## _wp15256 — Section 1–4 (If the Fed Acts, How Do You React? The Liftoff Effect on Capital Flows)

## Source details

**Canonical URL:** [_wp15256 — Section 1–4 (If the Fed Acts, How Do You React? The Liftoff Effect on Capital Flows)](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15256.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15256.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15256.pdf.json)

---

### Abstract, motivation, and contribution
- After more than six years of ultra-low interest rates, a Fed liftoff is considered imminent; the paper measures the spillover or “liftoff” effect of previous Fed liftoffs on capital flows using history.
- Sample and scope:
  - Dynamic panel of 48 countries: 27 advanced market (AM) and 21 emerging market (EM) economies.
  - Sample period: 1982Q1-2006Q4.
  - Quarterly frequency to capture timing of liftoff effects.
- Key high-level findings:
  - The liftoff effect on capital flows (total private, portfolio) is significantly higher for EM than AM.
  - EM capital flows are hit indiscriminately one quarter before liftoff, implying markets price in liftoff before the event.
  - Over time, variation among EM increases as country-specific policy responses/frameworks can dampen market reactions to some extent.
- Contribution:
  - Focuses specifically on liftoff episodes (first increases that start hiking cycles) and interacts liftoff timing dummies with push/pull variables to identify extra sensitivity during liftoffs.

### Data, empirical strategy, and liftoff framework
- Push (external) variables included:
  - U.S. interest rate; U.S. consumer prices (year-over-year growth); U.S. GDP Growth (year-over-year growth); risk aversion index (volatility of S&P 500 index returns; VIX in robustness); commodity price index (year-over-year growth).
  - S&P index volatility: quarterly average of twelve-month rolling standard deviation of S&P index annual returns.
- Pull (domestic) variables included:
  - Domestic interest rates; consumer prices (year-over-year growth); real effective exchange rate (year-over-year growth); GDP growth (year-over-year growth).
- Liftoff episodes identified (first hikes after flat/declining policy rate):
  - Five episodes and liftoff quarters: March 1983 (Q1 1983); January 1987 (Q1 1987); February 1994 (Q1 1994); June 1999 (Q2 1999); June 2004 (Q2 2004).
- Three-step empirical approach:
  1. Unconstrained generic model for capital flows: fixed effects, push/pull factors, lagged dependent variables.
  2. Saturated model: introduce five liftoff time dummies (d_{-2}, d_{-1}, d_{0}, d_{+1}, d_{+2}) and interact with variables (initially the U.S. interest rate).
  3. Restricted model: delete nonsignificant interactions via log-likelihood ratio tests.

### Main empirical findings and quantified liftoff effects
- EM vs AM:
  - Liftoff effect significantly larger for EM than AM; liftoff effect largely absent or smaller in AM.
  - EM hit predominantly one quarter before liftoff (pricing-in).
- Timing:
  - Significant liftoff effect of the U.S. interest rate one quarter before liftoff for EM for both total and portfolio flows.
  - Liftoff-quarter effects often insignificant, consistent with markets pricing in hikes in advance.
- Estimated magnitudes (selected):
  - For every 100bps of the U.S. interest rate, one quarter prior to liftoff:
    - 0.72 percent of GDP net total outflow.
    - 0.33 percent of GDP net portfolio outflow.
  - Simulated one standard deviation shock in the U.S. interest rate (exercise 1) decreases flows by:
    - 2.0 percent of GDP for total flows.
    - 0.9 percent of GDP for portfolio flows.
    - These equal around 0.3 standard deviations of the respective data.
  - Using average U.S. rate values from previous five episodes (exercise 2) implies prior liftoffs resulted in net outflows of:
    - 3.5 percent of GDP for total flows.
    - 1.6 percent of GDP for portfolio flows.
    - These equal 0.5 and 0.4 standard deviations, respectively.
- Representative coefficient estimates (EM):
  - U.S. rate × one quarter before liftoff: -0.890*** (standard error 0.340) and -0.721** (standard error 0.310) in different specifications (Table 1 columns reported).
  - U.S. rate × two quarters before liftoff: -0.384* (standard error 0.228) in one specification.
  - U.S. rate × one quarter after liftoff: positive coefficients reported (e.g., 0.436*, standard error 0.223), indicating partial payback post-liftoff though smaller than pre-liftoff loss.

### Robustness checks and data handling
- Robustness variants:
  - Alternative risk-aversion measure: VIX used.
  - Instrumenting potentially endogenous regressors (real effective exchange rates, inflation) with lags in EM regressions.
  - Handling extreme data points (notably Brazil): multiple exclusion tests (policy rate > 30 percent; inflation > 30 percent; excluding Brazil; using Brazil series since 1998).
- Robustness outcomes:
  - Liftoff effect on EM portfolio flows robust to all alternative specifications; coefficients often larger (reported results conservative).
  - EM total flows results less robust in some specifications.
  - Some policy-rate and inflation effects lose significance when extremes are excluded.

### Domestic policies and frameworks — empirical findings
- Policy dimensions examined:
  - Domestic policy rate (policy response).
  - Monetary Policy Independence (MI): reciprocal of 12-month rolling correlation between domestic policy rate and U.S. policy rate; index normalized between 0 and 1 (higher = more independence). If domestic rate constant over 12 months, MI assigned 0.5.
  - Exchange Rate Stability (ERS): 12-month rolling standard deviation of monthly exchange rate change vs USD, normalized between 0 and 1 (higher = more stable/fixed).
  - Capital Account Openness: Chinn-Ito Index normalized 0–1 (higher = more open).
  - Budget Surplus (quarter-over-quarter, share of GDP).
- Statistical summary:
  - Policy variables often insignificant during normal times but significant when interacted with liftoff dummies — policies/frameworks matter primarily during liftoff episodes.
  - Monetary policy independence:
    - Keeping MI helps increase net capital flows during liftoff episodes. Positive liftoff-interaction coefficients for MI reported for EM (e.g., large positive coefficients in timing columns with statistical significance in several specifications; Table 5 excerpts).
  - Domestic policy rates:
    - Weak/no consistent evidence that raising policy rates during liftoff episodes mitigates outflows; significance often disappears under robustness checks.
  - Budget surplus (quarter-over-quarter):
    - Improvement in quarterly budget surplus significant when interacted with liftoff dummies; robust to checks.
    - Effects do not hold when budget surplus measured year-over-year.
  - Capital account openness:
    - No evidence that more open capital accounts lead to worse outcomes prior to/during liftoff; openness associated with quicker recovery one or two quarters after liftoff.
  - Exchange rate stability:
    - Mixed results for EM: more stable exchange rates associated with more total outflows two quarters before liftoff and during liftoff quarter; also associated with portfolio inflows one quarter prior and two quarters post liftoff in some specifications.
    - For AM, more flexible regimes fare better post-liftoff for attracting flows.

### Magnitude assessment of policy responses (two exercises; EM focus)
- Exercise 1 — one standard deviation shocks (EM, selected results):
  - Total flows:
    - Before liftoff: net outflow from one s.d. U.S. rate shock ≈ 2.4 percent of GDP (0.3 s.d. of the data).
    - A one s.d. quarterly improvement in budget surplus could attract net inflow ≈ 1.8 percent of GDP before liftoff.
    - During liftoff quarter: quarterly budget improvement attracts ≈ 3.7 percent of GDP.
    - After liftoff: MI attracts ≈ 2.1 percent of GDP; open capital accounts attract ≈ 1.3 percent of GDP.
  - Portfolio flows:
    - Before liftoff: net outflow from one s.d. U.S. rate shock ≈ 1.7 percent of GDP (0.5 s.d. of the data).
    - During liftoff quarter: MI can attract ≈ 0.6 percent of GDP; more fixed ERS associated with ≈ 0.6 percent of GDP net outflow.
    - After liftoff: budget improvement attracts ≈ 1.3 percent of GDP; ERS attracts ≈ 0.8 percent; openness attracts ≈ 0.5 percent.
- Exercise 2 — using average values from five previous liftoff episodes (EM, selected results):
  - Total flows:
    - Before liftoff: net outflow due to the U.S. rate ≈ 4.4 percent of GDP (0.6 s.d. of total flows).
    - Policy variables on average not strong enough to fully offset the U.S. rate impact before liftoff.
    - During liftoff quarter: budget improvement can yield ≈ 1.4 percent of GDP extra inflow.
    - Post liftoff: MI attracts ≈ 3.4 percent of GDP; capital account openness attracts ≈ 1.3 percent of GDP.
  - Portfolio flows:
    - Before liftoff: U.S. rate impact ≈ -3.0 percent of GDP (0.8 s.d. of data).
    - MI and ERS help before liftoff but cannot offset U.S. rate impact.
    - During liftoff quarter: MI can attract ≈ 0.3 percent of GDP.
    - Post liftoff: openness ≈ 0.5 percent; ERS ≈ 0.2 percent; budget improvement ≈ 0.1 percent.

### Policy implications and concluding thoughts
- Core takeaways:
  - EM capital flows are substantially more sensitive to Fed liftoffs than AM, with most EM outflows occurring one quarter before liftoff as markets price in hikes.
  - Domestic policies and frameworks can only partly dampen negative liftoff effects; some measures help more post-liftoff than pre-liftoff.
- Specific policy messages:
  - Monetary policy independence can help attract capital flows and provide policy leeway, but increasing MI requires time and may be difficult during currency pressure.
  - Fiscal policy: quarter-over-quarter improvement in budget surplus can signal commitment and reduce outflows in credible countries; effect is timely and robust in regressions.
  - Capital account openness does not worsen pre-liftoff vulnerability and can aid recovery post-liftoff.
  - Trade-offs: measures that attract capital flows (e.g., fiscal consolidation) may be pro-cyclical amid slowing EM growth.
- Caveats:
  - Analysis covers five policy areas and five historical liftoff episodes up to 2006; results may not capture post-2006/global financial crisis dynamics.
  - Other vulnerability factors (elevated current account deficits, high inflation, weak growth, low reserves) and macroprudential policies are relevant but outside the paper’s detailed scope.

*Source: WP/15/256 — If the Fed Acts, How Do You React? The Liftoff Effect on Capital Flows, Swarnali Ahmed, IMF Working Paper (sections 1–4 content provided).*

### Section 1

### _wp15256 - Section 1

### Abstract and key takeaway
- After more than six years of ultra-low interest rates, a Fed liftoff (rate hike) is just a matter of time; the paper uses history to measure the spillover or “liftoff” effect of previous Fed liftoffs on capital flows.
- Using a dynamic panel framework covering 48 countries (27 AM, 21 EM) over the period 1982Q1-2006Q4, the paper finds:
  - The liftoff effect on capital flows (total private, portfolio) is significantly higher for emerging market economies (EM) than advanced market economies (AM).
  - EM capital flows are hit indiscriminately one quarter before liftoff, suggesting markets usually price in the liftoff before the actual event.
  - Over time, variation among EM increases as country-specific policy responses/frameworks can to some extent dampen market reactions.
- The findings are similar to events during the taper tantrum episode and suggest history can provide useful guidance even when current circumstances differ.

### Introduction: motivation and contribution
- Motivation:
  - Concern over ripple effects on EM from an imminent Fed liftoff, especially given declining EM growth prospects and memories of the 2013 taper tantrum.
- Contribution:
  - First attempt (to the author's knowledge) to focus specifically on liftoff episodes to determine the extra sensitivity of capital flows during such times.
  - Builds on literature on determinants of capital flows, distinguishing push (external) and pull (domestic) factors, and complements work on monetary policy spillovers.

### Data, sample, and estimation framework
- Sample and frequency:
  - Dynamic panel of 48 countries: 27 advanced market (AM) and 21 emerging market (EM) economies.
  - Sample period: 1982Q1-2006Q4.
  - Quarterly data used to capture timing of liftoff effects.
- Estimation and inference:
  - Dynamic panel framework with lagged dependent variables (or AR() errors where appropriate).
  - Robust standard errors: White cross-section covariance method.
  - Breusch-Godfrey tests used where needed to check for serial correlation.
- Push (external) variables included:
  - U.S. interest rate, U.S. consumer prices (year-over-year growth), U.S. GDP Growth (year-over-year growth), index for risk aversion (volatility of S&P 500 index returns), commodity price index (year-over-year growth).
  - Note: S&P index volatility is the quarterly average of twelve-month rolling standard deviation of S&P index annual returns; VIX used in robustness checks.
- Pull (domestic) variables included:
  - Domestic interest rates, consumer prices (year-over-year growth), real effective exchange rate (year-over-year growth), GDP growth (year-over-year growth).

### Empirical strategy — three-step approach
- Step 1: Build a generic model for capital flows (unconstrained specification) where net inflows (total private or portfolio) as a share of nominal GDP are a function of fixed effects, external (push) factors, domestic (pull) factors, and s lagged dependent variables.
- Step 2: Construct a saturated model by introducing lifecycle (liftoff) interactions:
  - Definition: liftoff = first increase in the U.S. interest rate that starts a hiking cycle after a period of declining or constant policy rate.
  - Five liftoff episodes identified (1982–2006): March 1983, January 1987, February 1994, June 1999, June 2004.
  - Corresponding liftoff quarters: Q1 1983, Q1 1987, Q1 1994, Q2 1999, Q2 2004.
  - Five time dummies created for each liftoff episode:
    - Two quarters before liftoff = 1 and =0 otherwise.
    - One quarter before liftoff = 1 and =0 otherwise.
    - Liftoff quarter = 1 and =0 otherwise.
    - One quarter post liftoff = 1 and =0 otherwise.
    - Two quarters post liftoff = 1 and =0 otherwise.
  - Strategy: interact these time dummies with relevant variables (initially the U.S. interest rate) to capture the liftoff effect and its timing.
- Step 3: Reduce the saturated model by deleting nonsignificant interaction terms after appropriate tests to obtain a restricted model.

### Liftoff effect: measurement and timing
- The liftoff effect of a variable is measured by the coefficient on the interactive term of the liftoff time dummy and that variable.
- Time dummies allow identification of whether the liftoff effect occurs before, during, and/or after liftoff; pricing-in by markets is explicitly tested via pre-liftoff dummies.
- Algebraic notation in the paper denotes the five dummy variables as d_j with j = -2, -1, 0, +1, +2 representing two quarters before liftoff through two quarters after liftoff.

### Main empirical findings reported in Section I and II preview
- Liftoff effect on capital flows is significantly higher for EM than AM.
- EM capital flows show a pronounced response one quarter before liftoff, indicating markets price in the liftoff prior to the actual rate hike.
- Before liftoff, domestic policy cannot do much to prevent the liftoff effect; there is no evidence that more open capital accounts make countries more susceptible pre-liftoff.
- Post-liftoff, some country-specific policy responses and frameworks can dampen negative market reactions; examples highlighted later in the paper include maintaining monetary policy independence and improving near-term budget deficits, while countries with open capital accounts seem to recover more quickly over time.

### Relation to literature and methodological choices
- Places the analysis within three related strands:
  - Studies of extreme capital flow episodes (sudden stops and surges).
  - Studies using full-sample approaches to identify longer-term determinants of capital flows.
  - Studies on spillovers of monetary policy shocks across asset classes.
- Methodological choices:
  - Unconstrained model presented (separate components) rather than constrained model (real rate differentials and GDP growth differentials).
  - Quarterly frequency chosen due to data constraints on higher-frequency coverage, especially for earlier years in EM.

*Source: WP/15/256 — If the Fed Acts, How Do You React? The Liftoff Effect on Capital Flows, Swarnali Ahmed, IMF Working Paper (section content provided).*

### Section 2

### _wp15256 - Section 2

### Step 3: The Restricted Model from the Saturated Model
- Not all liftoff variables (time dummies interacted with the U.S. rate) are significant.
- Insignificant interactive dummies and corresponding time dummies are removed using a log-likelihood ratio test to choose the restricted model over the saturated model.
- Methodological inspiration and references: Cox and Snell (1974); Hoover and Perez (1999); Hendry and Krolzig (2005); exposition in Hendry and Nielsen (2007).

### Robustness Checks
- Alternative global risk-aversion measure: VIX index used instead of volatility of S&P 500 returns for robustness.
- Potential endogeneity: real effective exchange rates and inflation rates (when used as dependent variables, particularly in EM) are instrumented using their lags for EM regressions.
  - Some coefficients increase in magnitude and significance when instruments are used.
- Handling extreme data points in EM (noting Brazil’s longer series for policy rate/SELIC contains extreme points):
  - Alternative specifications run as robustness checks:
    - 1) Excluding data points where policy rate > 30 percent.
    - 2) Excluding data points where policy rates or inflation rates > 30 percent.
    - 3) Excluding data points where policy rates or inflation or commodity prices > 30 percent.
    - 4) Excluding Brazil from the list of countries.
    - 5) Using the Brazil policy rate series starting from 1998 (shorter series).
- Main findings are robust to these specifications unless otherwise stated; discrepancies are discussed where robustness fails.

### Model Results (EM focus)
- Dependent variables:
  - Net total flows, as a share of GDP.
  - Net portfolio flows, as a share of GDP.
- Independent variables grouped into:
  - External factors (push factors).
  - Domestic factors (pull factors).
  - Liftoff variables (dummies interacted with U.S. interest rate).
- Key empirical findings:
  - Risk aversion and domestic factors are significant determinants of capital flows in initial specifications.
  - Excluding extreme data points can render policy rates and inflation insignificant for total flows, and inflation and growth insignificant for portfolio flows.
  - Coefficients of interactive terms (liftoff effects) are similar across specifications including/excluding extreme data points.
  - Liftoff effect timing:
    - Significant liftoff effect of the U.S. interest rate one quarter before liftoff.
    - Interaction capturing one quarter before liftoff and the U.S. interest rate is significant for both total and portfolio flows.
  - Magnitudes:
    - For every 100bps of the U.S. interest rate, there is 0.72 and 0.33 percent of GDP of net total and net portfolio outflow, respectively, one quarter prior to the liftoff.
- AM (Advanced Markets) comparison:
  - Liftoff effect absent or smaller in AM (table 2; figures 3 and 4).
  - Possible reasons:
    - Divergent reactions during risk sell-off episodes (EM typically experience outflows; AM, especially safe havens, may receive inflows).
    - Currency denomination differences: AM tend to borrow in domestic currency; most EM borrow in US dollars, increasing vulnerability to U.S. rate increases.
- Robustness summary:
  - Liftoff effect on EM portfolio flows robust to all alternative specifications; coefficients often larger (reported results regarded as conservative).
  - EM total flows results are less robust in some specifications.

### Using Model Results to Get a Sense of Magnitude
- Two exercises to approximate extra total impact due to liftoff:
  1. Multiply coefficients by one standard deviation of the U.S. rate to simulate a one standard deviation shock.
  2. Multiply coefficients by average values of the U.S. rate in the past five episodes (quarter-by-quarter before liftoff) to estimate extra total impact due to the level of the U.S. rate.
- Caution: exercises give approximate magnitudes and should not be over-interpreted.
- Computation results:
  - One standard deviation shock in the U.S. interest rate would decrease flows by:
    - 2.0 percent of GDP for total flows.
    - 0.9 percent of GDP for portfolio flows.
  - In terms of standard deviations of the respective data, these equal around 0.3 standard deviations.
  - Using averages of the U.S. rate for every quarter before liftoff in the sample implies previous liftoffs resulted in net outflows of:
    - 3.5 percent of GDP for total flows.
    - 1.6 percent of GDP for portfolio flows.
  - In standard deviation terms, these amount to 0.5 and 0.4, respectively.
  - For total flows, there appears to be some payback one quarter post-liftoff, but magnitude is less than the pre-liftoff loss.

### Why a Liftoff Effect Prior to Liftoff?
- Empirical conclusion: liftoff effect is non-negligible and tends to kick in one or two quarters prior to liftoff, not during the liftoff quarter itself.
- Two explored explanations:
  1. Market expectations prior to liftoff.
  2. Macroeconomic conditions leading to liftoff.
- Evidence on market expectations and events:
  - Markets often anticipate rate hikes; they price in liftoff and reposition funds beforehand.
  - June 2004 and June 1999 episodes: markets appeared to expect hikes (news reports cited).
  - February 1994: market surprised by timing/magnitude, though there were macroeconomic clues (President Bill Clinton’s State of the Union, improving indicators; Greenspan’s hints).
- Box I (selected news/clues prior to liftoff):
  - June’2004:
    - 25bps hike was first in nearly 4 years; markets were expecting more as core CPI had risen by 3.3 percent in the months leading to the decision [Chen, Mancini-Griffoli and Sahay (2014) based on money.cnn news, 06/30/04].
  - June’1999:
    - Investors were “all but certain” the Fed would lift short-term interest rates next week to keep inflation from accelerating [money.cnn, five days prior to liftoff, “Rate hike is on the way”].
  - February’1994:
    - Market was caught by surprise; macro clues included strong auto sales, record home sales, and “1.6m private sector jobs in 1993” (Statement in President Bill Clinton’s State of the Union, January 1994). Greenspan had been hinting at potential rate hikes; surprise was mainly timing and magnitude [Allianz Global Investors, October 2014].
- Macro conditions:
  - Improving labor market (unemployment, non-farm payrolls), rising industrial production, and growth in cyclical sectors typically preceded liftoffs.
  - For liftoff quarters, year-over-year GDP growth and/or inflation were picking up, signaling the Fed’s dual mandate would entail a rate hike.
- Interpretation:
  - Liftoffs are usually priced in before the actual event; this explains insignificant liftoff-quarter effect and significant pre-liftoff effect.
  - Findings align with literature on negative spillovers from unanticipated U.S. rate hikes (IMF 2011a; Kuttner, 2001).

### III. THE LIFTOFF EFFECT OF DOMESTIC POLICIES — Introduction
- Central question: Can domestic policies mitigate the negative liftoff effect from a Fed liftoff?
- Conceptual division:
  - Policy responses (near-term actions).
  - Policy framework (underlying structural attributes).
  - Policy response in a country is a function of its policy framework.
- Approach:
  - Describe five domestic policy responses/framework areas.
  - Empirical strategy to include them in regression specification.
  - Report results and discuss relative magnitudes of liftoff effects from U.S. rate hike versus policy responses/framework.

### A. The Policies (five areas examined)
- Domestic Policy Rates:
  - Question: Can increasing domestic policy rates retain capital flows by maintaining interest differentials?
  - Empirical capture: use same policy rate as in generic model and include interactive term with liftoff dummies to capture extra liftoff effect related to domestic policy rates.
- Monetary Policy Independence (MI):
  - Concept: Whether domestic rates should chase the Fed or follow domestic inflation objective.
  - Measurement: Reciprocal of the 12-month rolling correlation between each country’s policy rate and the U.S. policy rate (following Aizenman, Chinn and Ito (2013) with modifications).
    - By construction, maximum value is 1 and minimum value is 0. Higher value = more monetary policy independence.
    - Differences from Aizenman, Chinn and Ito (2013):
      - Base rate is always the U.S. policy rate.
      - Uses 12-month rolling correlation, then takes quarterly average (rather than annual correlation).
    - Caveats:
      - When a country’s interest rate is constant over the 12-month rolling window, correlation is undefined; MI index assigned value of 0.5 for these cases (following Aizenman, Chinn and Ito (2013)).
      - Common shocks may comove domestic and U.S. rates (less of an issue for this paper since analysis ends in 2006 and excludes the global financial crisis period).
- Exchange Rate Stability (ERS):
  - Measurement: 12-month rolling standard deviation of monthly exchange rate change (bilateral to the U.S. dollar), normalized between 0 and 1.
  - Interpretation: Higher value indicates more stable exchange rate movement (more fixed regime).
  - Difference from Aizenman, Chinn and Ito (2013): uses bilateral vis-à-vis U.S. dollar and 12-month rolling standard deviation rather than annual standard deviations.
- Capital Account Openness:
  - Measurement: Chinn-Ito Index (based on IMF AREAER information), normalized between 0 and 1; higher value = more open to cross-border capital transactions.
- Budget Surplus (Deficit):
  - Measurement: Quarter-over-quarter budget surplus, expressed as a share of GDP.
  - Interpretation: An increase implies more budget surplus (greater fiscal discipline), which could signal to markets in the face of a U.S. liftoff.

### B. Empirical Strategy
- A 3-step empirical approach is used for incorporating domestic policy responses and frameworks, mirroring the 3-step approach used to measure the U.S. liftoff effect.

*Source: _wp15256 - Section 2*

### Section 3

### _wp15256 - Section 3

### Methodology: three-step model construction
- Step 1: Start with the reduced model from the previous section containing:
  - country specific fixed effects, “push” and “pull” factors, lagged dependent variables;
  - the U.S. rate interacted with dummy variables (only the significant ones) and the corresponding time dummy variables;
  - all variables representing policy responses/framework, except budget surplus.
- Step 2: Build the saturated model by including all possible liftoff effects of all policy variables:
  - for each policy variable, include interactive terms between the policy variable and each time dummy (five interactive terms per policy variable, showing interaction with two quarters before liftoff, one quarter before liftoff, liftoff quarter, one quarter after liftoff, and two quarters after liftoff);
  - include all the time dummy variables;
  - interpret the liftoff effect of a variable as the coefficient of the interactive term (time dummy × variable), which identifies effects before, during, and/or after liftoff.
- Step 3: Obtain the restricted version from the saturated model by deleting interactive terms that are not significant:
  - perform log-likelihood ratio tests to confirm preference for the reduced model over the saturated model.
- Budget surplus special treatment:
  - regressions for budget surplus are run separately due to a considerably shorter data set;
  - the three-step approach is applied: starting model includes country fixed effects, “push” and “pull” factors, lagged dependent variables, significant U.S. rate × dummy interactions, corresponding time dummies, and budget surplus; extended to include budget surplus × time dummy interactions; final restricted version retains only statistically significant interactions with log-likelihood ratio tests.

### Results — statistical significance (overview)
- General observations:
  - The liftoff effects of policies are substantially higher for EM than AM.
  - The liftoff effect of the U.S. rate typically kicks in before liftoff; variables representing policy responses/framework usually act during or after liftoff.
  - Policy variables (except policy rates) are often not significant on their own (i.e., not significant during normal times) but are significant when interacted with liftoff dummies, implying policy responses/framework can matter during liftoff episodes.
- Specific policy-framework/response findings:
  - Monetary policy independence:
    - Keeping monetary policy independence can help increase net capital flows during liftoff episodes.
    - Statistical significance may partly reflect indirect impact of lower output volatility on capital flows.
    - Policy implication: adjust monetary policy according to domestic inflation objectives; unwarranted rate hikes to stem outflows risk slowing the economy and undermining investor confidence.
  - Domestic policy rates:
    - Some interactive terms between domestic policy rates and dummy variables are significant in baseline results but do not survive robustness tests.
    - Policy rates do not remain significant when extreme independent-variable data points are removed.
    - Conclusion: very weak or no evidence that raising policy rates helps during liftoff episodes.
  - Budget surplus:
    - Quarter-over-quarter change in budget surplus (as a share of GDP) is significant when interacted with liftoff dummies, even though near-term budget surplus is not significant without interaction.
    - This result is robust to robustness checks.
    - The result does not hold when budget surplus is expressed as year-over-year change.
    - Possible interpretation: quarter-over-quarter improvement can serve as an early signal of policy commitment; consistent with evidence that emerging markets that acted early and decisively fared better during the taper tantrum episode.
  - Capital account openness:
    - No statistical evidence that more open capital accounts are associated with worse outcomes prior to or during liftoff.
    - Countries with more open capital accounts recover quicker: statistically significant positive liftoff effect one or two quarters after liftoff.
  - Exchange rate stability:
    - For AM, exchange rate stability is the only variable showing significant liftoff effect; more flexible regimes fare better in attracting capital flows post liftoff.
    - For EM, results are mixed and not necessarily significant in magnitude:
      - More stable exchange rates associated with more net total outflows two quarters before liftoff and during liftoff quarter.
      - More stable exchange rates associated with more portfolio outflows during liftoff quarter.
      - More stable exchange rates associated with more portfolio inflows one quarter prior to liftoff and two quarters post liftoff.

### Can policies mitigate the negative liftoff effect of the U.S. rate? — magnitude assessment
- Two exercises performed (EM focus):
  1. Extra impact on capital flows during liftoff episodes from one standard deviation shock of each variable.
     - Use standard deviation for entire sample, 1982-2006, for all variables.
     - For domestic policy rate, use standard deviation from subsample where policy rates < 30 percent.
     - Coefficients for liftoff effects reduced to three time frames: before liftoff, during liftoff, post liftoff. Only significant coefficients are used; averages taken across significant adjacent quarters as described.
     - Express extra impact on total and portfolio flows both as share of GDP and as standard deviation of dependent-variable data.
  2. Extra impact using average values of each variable from past five liftoff episodes:
     - Multiply the same relevant coefficients by the average actual values of the variables in the previous five liftoff episodes.
     - Express impact as share of GDP and as standard deviation of dependent-variable data.

- Exercise 1: one standard deviation shocks — key magnitudes (EM)
  - Total flows:
    - Before liftoff: net outflow due to one standard deviation shock in the U.S. rate ≈ 2.4 percent of GDP (0.3 standard deviation of the data).
    - Only a one standard deviation shock of quarterly improvement in budget could help before liftoff: attracts net inflow of 1.8 percent of GDP.
    - During liftoff quarter: improvement in quarterly budget surplus attracts net inflows of around 3.7 percent of GDP.
    - After liftoff: monetary policy independence attracts net inflows of 2.1 percent of GDP; open capital accounts attract net inflows of 1.3 percent of GDP.
  - Portfolio flows:
    - Before liftoff: net outflow due to one standard deviation shock in the U.S. rate ≈ 1.7 percent of GDP (0.5 standard deviation of the data).
    - Before liftoff: near-term budget shock does not have impact; monetary policy independence and exchange rate stability shocks can help to some extent.
    - During liftoff quarter: maintaining monetary policy independence can attract net inflow of 0.6 percent of GDP; more fixed exchange rate associated with net outflow of 0.6 percent of GDP.
    - After liftoff: near-term budget improvement attracts 1.3 percent of GDP; exchange rate stability attracts 0.8 percent of GDP; capital account openness attracts 0.5 percent of GDP.

- Exercise 2: using average values from five previous liftoff episodes — key magnitudes (EM)
  - Total flows:
    - Before liftoff: net outflow due to the U.S. rate ≈ 4.4 percent of GDP (0.6 standard deviations of total flows).
    - Policy variables: none strong enough to fully offset the negative impact of the U.S. rate before liftoff.
    - Near-term budget: using actual averages, the extra impact on total flows is negative (near-term budget on average worsened in the five episodes).
    - During liftoff quarter: budget improvement can yield ≈ 1.4 percent of GDP extra inflow.
    - Post liftoff: monetary policy independence attracts ≈ 3.4 percent of GDP; capital account openness attracts ≈ 1.3 percent of GDP.
  - Portfolio flows:
    - Before liftoff: extra impact due to the U.S. rate ≈ -3.0 percent of GDP (0.8 standard deviation of data).
    - Monetary policy independence and exchange rate stability help before liftoff but cannot offset the U.S. rate impact.
    - During liftoff quarter: monetary policy independence can attract ≈ 0.3 percent of GDP.
    - Post liftoff: capital account openness ≈ 0.5 percent of GDP; exchange rate stability ≈ 0.2 percent of GDP; near-term budget improvement ≈ 0.1 percent of GDP.

### Policy implications and concluding thoughts
- Overall findings:
  - Liftoff effects substantially higher for EM than AM; substantial net outflow, particularly for EM portfolio flows, occurs before liftoff in anticipation.
  - Policy responses/framework can only to some extent dampen market reactions.
- Policy takeaways:
  - Monetary policy independence can help attract capital flows and provide greater leeway, but increasing independence requires time and may be difficult when EM currencies are under pressure.
  - Fiscal policy: reducing near-term budget deficit (quarter-over-quarter improvement in budget surplus) can signal commitment to fundamentals and somewhat reduce capital outflows for countries with credibility and fiscal sustainability.
  - Capital account openness does not lead to more outflows during or before liftoff; it can help attract capital flows after Fed liftoff.
  - Some policies that attract capital flows (e.g., improving fiscal sustainability) can be pro-cyclical amid slowing EM growth, posing tradeoffs.
- Caveats and limits:
  - The paper focuses on five policy framework/response areas; it is not exhaustive.
  - Other factors identified in recent literature can affect external vulnerabilities, such as elevated current account deficits, high inflation, weak growth prospects, and relatively low reserves.
  - Macroprudential policies (not addressed in this paper) have been found in other studies to reduce credit booms and slow inflows, and could be relevant policy tools.
- Final synthesis:
  - Before liftoff, EM capital flows are indiscriminately hit across countries; over time, variation emerges as individual policy responses/frameworks can somewhat dampen market reaction.
  - Some EM face tough policy tradeoffs given limited immediate scope to change monetary policy independence and the pro-cyclicality of fiscal consolidation in a slowing growth environment.

*Source: _wp15256 - Section 3*

### Section 4

### _wp15256 - Section 4

### Empirical strategy and liftoff framework
- Step 1: Generic Model for Capital Flows
  - External “push” factors and domestic “pull” factors
  - Identify the variables using literature
  - Lagged dependent variables, country fixed effects
- Step 2: Capture the Fed Liftoff Effect (Saturated/Unrestricted Model)
  - Include time dummies to capture the liftoff effect
  - Include interactive terms – U.S. rates with all the liftoff time dummies
- Step 3: The Restricted Version
  - Using log likelihood ratio tests, reduce the model by deleting the interactive terms that are not significant
  - References for model selection: Cox and Snell (1974), Hendry (1995), Hendry and Krolzig (2005), Hendry and Nielsen (2007)
- Liftoff timing and measurement
  - One quarter before liftoff: average of the U.S. rate one quarter before liftoff for the five previous liftoff episodes (noted in figure captions)

### Main empirical findings (selected coefficient results)
- Table 1 — Emerging Markets (Dependent Variable: Net Total Flows and Net Portfolio Flows, in percent GDP)
  - External Factors
    - US Interest Rate coefficients (Net Total Flows columns): -0.158, -0.078, -0.163; (Net Portfolio Flows columns): -0.048, -0.018, -0.026
      - Standard errors (below): 0.131, 0.157, 0.134, 0.092, 0.110, 0.090
    - US Consumer Prices coefficients: 0.310, 0.247, 0.309, -0.118, -0.060, -0.120
      - Standard errors: 0.341, 0.351, 0.341, 0.259, 0.271, 0.257
    - US GDP Growth coefficients: -0.027, -0.107, -0.021, 0.039, -0.012, 0.017
      - Standard errors: 0.166, 0.174, 0.161, 0.111, 0.124, 0.110
    - Risk Aversion coefficients: -0.095**, -0.143**, -0.109**, -0.026, -0.060, -0.043
      - Standard errors: 0.048, 0.059, 0.051, 0.032, 0.037, 0.032
    - Commodity Prices coefficients: 0.002, 0.002, 0.001, 0.002, -0.003, 0.002
      - Standard errors: 0.010, 0.011, 0.010, 0.009, 0.009, 0.009
  - Domestic Factors
    - Interest Rate coefficients: 0.004***, 0.004***, 0.004***, 0.003***, 0.003***, 0.003***
      - Standard errors: 0.001 (all)
    - Consumer Prices coefficients: -0.004***, -0.004***, -0.004***, -0.001, -0.002**, -0.001
      - Standard errors: 0.001 (all)
    - REER coefficients: 0.064***, 0.061***, 0.061***, 0.024***, 0.025***, 0.024***
      - Standard errors: 0.012, 0.013, 0.012, 0.009, 0.009, 0.009
    - GDP Growth coefficients: 0.256***, 0.239***, 0.247***, 0.014, 0.015, 0.009
      - Standard errors: 0.058, 0.060, 0.059, 0.029, 0.029, 0.029
  - Liftoff Effect (US Interest Rate interacted with:)
    - Two quarters before liftoff: -0.384* and -0.095 (standard errors 0.228 and 0.156)
    - One quarter before liftoff: -0.890***, -0.721**, -0.398**, -0.333*** (standard errors 0.340, 0.310, 0.176, 0.134)
    - Liftoff quarter: -0.319 and 0.110 (standard errors 0.274 and 0.169)
    - One quarter after liftoff: 0.436*, 0.517**, 0.073 (standard errors 0.223, 0.217, 0.144)
    - Two quarters after liftoff: 0.033 and -0.034 (standard errors 0.189 and 0.180)
  - Model features
    - Dummies for liftoff variables (individual time dummies): N Y Y N Y Y (across columns)
    - Number of Lagged Dependent Variables: 8 8 8 4 4 4
    - Fixed Effects: Y Y Y Y Y Y
    - Observations: 966, 966, 966, 973, 973, 973
  - Note on significance: ***, ** and * indicate statistically significant at 1%, 5% and 10% percent, respectively.

- Table 2 — Advanced Markets (Dependent Variable: Net Total Flows and Net Portfolio Flows, in percent GDP)
  - External Factors
    - US Interest Rate coefficients for Net Total Flows: 0.009, 0.022, -0.006; for Net Portfolio Flows: -0.229*, -0.081, -0.101
      - Standard errors: 0.109, 0.118, 0.106, 0.132, 0.165, 0.139
    - Risk Aversion: -0.070*, -0.092**, -0.074*, -0.014, -0.057, -0.041 (standard errors 0.039, 0.046, 0.039, 0.056, 0.062, 0.056)
  - Domestic Factors
    - Interest Rate coefficients (Net Total Flows): 0.118, 0.113, 0.123; (Net Portfolio Flows): 0.227***, 0.228***, 0.231***
      - Standard errors: 0.090, 0.090, 0.091, 0.075, 0.078, 0.077
    - GDP Growth coefficients (Net Total Flows): 0.151**, 0.150**, 0.154**; (Net Portfolio Flows): 0.081, 0.073, 0.076
      - Standard errors: 0.070, 0.070, 0.070, 0.115, 0.116, 0.115
  - Liftoff Effect (US Interest Rate interacted with:)
    - Two quarters before liftoff: -0.169, -0.462*, -0.406* (standard errors 0.155, 0.256, 0.228)
    - One quarter before liftoff and liftoff quarter reported with coefficients and standard errors in table (partial listing)
    - One quarter after liftoff: 0.293**, 0.305***, 0.167 (standard errors 0.117, 0.117, 0.168)
    - Two quarters after liftoff: -0.223*, -0.217*, -0.667*, -0.660* (standard errors 0.125, 0.121, 0.388, 0.370)
  - Model features
    - Dummies for liftoff variables (individual time dummies): N Y Y N Y Y
    - Number of Lagged Dependent Variables: reported as 8 10 10 12 12 12 across columns
    - Fixed Effects: Y Y Y Y Y Y
    - Observations: 1512, 1496, 1496, 1459, 1459, 1459

### Policy variables and their estimated impacts (selected results)
- Table 5 — Emerging Markets (Policy regressions; Dependent variables in percent GDP)
  - Net Total Flows: (Number of lagged dependent variables = 8; observations = 965)
    - Policy Rate coefficients: 0.001*** and 0.003*** (standard errors 0.000, 0.000)
    - Monetary Policy Independence coefficients: -0.477, 2.161*, 7.341*** (standard errors 0.445, 1.300, 2.490)
    - Exchange Rate Stability coefficients: 0.795, -3.928*, -10.324*** (standard errors 1.637, 2.088, 1.933)
    - Capital Account Openness coefficients: -0.074, 3.976*** (standard errors 0.788, 1.162)
    - Timing columns include: No Interactive Term; Two Quarters Before Liftoff; One Quarter Before Liftoff; Liftoff Quarter; One Quarter After Liftoff; Two Quarters After Liftoff
  - Net Portfolio Flows: (AR(4) included; observations = 912)
    - Policy Rate coefficients: 0.000***, 0.003***, -0.020* (standard errors 0.000, 0.000, 0.011)
    - Monetary Policy Independence: -0.326, 2.530***, 2.228**, -2.153*, 2.182*** (standard errors 0.396, 0.700, 0.926, 1.227, 0.663)
    - Exchange Rate Stability: 2.044, 5.650**, -3.303***, 4.675*** (standard errors 1.468, 2.729, 1.020, 1.210)
    - Capital Account Openness: -0.539, 1.538** (standard errors 0.674, 0.740)
- Advanced Markets (Table 5 and Table 6 excerpts)
  - Net Total Flows (observations = 1170)
    - Policy Rate coefficients: 0.078, -0.328** (standard errors 0.115, 0.143)
    - Exchange Rate Stability reported with large coefficients and standard errors, e.g., -9.172, 111.521***, 296.593**, -184.509***, -54.320** (standard errors 7.808, 36.310, 115.075, 69.329, 23.930)
    - Capital Account Openness: -3.225*, 5.577** (standard errors 1.827, 2.882)
  - Net Portfolio Flows (Table 6, observations = 800; AR(16) included)
    - Policy Rate coefficients: 0.410*, 0.572* (standard errors 0.222, 0.311)
    - Exchange Rate Stability coefficients: -46.259, -392.574* , -525.911*** (standard errors 31.661, 222.604, 166.203)
    - Capital Account Openness: -5.543, 3.907 (standard errors not shown in excerpt)

- Fiscal variable (Table 7 excerpt)
  - Quarterly Change in Budget (surplus/deficit +/-, dependent variable in percent GDP)
    - Change in Budget coefficients: -0.109, -0.072, -0.034, 0.044 (standard errors 0.076, 0.056, 0.048, 0.076)
    - Change in Budget interacted with Two quarters before liftoff: 0.491*** (standard error 0.161)
    - Change in Budget interacted with Liftoff quarter: 1.023*** (standard error 0.329)
    - Change in Budget interacted with One quarter after liftoff: 0.370***, 0.418** (standard errors 0.084, 0.173)
    - Change in Budget interacted with Two quarters after liftoff: 0.378* (standard error 0.214)
    - Number of Lagged Dependent Variables listed: 6 4 8 4 (across columns)
    - Fixed Effects: Y Y Y Y
    - Observations: 620, 624, 1202, 1213

### Figures: qualitative and quantitative highlights
- Figures present:
  - Empirical strategy flow (three-step approach described above)
  - The Fed liftoff episodes charting Federal Funds Effective Rate across 1982–2006 liftoff episodes
  - Quantified liftoff effects:
    - "The Extra Sensitivity of Flows to the U.S. Interest Rate One Quarter Before Liftoff" shown for EMAM (per every 100bps of the U.S. rate; coefficients and significance levels displayed)
    - Decreases in flows expressed as share of GDP for 1 percent increase in the U.S. rate and for one standard deviation shock in the U.S. rate (figures show negative coefficients and statistical significance indicated by ***, **, *)
  - Time series charts for U.S. Real GDP Growth (percentage, year-over-year) and U.S. Inflation Rate (percentage, year-over-year) with liftoff quarters indicated by red squares
  - Policy variable interactions during Prelift / Lift / PostLift periods showing estimated extra impacts during liftoff episodes:
    - Variables displayed: U.S. Rate, Policy Rate, MI (Monetary Independence?), ERS (Exchange Rate Stability/KA?), Open (Capital Account Openness), Change in Budget
    - Outcome metrics visualized as Share of GDP and # of stdevs (coefficient*one std. of variable) for EM Total Flows, EM Portfolio Flows, AM Total Flows, AM Portfolio Flows

### Appendix: sample and data sources
- List of countries (Emerging Markets and Advanced Markets) with country codes (selected examples shown in appendix table)
  - Emerging Markets include: Brazil 223, Bulgaria 918, Chile 228, Colombia 233, Croatia 960, Hungary 944, India 534, Indonesia 536, Latvia 941, Lithuania 946, Macedonia, FYR 962, Mexico 273, Peru 293, Philippines 566, Poland 964, Romania 968, Russian Federation 922, South Africa 199, Thailand 578, Turkey 186, Ukraine 926, Slovak Republic 936, Slovenia 961 (listed in appendix)
  - Advanced Markets include: Australia 193, Austria 122, Canada 156, China, P.R.: Hong Kong 532, Czech Republic 935, Denmark 128, Finland 172, France 132, Germany 134, Greece 174, Iceland 176, Ireland 178, Israel 436, Italy 136, Japan 158, Korea, Republic of 542, Malta 181, Netherlands 138, New Zealand 196, Portugal 182, Singapore 576, Spain 184, Sweden 144, Switzerland 146, United Kingdom 112 (listed in appendix)
- Data sources for indicators
  - Total private flows and portfolio flows, expressed as a share of GDP: Financial Flows Analytics database compiled from the IMF’s Balance of Payments Statistics, International Financial Statistics, and World Economic Outlook databases, World Bank’s World Development Indicators database, Haver Analytics, CEIC Asia database, CEIC China database, and national sources.
  - U.S. interest rate: Federal Reserve (FRED)
  - U.S. consumer prices: IFS
  - U.S. real GDP growth: U.S. Bureau of Economic Analysis (BEA)
  - S&P 500 total return index: S&P (downloaded using Haver Analytics)
  - VXO index: WSJ (downloaded using Haver Analytics)
  - Commodity price index: IMF
  - Domestic interest rate, domestic consumer prices, domestic real GDP, real effective exchange rate, exchange rates vis-à-vis dollar: IFS, National Sources
  - Capital account openness index: Chinn and Ito (2006)
  - Budget surplus, as a share of GDP: National sources, downloaded using Haver Analytics

*Content derived from _wp15256 - Section 4 (source PDF: _wp15256 - Section 4).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15256.pdf_