## wp17222 - References

## Source details

**Canonical URL:** [wp17222 - References](https://www.imf.org/-/media/files/publications/wp/2017/wp17222.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2017/wp17222.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2017/wp17222.pdf.json)

---

### I. Introduction — purpose, context, and high-level findings
- Purpose:
  - Investigates how global liquidity (GL) affects macro-financial variables and policies in emerging-market economies (EMEs).
  - Decomposes GL into three momenta: policy-driven liquidity, market-driven liquidity, and risk averseness, and compares responses between inflation targeting (IT) and non-IT EMEs.
- Context and literature:
  - GL linked to interest rates, asset prices, inflation, and capital flows (selected cited works: Frankel et al., 2002; Edwards, 2010, 2015; Kim and Yang 2009; di Giovanni and Shambaugh, 2008; Valente, 2009; Kim and Shin, 2016; Rigobon and Sack, 2004; Bluedorn and Bowdler, 2011; Ehrmann and Fratzscher 2009; Ammer et al., 2010; Wongswan, 2009; Berger and Harjes, 2009; Cerutti et al., 2014; Cerutti et al., 2015).
  - U.S. Federal Reserve’s post-GFC policy actions are instrumental in gauging U.S. monetary policy impacts on EMEs (Glick and Leduc, 2012; Bauer and Neely, 2014).
  - Prior GL measurement approaches include single-country or group measures (D’Agostino and Surico, 2007; Bruno and Shin, 2015; Chen et al., 2012; Eickmeier et al., 2013).
- Key high-level findings previewed:
  - Positive GL shocks from G5 policies or financial markets induce EMEs to reduce policy rates and increase foreign reserves, with policy responses focused on external fronts.
  - Heightened risk averseness that accompanies capital outflows causes EMEs to run down foreign reserves and furnish foreign-currency liquidity.
  - Overall, increases in global liquidity generate positive spillovers on equity markets and output; liquidity reversals due to heightened risk averseness cause negative spillovers.
  - Responses of macro-financial variables to GL shocks are less volatile in IT countries than in non-IT countries.
- Data and sample for EME analysis:
  - Panel FAVAR model incorporates panel data of 10 EMEs for 1995:Q1-2014:Q3.

### II. Empirical modeling of global liquidity transmission
- Two-stage empirical approach:
  - Stage 1: Derive GL momenta from a static factor model (principal components) using AEs’ financial data. Equation: X_t = Λ F_t + u_t.
  - Stage 2: Add shocks to the momenta (v_t) as exogenous variables to a panel VAR model of EME variables (Y_t), integrating Stock and Watson (2005) dynamic factor ideas into a panel VAR. VAR system: Y_t = A(L) Y_{t−1} + B(L) v_t + ε_t; F_t = C(L) F_{t−1} + v_t.
- Model choices and motivations:
  - Panel FAVAR extends panel VAR literature to measure dynamic impacts of AE-originating GL momenta on EME macro-financial variables.
  - Factors act as conditioning information (in the spirit of Bernanke et al., 2005) to proxy for otherwise missing variables and to address empirical anomalies.
  - GL momenta shocks (v_t) are assumed independent of individual EME shocks (ε_t).
- Data characteristics and preprocessing:
  - Use both price and quantity data to account for monetary policy at the zero lower bound.
  - Factor models aggregate large, heterogeneous global data sets; sign restrictions employed for identification.
  - Outliers replaced by medians of previous observations (Stock and Watson, 2005).
  - Hodrick-Prescott filter applied to interest rates to remove trending behavior and render series stationary.
  - Financial variables regressed on GDP growth and producer-price inflation; residuals used to net out macroeconomic fundamentals.
  - Each variable weighted by GDP volume to account for size differences among AEs.

### III. Deriving global liquidity momenta — sample, extraction, and identification
- Sample and variables for factor extraction:
  - Quarterly data for 1990:Q1-2014:Q3 from the G5: the U.S., the U.K., France, Germany, and Japan.
  - Nine variables per country: overnight call rates, government bill rates, real exchange rates, lending-rate spread against overnight call rates or policy rates, the monetary base, private domestic credit, international claims, stock prices, and stock market volatility.
- Factor extraction method:
  - Principal components analysis; select three principal components per Ahn and Horenstein (2013).
  - The three factors explain 48 percent of the variability of the underlying AE data:
    - policy-driven liquidity explains 16%;
    - market-driven liquidity explains 13%;
    - risk averseness explains 18%.
  - Minimal sign restrictions imposed for identification; selection of factor candidate closest to median as per Fry and Pagan (2011).
- Sign restrictions used (applied to the U.S. only in pinning down Λ in equation (1)):
  - Policy-Driven Liquidity:
    - M0 (+)
    - Lending rate spread (+)
    - Treasury bill rate (−)
    - Real interest rate (−)
  - Market-Driven Liquidity:
    - Private Domestic Credit (+)
    - Stock Price (+)
  - Risk Averseness:
    - Private Domestic Credit (−)
    - Stock Volatility (+)
  - Note: “A positive sign attached to a variable means that the GL momentum in that row rises with the level of the variable.”
- Economic interpretation and rationale for restrictions:
  - Policy-driven liquidity identified with monetary policy stance and base money expansions (M0), relevant under unconventional monetary policy and at the zero lower bound.
  - Lending-rate spread restriction recognizes that, prior to unconventional policy, expansion in policy-driven liquidity can widen the spread by reducing funding cost of banks through short-term instruments.
  - Market-driven liquidity tied to endogenous banking-sector credit expansion and higher stock prices associated with market liquidity.
  - Risk averseness captures investor risk appetite; higher risk averseness associated with reduced private credit and increased stock volatility.

### Global liquidity momenta before and during the GFC
- Three momenta indicate an overall easing in GL for several years in the run-up to the GFC.
- With the culmination of the GFC in 2008:Q4:
  - the policy-driven factor dropped to its lowest point;
  - the market-driven factor slid down with the Lehman Brothers collapse (2008:Q3);
  - the risk averseness factor reached its highest point.
- Catalyzing effects during the GFC quarter:
  - Confluence of sudden deteriorations in major central banks’ liquidity supply, sudden deteriorations in market-driven liquidity, and a sharp spike in risk averseness.
  - Supply of liquidity by major central banks during this quarter was not sufficient to backstop the sudden deterioration of output and weak prices.
  - Real interest rates (after controlling for producer price inflation and output growth) pointed to a record high.
- Annotation note:
  - Footnote: The validity of these restrictions remains intact even after the federal funds rate was set at 025 basis points in December 2008, while the bank prime loan rate in the U.S. remained unchanged at 3.25 percent for 2009:Q1-2014:Q3.

### GL momenta — definitions and key properties
- Three GL momenta are extracted from AE monetary and financial series and standardized.
- Risk averseness GL factor:
  - Composite extracted from nine financial variables including cross-border and domestic-credit flows as well as stock price volatility for the G5.
  - Simple correlation with the CBOE Volatility Index (VIX) is 0.75.
- Market-driven GL factor:
  - Positively correlated with the domestic credit growth of the U.S., which is negatively associated with domestic credit growth of other G5 countries except the U.K.
- Correlation between market-driven and risk averseness GL factors is very low at −0.04.
- Historical episodes:
  - Onset of the 1991 recession (1991:Q1), dot-com bubble bust (2000:Q1), Lehman Brothers Bankruptcy (2008:Q3).
  - At the onset of the GFC (2009:Q1) the size of the policy-driven GL shock was as big as 3.7 standard deviations.

### EME panel data, variables, estimation details
- EME panel sample period: 1995:Q1-2014:Q3.
- EME countries in the IT panel: the Czech Republic, Hungary, Israel, Korea, Mexico, the Philippines, Poland, Romania, Thailand, and Turkey.
- Model variables (quarterly, eight variables):
  - real GDP growth
  - CPI inflation
  - stock price growth
  - nominal effective exchange rate (NEER) growth
  - current account balance (as percent of GDP)
  - capital inflows (as percent of GDP)
  - foreign reserves (as percent of GDP)
  - overnight call rates
- Lag structure selected by the Akaike information criterion: two autoregressive lags and one contemporaneous and two lagged terms of the GL shock.
- Estimation: equation-by-equation least squares for the EME panel.
- Error bands for impulse responses constructed via a modified Bayesian Monte Carlo integration method that:
  - (i) resamples GL factors,
  - (ii) re-estimates equation (2) per realization,
  - (iii) randomly picks a parameter set via Bayesian Monte Carlo integration,
  - (iv) draws an impulse response;
  - confidence intervals and mean values obtained from the empirical distribution of impulse responses.
- For real GDP, CPI, stock prices and exchange rates, responses of respective growth variables are accumulated from the model to obtain responses in their levels.

### Forecast error variance decomposition — aggregate findings (three-year horizon)
- About 10 percent of real GDP growth variability for the three-year horizon is attributable to GL momenta.
- About 13 percent of capital inflows variability is attributable to GL momenta.
- About 30 percent of stock price variability is attributable to GL momenta.
- NEER variability appears mostly driven by domestic elements; spillovers from AEs to EMEs’ NEER are generally smaller than spillovers to exchange rates against the U.S. dollar.
- Policy-driven and risk averseness GL momenta play more important roles in movements of EME variables than the market-driven GL momentum, except for current account, foreign reserves, and overnight call rates which show substantial variability from the market-driven GL momentum.
- Policy-driven GL dominates market-driven GL in impacts on EME key variables except current account.

- Table of three-year forecast error variance decomposition (percent) — aggregate contributions (Table 2):
  - Exchange Rate: All Factors 10.0; Policy-Driven Liquidity 2.9; Market-Driven Liquidity 1.9; Risk Averseness 5.2
  - Capital Inflow: All Factors 13.2; Policy-Driven Liquidity 2.5; Market-Driven Liquidity 1.9; Risk Averseness 8.8
  - Stock Price: All Factors 29.6; Policy-Driven Liquidity 12.1; Market-Driven Liquidity 6.9; Risk Averseness 10.9
  - Real GDP: All Factors 10.5; Policy-Driven Liquidity 3.3; Market-Driven Liquidity 2.3; Risk Averseness 4.9
  - CPI: All Factors 6.8; Policy-Driven Liquidity 2.3; Market-Driven Liquidity 1.8; Risk Averseness 2.6
  - Current Account: All Factors 11.8; Policy-Driven Liquidity 3.4; Market-Driven Liquidity 5.1; Risk Averseness 3.4
  - Foreign Reserves: All Factors 10.7; Policy-Driven Liquidity 4.6; Market-Driven Liquidity 3.9; Risk Averseness 2.4
  - Overnight Call Rate: All Factors 7.1; Policy-Driven Liquidity 3.0; Market-Driven Liquidity 2.7; Risk Averseness 1.5

### EME responses to a policy-driven GL shock
- Shock size:
  - one-standard deviation of residuals from equation (3) corresponds to about the 89th percentile of the shock distribution;
  - years with shocks > one-standard deviation: 1992, 1993, 2001, 2002, 2003, 2009 and 2011.
  - The size at 2009:Q1 was 3.7 standard deviations.
- Key responses to a positive policy-driven GL shock (one-standard deviation):
  - Output: increases by 0.25 percentage point three years after the shock.
  - Stock prices: increase by 6.5 percentage point three years after the shock.
  - Local currency: appreciation.
  - Current account: downward pressure (worsens) due to appreciation and stimulated domestic demand.
  - CPI: upward pressure from liquidity expansion and output; inflationary effect diffused over time by exchange rate appreciation.
  - Policy response: EME authorities cut policy rates and absorb incoming liquidity into foreign reserves; despite policy efforts, local currencies appreciate and stock markets boom.

### EME responses to a market-driven GL shock
- Shock size: one-standard deviation corresponds to the 81st percentile.
- Typical transmission narrative:
  - Increased market liquidity and lower funding costs boost stock prices strongly.
  - Lower funding costs lead policy rates to “catch up” by falling.
  - Working-capital channel: lower funding costs increase export competitiveness and current account balances, leading to foreign reserve accumulation.
  - Disinflationary pressures from lower funding costs largely offset inflationary pressures from rising GDP and depreciations associated with capital outflows.
- Magnitude and persistence of some variable responses to market-driven GL shocks are less pronounced than for policy-driven shocks.

### EME responses to a risk averseness GL shock
- Shock size:
  - one-standard deviation corresponds to the 92nd percentile;
  - three years before and after the GFC saw shocks > this level (1998, 2002, 2011).
  - At the height of the GFC, the shock to risk averseness amounted to 6.5 standard deviations.
- Key responses to heightened risk averseness:
  - Output: weakens.
  - CPI: weakens (lower inflation).
  - Stock prices: fall.
  - Local currencies: depreciate.
  - Current account: rises (possible weakening of domestic absorption and improved price competitiveness from depreciation).
  - Policy response: EMEs can deploy foreign reserves; policy-rate responses vary across EMEs (some raised rates during episodes like the 2013 QE tantrum, others did not; some lowered policy rates with a lag).

### Comparison of IT versus non-IT EMEs
- Non-IT panel construction:
  - Data from 10 sample countries prior to their adoption of IT plus three additional non-IT countries (Hungary, India, Malaysia); country-level demeaning applied.
- Selected three-year forecast error variance decomposition (non-IT panel) — selected figures (percent):
  - Exchange Rate: All Factors 9.0; Policy-Driven Liquidity 3.1; Market-Driven Liquidity 3.1; Risk Averseness 3.1
  - Capital Inflow: All Factors 14.4; Policy-Driven Liquidity 2.6; Market-Driven Liquidity 2.8; Risk Averseness 9.1
  - Stock Price: All Factors 21.9; Policy-Driven Liquidity 11.2; Market-Driven Liquidity 4.6; Risk Averseness 6.1
  - Real GDP: All Factors 14.4; Policy-Driven Liquidity 5.5; Market-Driven Liquidity 4.2; Risk Averseness 4.6
  - CPI: All Factors 8.0; Policy-Driven Liquidity 1.6; Market-Driven Liquidity 2.7; Risk Averseness 3.7
  - Foreign Reserves: All Factors 16.0; Policy-Driven Liquidity 1.9; Market-Driven Liquidity 3.0; Risk Averseness 11.1
  - Overnight Call Rate: All Factors 13.5; Policy-Driven Liquidity 4.2; Market-Driven Liquidity 4.9; Risk Averseness 4.1
- Notable differences and robustness checks:
  - Stock price movements in the non-IT panel are less influenced by GL shocks than in the IT panel.
  - Overnight call rates and foreign reserves in the non-IT panel show disproportionately high variability attributable to GL factors relative to the IT panel.
  - Overall, IT EMEs fare better than non-IT EMEs in terms of expected volatility of macro-financial variables:
    - (i) non-IT countries have higher volatility in exchange rates and capital flows;
    - (ii) policy-variable variability is greater in non-IT EMEs, partly due to higher responsiveness to GL shocks;
    - (iii) non-IT EMEs show higher volatility of output and inflation.
  - Counterfactual analysis suggests higher volatility in non-IT EMEs is largely attributable to greater domestic shocks including monetary policy shocks rather than differences in structural parameters.

### Policy-relevant synthesis and conclusions
- Policy-driven GL expansions:
  - Boost EME stock prices and output via capital inflows and local-currency appreciation.
  - EME authorities typically reduce policy rates and accumulate foreign reserves to absorb incoming liquidity, but partial accommodation (appreciation and equity booms) still occurs.
- Market-driven GL expansions:
  - Lower funding costs and increase market liquidity, lifting stock markets and improving export competitiveness and current account balances.
- Heightened risk averseness:
  - Triggers capital outflows, weakens output, CPI, stock prices and local currencies; raises current accounts; elicits heterogeneous policy responses across EMEs.
- Inflation-targeting (IT) regimes:
  - Associated with lower macro-financial volatility in EMEs relative to non-IT peers, suggesting IT helps reduce susceptibility of inflation, GDP, and policy measures to GL shocks.

### Appendix A — data and correlations among selected variables
- Coverage period: 1990:Q1 to 2014:Q3.
- Countries for underlying AE financial data: U.S., the U.K., Japan, Germany, and France.
- Price measures used: overnight call rates, Treasury bill rates, real interest rates (overnight call rates minus CPI inflation rates), lending rate spread (lending rate minus overnight call rate), stock prices, stock price volatility.
- Quantity measures used: monetary base, private domestic credit, international claims.
- Data frequency and transformations:
  - All data except for interest rates are seasonally adjusted and differenced quarter over quarter if necessary.
  - For EMEs: seasonally adjusted quarter-over-quarter differences used for private domestic credit, international claims, stock prices, and the monetary base.
  - Overnight call rates, Treasury bill rates, and real interest rates are filtered to remove downward trends.
- AE data sources listed: IFS, Bank of Japan, Bloomberg, BIS, DataStream, Bank of England.
- EME data sources listed: DataStream, CEIC, Bloomberg, BIS, IFS.
- Capital flows (EMEs): sums of inbound direct investments, inbound portfolio investments and inbound other investments from the IFS.
- Correlation matrix entries (1990:Q1-2014:Q3) — row-wise with column headers: Market-driven GL, Risk Averseness GL, VIX, DC Growth (G5 Avg), DC Growth (U.S.), SMV (G5 Avg), SMV (U.S.):
  - Market-driven GL: 1.00, −0.04, 0.06, −0.32, 0.56, 0.03, 0.18
  - Risk averseness GL: −0.04, 1.00, 0.75, −0.18, −0.06, 0.79, 0.77
  - VIX: 0.06, 0.75, 1.00, −0.14, 0.02, 0.92, 0.89
  - DC Growth (Avg): −0.32, −0.18, −0.14, 1.00, 0.17, −0.08, −0.20
  - DC Growth (U.S.): 0.56, −0.06, 0.02, 0.17, 1.00, 0.08, 0.14
  - SMV (Avg): 0.03, 0.79, 0.92, −0.08, 0.08, 1.00, 0.92
  - SMV (U.S.): 0.18, 0.77, 0.89, −0.20, 0.14, 0.92, 1.00

### Appendix B — non-IT EME responses (note)
- Appendix B presents non-IT EME responses of key macro-financial variables in percentage change to a positive, one-standard-deviation shock to each GL momentum (figure not reproduced here).

### Appendix C — comparison of forecast errors between IT and non-IT countries
- Methodology summary:
  - Table A2 reports three-year forecast errors (in standard deviation) using the estimated FAVAR model for IT and non-IT panels.
  - Counterfactual exercise for non-IT EMEs: replace A(L), B(L), and covariance of domestic shock tε in the non-IT estimates with IT counterparts to assess contributions to forecast errors.
  - Finding: replacement of the variance-covariance matrix reduces forecast errors more than replacing A(L) or B(L) for most variables.
- Table A2 — Three-year forecast error variance (in standard deviation):
  - Exchange Rate — IT 3.3 — Non-IT 3.7 — A(L) replaced 3.7 — B(L) replaced 3.8 — Cov(tε) replaced 3.5
  - Capital Inflow — IT 2.1 — Non-IT 2.3 — A(L) replaced 2.3 — B(L) replaced 2.3 — Cov(tε) replaced 2.2
  - Stock Price — IT 12.9 — Non-IT 17.0 — A(L) replaced 16.8 — B(L) replaced 16.8 — Cov(tε) replaced 13.8
  - Real GDP — IT 1.0 — Non-IT 1.3 — A(L) replaced 1.3 — B(L) replaced 1.3 — Cov(tε) replaced 1.1
  - CPI — IT 1.2 — Non-IT 1.7 — A(L) replaced 2.0 — B(L) replaced 1.7 — Cov(tε) replaced 1.3
  - Current Account — IT 1.1 — Non-IT 1.5 — A(L) replaced 1.3 — B(L) replaced 1.5 — Cov(tε) replaced 1.4
  - Foreign Reserves — IT 5.2 — Non-IT 8.8 — A(L) replaced 6.6 — B(L) replaced 8.3 — Cov(tε) replaced 7.6
  - Overnight Call Rate — IT 1.5 — Non-IT 2.3 — A(L) replaced 2.9 — B(L) replaced 2.2 — Cov(tε) replaced 1.4
- Interpretive summary:
  - Replacing the variance-covariance matrix (Cov(tε)) in the non-IT panel with the IT counterpart produces the largest reductions in three-year forecast errors for most variables relative to replacing A(L) or B(L).

*Source: wp17222 — References and Appendices (IMF Working Paper content provided).*

### References .............................................................................................................

### wp17222 - References

### I. INTRODUCTION
- Purpose:
  - Investigates how global liquidity (GL) affects macro-financial variables and policies in emerging-market economies (EMEs).
  - Decomposes GL into three momenta: policy-driven liquidity, market-driven liquidity, and risk averseness, and compares responses between inflation targeting (IT) and non-IT EMEs.
- Context and literature:
  - GL has been linked to interest rates, asset prices, inflation, and capital flows (references include Frankel et al., 2002; Edwards, 2010, 2015; Kim and Yang 2009; di Giovanni and Shambaugh, 2008; Valente, 2009; Kim and Shin, 2016; Rigobon and Sack, 2004; Bluedorn and Bowdler, 2011; Ehrmann and Fratzscher 2009; Ammer et al., 2010; Wongswan, 2009; Berger and Harjes, 2009; Cerutti et al., 2014; Cerutti et al., 2015).
  - The U.S. Federal Reserve’s post-GFC policy actions are instrumental in gauging U.S. monetary policy impacts on EMEs (Glick and Leduc, 2012; Bauer and Neely, 2014).
  - Prior approaches to measuring GL include single-country or group measures (D’Agostino and Surico, 2007; Bruno and Shin, 2015; Chen et al., 2012; Eickmeier et al., 2013).
- Key high-level findings previewed:
  - Positive GL shocks from G5 policies or financial markets induce EMEs to reduce policy rates and increase foreign reserves, focusing policy responses on external fronts.
  - Heightened risk averseness that accompanies capital outflows causes EMEs to run down foreign reserves and furnish foreign-currency liquidity.
  - Overall, increases in global liquidity generate positive spillovers on equity markets and output; liquidity reversals due to heightened risk averseness cause negative spillovers.
  - Responses of macro-financial variables to GL shocks are less volatile in IT countries than in non-IT countries.
- Data and sample for EME analysis:
  - Panel FAVAR model incorporates panel data of 10 EMEs for 1995:Q1-2014:Q3.

### II. EMPIRICAL MODELING OF GLOBAL LIQUIDITY TRANSMISSION
- Two-stage empirical approach:
  - Stage 1: Derive GL momenta from a static factor model (principal components) using AEs’ financial data.
    - Equation reference: X_t = Λ F_t + u_t (equation (1)).
  - Stage 2: Add shocks to the momenta (v_t) as exogenous variables to a panel VAR model of EME variables (Y_t), integrating Stock and Watson (2005) dynamic factor ideas into a panel VAR.
    - VAR system referenced: Y_t = A(L) Y_{t−1} + B(L) v_t + ε_t  (equation (2)); F_t = C(L) F_{t−1} + v_t  (equation (3)).
- Model choices and motivations:
  - Use of panel FAVAR extends panel VAR literature to measure dynamic impacts of AE-originating GL momenta on EME macro-financial variables.
  - Factors act as conditioning information (in the spirit of Bernanke et al., 2005) to proxy for otherwise missing variables and to address empirical anomalies.
  - GL momenta shocks (v_t) are assumed independent of individual EME shocks (ε_t).
- Data characteristics and processing choices for GL measurement:
  - Use both price and quantity data to account for monetary policy at the zero lower bound.
  - Factor models used to aggregate large, heterogeneous global data sets; sign restrictions employed for identification.
  - Data filtering and preprocessing:
    - Outliers replaced by medians of previous observations following Stock and Watson (2005).
    - Hodrick-Prescott filter applied to interest rates to remove trending behavior and to render series stationary.
    - Financial variables regressed on GDP growth and producer-price inflation; residuals used to net out macroeconomic fundamentals.
    - Each variable weighted by GDP volume to account for size differences among AEs.

### III. DERIVING GLOBAL LIQUIDITY MOMENTA
- Sample and variables for factor extraction:
  - Quarterly data for 1990:Q1-2014:Q3 from the G5: the U.S., the U.K., France, Germany, and Japan.
  - Nine variables per country: overnight call rates, government bill rates, real exchange rates, lending-rate spread against overnight call rates or policy rates, the monetary base, private domestic credit, international claims, stock prices, and stock market volatility.
- Factor extraction method:
  - Principal components analysis; select three principal components per Ahn and Horenstein (2013).
  - The three factors explain 48 percent of the variability of the underlying AE data.
    - Contribution to variability: policy-driven liquidity explains 16%, market-driven liquidity 13%, and risk averseness 18%.
  - Minimal sign restrictions imposed for identification; selection of factor candidate closest to median as per Fry and Pagan (2011).
- Sign restrictions used (applied to the U.S. only in pinning down Λ in equation (1)):
  - Policy-Driven Liquidity:
    - M0 (+)
    - Lending rate spread (+)
    - Treasury bill rate (−)
    - Real interest rate (−)
  - Market-Driven Liquidity:
    - Private Domestic Credit (+)
    - Stock Price (+)
  - Risk Averseness:
    - Private Domestic Credit (−)
    - Stock Volatility (+)
  - Note: “A positive sign attached to a variable means that the GL momentum in that row rises with the level of the variable.”
- Economic interpretation and rationale for restrictions:
  - Policy-driven liquidity identified with monetary policy stance and base money expansions (M0). Base money plays a prominent role under unconventional monetary policy and at the zero lower bound.
  - Lending-rate spread restriction acknowledges that, prior to unconventional policy, expansion in policy-driven liquidity can widen the spread by reducing funding cost of banks through short-term instruments.
  - Market-driven liquidity tied to endogenous banking-sector credit expansion and higher stock prices associated with market liquidity.
  - Risk averseness captures investor risk appetite; higher risk averseness is associated with reduced private credit and increased stock volatility.
- Factor construction and qualitative dynamics:
  - Figures (not reproduced here) show that policy-driven liquidity expanded rapidly after the early-2000s recessions until the Fed raised its policy rate from 1 percent in mid-2004 to 5.25 percent in June 2006, then decreased after mid-2007.
  - During that tightening cycle, market-generated liquidity continued to expand to its peak in early- (text continued beyond provided excerpt).

*Source: wp17222 - References (IMF Working Paper content provided).*

### 2007. Meanwhile, the risk averseness of market participants declined. The confluence of these

### wp17222 - 2007. Meanwhile, the risk averseness of market participants declined. The confluence of these

### Global liquidity momenta before and during the GFC
- Three momenta (policy-driven, market-driven, and risk averseness) suggest an overall easing in GL for several years in the run-up to the GFC.
- With the culmination of the GFC in 2008:Q4:
  - the policy-driven factor dropped to its lowest point;
  - the market-driven factor slid down with the Lehman Brothers collapse (2008:Q3);
  - the risk averseness factor reached its highest point.

### Catalyzing effects during the GFC quarter
- The debacle of the GFC appears to accompany the catalyzing effects of:
  - sudden deteriorations in major central banks’ liquidity supply;
  - sudden deteriorations in market-driven liquidity;
  - a sharp spike in risk averseness.
- First, the supply of liquidity by major central banks during this quarter was not sufficient to backstop the sudden deterioration of output and weak prices.
- Especially, real interest rates (after controlling for producer price inflation and output growth) pointed to a record high.

### Methodological and annotation notes
- Notes reference Figure 1. Global liquidity momenta and state that three momenta are derived from principal component analysis and identified by the sign restrictions in Table.
- Footnote text: The validity of these restrictions remains intact even after the federal funds rate was set at 025 basis points in December 2008, while the bank prime loan rate in the U.S. remained unchanged at 3.25 percent for 2009:Q1-2014:Q3.

*Source: wp17222 - 2007. Meanwhile, the risk averseness of market participants declined. The confluence of these (IMF working paper PDF).*

### 1. All momenta are standardized. The solid blue lines depict the medians of all candidates as the corresponding

### wp17222 - 1. All momenta are standardized. The solid blue lines depict the medians of all candidates as the corresponding

### Global liquidity (GL) momenta — definitions and key properties
- Three GL momenta are extracted from advanced-economy (AE) monetary and financial series and standardized.
- The risk averseness GL factor is a composite extracted from nine financial variables including cross-border and domestic-credit flows as well as stock price volatility for the G5; its simple correlation with the CBOE Volatility Index (VIX) is 0.75.
- The market-driven GL factor (second panel) is positively correlated with the domestic credit growth of the U.S., which is negatively associated with domestic credit growth of other G5 countries except the U.K.
- The correlation between the market-driven and risk averseness GL factors is very low at 0.04.
- Historical episodes noted: onset of the 1991 recession (1991:Q1), dot-com bubble bust (2000:Q1), Lehman Brothers Bankruptcy (2008:Q3). At the onset of the GFC (2009:Q1) the size of the policy-driven GL shock was as big as 3.7 standard deviations.

### Data, sample, and econometric specification
- EME panel sample period: 1995:Q1-2014:Q3.
- EME countries in the IT panel: the Czech Republic, Hungary, Israel, Korea, Mexico, the Philippines, Poland, Romania, Thailand, and Turkey.
- Model variables (quarterly, eight variables):
  - real GDP growth
  - CPI inflation
  - stock price growth
  - nominal effective exchange rate (NEER) growth
  - current account balance (as percent of GDP)
  - capital inflows (as percent of GDP)
  - foreign reserves (as percent of GDP)
  - overnight call rates
- Lag structure selected by the Akaike information criterion: two autoregressive lags and one contemporaneous and two lagged terms of the GL shock.
- Estimation: equation-by-equation least squares for the EME panel.
- Error bands for impulse responses constructed via a modified Bayesian Monte Carlo integration method that (i) resamples GL factors, (ii) re-estimates equation (2) per realization, (iii) randomly picks a parameter set via Bayesian Monte Carlo integration, (iv) draws an impulse response; confidence intervals and mean values obtained from the empirical distribution of impulse responses.
- For real GDP, CPI, stock prices and exchange rates, responses of respective growth variables are accumulated from the model to obtain responses in their levels.

### Forecast error variance decomposition — aggregate findings
- About 10 percent of real GDP growth variability for the three-year horizon is attributable to GL momenta.
- About 13 percent of capital inflows variability is attributable to GL momenta.
- About 30 percent of stock price variability is attributable to GL momenta.
- NEER variability appears mostly driven by domestic elements; spillovers from AEs to EMEs’ NEER are generally smaller than spillovers to exchange rates against the U.S. dollar.
- Policy-driven and risk averseness GL momenta play more important roles in movements of EME variables than the market-driven GL momentum, except for current account, foreign reserves, and overnight call rates which show substantial variability from the market-driven GL momentum.
- Policy-driven GL dominates market-driven GL in impacts on EME key variables except current account.

### Table of three-year forecast error variance decomposition (percent)
- Aggregate contributions (Table 2)
  - Exchange Rate: All Factors 10.0; Policy-Driven Liquidity 2.9; Market-Driven Liquidity 1.9; Risk Averseness 5.2
  - Capital Inflow: All Factors 13.2; Policy-Driven Liquidity 2.5; Market-Driven Liquidity 1.9; Risk Averseness 8.8
  - Stock Price: All Factors 29.6; Policy-Driven Liquidity 12.1; Market-Driven Liquidity 6.9; Risk Averseness 10.9
  - Real GDP: All Factors 10.5; Policy-Driven Liquidity 3.3; Market-Driven Liquidity 2.3; Risk Averseness 4.9
  - CPI: All Factors 6.8; Policy-Driven Liquidity 2.3; Market-Driven Liquidity 1.8; Risk Averseness 2.6
  - Current Account: All Factors 11.8; Policy-Driven Liquidity 3.4; Market-Driven Liquidity 5.1; Risk Averseness 3.4
  - Foreign Reserves: All Factors 10.7; Policy-Driven Liquidity 4.6; Market-Driven Liquidity 3.9; Risk Averseness 2.4
  - Overnight Call Rate: All Factors 7.1; Policy-Driven Liquidity 3.0; Market-Driven Liquidity 2.7; Risk Averseness 1.5

### EME responses to a policy-driven GL shock
- Shock size: one-standard deviation of residuals from equation (3) corresponds to about the 89th percentile of the shock distribution; years with shocks > one-standard deviation: 1992, 1993, 2001, 2002, 2003, 2009 and 2011. The size at 2009:Q1 was 3.7 standard deviations.
- Key responses to a positive policy-driven GL shock (one-standard deviation):
  - Output: increases by 0.25 percentage point three years after the shock.
  - Stock prices: increase by 6.5 percentage point three years after the shock.
  - Local currency: appreciation.
  - Current account: downward pressure (worsens) due to appreciation and stimulated domestic demand.
  - CPI: upward pressure from liquidity expansion and output; inflationary effect diffused over time by exchange rate appreciation.
  - Policy response: EME authorities cut policy rates and absorb incoming liquidity into foreign reserves; despite policy efforts, local currencies appreciate and stock markets boom.

### EME responses to a market-driven GL shock
- Shock size: one-standard deviation corresponds to the 81st percentile.
- Typical transmission narrative:
  - Increased market liquidity and lower funding costs boost stock prices strongly.
  - Lower funding costs lead policy rates to “catch up” by falling.
  - Working-capital channel: lower funding costs increase export competitiveness and current account balances, leading to foreign reserve accumulation.
  - Disinflationary pressures from lower funding costs largely offset inflationary pressures from rising GDP and depreciations associated with capital outflows.
- Magnitude and persistence of some variable responses to market-driven GL shocks are less pronounced than for policy-driven shocks.

### EME responses to a risk averseness GL shock
- Shock size: one-standard deviation corresponds to the 92nd percentile; three years before and after the GFC saw shocks > this level (1998, 2002, 2011). At the height of the GFC, the shock to risk averseness amounted to 6.5 standard deviations.
- Key responses to heightened risk averseness:
  - Output: weakens.
  - CPI: weakens (lower inflation).
  - Stock prices: fall.
  - Local currencies: depreciate.
  - Current account: rises (possible weakening of domestic absorption and improved price competitiveness from depreciation).
  - Policy response: EMEs can deploy foreign reserves; policy-rate responses vary across EMEs (some raised rates during episodes like the 2013 QE tantrum, others did not; some lowered policy rates with a lag).

### Comparison of IT versus non-IT EMEs
- Non-IT panel construction: data from 10 sample countries prior to their adoption of IT plus three additional non-IT countries (Hungary, India, Malaysia); country-level demeaning applied.
- Table 3 (non-IT three-year forecast error variance decomposition) selected figures (percent):
  - Exchange Rate: All Factors 9.0; Policy-Driven Liquidity 3.1; Market-Driven Liquidity 3.1; Risk Averseness 3.1
  - Capital Inflow: All Factors 14.4; Policy-Driven Liquidity 2.6; Market-Driven Liquidity 2.8; Risk Averseness 9.1
  - Stock Price: All Factors 21.9; Policy-Driven Liquidity 11.2; Market-Driven Liquidity 4.6; Risk Averseness 6.1
  - Real GDP: All Factors 14.4; Policy-Driven Liquidity 5.5; Market-Driven Liquidity 4.2; Risk Averseness 4.6
  - CPI: All Factors 8.0; Policy-Driven Liquidity 1.6; Market-Driven Liquidity 2.7; Risk Averseness 3.7
  - Foreign Reserves: All Factors 16.0; Policy-Driven Liquidity 1.9; Market-Driven Liquidity 3.0; Risk Averseness 11.1
  - Overnight Call Rate: All Factors 13.5; Policy-Driven Liquidity 4.2; Market-Driven Liquidity 4.9; Risk Averseness 4.1
- Notable differences and robustness checks:
  - Stock price movements in the non-IT panel are less influenced by GL shocks than in the IT panel.
  - Overnight call rates and foreign reserves in the non-IT panel show disproportionately high variability attributable to GL factors relative to the IT panel.
  - Overall, IT EMEs fare better than non-IT EMEs in terms of expected volatility of macro-financial variables: (i) non-IT countries have higher volatility in exchange rates and capital flows; (ii) policy-variable variability is greater in non-IT EMEs, partly due to higher responsiveness to GL shocks; (iii) non-IT EMEs show higher volatility of output and inflation.
  - Counterfactual analysis suggests higher volatility in non-IT EMEs is largely attributable to greater domestic shocks including monetary policy shocks rather than differences in structural parameters.

### Synthesis of policy-relevant conclusions
- Policy-driven GL expansions:
  - Boost EME stock prices and output via capital inflows and local-currency appreciation.
  - EME authorities typically reduce policy rates and accumulate foreign reserves to absorb incoming liquidity, but partial accommodation (appreciation and equity booms) still occurs.
- Market-driven GL expansions:
  - Lower funding costs and increase market liquidity, lifting stock markets and improving export competitiveness and current account balances.
- Heightened risk averseness:
  - Triggers capital outflows, weakens output, CPI, stock prices and local currencies; raises current accounts; elicits heterogeneous policy responses across EMEs.
- Inflation-targeting (IT) regimes:
  - Associated with lower macro-financial volatility in EMEs relative to non-IT peers, suggesting IT helps reduce susceptibility of inflation, GDP, and policy measures to GL shocks.

*Source: wp17222 - 1. All momenta are standardized. The solid blue lines depict the medians of all candidates as the corresponding*

### REFERENCES

### REFERENCES

### Key themes in the cited literature
- Asset pricing and liquidity risk: Acharya and Pedersen (2005) “Asset Pricing with Liquidity Risk.”  
- Factor determination and factor models: Ahn and Horenstein (2013); Bai and Ng (2002); Breitung and Eickmeier (2005); Stock and Watson (2003, 2005).  
- International transmission of U.S. monetary policy and global spillovers: Ammer, Vega, and Wongswan (2010); Bernanke, Boivin, and Eliasz (2005); Kim (2001); Kim and Yang (2009); Glick and Leduc (2012); Wongswan (2009); Valente (2009).  
- Global liquidity, cross-border banking, and capital flows: Bruno and Shin (2015); Cerutti, Claessens, and Ratnovski (2014); Chen et al. (2012); Eickmeier, Gambacorta, and Hofmann (2013).  
- Monetary policy frameworks and inflation targeting in emerging markets: Brito and Bystedt (2010); Gonçalves and Salles (2008); Roger (2010); Rose (2007, 2014); Hammond (2012).  
- Methods for VAR, FAVAR, DSGE, and Bayesian estimation: Bernanke, Boivin, and Eliasz (2005); Christiano, Trabandt, and Walentin (2011); Kadiyala and Karlsson (1997); Kloek and van Dijk (1978); Kilian (1998); Sims and Zha (1999); Fry and Pagan (2011).  
- Financial frictions, systemic crises, and banking crises databases: Brunnermeier, Eisenbach, and Sannikov (2012); Laeven and Valencia (2013).  
- Studies on international monetary coordination, Mundellian trilemma, and policy independence: Rey (2016); Taylor (2013a, 2013b, 2016); Edwards (2010, 2015); Frankel, Schmukler, and Serven (2002).  
- Empirical studies on global liquidity measures, drivers, and forecasting asset prices: D’Agostino and Surico (2007); Darius and Radde (2010); Sun and Psalida (2011); Chen et al. (2012).

### Appendix A — Macro-Financial Data Used in Estimation and Correlations among Selected Variables
- Coverage period: 1990:Q1 to 2014:Q3.  
- Countries for underlying AE financial data: U.S., the U.K., Japan, Germany, and France.  
- Price measures of liquidity used: overnight call rates, Treasury bill rates, real interest rates (overnight call rates minus CPI inflation rates), lending rate spread (lending rate minus overnight call rate), stock prices, stock price volatility.  
- Quantity measures of liquidity used: monetary base, private domestic credit, international claims.  
- Data frequency and transformations:
  - All data except for interest rates are seasonally adjusted and differenced quarter over quarter if necessary.  
  - For EMEs: seasonally adjusted quarter-over-quarter differences used for private domestic credit, international claims, stock prices, and the monetary base.  
  - Overnight call rates, Treasury bill rates, and real interest rates are filtered to remove downward trends.  
- AE data sources (as listed): IFS, Bank of Japan, Bloomberg, BIS, DataStream, Bank of England.  
- EME data sources (as listed): DataStream, CEIC, Bloomberg, BIS, IFS.  
- Capital flows (EMEs): sums of inbound direct investments, inbound portfolio investments and inbound other investments from the IFS.

### Appendix A — Correlations (Table A1)
- Notes: Simple correlations for the 1990:Q1-2014:Q3 period between market-driven GL factor, risk averseness GL factor, VIX, domestic credit (DC) of the U.S. and G5 average (Avg), and stock market volatility (SMV) of the U.S. and G5 average.
- Reported correlation matrix entries (row-wise with corresponding column headers: Market-driven GL, Risk Averseness GL, VIX, DC Growth (G5 Avg), DC Growth (U.S.), SMV (G5 Avg), SMV (U.S.)):
  - Market-driven GL: 1.00, −0.04, 0.06, −0.32, 0.56, 0.03, 0.18
  - Risk averseness GL: −0.04, 1.00, 0.75, −0.18, −0.06, 0.79, 0.77
  - VIX: 0.06, 0.75, 1.00, −0.14, 0.02, 0.92, 0.89
  - DC Growth (Avg): −0.32, −0.18, −0.14, 1.00, 0.17, −0.08, −0.20
  - DC Growth (U.S.): 0.56, −0.06, 0.02, 0.17, 1.00, 0.08, 0.14
  - SMV (Avg): 0.03, 0.79, 0.92, −0.08, 0.08, 1.00, 0.92
  - SMV (U.S.): 0.18, 0.77, 0.89, −0.20, 0.14, 0.92, 1.00

### Appendix B — Responses of Non-IT EMEs to GL Shocks
- Notes (as provided): Figure shows the non-IT EME responses of key macro-financial variables in percentage change to a positive, one-standard-deviation shock to each GL momentum. See notes (a) and (b) to Figure 3. (Figure not reproduced here.)

### Appendix C — Comparison of Forecast Errors between IT and Non-IT Countries
- Methodology summary (as provided):
  - Table A2 reports three-year forecast errors (in standard deviation) using the estimated FAVAR model comprising equations (2) and (3) for the IT and non-IT panels (first two columns).  
  - Counterfactual exercise for non-IT EMEs: in the estimated result of equation (2) of the non-IT panel, the estimate of A(L) matrix (third column), B(L) matrix (fourth column), and covariance of domestic shock tε (fifth column) are replaced by the counterpart of the IT panel estimation.  
  - Finding from the counterfactual: replacement of the variance-covariance matrix reduces the forecast errors more than other alternatives for most variables.
- Table A2 — Three-year forecast error variance (in standard deviation)
  - Variable — IT — Non-IT — A(L) replaced — B(L) replaced — Cov(tε) replaced
  - Exchange Rate — 3.3 — 3.7 — 3.7 — 3.8 — 3.5
  - Capital Inflow — 2.1 — 2.3 — 2.3 — 2.3 — 2.2
  - Stock Price — 12.9 — 17.0 — 16.8 — 16.8 — 13.8
  - Real GDP — 1.0 — 1.3 — 1.3 — 1.3 — 1.1
  - CPI — 1.2 — 1.7 — 2.0 — 1.7 — 1.3
  - Current Account — 1.1 — 1.5 — 1.3 — 1.5 — 1.4
  - Foreign Reserves — 5.2 — 8.8 — 6.6 — 8.3 — 7.6
  - Overnight Call Rate — 1.5 — 2.3 — 2.9 — 2.2 — 1.4
- Interpretive summary (as provided in source):
  - The replacement of the variance-covariance matrix (Cov(tε)) in the non-IT panel with the IT counterpart produces the largest reductions in three-year forecast errors for most variables relative to replacing A(L) or B(L).

*Source: REFERENCES and APPENDIX sections from wp17222 - REFERENCES (1990:Q1–2014:Q3 data and tables as provided).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2017/wp17222.pdf_