## wp1756

## Source details

**Canonical URL:** [wp1756](https://www.imf.org/-/media/files/publications/wp/2017/wp1756.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2017/wp1756.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2017/wp1756.pdf.json)

---

### Introduction: scope and approach
- Studies domestic and cross-border spillover effects of unconventional monetary and exchange rate policies.
- Combines three empirical approaches:
  - Annual cross-country regressions linking policies to current accounts.
  - Daily regressions exploring spillovers from US economic and monetary policy announcements to international financial assets.
  - A stylized macroeconomic model with imperfect asset substitution to rationalize empirical findings.
- Sample period for annual regressions: 1985 through 2014.
- Cross-country panel: up to 2,088 observations for 141 countries.

### Main empirical findings (annual panel)
- Net official flows (NOF), proxied by purchases of foreign exchange reserves and SWF assets:
  - Large and robust positive effect on a country’s current account when capital mobility is low.
  - Moderate effect when capital mobility is high.
- Quantitative easing (QE), proxied by central bank acquisitions of domestic assets:
  - Small but significant effect on a country’s current account when capital mobility is low.
  - Effect declines to near zero when capital mobility is high.
- Spillovers:
  - Unconventional policy actions in one country affect current account balances of other countries.
  - Stronger effects for countries more integrated with global financial markets.
- Fiscal policy:
  - Large effect on the current account when capital mobility is high (noted as a new result in this literature).

### Empirical specification, data, and key variables
- Dependent variables:
  - CAX: current account excluding net investment income.
  - NPFX: net private flows excluding net investment income.
- Key regressors:
  - NOF: net official flows (acquisition/disposition of assets and liabilities denominated in foreign currency by public-sector institutions).
  - NOA: stock of net official assets.
  - MOB: Aizenman, Chinn, and Ito (2015) measure of legal restrictions on capital mobility, normalized to [0-1] (higher value = fewer restrictions).
- Auxiliary variables (AUX) include:
  - MOB.
  - Lagged PPP GDP per capita relative to the United States.
  - 10-year forward change in the old-age dependency ratio.
  - Lagged real GDP growth rate over the previous 5 years.
  - Net energy exports relative to GDP.
  - Cyclically adjusted fiscal balance relative to GDP.
- Econometric practice:
  - Coefficient standard errors robust to heteroskedastic and first-order autoregressive errors.
  - New instruments developed to identify exogenous variation in NOF and address endogeneity.

### Instrumentation, endogeneity, and identification
- Key endogeneity concern: NOF may respond endogenously to shocks to current account balances and net private flows.
- Instruments used:
  - Incidence of a financial or currency crisis in the previous three years.
  - Portion of NOF that is not related to foreign exchange reserves (captures SWF-related asset flows and development loans).
- Instrument diagnostics:
  - One of the two non-reserves flow instruments is significant in first-stage regressions; the two crisis instruments are sometimes individually and always jointly significant.
  - F-test statistic values are significantly larger than 10.
  - Angrist-Pischke first-stage chi-squared statistic rejects null that NOF is unidentified.
  - First-stage R2 values are around 0.6.
- Modeling choices:
  - MOB is lagged in all regressions, including in interaction terms.
  - Interactions between all auxiliary variables and MOB are included.
  - QE is cyclically adjusted and lagged to attenuate endogeneity biases.

### QE definition, potential biases, and measurement
- QE measure: increase in central bank domestic assets (captures central bank balance-sheet expansion that takes risk off domestic market participants).
- Potential biases:
  - Negative bias if monetary expansions combat declines in growth associated with declining current account balances.
  - Positive bias from omitted variables if a negative shock to domestic demand simultaneously causes QE expansion and import declines.
- Mitigation: cyclically adjust QE and use lagged values; adjusted QE is the residual from a regression of the change in central bank domestic assets on the level and change of the output gap, the change in nominal GDP, and NOF.

### SPILL variable and allocation hypotheses
- Six spillover allocation hypotheses tested (based on aggregate global NOF divided by world GDP):
  1. Financial integration based on cross-border financial transactions.
  2. Capital mobility (MOB).
  3. Financial development based on Sahay et al. (2015).
  4. Reserve currency shares (IMF COFER database).
  5. Economic size (nominal GDP).
  6. Country’s stage of economic development (PPP GDP per capita).
- Final SPILL term: financial integration measure multiplied by aggregate global NOF divided by world GDP.
- Empirical result: only one spillover term has a statistically significant coefficient; median financial integration measure = 0.1, so a coefficient of about -20 on SPILL implies that moving from the median to twice the median lowers a country’s current account by twice the value of world NOF divided by world GDP, or around 2 percent of GDP.
  - Note: In 2013, the ratio of world NOF to world GDP was about 1 percent.
  - Around 90 percent of observations of the financial integration measure take values less than twice the median value.

### Baseline regression results (two-stage least squares)
- Sample is 2.6 times bigger than Bayoumi, Gagnon, and Saborowski (2015), adding four years and many low-income countries.
- First-stage: instruments relevant (see diagnostics above).
- Key second-stage quantitative results (selected):
  - Estimated effect of NOF on current account when capital mobility is lowest (α1):
    - 0.7 in column 1 (CAX).
    - 0.8 in column 2 (NPFX).
  - Overall effect of NOF when mobility is highest = α1 + α2:
    - 0.06 in column 1.
    - 0.52 in column 2.
  - Effect of NOA on current account: close to zero under low mobility and rises to around 3 percent under high mobility.
  - Fiscal balance coefficient when capital is highly mobile: 0.6.
- Inclusion of interactions between control variables and MOB leaves NOF and lagged NOA coefficients broadly unaffected.
- Full baseline with QE, QE*MOB interaction, and SPILL: coefficients on NOF and NOA remain stable.

### QE coefficient and interpretation (baseline)
- QE variable is significant with a coefficient close to 0.25.
- QE interaction with MOB: marginally significant negative coefficient of about the same magnitude (≈0.25), implying:
  - QE has a small but significant effect on the current account when capital mobility is low.
  - QE effect is close to zero as capital mobility approaches its upper bound.

### Robustness checks and influential-observation analysis
- Robustness (high-level):
  - OLS, alternative instruments, GDP-weighting, robust regressions, splits by exchange rate regime and trade openness — NOF coefficients remain positive and significant across many specifications.
  - QE and QE*MOB broadly similar to baseline; QE appears to have little or no impact on current account when capital is highly mobile in weighted regressions.
  - SPILL significance varies across robustness checks; often negative and significant but sometimes becomes no longer significant.
- Influential observations for NOF:
  - Azerbaijan (2008, 2010, 2011); Republic of Congo (1994).
  - Kuwait 1991 dropped due to scale.
  - Dropping four most positive and four most negative influential observations produced essentially no change in QE coefficient — QE remained statistically significant.

### Fiscal balances, net official flows, and energy exporters
- Governments’ decisions to save or not save energy revenues abroad via NOF drive current account outcomes; net energy exports per se have little direct effect once NOF and fiscal balance are controlled.
- Examples:
  - Norway and Saudi Arabia: fiscal balance almost identical to NOF; sum of NOF and fiscal coefficients ≈ 0.8 with high mobility and ≈ 0.9 with low mobility.
  - Algeria: lowest capital mobility among examples; current account closer to NOF than fiscal balance.

### Dynamics and interpretation
- Exchange rate effects of intervention expected to be fast; exchange rate-to-trade/current-account effects typically gradual (~two years).
- Annual data coefficients capture long-run effects rather than immediate effects.
- Residual first-order autocorrelation in baseline regression ≈ 0.7.
- Oil exports have significant temporary effects, boosting current account and SWF flows contemporaneously.

### Daily regression findings (US announcements and high-frequency spillovers)
- Sample: November 2008 to July 2015.
- Benchmark regression: ΔX_it = α_i + (β1,i + β2,i D_FOMC,t) Δy_t + u_it
  - X vector: (1) 10-year sovereign local-currency bond yield, (2) log stock market index, (3) log exchange rate (dollars per unit of non-US currency).
  - Δy_t = change in 10-year US Treasury yield.
  - D_FOMC,t = dummy for FOMC statement/minutes or FOMC Chair speech; separate regressions use dummies for US nonfarm payrolls or ISM PMI releases.
- Key daily-data findings:
  - Sovereign yields: β1 positive and significant for almost all advanced economies and many emerging markets; β2 on FOMC days usually has opposite sign to β1, reducing total co-movement but often leaving β1 + β2 significant.
  - Stock prices: near-universal positive and significant response to US bond-yield rises (β1 positive); β2 (FOMC-day additional effect) almost always negative and significant, reducing spillovers on announcement days; β1 + β2 often remains significant.
  - Exchange rates: most currencies depreciate against the dollar when US yields rise; typical magnitude: one percentage point increase in US bond yields typically causes a one to three percent depreciation in most currencies (noting a one percentage point daily move is much larger than typical daily moves).
    - Pegged currencies show smaller effects; safe-haven currencies can appreciate.
  - Persistence: spillover effects for yields and stocks are persistent; spillovers reduced around monetary policy announcement days.
  - Pre-QE sample (January 2001–November 2008): similar patterns; co-movements on non-announcement days are larger after 2008 than before.

### Daily-regression cross-country determinants and selected betas
- Country characteristics considered: MOB; bank assets to GDP; exchange rate volatility; sovereign risk; exports to the United States as share of GDP.
- Main panel findings:
  - Sovereign yields co-move more with US Treasury yields in economies with high MOB or deeper financial systems (β12 positive and significant).
  - Sovereign yields in countries with higher sovereign risk co-move less on non-FOMC days but have higher additional spillovers on FOMC days.
  - Bond yields in economies with more flexible currencies co-move less with US yields.
  - Trade linkages with the US are positive and significant in explaining sovereign-yield co-movement.
  - Stocks: co-movement larger in high MOB, deeper financial systems, and more flexible-exchange-rate economies; trade linkages counterintuitively reduce stock spillovers.
  - Exchange rates: country factors have limited power to explain additional effects on FOMC/data days.
- Selected country betas (examples):
  - Sovereign yields: Canada β1 = 0.613***; β2 (FOMC) = -0.061**; β1 + β2 (FOM) = 0.539***.
  - Sovereign yields: United Kingdom β1 = 0.457***; β2 (FOMC) = -0.286***; β1 + β2 (FOM) = 0.179***.
  - Stocks: Belgium β1 = 0.101***; β2 (FOMC) = -0.055**; β1 + β2 (FOM) = 0.037***.
  - Exchange rates: Japan β1 = 0.033***; β2 (FOMC) = -0.002; β1 + β2 (FOM) = 0.033***.
  - Notable negatives: Greece sovereign-yield β1 = -0.672***; Russia sovereign-yield β1 = -0.164*.

### Stylized model (imperfect asset substitution): setup and parameters
- Framework: small open-economy macro model with imperfect asset substitution; ignores inflation; variables expressed as deviations from steady state; period ≥ one year.
- Key endogenous variables: BS, FBS, BL, FBL, CA, DD, Y, RS, RL, E, CBS, MB.
- Key exogenous variables: RSF, RLF, QE, NOF, XS, XL, and shocks.
- Units:
  - Interest rates in percentage points.
  - Exchange rate in percent (E increase = appreciation).
  - Other variables as proportion of GDP (in percent).
- Core behavioral and identity equations provided (short- and long-term bond demand, CA, IS, monetary policy, market identities).
- Baseline parameter values:
  - Current account parameters (d1, d2) = (0.25, 0.25).
  - Absorption parameters (e1, e2) baseline = (0.50, 0.50); alternatives include (0.25, 0.25), (0.25, 0.50), (1.0, 1.0).
  - Monetary parameter f1 = 1; alternative f1 = 2.
  - ax parameters: a1=a2=a3=a4 baseline = 0.1; alternatives = 0.01 and 1.
  - bx parameters baseline = 1; alternatives = 0.1 and 10.
  - cx parameters c2=c4 baseline = 0.1 for low capital mobility and 1 for high capital mobility; alternatives = 0.01 and 10.

### Model implications — NOF, QE, and spillovers
- NOF effects:
  - NOF increases CA in both flexible and fixed exchange rate regimes.
  - NOF effect larger when capital mobility is low.
  - Income effects: NOF increases income under flexible exchange rates; NOF decreases income under fixed exchange rates.
  - Relation to empirical magnitudes:
    - At the lowest level of capital mobility, each dollar of NOF raises CA about 75 cents.
    - At the highest level of mobility, effect declines to around 20 cents.
    - Lagged NOA increases CA by about 4 cents on the dollar when capital mobility is high.
- QE effects:
  - QE has small and ambiguously signed effect on CA under flexible rates; negative effect when exchange rate fixed via NOF.
  - Two channels under flexible rate:
    - Channel 1: QE lowers long-term rates → depreciates exchange rate → raises CA.
    - Channel 2: QE lowers long-term rates → increases domestic demand → lowers CA.
  - When exchange rate fixed via RS, RS rises to offset QE stimulus.
  - Annual regressions consistent: QE has no significant effect on CA when capital mobility is high; modest positive impact when mobility is low.
- Spillovers:
  - Foreign NOF shocks can be offset by domestic NOF; when not offset, foreign NOF effects are equal in magnitude and opposite in direction to domestic NOF.
  - QE and conventional monetary policy operate through different rate channels (RSF and RLF); QE operates primarily through RLF when RSF near zero lower bound.

### Relation between model and daily regressions
- Interpretation of shocks:
  - Non-FOMC-day rises in US bond yields treated as revisions to projected US economic activity (“good news”) → increase RLF and UCA.
  - FOMC-day bond yield changes treated as news about unanticipated UMP changes.
- Model predictions consistent with daily results:
  - Good-news foreign shock raises foreign bond yields and income; daily β1 coefficients confirm bond yield and stock responses.
  - Tighter-than-expected foreign UMP (FOMC days) implies smaller or opposite effects on bond yields and stocks; daily β2 coefficients confirm smaller or opposite signs.
  - Exception: model predicts negative additional effect of UMP shocks on foreign exchange rates; daily regressions do not find clear additional negative effect on exchange rates.

### Policy implications
- Many foreign central banks often follow US monetary policy, producing large co-movements in bond yields and small exchange-rate responses.
- Policy toolkit in world of imperfect asset substitution:
  - NOF can neutralize effects of foreign capital inflows into short-term bonds.
  - NOF can insulate a country from shocks to foreign short-term interest rates if chosen as policy instrument.
  - Combination of NOF and QE can fully offset effects of foreign inflows into long-term bonds and changes in foreign long-term rates.
  - These tools allow pursuit of objectives on exchange rates, current accounts, or bond yields while preserving monetary policy for income stabilization.
- Caveats and constraints:
  - Central banks may be uncomfortable taking large net positions in long-term bonds and foreign exchange.
  - Political constraints may limit full use of fiscal policy, capital controls, and macro-prudential measures.
  - Model is linear and simple; cannot fully assess benefits, costs, and risks at scale.

### Conclusions — principal empirical and theoretical findings
- NOF findings:
  - Large direct effect on a country’s CA when capital mobility is low; effect diminishes as mobility rises but remains significant at high mobility.
  - Important additional effect through lagged stock of net official assets.
  - Quantitative examples:
    - At lowest capital mobility, $1 of NOF → ~75 cents increase in CA.
    - At highest mobility, ~20 cents.
    - NOA adds ~4 cents on the dollar when mobility high.
- QE findings:
  - QE has no significant effect on CA when capital mobility is high; modest but significant positive impact when capital mobility is low.
- Spillovers:
  - Official flows spill over to other countries proportionally to international financial integration.
  - US economic developments raise US bond yields, raise foreign bond yields, increase foreign stock prices, and depreciate foreign currencies.
  - Spillovers often greater on foreign bond yields than on foreign exchange rates — suggesting many foreign central banks follow the Fed to stabilize exchange rates.
  - Pure monetary shocks have small and ambiguous effects on CA and activity for countries with high capital mobility.
- Policy message:
  - Foreign exchange intervention (NOF) and QE at home can be useful responses to foreign policies of similar type; negative spillovers on CA can be fully offset by increasing domestic NOF; spillovers onto bond yields can be offset by countervailing QE.
  - Practical and political constraints may limit full use of these tools; model provides framework to compare options.

*Source: IMF Working Paper (wp1756), Appendix Tables and Sections summarized above.*

### Appendix Tables

### Appendix Tables

### Introduction: scope and approach
- Studies domestic and cross-border spillover effects of unconventional monetary and exchange rate policies.
- Combines three empirical approaches:
  - Annual cross-country regressions linking policies to current accounts.
  - Daily regressions exploring spillovers from US economic and monetary policy announcements to international financial assets.
  - A stylized macroeconomic model with imperfect asset substitution to rationalize empirical findings.
- Sample period for annual regressions: 1985 through 2014.
- Cross-country panel: up to 2,088 observations for 141 countries.

### Main empirical findings (annual panel)
- Net official flows (NOF), proxied by purchases of foreign exchange reserves and SWF assets, have:
  - A large and robust positive effect on a country’s current account when capital mobility is low.
  - A moderate effect when capital mobility is high.
- Quantitative easing (QE), proxied by central bank acquisitions of domestic assets, has:
  - A small but significant effect on a country’s current account when capital mobility is low.
  - An effect that declines to near zero when capital mobility is high.
- Spillovers: unconventional policy actions in one country affect current account balances of other countries, with stronger effects for countries more integrated with global financial markets.
- Fiscal policy: a large effect on the current account when capital mobility is high (noted as a new result in this literature).

### Empirical specification and data details
- Dependent variables:
  - CAX: current account excluding net investment income.
  - NPFX: net private flows excluding net investment income.
- Key regressors:
  - NOF: net official flows (acquisition/disposition of assets and liabilities denominated in foreign currency by public-sector institutions).
  - NOA: stock of net official assets.
  - MOB: Aizenman, Chinn, and Ito (2015) measure of legal restrictions on capital mobility, normalized to [0-1] (higher value = fewer restrictions).
- Auxiliary variables (AUX) include:
  - MOB.
  - Lagged PPP GDP per capita relative to the United States.
  - 10-year forward change in the old-age dependency ratio.
  - Lagged real GDP growth rate over the previous 5 years.
  - Net energy exports relative to GDP.
  - Cyclically adjusted fiscal balance relative to GDP.
- Econometric practice:
  - Coefficient standard errors robust to heteroskedastic and first-order autoregressive errors.
  - New instruments developed to identify exogenous variation in NOF and address endogeneity.

### Key equations (structure and interpretation)
- Baseline specification links CAX and NPFX to NOF, NOA, interactions with MOB, auxiliary controls, year fixed effects, and country fixed effects.
- Interpretation of parameters:
  - α1: effect of NOF on the current account (CAX).
  - α2: differential effect of NOF depending on capital mobility (MOB).
  - β1: effect of lagged NOA on the current account.
  - β2: differential effect of NOA with higher capital mobility.
- Balance of payments (BOP) identity implication:
  - Any effect of NOF on the current account that is less than 1 must show up as a negative effect on net private flows (NPFX).
  - If α1 = 0 (NOF has no effect on CAX), NOF must cause a one-for-one reduction of NPFX (subject to measurement errors and errors and omissions).

### Daily regression findings (US announcements and high-frequency spillovers)
- Strong spillovers from US bond yield moves to prices of foreign financial assets immediately following US economic announcements, consistent with positive news about future US activity being good for other countries.
- Smaller spillovers when US bond yield changes are associated with monetary policy revisions (pure monetary shocks).
  - Interpretation: monetary shocks lack accompanying good news about US activity; thus higher bond yields on those days are not good news for foreign economies.
- Monetary shocks have only a small effect on foreign currencies, limiting the channel through which US unconventional monetary policy could reduce current account balances abroad.

### Stylized model implications (imperfect asset substitution)
- Imperfect asset substitution is central; it:
  - Explains how unconventional monetary policy can operate when short-term interest rates do not move.
  - Helps reconcile observed cross-country asset price correlations that standard models with perfect substitution cannot.
- Model supports empirical hierarchy of effects:
  - NOF (unconventional exchange rate policy) has a larger and more consistent effect on the current account than QE.
  - Stronger US activity unambiguously raises foreign activity and foreign current account balances.
  - Pure monetary shocks have small and ambiguous effects on foreign current accounts and activity.
- Policy toolkit implications:
  - Imperfect asset substitution provides additional policy instruments: in addition to interest rates, central banks can use NOF and QE to target exchange rates, current accounts, or bond yields.

*Source: wp1756 - Appendix Tables (excerpt of paper text).*

### 2.1 and upward in Equation 2.2, which helps to put a range on its true value, and makes the average

### wp1756 - 2.1 and upward in Equation 2.2, which helps to put a range on its true value, and makes the average

### Instrumentation, endogeneity, and identification
- Key endogeneity concern: NOF (net official flows) may respond endogenously to shocks to current account balances and net private flows (stabilization of exchange rate vs. responses to private financial shocks).
- Instruments used to isolate exogenous variation in NOF:
  - Incidence of a financial or currency crisis in the previous three years (captures higher propensity to build reserves for precautionary reasons).
  - Portion of NOF that is not related to foreign exchange reserves (captures SWF-related asset flows and development loans).
- Instrument performance and diagnostics:
  - One of the two non-reserves flow instruments is significant in first-stage regressions; the two crisis instruments are sometimes individually and always jointly significant.
  - F-test statistic values are significantly larger than 10 (rejecting instrument irrelevance).
  - Angrist-Pischke first-stage chi-squared statistic rejects null that NOF is unidentified.
  - First-stage R2 values are around 0.6.
- Modeling choices:
  - MOB (capital mobility index) is lagged in all regressions, including in interaction terms, even when the interacted variable is not lagged.
  - Interactions between all auxiliary variables and MOB are included to allow effects to vary with capital mobility.
  - QE is defined as an increase in central bank domestic assets; QE is cyclically adjusted and lagged in regressions to attenuate endogeneity biases.

### QE definition, potential biases, and measurement
- QE measure = increase in central bank domestic assets (captures central bank balance-sheet expansion that takes risk off domestic market participants).
- Potential biases when estimating QE effect on current account:
  - Negative bias if countries use monetary expansions to combat declines in growth associated with declining current account balances.
  - Positive bias from omitted variables if a negative shock to domestic demand simultaneously causes QE expansion and import declines (raising current account).
- Mitigation: cyclically adjust QE and use lagged values; adjusted QE is the residual from a regression of the change in central bank domestic assets on the level and change of the output gap, the change in nominal GDP, and NOF.

### Spillovers, SPILL variable, and hypotheses tested
- Theory: any effect of QE or NOF on a purchaser country’s current account must have an equal and opposite effect on the rest of the world; allocation and magnitude of spillovers are tested.
- Six spillover allocation hypotheses tested (based on aggregate global NOF divided by world GDP):
  1. Financial integration based on cross-border financial transactions.
  2. Capital mobility (MOB).
  3. Financial development based on Sahay et al. (2015).
  4. Reserve currency shares (IMF COFER database).
  5. Economic size (nominal GDP).
  6. Country’s stage of economic development (PPP GDP per capita).
- Final SPILL term used: financial integration measure multiplied by aggregate global NOF divided by world GDP; financial integration measured as share of gross private financial transactions in total current and financial transactions in the balance of payments.
- Empirical result: only one spillover term has a statistically significant coefficient; median value of financial integration measure is 0.1, so a coefficient of about -20 on SPILL implies that moving from the median to twice the median lowers a country’s current account by twice the value of world NOF divided by world GDP, or around 2 percent of GDP.
  - Note: In 2013, the ratio of world NOF to world GDP was about 1 percent.
  - Around 90 percent of observations of the financial integration measure take values less than twice the median value.

### Baseline regression results (two-stage least squares; Equations 2.1 and 2.2)
- Sample is 2.6 times bigger than Bayoumi, Gagnon, and Saborowski (2015), adding four years and many low-income countries.
- First-stage: instruments relevant (see diagnostics above).
- Second-stage highlights:
  - Coefficients on NOF and NOA broadly similar to previous paper.
  - Relative GDP now carries expected negative coefficient.
  - Estimated effect of NOF on current account when capital mobility is lowest (α1):
    - 0.7 in column 1 (current account excluding investment income).
    - 0.8 in column 2 (net private flows).
  - Overall effect of NOF when mobility is highest = α1 + α2:
    - 0.06 in column 1.
    - 0.52 in column 2.
  - Effect of NOA on current account: close to zero under low mobility and rises to around 3 percent under high mobility.
- Inclusion of interactions between control variables and MOB (columns 3 and 4) leaves NOF and lagged NOA coefficients broadly unaffected; many auxiliary variable effects are conditioned by capital mobility (e.g., trend growth and fiscal balance more important when capital is more mobile; net energy exports larger effect when private capital is less mobile).
- Fiscal balance coefficient when capital is highly mobile: 0.6.
- Full baseline with QE, QE*MOB interaction, and SPILL: coefficients on NOF and NOA remain stable.

### QE coefficient and interpretation
- QE variable is significant with a coefficient close to 0.25.
- QE interaction with MOB: marginally significant negative coefficient of about the same magnitude (≈0.25), suggesting:
  - QE has a small but significant effect on the current account when capital mobility is low.
  - QE effect is close to zero as capital mobility approaches its upper bound.
- Interpretation: QE has larger effect when capital is less mobile — somewhat surprising and investigated further with robustness checks and influential-observation analysis.

### Robustness checks and alternative specifications (Table 2.2 summary)
- Column summaries (high-level):
  - Column 1: baseline (average of last two columns in Table 2.1).
  - Column 2: OLS version — coefficients similar to baseline; NOF biases in either direction appear to cancel out in aggregate.
  - Column 3: replace instruments with full set of country dummies (dropping countries with <5 observations) — NOF coefficient increases somewhat; interaction magnitude also increases but average effect only slightly larger for high MOB.
  - Column 4: GDP-weighted regression (weights sum to 1 in a given year) — results broadly unchanged; QE and QE*MOB similar to baseline and imply little or no impact of QE on current account when capital is highly mobile (relevant to major economies); spillover effect becomes no longer significant.
  - Column 5: robust regression to reduce outlier influence — NOF coefficients little changed; QE and SPILL somewhat smaller and less significant.
  - Columns 6–7: split by exchange rate regime (IMF de facto classification) — QE coefficient smaller under fixed rates than flexible rates, roughly same under flexible rates as baseline; difference across regimes not statistically significant.
  - Columns 8–9: split by trade openness — openness has modest effects on NOF and NOA coefficients; SPILL coefficient much larger in open economies than closed ones; QE appears to have a larger effect on current account in more open economies, but this disappears with high capital mobility.
- Alternative QE measures (unreported regressions): current unadjusted and cyclically adjusted changes in central bank domestic assets; lagged/current changes in cyclically adjusted monetary base — all have smaller (sometimes slightly negative) coefficients than baseline.
- Adding dummy for current financial crisis: little effect on QE (or other) coefficients.
- Adding macro controls (four alternative regressions):
  1. growth rate of nominal GDP,
  2. changes in output gap and GDP deflator,
  3. cyclically adjusted growth rate of monetary base,
  4. lagged cyclically adjusted growth rate of monetary base.
  - None of these noticeably affect coefficients on NOF, NOA, QE, and SPILL.

### Influential-observation analysis
- Method: dfbeta to identify observations with largest marginal effect on coefficients.
- NOF influential observations:
  - Three of four most influential: Azerbaijan in 2008, 2010, and 2011 (managed exchange rate; relatively closed capital account; rapid growth of net energy exports; State Oil Fund outflows beginning 2008).
  - Republic of Congo 1994: non-reserves official development loans with fixed exchange rate (CFA franc zone) and low capital mobility — loans likely exogenous to exchange rate pressures.
  - Note: Kuwait 1991 (massive SWF drawdown) was dropped due to scale and potential nonlinear effects.
  - Empirical implication: NOF is a more important determinant of current account than net energy exports in some cases.
- QE influential observations (positive direction):
  - Malaysia 2008 (liquidity injection in 2007 amid global financial stress; policy rate behavior noted).
  - Angola 2001 (central bank sold domestic assets to buy foreign assets; positive QE coefficient suggests sale of domestic assets tended to lower current account).
  - Thailand 1998 and 1999 (central bank lending/recapitalization during Asian crisis; quantitative monetary program to ease velocity changes; policy rate raised 200 basis points to 12.5 percent in 1997 and held until 1999).
- Robustness to influential observations: dropping four most positive and four most negative influential observations produced essentially no change in QE coefficient — it remained statistically significant.

### Fiscal balances, net official flows, and energy exporters
- Empirical finding: governments’ decisions to save or not save energy revenues abroad via NOF drive current account outcomes; net energy exports per se have little direct effect on current account once NOF and fiscal balance are controlled.
- Examples:
  - Norway and Saudi Arabia: fiscal balance almost identical to NOF; current account moves closely with NOF and fiscal balance; sum of NOF and fiscal coefficients ≈ 0.8 with high mobility and ≈ 0.9 with low mobility.
  - Algeria: lowest capital mobility among examples; fiscal balance lower than NOF and current account closer to NOF than fiscal balance.
- Conclusion: current account follows NOF and fiscal balance, not net energy exports; the government’s choice in using energy revenues matters.

### Dynamics and interpretation of coefficients
- Timing: exchange rate effects of intervention expected to be fast; exchange rate-to-trade/current-account effects typically gradual (~two years).
- Annual data interpretation: coefficients capture long-run effects rather than immediate effects; dynamics differ across countries and variables, making precise dynamic modeling difficult.
- Residual behavior: first-order autocorrelation of residuals in baseline regression is around 0.7.
- Temporary effects: oil exports have significant temporary effects in many countries, boosting current account and SWF flows contemporaneously; explains apparent lack of lag between SWF flows and current account in some figures.

### Daily-data analysis of US unconventional monetary policy (UMPs) and financial-market spillovers
- Motivation: circumvent limitations of annual regressions (few QE observations; GDP/current account not observed daily) by using daily international asset-price responses to US 10-year Treasury yield changes.
- Sample: November 2008 to July 2015 (FOMC used unconventional policies; conventional short-term rate pinned at zero lower bound).
- Benchmark country-level regression (Equation 3.1): ΔX_it = α_i + (β1,i + β2,i D_FOMC,t) Δy_t + u_it
  - X vector: (1) 10-year sovereign local-currency bond yield, (2) log stock market index, (3) log exchange rate (dollars per unit of non-US currency).
  - Δy_t = change in 10-year US Treasury yield.
  - D_FOMC,t = dummy for FOMC statement/minutes or FOMC Chair speech; separate regressions use dummies for US nonfarm payrolls or ISM PMI releases.
  - β1 captures effect on non-FOMC days (economic-news-driven yield changes); β2 captures additional response on FOMC (or data) days (monetary-policy-driven yield changes).
- Key daily-data findings:
  - Sovereign yields: β1 positive and significant for almost all advanced economies and many emerging markets; some emerging markets (Hungary, Russia, Greece) show significant negative correlations (default-risk dynamics).
  - On FOMC days, β2 usually has opposite sign to β1, reducing the total co-movement (β1 + β2 closer to zero but often still significant).
  - Stock prices: nearly universal positive and significant response to US bond-yield rises (β1 positive); β2 (FOMC-day additional effect) almost always negative and significant in many countries (reducing spillovers on announcement days); total effect on FOMC days (β1 + β2) remains significant for many countries.
  - Exchange rates: most currencies depreciate against the dollar when US yields rise; economic magnitude small — a one percentage point increase in US bond yields typically causes a one to three percent depreciation in most currencies (a one percentage point daily move is much larger than typical daily moves).
    - Pegged currencies (e.g., Hong Kong dollar, Chinese renminbi) show smaller effects; safe-haven currencies (Swiss Franc, Japanese Yen) can move positively (appreciate).
  - Persistence: spillover effects for yields and stocks are persistent; spillovers are reduced around monetary policy announcement days.
  - Pre-QE sample (January 2001–November 2008) shows similar patterns; co-movements on non-announcement days are larger after 2008 than before.
- Implication: effects of US unconventional monetary policy on exchange rates are small, implying likely small effects on US current account.

### Cross-country determinants of daily financial spillovers (panel regressions)
- Specification: allow β coefficients to vary with country characteristics C_i (Equation 3.2 and 3.3); country characteristics considered separately to avoid collinearity.
- Country characteristics used (averaged over sample period):
  - Capital mobility (MOB).
  - Financial depth: ratio of bank assets to GDP.
  - Exchange rate regime: volatility of exchange rate vs. US dollar.
  - Sovereign risk: yield of sovereign bonds.
  - Trade linkages: exports to the United States as share of domestic GDP.
- Main findings:
  - Sovereign yields co-move more with US Treasury yields in economies with high capital mobility or deeper financial systems (β2 positive and significant at 1 percent).
  - Sovereign yields in countries with higher sovereign risk co-move less with US Treasury yields, but their additional spillover on FOMC days is higher.
  - Bond yields in economies with high-volatility (more flexible) currencies co-move less with US yields.
  - Trade linkages with the US are positive and significant in explaining sovereign-yield co-movement (closer trade linkages → more co-movement).
  - Stocks: stock prices in high MOB, deeper financial systems, and more flexible-exchange-rate economies co-move more with US Treasury yields; co-movement weaker where sovereign bonds are risky; trade linkages counterintuitively reduce stock spillovers.
  - Exchange rates: capital mobility and exports to US have little power for cross-country differences in β1; financial depth offsets negative effect and sovereign risk increases it; country factors have little explanatory power for additional effects on FOMC/data days (capital mobility marginally offsets negative effect on FOMC days).

*Italic: Source — content from wp1756 (sections summarized above).*

### Section  2  of  this  paper  examined  the  direct  and  spillover  effects  of  net  official  flows  and

### wp1756 - Section  2  of  this  paper  examined  the  direct  and  spillover  effects  of  net  official  flows  and

### Model specification
- Framework: small open-economy macro model with imperfect asset substitution; ignores inflation; variables expressed as deviations from steady state; period is at least one year.
- Key endogenous variables: BS, FBS, BL, FBL, CA, DD, Y, RS, RL, E, CBS, MB.
- Key exogenous variables: RSF, RLF, QE, NOF, XS, XL, and shocks ݑ.
- Units and interpretation:
  - Interest rates in percentage points.
  - Exchange rate in percent (E increase = appreciation).
  - Other variables as proportion of GDP (in percent).
  - Variables expected to return to steady state in periods 2, ... , n (n = term of long bond); focus on period 1.
- Core behavioral equations (as presented):
  - Private domestic demand for domestic short-term bonds:
    - BS = a1*RS + b1*(RS – n*RL) + ݑ
  - Foreign demand for domestic short-term bonds:
    - FBS = a2*(RS–E) + b2*(RS – n*RL) + c2*(RS – E – RSF) + ݑ
  - Private domestic demand for domestic long-term bonds:
    - BL = a3*n*RL + b3*(n*RL – RS) + ݑ
  - Foreign demand for domestic long-term bonds:
    - FBL = a4*(n*RL – E) + b4*(n*RL – RS) + c4*(n*RL – E – n*RLF) + ݑ
  - Current account:
    - CA = –d1*Y – d2*E + ݑ
  - Absorption / domestic demand (IS curve):
    - DD = –e1*RS – e2*n*RL + ݑ
  - Monetary policy (Taylor rule, countercyclical):
    - RS = f1*Y + ݑ
- Identities:
  - Short-term bond market: XS = BS + FBS + CBS (CBS = central bank purchases of short-term bonds).
  - Long-term bond market: XL = BL + FBL + QE (QE = central bank purchases of long-term bonds).
  - Balance of payments: CA = NOF – FBS – FBL (NOF = central bank purchases of foreign assets).
  - GDP identity: Y = DD + CA.
  - Central bank balance sheet: MB = CBS + QE + NOF (MB = growth of monetary base).
- Parity/expectations notes:
  - Short-run interest parity deviations allowed: RS = E + RSF; RL = E/n + RLF would hold under parity.
  - Pure expectations model: RL = RS/n (not enforced).

### Baseline parameter values (as set for numerical solutions)
- Current account parameters (d1, d2) = (0.25, 0.25).
- Absorption parameters (e1, e2) baseline = (0.50, 0.50); alternative sets = (0.25, 0.25), (0.25, 0.50), (1.0, 1.0).
- Monetary parameter f1 = 1. Alternative f1 = 2.
- ax parameters: a1=a2=a3=a4 baseline = 0.1; alternatives = 0.01 and 1.
- bx parameters: b1=b2, b3=b4, b1=b3, b2=b4 baseline = 1; alternatives = 0.1 and 10.
- cx parameters: c2=c4; baseline values = 0.1 for low capital mobility and 1 for high capital mobility; alternatives = 0.01 and 10.

### Rationale and modeling choices
- Imperfect asset substitution introduced to capture portfolio balance channel central to UMP and observed high correlations in bond yields with weak support for interest rate parity.
- Model linear and focused on temporary but persistent shocks; closed-form solutions not available, solved numerically to draw qualitative conclusions.
- Supply of domestic assets is exogenous; central bank finances purchases by issuing zero-interest monetary base.

### Model properties and relation to annual regression results — Effects of NOF
- NOF increases the current account (CA) in both flexible and fixed exchange rate regimes:
  - Flexible regime channel: exchange rate depreciation raises CA.
  - Fixed regime (via RS): interest rates rise, choking off domestic demand, raising CA.
  - If NOF used to fix the exchange rate (NOF endogenous), NOF unavailable as separate tool; moving the exchange rate would produce effects consistent with flexible-rate results.
- NOF effect is larger when capital mobility is low (robust result; supported strongly in annual regressions).
- Income effects:
  - NOF increases income under flexible exchange rates.
  - NOF decreases income under fixed exchange rates (reflecting higher RS to maintain the peg).
- Relation to empirical magnitudes (from conclusions):
  - At the lowest level of capital mobility, each dollar of NOF raises the CA about 75 cents.
  - At the highest level of mobility, effect declines to around 20 cents.
  - Lagged stock of net official assets (NOA) increases CA by about 4 cents on the dollar when capital mobility is high.

### Model properties and relation to annual regression results — Effects of QE
- QE effects on current account (CA):
  - Flexible exchange rate or fixed via RS: QE has small and ambiguously signed effect on CA.
  - Fixed via NOF: QE has a negative effect on CA.
  - Mechanisms under flexible rate:
    - Channel 1: QE lowers long-term rates → depreciates exchange rate → boosts exports (raises CA).
    - Channel 2: QE lowers long-term rates → increases domestic demand → boosts imports (lowers CA).
    - These channels can offset.
  - When exchange rate fixed via NOF, only domestic demand channel operates → negative CA effect.
  - When exchange rate fixed via RS, RS rises to keep exchange rate fixed and offsets QE stimulus.
- Empirical consistency:
  - Annual regressions: QE has essentially no effect on CA under high capital mobility.
  - Small positive effect under low capital mobility (possibly reflecting heterogeneity across advanced vs emerging markets).
- Income effects:
  - Model suggests QE raises income under flexible exchange rate and under exchange rate fixed using NOF.
  - Effects larger when capital mobility is lower.
  - High capital mobility dampens QE impact on bond yields and hence on income.
  - When exchange rate fixed via RS, RS rises to offset downward pressure on exchange rate, damping income effects.

### Spillovers of NOF, QE, and conventional monetary policy
- Spillovers of foreign NOF:
  - Foreign NOF modeled as shock to UFBS.
  - UFBS effects can be exactly offset by equal movements in domestic NOF.
  - Under fixed exchange rate via NOF, UFBS shocks have no effect on CA or income (fully offset).
  - When UFBS not offset, its effects are equal in magnitude and opposite in direction to those of domestic NOF.
  - Model does not address global allocation determinants of NOF.
- Spillovers of QE and conventional monetary policy:
  - RSF and RLF represent rest-of-world monetary policy.
  - Conventional policy operates primarily through RSF, also through RLF.
  - QE and unconventional policies operate primarily through RLF, especially when RSF near zero lower bound.
  - Rest-of-world monetary policy likely associated with movements in exogenous current account component, UCA.

### Relation to daily regression results — Interpretation of shocks
- Distinction:
  - Non-FOMC-day rises in foreign (US) bond yields interpreted as revisions to projected US economic activity (“good news”) → increase RLF and UCA.
  - FOMC-day bond yield changes interpreted as news about unanticipated UMP changes (tighter/looser UMP).
- Model implications for good-news foreign shock (flexible exchange rate):
  - Raises foreign bond yield, raises income, uncertain effect on exchange rate.
  - Capital mobility amplifies effects on yields and income; reduces or makes more negative exchange rate effect.
  - Daily regressions (β1 coefficients) confirm model implications for bond yields and income (if stock price effects reflect expected income).
  - Quantitative results indicate RLF effect typically dominates exchange-rate UCA effect, leading to depreciation.
- Model implications for tighter-than-expected foreign UMP (FOMC days, focus on β2):
  - Tighter UMP implies increase in RLF and decrease in UCA.
  - Relative to good-news shock: raises local bond yield by less or may reduce it; raises local stock price by less or may reduce it; reverses any exchange-rate appreciation or makes depreciation larger.
  - Daily regression β2 coefficients confirm smaller or opposite effects on bond yields and stock prices.
  - Exchange rate regressions found no significant FOMC-day difference—conclusive evidence lacking.
  - Model predicts capital mobility increases local bond yield response and worsens income/exchange rate responses; daily regressions reject this for bond yields and exchange rates (marginal rejection for exchange rates at 10 percent level).
- Overall: daily regressions broadly consistent with model, with primary exception being lack of clear additional negative effect of US UMP shocks on foreign exchange rates.

### Policy implications
- Empirical and model evidence suggest many foreign central banks often follow US monetary policy, producing large co-movements in bond yields and small exchange-rate responses.
- Policy toolkit in world of imperfect asset substitution:
  - NOF can neutralize effects of foreign capital inflows into short-term bonds (including spillovers of NOF).
  - NOF can insulate a country from shocks to foreign short-term interest rates if chosen as policy instrument.
  - Combination of NOF and QE can fully offset effects of foreign inflows into long-term bonds and changes in foreign long-term rates.
  - These tools allow pursuit of objectives on exchange rates, current accounts, or long-term yields while preserving monetary policy for income stabilization.
- Caveats and constraints:
  - Central banks may be uncomfortable taking large net positions in long-term bonds and foreign exchange.
  - Political constraints may limit full use of fiscal policy, capital controls, and macro-prudential measures.
  - Model is linear and simple; cannot fully assess benefits, costs, and risks at scale, but highlights range of policy options.

### Conclusions (summary of principal empirical and theoretical findings)
- Paper investigates direct effects and spillovers of unconventional monetary and exchange-rate policies using annual regressions (direct effects of official purchases on CA) and daily regressions (spillovers of US unconventional policy on foreign financial prices).
- Findings consistent with a small-economy imperfect substitution model.
- NOF findings:
  - Large direct effect on a country’s CA when capital mobility is low; effect diminishes as mobility rises but remains significant at high mobility.
  - Important additional effect through lagged stock of net official assets.
  - Quantitative examples: at lowest capital mobility, $1 of NOF → ~75 cents increase in CA; at highest mobility, ~20 cents; NOA adds ~4 cents on the dollar when mobility high.
- QE findings:
  - QE has no significant effect on CA when capital mobility is high; modest but significant positive impact when capital mobility is low.
  - Puzzle: significant effect only under low capital mobility—may reflect differences between major advanced economies and emerging markets.
- Spillovers:
  - Effects of official flows spill over to other countries proportionally to degree of international financial integration; effect moderately robust but less precise for US and euro area (US receives relatively more spillovers, not statistically significant).
  - US economic developments raise US bond yields, raise foreign bond yields, increase foreign stock prices, and depreciate foreign currencies.
  - Stronger US activity unambiguously raises foreign activity and foreign CA in the model; spillovers stronger with higher capital mobility and deeper financial markets.
  - Spillovers often greater on foreign bond yields than on foreign exchange rates — suggesting many foreign central banks follow the Fed to stabilize exchange rates.
  - US bond yield increases associated with tighter-than-expected future monetary policy have smaller effects on foreign financial variables, consistent with roughly neutral effects on foreign activity and CA; pure monetary shocks have small and ambiguous effects on CA and activity for countries with high capital mobility.
- Policy message:
  - Foreign exchange intervention (NOF) and QE at home can be useful responses to foreign policies of similar type; negative spillovers on CA can be fully offset by increasing domestic NOF; spillovers onto bond yields can be offset by countervailing QE.
  - Practical and political constraints may limit full use of these tools; model provides framework to compare options.

*Italic source attribution: IMF Working Paper (wp1756), Section 3–5 content as provided.*

### REFERENCES

### REFERENCES

### Bibliographic citations
- Adler, Gustavo, Noemie Lisack, and Rui Mano. 2015. Unveiling the Effects of Foreign Exchange Intervention:  A  Panel  Approach.  IMF  Working  Paper  No.  WP/15/130.  Washington:  International Monetary Fund.
- Aizenman,    Joshua,    Menzie    Chinn,    and    Hiro    Ito.    2015.    The    Trilemma    Indexes.    http://web.pdx.edu/~ito/trilemma_indexes.htm (accessed August 17, 2015).
- Bartolini,  Leonard,  Linda  Goldberg,  and  Adam  Sacarny.  2008.  How  Economic  News  Moves  Markets. Current Issues in Economics and Finance 14(6): 1-7.
- Bayoumi, Tamim, and Trung Bui. 2010. Deconstructing the International Business Cycle: Why Does A U.S. Sneeze Give The Rest Of The World A Cold? IMF Working Papers WP/10/239. Washington: International Monetary Fund.
- ____________,   and   Christian   Saborowski.   2014.   Accounting   for   Reserves.   Journal   of   International Money and Finance 41: 1-29.
- ____________ Joseph Gagnon, and Christian Saborowski. 2015. Official Financial Flows, Capital Mobility, and Global Imbalances. Journal of International Money and Finance 52 (April): 146-74.
- Blanchard,  Olivier,  Gustavo  Adler,  and  Irineu  de  Carvalho  Filho.  2015.  Can  Foreign  Exchange  Intervention  Stem  Exchange  Rate  Pressures  from  Global  Capital  Flow  Shocks?  IMF  Working Papers WP/15/159. Washington: International Monetary Fund.
- Bowman,  David,  Juan  M.  Londono,  and  Horacio  Sapriza.  2014.  US  Unconventional  Monetary  Policy and Transmission to Emerging Market Economies. Journal of International Money and Finance 55: 57-59.
- Carstens,  Agustin.  2015.  Challenges  for  Emerging  Economies  in  the  Face  of  Unconventional  Monetary Policies in Advanced Economies. Stavros Niarchos Foundation Lecture, April 20. Washington: Peterson Institute for International Economics.
- Chinn,  Menzie,  and  Eswar  Prasad.  2003.  Medium-Term  Determinants  of  Current  Accounts  in  Industrial  and  Developing  Countries:  An  Empirical  Exploration.  Journal  of  International Economics 59: 47-76.
- Gagnon, Joseph. 2012. Global Imbalances and Foreign Asset Expansion by Developing-Economy Central Banks. Working Paper No. 12-5. Washington: Peterson Institute for International Economics.
- Gagnon,  Joseph.  2013.  The  Elephant  Hiding  in  the  Room:  Currency  Intervention  and  Trade  Imbalances.  Working  Paper  No.  13-2.  Washington:  Peterson  Institute  for  International Economics.
- ____________,  Matthew  Raskin,  Julie  Remache,  and  Brian  Sack.  2011.  The  Financial  Market  Effects  of  the  Federal  Reserve’s  Large-Scale  Asset  Purchases.  International  Journal  of  Central Banking 7: 3-44.
- Hausman, Joshua, and Jon Wongswan. 2011. Global asset prices and FOMC announcements. Journal of International Money and Finance 30 (3): 547-71.
- Hofmann, Boris, and Elod Takats. 2015. International Monetary Spillovers. BIS Quarterly Review (September): 105-18.
- International  Monetary  Fund.  2015.  2015  Spillover  Report.  July  23.  Washington:  International Monetary Fund.
- Laeven,  Luc,  and  Fabián  Valencia.  2012.  Systemic  Banking  Crises  Database:  An  Update.  IMF Working Paper WP/12/163. Washington: International Monetary Fund.
- Londono, Juan, and Hao Zhou. 2016. Variance Risk Premiums and the Forward Premium Puzzle. Journal of Financial Economics (forthgoming).
- Rogers,  John,  Chiara  Scotti,  and  Jonathan  Wright.  2014.  Evaluating  Asset-Market  Effects  of  Unconventional  Monetary  Policy:  A  Cross-Country  Comparison.  International  Finance Discussion  Papers  No.  1101.  Washington:  Board  of  Governors  of  the  Federal  Reserve  System.
- Sahay,  Ratna,  Martin  Cihak,  Papa  N’Diaye,  Adolfo  Barajas,  Ran  Bi,  Diana  Ayala,  Yuan  Gao,  Annette Kyobe, Lam Nguyen, Christian Saborowski,  Katsiaryna  Svirydzenka,  and  Seyed  Yousefi.  2015.  Rethinking  Financial  Deepening:  Stability  and  Growth  in  Emerging  Markets. IMF Staff Discussion Note 15/08. Washington: International Monetary Fund.

### Appendix: Data sources and definitions
- Sources: IMF, Balance of Payments Statistics version 6 (BOP); IMF, International Financial Statistics (IFS); IMF, Monetary and Financial Statistics; IMF, World Economic Outlook (WEO); United Nations, World Population Prospects 2010 (UN); World Bank, World Development Indicators (WDI); World Bank, Worldwide Governance Indicators (WGI); Norway, Norges Bank; and Singapore, Ministry of Finance. Data on capital controls and exchange rate volatility are from Aizenman, Chinn, and Ito (2015). Daily financial data are from Bloomberg.
- Annual Regressions:
  - CAX: The BOP current account balance minus BOP net investment income.
  - NPFX: BOP net financial account flows minus BOP net investment income minus NOF.
  - NOF: Based on the BOP data, NOF is the sum of reserves flows and net portfolio investment and other investment flows for central bank and general government except that portfolio liability flows are set at zero for advanced economies because they do not borrow significantly in foreign currency. Nonreserve flows data for Norway are from the Norges Bank website for the Norwegian Pension Fund (Global). Debt forgiveness is removed from NOF but not NOA.
  - NOA: Based on the IIP data, NOA is defined as the stock version of NOF. Missing values of liabilities are filled in from the World Bank’s external debt data. In countries in which less data is available for the stock than the flow variable, we use perpetual inventory to project NOA backwards. For Norway, NOA is the sum of reserves and Pension Fund (Global) assets.
  - GDP: Nominal GDP in US dollars and in local currency, and real GDP, are from WEO.
  - MOB: Capital controls index available at Aizenman, Chinn, and Ito (2015).
  - QE: Central bank domestic assets. Source: IMF Monetary and Financial Statistics (MFS) and International Financial Statistics (IFS).
  - SPILL: Global Financial Integration multiplied with the sum of NOF across countries and divided by the sum of trend GDP across countries.
- Relative PPP:
  - GDP Per Capita: WEO (relative to US level). We set this as missing before 1996 for European transition economies.
- Appendix: Data sources and definitions (cont’d)
  - Aging: 10-year forward change in ratio of elderly to working age population. Historical elderly ratios through 2010 are from WDI. Ratios for 2020 and 2020 are from UN and are interpolated and extrapolated in order to create 10-year changes for 2001–15.
  - Growth: 5-year moving average of growth rate of real GDP based on WEO. We corrected an error in Malta real GDP using IFS data. We set real GDP growth as missing for European transition economies before 1996.
  - Net  Energy  Exports: Difference between energy production and consumption in tons of oil equivalent (WDI), converted into dollars using Brent oil price (IFS) assuming

*Source: wp1756 - REFERENCES (IMF Working Paper wp1756).*

### 7.33 barrels per ton.

### wp1756 - 7.33 barrels per ton.

### Data sources and variable definitions
- 7.33 barrels per ton.
- Fiscal Balance: General government balance in percent of GDP (WEO) is cyclically adjusted as the residual in a panel regression of the fiscal balance on the level and change of the GDP gap with no country or year effects. The GDP gap is the difference between log real GDP and its 11-year centered moving average using WEO forecasts for 2015–18. A missing value for South Africa in 2005 is interpolated.
- Global Financial Integration: Defined as the ratio of BOP private financial account transactions divided by the sum of financial and current account transactions.
- Non-reserve Flows: NOF minus reserve assets flow divided by trend GDP.
- Crisis: Dummy = 1 if country experienced a financial or currency crisis in the previous three years. Source: Laeven and Valencia (2011).
- Trade Openness: Exports of goods and services plus imports of goods and services divided by trend GDP.
- Exchange rate regime: IMF de facto (coarse) index from www.carmenreinhart.com and Aizenman-Chinn-Ito (2015) rolling measure of ER volatility.
- Scaling by trend GDP: 11-year centered moving average of nominal GDP in US dollars (WEO), including forecast data through 2018.

Daily-regression variables (Bloomberg unless noted):
- Currency flexibility: Sample period daily standard deviation of the exchange rate w.r.t. the U.S. dollar.
- Exchange rate: Log of the local exchange rate (in dollars per unit of non U.S. currency).
- (Exports-to-U.S.)/GDP: Sample average ratio of exports to the United States to domestic GDP based on annual regression data.
- FOMC Dummy (D_(U.S.,t)): 1 on days FOMC released policy statement, minutes, or there was a monetary policy speech by the FOMC Chair; 0 otherwise. (Constructed by authors)
- Econ Dummy (D_(U.S.,t)): 1 on days of U.S. nonfarm payrolls or ISM PMI releases; 0 otherwise. (Constructed by authors)
- MOB: Sample average of MOB from the annual regression data.
- Bank assets to GDP: Average from 2008–12 from Helgi Library.
- Sovereign bond yield: Daily yield of 10-year constant maturity sovereign local-currency bond.
- Sovereign risk: Sample-period average sovereign bond yields.
- Stock price: Log of the daily local stock market price index.
- U.S. sovereign yield: Daily yield of 10-year constant maturity U.S. Treasury bond.

### Baseline regressions — First-stage (NOF instrument)
- Key coefficient: Nonreserve Flows = 0.851*** (columns (1)–(4)); 0.868*** in columns (5)–(6).
- Interaction with MOB: -0.108 (columns (1)–(4)); -0.133 (columns (5)–(6)).
- Crisis coefficient: 0.004 in reported specifications.
- Interaction with MOB (Crisis): 0.004 (columns (1)–(4)); 0.008 (columns (5)–(6)).
- R-squared = 0.59 (columns (1)–(4)); 0.59 (columns (5)–(6)).
- F-test = 68.5 (p = 0.000) in many specifications; 55.8 (p = 0.000) in columns (5)–(6).
- AP Chi-sq test = 69.9 (p = 0.000) in many specifications; 58.1 (p = 0.000) in columns (5)–(6).

First-stage for NOF interaction:
- Nonreserve Flows = 0.032.
- Interaction with MOB = 0.709*** (columns (1)–(4)); 0.718*** (columns (5)–(6)).
- Crisis = -0.001 to -0.002 across columns.
- Interaction with MOB (Crisis) = 0.009* (columns (1)–(4)); 0.015** (columns (5)–(6)).
- R-squared ≈ 0.61 (columns (1)–(4)); 0.60 (columns (5)–(6)).
- F-test = 56.0 (p = 0.000) (columns (1)–(4)); 43.3 (p = 0.000) (columns (5)–(6)).
- AP Chi-sq test = 98.6 (p = 0.000) (columns (1)–(4)); 87.3 (p = 0.000) (columns (5)–(6)).

### Baseline regressions — Second-stage (selected coefficients and stats)
- Reporting across CAX and NPFX specifications (columns and pairs shown in table):
  - MOB, lagged: coefficients shown as 0, -0.002, -0.007, -0.017, 0.011, 0.01 (with zeros for standard errors in the table where noted).
  - MOB squared and lagged: 0.039**, 0.061***, 0.035*, 0.046** (where reported).
  - Global Financial Integration: 0.140**, 0.146* (in some columns).
  - Relative GDP PPP pc, lagged: -0.011, -0.026**, 0.02, 0.040*, 0.012, 0.027.
  - Aging: 2.165***, 2.211***, 3.620**, 2.165, 2.831, 1.37 (with noted standard errors).
  - Interaction with MOB (Aging): coefficients reported including -2.989, -1.117, -0.59, 0.564 (with large standard errors).
  - Growth, lagged: -0.186***, -0.233***, 0.021, 0.078, 0.101, 0.179*.
  - Interaction with MOB (Growth): -0.522***, -0.731***, -0.752***, -0.999***.
  - Net Energy Exports: 0.171***, 0.143***, 0.270***, 0.253***, 0.280***, 0.259***.
  - Interaction with MOB (Net Energy Exports): -0.245***, -0.277***, -0.251***, -0.268***.
  - Fiscal balance: 0.392***, 0.349***, 0.212***, 0.128, 0.176**, 0.132.
  - Interaction with MOB (Fiscal balance): 0.347***, 0.425***, 0.421***, 0.459***.
  - NOF: 0.701***, 0.780***, 0.683***, 0.742***, 0.724***, 0.775***.
  - Interaction with MOB (NOF): -0.643***, -0.26, -0.622***, -0.215, -0.715***, -0.421*.
  - NOA, lagged: 0.007, -0.021**, 0.004, -0.026***, 0.006, -0.027***.
  - Interaction with MOB (NOA): 0.032*, 0.03, 0.044***, 0.048**, 0.040**, 0.059**.
  - QE, lagged: 0.231**, 0.255** (in some columns).
  - Interaction with MOB (QE): -0.302*, -0.22 (in some columns).
  - SPILL: -18.638***, -24.077*** (with standard errors 5.4 and 5.8).
- R-squared values across columns: 0.457, 0.239, 0.493, 0.308, 0.525, 0.376.
- Observations: 2088 (in four columns), 1745 (in two columns).
- Significance notation: * p<0.1, ** p<0.05, *** p<0.01.

### Robustness checks (Table 2.2) — highlights
- Baseline NOF coefficient: 0.750*** (column (1) Baseline).
- Alternative instrument NOF: 0.674*** (column (2) OLS/Alt Instr).
- Weighted and robust specifications show NOF coefficients: 0.970*** (col (3)), 0.765*** (col (4)).
- Interaction with MOB (averaged across CAX/NPFX): -0.568** (col (1)), -0.408*** (col (2)), -0.7305** (col (3)), -0.4135** (col (4)), -0.751*** (col (5)), -0.981** (col (6)), -0.3565 (col (7)), -0.52*** (col (8)), -0.5385*** (col (9)).
- QE, lagged: 0.243** (col (1)), 0.248** (col (2)), 0.269** (col (3)), 0.2165** (col (4)), with interaction with MOB often negative (e.g., -0.279* in col (2)).
- SPILL consistently negative and significant in many robustness checks: -21.4*** (col (1)), -21.5*** (col (2)), -21.2*** (col (3)), -11.8*** (col (5)), -16.8* (col (6)), -14.9*** (col (7)), -39.2*** (col (8)).
- R-squared across robustness columns: 0.45, 0.37, 0.46, 0.64, 0.50, 0.52, 0.46, 0.51, 0.43.
- Observations across columns vary: 1745, 1755, 1699, 1745, 1745, 650, 1095, 873, 872.
- Note: Coefficients and standard errors are averages across CAX and NPFX regressions.

### Correlations in energy exporters (2000–14, percent of GDP) (Table 2.3)
- Algeria: NOF-Energy 0.92; NOF-Fiscal 0.90; NOF-CAX 0.98; Fiscal-CAX 0.94; Energy-CAX 0.91.
- Norway: NOF-Energy 0.73; NOF-Fiscal 0.90; NOF-CAX 0.80; Fiscal-CAX 0.86; Energy-CAX 0.58.
- Saudi Arabia: NOF-Energy 0.87; NOF-Fiscal 0.90; NOF-CAX 0.94; Fiscal-CAX 0.90; Energy-CAX 0.88.
- Yemen: NOF-Energy 0.74; NOF-Fiscal 0.86; NOF-CAX 0.87; Fiscal-CAX 0.93; Energy-CAX 0.89.
- Colombia: NOF-Energy 0.37; NOF-Fiscal 0.62; NOF-CAX 0.35; Fiscal-CAX -0.03; Energy-CAX -0.34.
- Indonesia: NOF-Energy -0.23; NOF-Fiscal 0.41; NOF-CAX 0.66; Fiscal-CAX 0.43; Energy-CAX -0.75.
- Venezuela: NOF-Energy 0.40; NOF-Fiscal 0.42; NOF-CAX 0.82; Fiscal-CAX 0.39; Energy-CAX 0.45.

### Daily spillover regressions — sovereign yields (Table 3.1) — selected country betas
- Canada: β1 = 0.613***; β2 (FOMC) = -0.061**; β1 + β2 (FOM) = 0.539***; β2 (Econ.) = 0.013; β1 + β2 (Econ) = 0.599***.
- United Kingdom: β1 = 0.457***; β2 (FOMC) = -0.286***; β1 + β2 (FOM) = 0.179***; β2 (Econ.) = 0.088; β1 + β2 (Econ) = 0.502***.
- Germany / Euro Area: Germany β1 = 0.434***; β2 (FOMC) = -0.228***; β1 + β2 (FOM) = 0.215***; β1 + β2 (Econ) for Euro Area = 0.461***.
- Japan: β1 = 0.125***; β2 (FOMC) = -0.023; β1 + β2 (FOM) = 0.120***; β2 (Econ.) = 0.040**; β1 + β2 (Econ) = 0.169***.
- Brazil: β1 = 0.153**; β2 (FOMC) = 0.344**; β1 + β2 (FOM) = 0.303**; β2 (Econ.) = 0.423**; β1 + β2 (Econ) = 0.395*.
- Notable negatives: Greece β1 = -0.672***; Russia β1 = -0.164*.

### Daily spillover regressions — stock prices (Table 3.2) — selected country betas
- Belgium: β1 = 0.101***; β2 (FOMC) = -0.055**; β1 + β2 (FOM) = 0.037***; β2 (Econ.) = 0.125; β1 + β2 (Econ) = 0.198.
- Euro Area / France / Germany: β1 ≈ 0.082*** (Euro Area), 0.082*** (France), 0.081*** (Germany); β2 (FOMC) negative small magnitudes.
- Japan: β1 = 0.065***; β2 (FOMC) = -0.023; β1 + β2 (FOM) = 0.046**; β2 (Econ.) = 0.037*; β1 + β2 (Econ) = 0.097***.
- Brazil: β1 = 0.071***; β2 (FOMC) = -0.068***; β1 + β2 (FOM) = 0.022; β2 (Econ.) = -0.003; β1 + β2 (Econ) = 0.076***.
- China: β1 = 0.011*; β2 (FOMC) = -0.045***; β1 + β2 (FOM) = -0.018.

### Daily spillover regressions — exchange rates (w.r.t. US dollar) (Table 3.3) — selected country betas
- Poland: β1 = -0.033***; β2 (FOMC) = 0.012; β1 + β2 (FOM) = -0.013; β2 (Econ.) = -0.018; β1 + β2 (Econ) = -0.040***.
- Japan: β1 = 0.033***; β2 (FOMC) = -0.002; β1 + β2 (FOM) = 0.033***; β2 (Econ.) = 0.011*; β1 + β2 (Econ) = 0.043***.
- Switzerland: β1 = 0.007*; β2 (FOMC) = -0.003; β1 + β2 (FOM) = 0.009; β2 (Econ.) = 0.006; β1 + β2 (Econ) = 0.015*.
- Many advanced economies cluster at β1 = -0.011*** with β2 (FOMC) = 0.003 and β1 + β2 (Econ) = -0.005 (Belgium, Euro Area members, etc.).
- Large negative exchange-rate β1 for several emerging markets: Hungary -0.033***, South Africa -0.030***, Russia -0.029***, Australia -0.028***, Sweden -0.027***, Norway -0.026***, Brazil -0.026***.

### Panel determinants of spillover effects (Tables 3.4a–3.4c)
- Sovereign yields (Table 3.4a):
  - MOB: 21.75*** for β12; -16.70*** for β22 (FOMC); -9.90* for β22 (Econ.).
  - Bank assets to GDP: 3.39*** for β12; -3.74** for β22 (FOMC).
  - Currency flexibility: -0.11** for β12; -0.01 for β22 (FOMC).
  - Sovereign risk: -4.31*** for β12; 4.36*** for β22 (FOMC).
  - (Exports-to-US)/GDP: 3.78** for β12; 1.92 for β22 (FOMC).
- Stock prices (Table 3.4b):
  - MOB: 5.05*** for β12; -1.50 for β22 (FOMC); 2.04 for β22 (Econ.).
  - Bank assets to GDP: 0.76*** for β12.
  - Currency flexibility: 0.07*** for β12.
  - Sovereign risk: -0.24*** for β12.
  - (Exports-to-US)/GDP: -0.75*** for β12.
- Exchange rates (Table 3.4c):
  - MOB: 0.39 for β12; 0.80* for β22 (FOMC); 0.47 for β22 (Econ.).
  - Bank assets to GDP: 0.20*** for β12.
  - Currency flexibility: -0.06*** for β12; 0.02* for β22 (FOMC).
  - Sovereign risk: -0.14*** for β12.

### Comparative effects (Tables 4.1–4.3) — qualitative signs across scenarios
- Table 4.1 (Effects on Bond Yield): signs reported for NOF, QE, RSF, RLF, UCA with distinctions for Low Mobility and High Mobility; cells indicate qualitative relationships (e.g., NOF, Low Mobility: ± < +; QE, Low Mobility: -; QE, High Mobility: -; RSF and RLF show mixed ±, +, 0 depending on scenario).
- Table 4.2 (Effects on Current Account): NOF generally + in Low and High Mobility; QE shows ±0 in many specifications; RSF and RLF generally +; UCA generally + but with mobility-dependent nuances.
- Table 4.3 (Effects on Income (GDP)): NOF generally + with mobility nuance (> then - in some columns); QE generally + with mobility-dependent effects and signs such as ±0 and <; RSF and RLF show + or mixed signs depending on mobility and specification; UCA generally + with mobility-dependent variations.
- Notes: First column based on equations 1–12; alternative columns change exogeneity assumptions for E or NOF. Equality and inequality signs placed when relationships hold for baseline and all alternative parameters.

### Miscellaneous figures and country examples
- Figures referenced: Central Bank Domestic Assets in Major QE Episodes; Current Accounts in Major QE Episodes; Azerbaijan Current Account and Flows; Nigeria Current Account and Flows; Rep. of Congo Current Account and Flows; Energy Exporters with Substantial NOF; Energy Exporters with Small NOF.
- Table/figure page references: Appendix data definitions conclude on page 36–37; regression tables and robustness checks across pages 38–49; spillovers and determinants pages 41–45; comparative effect tables pages 46–48; figures pages 49–51.

*Source: wp1756 - 7.33 barrels per ton.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2017/wp1756.pdf_