎝⎛ & 321  (as linearization of Equation 2; βs are functions of the home bias parameter)
- Final empirical equation employed (Equation 5, preserved as in source):
  - Equation 5: it it itit t tiUS it Y Y eeErr Y BF ε β β βα + ⎟⎟⎠⎞⎜⎜⎝⎛ +−+−+= ⎟⎠⎞⎜⎝⎛ + & 3121 )1/)(()( 
- Exchange rate expectations:
  - Forward prices unsuitable (covered interest parity makes R identically equal to 1).
  - Adaptive expectations (most recent change) yielded coefficients not significantly different from zero.
  - Adopted proxy: averages of private sector forecasts of exchange rate movements (Consensus Forecasts) where available.
  - Correlation coefficients between interest spread and Consensus expectations:
    - Period 1 (1995-2001): -.1431
    - Period 2 (2002-2006): -.3470
    - Whole Sample (1995-2006): -.2479

### Identification issues and estimation methods
- Endogeneity concern: spread (r_US - r_i) likely correlated with error term ε_it because higher bond inflows should lower spreads.
- Two approaches to mitigate identification problem:
  - heteroskedasticity-adjusted two-stage least squares (2SLS) estimator using the spread lagged one period as an instrument;
  - panel vector autoregressions (panel VAR) to capture possible lagged impacts.
- Panel estimation choices:
  - pooled estimator: consistent if explanatory variables uncorrelated with country-specific fixed effects;
  - fixed effects (“within”) estimator: consistent even if explanatory variables correlate with fixed effects but can be imprecise when cross-country variation dominates.
  - “Partial fixed effects” 2SLS: include country dummies only where fixed effects are significant (t-statistic > 1.80), leaving Canada as implicit omitted intercept in reported specification.

### Key empirical results (instrumental variables and panel VAR)
- Pooled 2SLS results (Table 4, dependent variable: bond flows as percent of own-country GDP; T-statistics in parentheses)
  - Period 1:
    - Interest rate spread (%) .161 (2.44)
    - Growth (% ann) .006 (0.27)
    - Expected Dollar Appreciation (% annual) .011 (1.88)
    - Additional coefficient .030 (1.68) noted in table (context preserved as in source)
  - Period 2:
    - Interest rate spread (%) .028 (0.23)
    - Growth (% ann) .096 (3.87)
    - Expected Dollar Appreciation (% annual) .049 (2.14)
    - Additional coefficient .009 (0.21) noted in table (context preserved)
  - Period 2 regression: 624 obs; R-squared = .0923; sigma = 6.515
- Interpretation of pooled 2SLS:
  - Statistically significant and substantial impact of interest rate spread on bond inflows in Period 2, not in Period 1 — consistent with increased elasticity of substitution and reduced home bias over time.
  - Own-country GDP growth significant in both periods and increases in magnitude from Period 1 to Period 2.
  - Exchange rate expectations have only marginal impact in Period 1 and almost vanish in Period 2.
- Fixed effects 2SLS results (Table 5):
  - Period 1:
    - Interest rate spread (%) .143 (2.19)
    - Growth (% ann) -.025 (-0.41)
    - Expected Dollar Appreciation .009 (1.56)
    - Additional .027 (1.34)
  - Period 2:
    - Interest rate spread (%) -.013 (-0.10)
    - Growth (% ann) .125 (1.02)
    - Expected Dollar Appreciation .042 (1.90)
    - Additional .003 (0.04)
  - Period 2 regression: 624 obs; R-squared = .1217; sigma = 6.409
  - Note: Period 2 estimated coefficient on interest rate spread rises in magnitude but t-statistic declines from 3.87 (pooled) to 1.02 (fixed effects), likely reflecting multicollinearity with country dummies.
- “Partial fixed effects” 2SLS (Table 6, country dummies used for Canada, China, Korea, and Switzerland):
  - Period 2 results:
    - Lag Bond Flows -.006 (-0.05)
    - Interest rate spread (%) .066 (2.72)
    - Growth (% ann) .043 (2.04)
    - Expected Dollar Appreciation (% ann) .017 (0.42)
  - Chi-squared test of maintained hypothesis that remaining country intercepts are zero: Prob > chi2 = 0.7346 (null not rejected).
  - These results support theoretical priors that interest rates and own-country growth positively influence bond flows and that average home bias declined from Period 1 to Period 2.
- Panel VAR results and impulse responses:
  - Panel VARs estimated for Period 1 and Period 2 with two lags; four endogenous variables: bond flows, GDP growth, interest rate spread, exchange rate expectations; fixed effects handled via Helmert forward mean-differencing; Monte Carlo used for confidence intervals.
  - Table 7 (Period 2 VAR residual correlations and standard deviations — preserved values):
    - Correlation w/ Bond Flows:
      - GDP Growth .1797
      - Spread .0115
      - Exp Dollar App .0679
    - Correlation w/ Spread:
      - GDP Growth -.0184
      - Bond Flows .0115
      - Exp Dollar App .3176
    - Correlation w/ Exp Dlr App:
      - GDP Growth .0544
      - Bond Flows .0679
      - Spread .3176
    - Standard deviations:
      - GDP Growth 22.7007
      - Bond Flows 6.7722
      - Spread 0.6826
      - Exp Dollar App 3.2630
  - Main VAR finding: dramatic increase in responsiveness of bond flows to interest rate changes between Periods 1 and 2.
    - Period 1 response to a spread shock: not positive.
    - Period 2 response: positive, statistically significant over the long run, and larger than in 2SLS regressions; response appears persistent (consistent with theory).
    - Response to exchange rate appreciation shock: significantly positive in Period 1 (in VAR), but vanished in Period 2.
  - Scaling implication preserved from source:
    - Based on world output ~ $44 trillion in 2005, of which about $12 trillion produced in the United States, the results suggest that a 1% reduction in the interest rate spread between the United States and the rest of the world would reduce bond inflows to the US by about $53 billion over the course of a year.

### Interpretation and overarching conclusions (as presented)
- Empirical evidence consistent with theoretical model implications:
  - Reduction in home bias (increase in t) between Period 1 and Period 2 increased the elasticity of substitution between domestic and foreign bonds.
  - This reduction in home bias made bond flows more responsive to interest rate differentials over time (notably in Period 2).
  - Own-country GDP growth increases demand for foreign bonds (even scaled as percent of own-country GDP).
  - Exchange rate expectations, as measured by available proxies (Consensus Forecasts or adaptive), show limited explanatory power in these estimations; possible reasons include inadequacies of expectation measures or investor belief that short-term exchange rate movements follow a random walk with no drift.
- Quantitative scale result: a 1% reduction in the interest rate spread against the US implies an estimated reduction in annual US bond inflows of about $53 billion (based on scaling to global GDP structure in 2005).

*Source: IMF working paper — Introduction (content unit: _wp07177 - Introduction).*

### Conclusion

### Conclusion

### Main findings
- Financial globalization and the decline in home bias have coincided with rising cross-border capital flows and greater foreign willingness to finance global imbalances.  
- Attempts to quantify the impact of declining home bias on fiscal policy and global imbalances have been limited; this paper addresses that gap by estimating parameters of the demand function for US bonds using cross-country time series on bond inflows.  
- Using two separate statistical approaches, the impact of interest rate spreads on bond flows into the United States was statistically significant in the 2002-2005 period, whereas it was not in the 1995-2001 period.  
- The elasticity of substitution between foreign and US bonds has increased, supporting the view that international financial integration has made it easier for nations to sustain larger current account deficits, while increasing the potential consequences of shifts in spreads for global financial stability.

### Statistical results and estimates
- Periods analyzed: 1995-2001 (Period 1) and 2002-2005 (Period 2).  
- Exchange rate expectations (proxied by Consensus Forecasts) predict capital flows in Period 1 but the effect vanishes in Period 2.  
- Domestic financial market expansion, proxied by GDP growth, is marginally significant in Period 1 and clearly significant in Period 2.  
- According to the higher of the two statistical estimates, a reduction in the average spread of the US interest rate over foreign interest rates of 1% would reduce bond inflows to the US by about $53 billion a year, out of total bond inflows that amounted to over $800 billion in 2005.

### Empirical coefficients (panel estimates without instrumental variables)
- Pooled regressions (Dependent variable: bond flows as percent of own-country GDP; T-Statistics in parentheses)
  - Period 1 (1995-2001):
    - Lagged Bond Flows: .160  (4.65)
    - Interest rate spread (%): .007 (0.22)
    - Own-country growth (% ann): .011 (1.70)
    - Expected dollar appreciation (% ann): .029 (1.62)
  - Period 2 (2002-2005):
    - Lagged Bond Flows: .028  (0.69)
    - Interest rate spread (%): .098 (1.80)
    - Own-country growth (% ann): .049 (4.42)
    - Expected dollar appreciation (% ann): .009 (0.17)

- Fixed effects regressions (Dependent variable: bond flows as percent of own-country GDP; T-Statistics in parentheses)
  - Period 1 (1995-2001):
    - Lagged Bond Flows: .142  (4.09)
    - Interest rate spread (%): -.015 (-0.22)
    - Own-country Growth (% ann): .009 (1.36)
    - Expected Dollar Appreciation (% ann): .027  (1.29)
    - F-Test that all ηi=0: Prob=.32
  - Period 2 (2002-2005):
    - Lagged Bond Flows: -.009 (-0.19)
    - Interest rate spread (%): .024  (0.14)
    - Own-country Growth (% ann): .044 (3.69)
    - Expected Dollar Appreciation (% ann): .034  (0.44)
    - F-Test that all ηi=0: Prob=.24

- Note: For the non-IV regressions, the F-test indicates that the null hypothesis that all country intercepts are zero and equal to each other cannot be rejected.

### Implications for carry trades, exchange-rate behavior, and adjustment dynamics
- The results provide insight into cross-border “carry trades”: investors borrowing in low-rate currencies to invest in higher-yielding currencies while taking on exchange rate risk.  
- The theoretical model implies capital movements should become more sensitive over time to interest rate differentials; empirical results confirm that relative interest rates have become a stronger factor behind capital movements.  
- There is little evidence that investors act as though they expect uncovered interest parity to hold. The Period 2 results are consistent with investors treating the trajectory of exchange rates as a random walk with no drift (though this is not the only logical interpretation).  
- If the theoretical model cited is correct, increased financial integration does not lessen the eventual degree of adjustment of exchange rates and current accounts needed to rectify imbalances, but only delays this adjustment.

*Source: _wp07177 - Conclusion*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2007/_wp07177.pdf_