## Central Bank Communication and Monetary Policy Surprises in Chile — WP/18/156 (wp18156)

## Source details

**Canonical URL:** [Central Bank Communication and Monetary Policy Surprises in Chile — WP/18/156 (wp18156)](https://www.imf.org/-/media/files/publications/wp/2018/wp18156.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2018/wp18156.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2018/wp18156.pdf.json)

---

### Abstract and core results
- Paper assesses predictability and effectiveness of Central Bank of Chile (CBC) communications via statements, minutes, and monetary policy reports (IPoM).
- Constructed indexes of monetary policy surprises using survey forecasts and short-term rate impacts of communication events.
- Key empirical findings:
  - CBC has been effective in forward guidance through statements and IPoM.
  - Policy actions are quite predictable, especially post the global financial crisis (GFC).
  - Equity prices and the exchange rate respond with the correct sign to monetary policy surprises, but results are not robust.
  - Minutes exhibit an asymmetric equity response: “dovish” minutes (minutes that decrease the 3-month swap rate on release) can have a negative impact on equity prices, suggesting information about the state of the economy rather than mere policy preferences.
  - A 100 bps monetary policy tightening shock implies a decline in economic activity (IMACEC) of about 2 pp. after one year; the response of inflation is more muted.

### Predictability of policy decisions (descriptive evidence, 2003–2017)
- Sample and counts:
  - Between 2003 and 2017: total of 180 monetary policy meetings (statements and minutes) and 53 IPoMs.
  - 60 meetings had a policy rate change (one every 3 meetings).
  - Median forecaster surprised 27 times (15 percent of meetings or about one every 7 meetings).
  - Only two cases (Dec 2009 and Jun 2010) where IPoM and statement were on the same day.
- Evolution and magnitudes:
  - Predictability increased over time: number of surprise meetings declined, as did the overall number of meetings with a change in the policy rate.
  - Mean absolute monetary policy surprise decreased from about 2 bps per meeting pre GFC to less than 1 bps post GFC.
  - Monthly mean absolute surprise for mean (median) forecast: about 6 bps (5 bps).
  - Mean (median) forecast absolute annual surprise: 48 (52) bps; excluding 2009: 28 (31) bps.
  - Average surprise during 2003-2017 has a loosening bias of about -3 bps.
  - Distribution skewed by GFC episodes: four major loosening surprises above 50 bps; biggest surprise in February 2009: 150 bps.
  - Sum of median- (mean-) forecast surprises between 2003 and 2017 is 526 (460) bps, of which 350 (327) bps are concentrated in 2009.
  - Correlation between monetary policy surprises and inflation expectations: 0.46.
  - Period 2008–2010 shows highest forecast disagreement (175 bps between most dovish and hawkish forecast).
- Predictability versus other central banks:
  - Predictability modestly better than the US Federal Reserve during the 1998-2012 period (standardized for relative policy rate volatility).
- Notable miscommunication episodes:
  - Jan 2004 and July 2007: central bank surprised markets by cutting rates (by 50 and 25 bps, respectively) when Bloomberg survey expected no cuts and showed no forecast uncertainty.
- Meeting timing specifics:
  - Statement published at 6 pm right after the monetary policy meeting (after markets close).
  - Minutes released with 45 days delay before 2006 and subsequently with 15 (11 business) days delay.
  - IPoM released 4 times a year starting in March, usually in the morning.
  - Since 2018, IPoM publication timing changed (outside sample).

### Efficacy on fixed-income markets (yield curve and forward guidance)
- Identification and regression controls:
  - High-frequency identification following Kuttner (2001), Gürkaynak et al (2005), Gertler and Karadi (2015), and Larraín (2007).
  - Regression controls: US interest rates, copper prices, country risk (∆CDS, ∆EMBI), two dummies for 2008 and 2011 FX interventions, and a possible time-trend in the constant.
- Surprise versus anticipated effects:
  - Anticipated policy changes: no significant impact on long-term or short-term rates.
  - Surprise policy changes: strong and highly significant impact on swap rates across maturities.
  - Magnitudes (from regressions):
    - A 100 bps surprise increases the 3-month swap rate by about 53 bps and explains about 64 percent of its variation.
    - Impact declines monotonically with maturity; impact on 10-year rate: 7 bps.
    - If surprise reflected only timing uncertainty (temporary 1 month): implied impact at most 33 bps for 3-month rate and 0.8 bps for 10-year rate (about 1/10th of observed).
    - Removing 2009Q1 increases impact on 3-month (1-year) rate to 76 (58) bps but reduces impact on 10-year rate.
- Forward guidance via statements and IPoM:
  - Statements (words) have significant effects on the yield curve even absent a policy rate surprise.
  - Regression evidence: impact coefficient of 3-month swap rate on yields in absence of surprises is >1 up to the 5-year yield and 0.9 for the 10-year yield (strongly significant) — interpreted as forward guidance from statements shifting the entire yield curve; surprise actions tend to tilt the curve.
  - IPoM provides substantial forward guidance: swap rates strongly co-move on IPoM release day; IPoM impact stronger than typical reaction over sample and stronger than statements in many specifications.
  - Minutes provide relatively little forward guidance beyond the statement; policy information content of minutes substantially less relevant for markets.

### Efficacy on exchange rate and equity markets
- Typical impact of a 100 bps tightening (statement) surprise:
  - Exchange rate: appreciates at impact by about 1.5 percent.
  - Equity prices: reaction has the correct sign but not significant in many specifications.
- Robustness and specification notes:
  - If monetary policy surprise proxied by movement in 3-month swap rate: impact on exchange rate no longer significant; fall in equity prices of about 1.6 percent becomes significant.
  - No significant equity or FX market reaction found for IPoM (IPoM shocks proxied by 3-month swap rate movement on IPoM dates), possibly due to short sample.
- Minutes and asymmetric equity response:
  - Minutes have a strong negative, mildly significant, impact on equity prices but only conditional on being “hawkish” (minutes that increase 3-month swap rate on release date).
  - For dovish minutes (reduce 3-month swap rate), the impact flips sign: more dovish minutes associated with lower equity prices.
  - Regression coefficients (Table VI, minutes regressions on 140 observations):
    - 3-month swap rate (+): IGPA coefficient -4.470* (standard error 2.264); FX coefficient -2.857 (standard error 1.727).
    - 3-month swap rate (-): IGPA coefficient 7.390*** (standard error 2.671); FX coefficient 5.959*** (standard error 2.037).
    - Constant: IGPA 0.224*** (0.0811); FX 0.112* (0.0619).
  - Interpretation: dovish minutes may convey information about the state of the economy rather than only Board members’ preferences.

### Monetary policy communication and inflation-forecast dispersion
- Objective and measures:
  - Assess whether monetary policy decisions or communication affect dispersion of 12-month-ahead inflation forecasts (interquartile range and standard deviation).
  - Sample statistics:
    - Interquartile range: average 0.28 point; standard deviation 0.10 point; maximum 0.68 point.
- Main regression findings (Table VII-A and VII-B):
  - Lagged dispersion and level of inflation both have positive coefficients and are highly significant across specifications.
  - Asymmetric surprise effects:
    - Median forecast surprise overall: negative and sometimes significant (e.g., -0.106** (0.0534) in specification (3)).
    - Median f. surprise (-) (dovish): -0.121** (0.0576) in specification (4) for interquartile range.
    - Mean f. surprise (-) (dovish): -0.122* (0.0642) in specification (6) for interquartile range.
    - For standard deviation measure, 3-m rate IPoM surprise: -0.358** (0.169) in specification (7); 3-m rate IPoM surpr. (-): -0.591*** (0.225) in specification (8).
  - Summary statements:
    - Hawkish surprises: not significant.
    - Dovish surprises: reduce inflation dispersion significantly in several specifications.
    - A -25 bps surprise from IPoM increases inflation dispersion by 0.15 points (reported in text); a -25 bps statement surprise increases the interquartile range by about 0.03 points.

### Macroeconomic impact — proxy-VAR results and identification
- Identification strategy:
  - Blend recursive VAR timing assumption with constructed monetary policy surprises used as an instrument for changes in the policy rate (proxy VAR a la Gertler and Karadi (2015)).
  - Constructed surprise series designed to be orthogonal to output and inflation forecasts.
- VAR specification:
  - Monthly VAR variables (all in logs): IMACEC (index of economic activity), CPI, constructed surprise series, TPM (monetary policy rate), IGPA (stock market index), peso-USD exchange rate.
  - Alternative specification in log-difference gives similar results.
- Key impulse-response findings to a 100 bps tightening shock (instrumented by surprise series):
  - Policy rate: hump-shaped reaction peaking at 30 bps; returns to original level after about 1 year.
  - Economic activity (IMACEC): peak decline of about 2 pp. after 1 year.
  - CPI: initial increase (a “price puzzle”) that reverts after about 2 months; CPI ends up 1 percent lower after 2-3 years.
  - Financial markets: stock market declines by about 0.8 percent after 3 months; exchange rate reaction not significant in VAR IRFs.
  - Additional reported result (not robust across specifications): a 100 bps tightening surprise typically results in an appreciation of the exchange rate at impact by about 1.5 percent; reaction of equity prices not significant though with the correct sign.

### Institutional features and communications toolkit
- Inflation-targeting and exchange-rate regime:
  - IT regime coupled with a free-floating exchange rate since September 1999.
  - CBC commitment (quoted): “to keep annual CPI inflation at around 3 percent most of the time, within a range of plus or minus 1 percentage point. To meet this target, the Bank focuses its monetary policy on keeping projected inflation at 3 percent annually over a policy horizon of around two years.” (CBC, 2017).
- Communications instruments and content:
  - Monthly monetary policy meetings followed by a short statement; minutes released with delay and include meeting discussion details, policy alternatives, arguments by Board members, and balance of votes.
  - Almost half the statements since 2011 included clear and concise conditional guidance about the likely future path of monetary policy.
  - IPoM is the Board’s main communication tool, released quarterly in Spanish and English, includes growth and inflation fan charts, and infrequently provides estimates of unobservable variables; does not provide a policy rate fan chart.

### Data, measurement, and special considerations
- Data sources:
  - Swap rates and, for robustness, government bond and central bank note yields.
  - Survey data from the central bank (Encuesta de Expectativas Económicas, EEE) and Bloomberg expectation survey; both filed every month.
- Survey timing and caveats:
  - EEE usually released about a week before the policy meeting and the CPI release; five exceptions identified (9/8/2003, 9/9/2008, 4/9/2009, 5/11/2009, and 9/10/2009) where EEE appears filled before the meeting.
  - Typo corrected in June 2007 Bloomberg survey (max value for the monetary policy rate changed from 9 to 6).
  - In two instances (Dec 2009 and Jun 2010) IPoM and statement timing prevents separate identification of their market impact.
- Choice of instruments:
  - Swap rates preferred for longer maturities; central bank yields may be more reliable up to one year.

### FX interventions and market reactions (two major episodes)
- April 10, 2008 intervention:
  - CBC announced program to accumulate reserves of up to USD 8 billion through the remainder of that year; program suspended at end of September after the demise of Lehman Brothers; Central Bank had accumulated about USD 5.75 billion when suspended.
  - Market reaction:
    - On the day after the announcement: the peso moved little, yields declined mildly at short term maturities, and the stock market gained modestly.
    - Two business days after: the exchange rate depreciated by more than 3 percent.
- January 4, 2011 intervention (extraordinary meeting):
  - CBC announced FX interventions lasting a year consisting in buying USD 12 billion in the spot market at an average of USD 50 million per day.
  - By end-2011: foreign reserves had climbed to more than USD 40 billion, equivalent to 17% of GDP.
  - After the announcement: the (multilateral) exchange rate appreciated by about 4 percent; the stock market improved; short-term rates declined; impact on long-term rates was more mixed.
- Interpretation and empirical treatment:
  - Portfolio effects were relatively modest while signaling effects were more visible.
  - Two dummies introduced to control for announcements of FX interventions in regressions.

### Robustness evidence (selected tables and results)
- Sub-sample 2010–2017 (Table VIII-A):
  - Unanticipated surprise coefficients remain large and significant at short maturities (e.g., 0.762*** (0.0816) for 3 month swap).
  - Anticipated surprise coefficients small or insignificant at long maturities in this subsample.
  - Observations: 95; R-squared vary by maturity from 0.539 down to 0.011.
- Full sample excluding 2009Q1 (Table VIII-B):
  - Unanticipated surprise coefficients remain large at short maturities (e.g., 0.709*** (0.0776) for 3 month swap).
  - Observations and fit vary by column; examples: Observations 135 (3 month); R-squared 0.474 (3 month).
- Forward guidance effects of statements (Table IX, 138 observations with 79 surprise meetings):
  - Regressions of longer maturities on 3-month swap show coefficients significantly greater than 1 at intermediate maturities (e.g., 1.525*** (0.283) for 1 year swap).
  - Interaction 3-mXdummy (surprise dummy interaction) negative and significant across maturities (e.g., -0.767*** (0.205) for 6 month relative to 3-m).
  - Unanticipated surprise (direct) significant at short maturities (e.g., 0.535*** (0.0449) for 3-m).
  - Results interpreted as forward guidance from statements shifting the yield curve; surprises tend to change the tilt.

### Summary of conclusions
- CBC predictability: generally high (except during the GFC) and has increased over time as the number of surprise meetings and the overall number of meetings with a change in the policy rate declined, even though macroeconomic volatility has not changed much.
- Efficacy:
  - Statements at times of monetary policy surprises have high efficacy on the yield curve.
  - IPoM provides substantial forward guidance, often inducing medium- and long-term yield changes superior to those from “surprise” meetings.
  - Minutes deliver a noisier and sometimes asymmetric signal for equities.
- Inflation expectations:
  - Inflation-forecast dispersion increases when markets are surprised by the CBC’s dovish tone in several specifications; dovish surprises often reduce measured dispersion in other specifications depending on the measure used.
- Macro effects:
  - Proxy-VAR results reaffirm that a 100 bps tightening shock induces a hump-shaped policy rate response and a peak IMACEC decline of about 2 pp. after 1 year; CPI initially rises then falls to 1 percent lower after 2-3 years.

*Source: WP/18/156 — Central Bank Communication and Monetary Policy Surprises in Chile (Sections 1–4).*

### Section 1

### Central Bank Communication and Monetary Policy Surprises in Chile — Section 1

### Abstract and core results
- Paper assesses predictability and effectiveness of Central Bank of Chile (CBC) communications via statements, minutes, and monetary policy reports (IPoM).
- Constructed indexes of monetary policy surprises using survey forecasts and short-term rate impacts of communication events.
- Key empirical findings:
  - CBC has been effective in forward guidance through statements and IPoM.
  - Policy actions are quite predictable, especially post the global financial crisis (GFC).
  - Equity prices and the exchange rate respond with the correct sign to monetary policy surprises, but results are not robust.
  - Minutes exhibit an asymmetric equity response: “dovish” minutes (minutes that decrease the 3-month swap rate on release) can have a negative impact on equity prices, suggesting information about the state of the economy rather than mere policy preferences.
  - A 100 bps monetary policy tightening shock implies a decline in economic activity (IMACEC) of about 2 pp. after one year; the response of inflation is more muted.

### Introduction: motivation and approach
- Communication became central to modern central banking to enhance accountability, credibility, and the management of private sector expectations.
- The paper evaluates:
  - Predictability of monetary policy actions (necessary for credible communication).
  - Efficacy of CBC communications by measuring impacts on yields, equity prices, and the exchange rate.
  - Macroeconomic impact of identified monetary policy shocks.
- Methodology overview:
  - Monetary policy surprises at regular meetings measured as unanticipated movement in the policy rate based on central bank and Bloomberg surveys.
  - Impact of IPoMs, statements, and minutes on the yield curve measured via daily changes in yields and swap rates at event dates (yields and swap rates first regressed on controls).

### Predictability and summary statistics
- Predictability findings:
  - Predictability of policy decisions relatively high except for 2009 when CBC cut rates substantially faster than expected.
  - Predictability increased over time: number of surprise meetings declined, as did the overall number of meetings with a change in the policy rate.
  - Mean absolute monetary policy surprise decreased from about 2 bps per meeting pre GFC to less than 1 bps post GFC.
  - Predictability modestly better than the US Federal Reserve during the 1998-2012 period (standardized for relative policy rate volatility).
  - Monetary policy surprises positively correlated to inflation expectations: 0.46.
- Sample counts and key dates:
  - Between 2003 and 2017: total of 180 monetary policy meetings (statements and minutes) and 53 IPoMs.
  - Only two cases (Dec 2009 and Jun 2010) where IPoM and statement were on the same day.
- Meeting and release timing specifics:
  - Statement published at 6 pm right after the monetary policy meeting (after markets close).
  - Minutes released with 45 days delay before 2006 and subsequently with 15 (11 business) days delay.
  - IPoM released 4 times a year starting in March, usually in the morning.
  - Since 2018, IPoM publication timing changed (outside sample).

### Efficacy: yield curve, equities, FX, and forecast dispersion
- Yield-curve effects:
  - Statements at times of monetary policy surprises significantly affect medium- and long-end of the yield curve.
  - Forward guidance from statements tends to shift the entire yield curve; surprise actions tend to tilt the short-end.
  - IPoM provides substantial forward guidance, inducing medium- and long-term yield changes often superior to those from “surprise” meetings.
  - Minutes provide relatively little forward guidance beyond the statement.
- FX and equity responses:
  - A 100 bps tightening surprise typically causes an exchange rate appreciation by about 1.5 percent on the same day of the surprise.
  - Reaction of equity prices is not significant though has the correct sign; both equity and FX results are not robust.
  - Minutes have a modest effect on the yield curve but appear to affect stock markets asymmetrically: dovish minutes can negatively impact equity prices, implying content about economic outlook.
- Inflation-forecast dispersion:
  - IPoMs and surprises appear to affect inflation-forecast dispersion only conditional on a “hawkish” tone.
  - A 25 bps tightening surprise induced by the release of minutes reduces inflation-forecast dispersion by 0.15 points.

### Macroeconomic impact (proxy-VAR results)
- Using a proxy-VAR a la Gertler and Karadi (2015):
  - A 100 bps monetary policy tightening shock induces a hump-shaped response for the policy rate.
  - Economic activity (IMACEC) declines with a peak decline of about 2 pp. after 1 year.
  - CPI response shows an initial increase (a “price puzzle”) that reverts after about 2 months and results in CPI being 1 percent lower after 2-3 years.

### The Central Bank of Chile’s inflation-targeting framework (institutional features)
- IT regime coupled with a free-floating exchange rate since September 1999.
- CBC’s transparency and long-run communication:
  - CBC states commitment: “to keep annual CPI inflation at around 3 percent most of the time, within a range of plus or minus 1 percentage point. To meet this target, the Bank focuses its monetary policy on keeping projected inflation at 3 percent annually over a policy horizon of around two years.” (CBC, 2017).
- Communications toolkit:
  - Monthly monetary policy meetings followed by a short statement; minutes released with delay.
  - Almost half the statements since 2011 included clear and concise conditional guidance about the likely future path of monetary policy.
  - Minutes historically include additional meeting discussion details, policy alternatives, arguments by Board members, and balance of votes.
  - IPoM is the Board’s main communication tool, released quarterly in Spanish and English, includes growth and inflation fan charts, and infrequently provides estimates of unobservable variables; does not provide a policy rate fan chart.

### Data, measurement, and special considerations
- Data sources:
  - Swap rates and, for robustness, government bond and central bank note yields.
  - Survey data from the central bank (Encuesta de Expectativas Económicas, EEE) and Bloomberg expectation survey; both filed every month.
- Survey and release caveats:
  - EEE usually released about a week before the policy meeting and the CPI release; five exceptions identified (9/8/2003, 9/9/2008, 4/9/2009, 5/11/2009, and 9/10/2009) where EEE appears filled before the meeting.
  - Typo corrected in June 2007 Bloomberg survey (max value for the monetary policy rate changed from 9 to 6).
  - In two instances (Dec 2009 and Jun 2010) IPoM and statement timing prevents separate identification of their market impact.
- Choice of instruments:
  - Swap rates preferred for longer maturities; central bank yields may be more reliable up to one year.

### FX interventions
- Since adoption of fully flexible exchange rate regime, CBC intervened in FX market on two major occasions: 2008 and 2011.
- Interventions aimed at increasing international reserves when the Chilean peso was strong and were very predictable.

*Source: WP/18/156 — Central Bank Communication and Monetary Policy Surprises in Chile (Section 1).*

### Section 2

### wp18156 - Section 2

### FX interventions and market reactions
- April 10 intervention (ordinary monetary policy meeting after market close): CBC announced program to accumulate reserves of up to USD 8 billion through the remainder of that year; program suspended at end of September after the demise of Lehman Brothers; Central Bank had accumulated about USD 5.75 billion when suspended.
- Market reaction to April 10 announcement:
  - On the day after the announcement: the peso moved little, yields declined mildly at short term maturities, and the stock market gained modestly.
  - Two business days after: the exchange rate depreciated by more than 3 percent.
- January 4, 2011 intervention (extraordinary meeting): CBC announced FX interventions lasting a year consisting in buying USD 12 billion in the spot market at an average of USD 50 million per day.
  - By end-2011: foreign reserves had climbed to more than USD 40 billion, equivalent to 17% of GDP.
  - After the announcement: the (multilateral) exchange rate appreciated by about 4 percent; the stock market improved; short-term rates declined; impact on long-term rates was more mixed.
- Interpretation: portfolio effects were relatively modest while signaling effects were more visible.
- Empirical treatment: two dummies introduced to control for announcements of FX interventions in regressions.

### Methodology: high-frequency identification and surprise decomposition
- Approach: construct series of monetary policy and forward guidance surprises using high frequency identification following Kuttner (2001), Gürkaynak et al (2005), Gertler and Karadi (2015), and Larraín (2007).
- Predictability concept: defined loosely as the inverse of the absolute policy rate surprise at the day of the meeting.
  - Main operationalization: difference between policy rate decided at the meeting and its survey forecast for that meeting.
  - Survey timing note: EEE usually filled 2-4 weeks prior to next meeting; Bloomberg survey released a few days before the meeting.
- Surprise decomposition for each forecast j:
  - ∆i_t = i_t − E_{j,t−1} i_t + E_{j,t−1} i_t − i_{t−1} = ∆i^{u}_{j,t} + ∆i^{a}_{j,t}
  - Construct un-anticipated (surprise) series using mean, median, maximum (most dovish), and minimum (most hawkish) expectations: ∆i^{u}_{a,t}, ∆i^{u}_{m,t}, ∆i^{u}_{dov,t}, ∆i^{u}_{hak,t}.
  - Predictability approximated by inverse of average-forecast surprise.
- Efficacy concept: market reaction to policy actions and events; inferred by looking at yield curve, equity, and FX market responses at times of statements, minutes, and IPoM releases.
- Regression control specification for assessing yield responses:
  - ∆i^{m}_{t} = c_{0,t} + c_{1} ∆i^{m,US}_{t} + c_{2} i^{10y,US}_{t−1} + c_{3} i^{1m,US}_{t−1} + c_{4} ∆ln(p_{c,t}) + c_{5} ∆CDS_{t} + c_{6} ∆EMBI_{t} + ε^{m}_{t}
  - Controls grouped as: 1) US interest rates, 2) copper prices, 3) country risk.
  - Two dummy variables added for the 2008 and 2011 FX interventions and a possible time-trend in c_{0,t}.

### Predictability: descriptive evidence (2003–2017)
- Sample: 180 monthly meetings between 2003 and 2017.
  - 60 meetings had a policy rate change (one every 3 meetings).
  - Median forecaster surprised 27 times (15 percent of meetings or about one every 7 meetings).
  - Of the 27 median-forecast surprises, in 6 occasions (about 5 percent of the times) forecasters were surprised by the TPM left unchanged (surprises conditional on a no policy change).
- Forecast disagreement and frequency of surprises:
  - Mean-forecast surprises are non-zero in about 55 percent of meetings.
  - Using mean forecast: frequency of meetings without a policy rate change that nonetheless surprised markets increases from 5 to 40 percent.
  - Conditional on a policy rate change, surprise frequency rises to 83 percent when using mean forecast.
  - Only in 2 instances did the CBC surprise markets with a move even in presence of no disagreement among forecasters.
- Distribution and bias:
  - Distribution of mean and median forecast surprises symmetric around zero but skewed by GFC episodes.
  - Four major loosening surprises above 50 bps; biggest surprise in February 2009: 150 bps.
  - Sum of median- (mean-) forecast surprises between 2003 and 2017 is 526 (460) bps, of which 350 (327) bps are concentrated in 2009.
  - Average surprise during 2003-2017 has a loosening bias of about -3 bps.
  - Only small tightening surprises recorded: biggest tightening shock 25 bps using median forecast and 50 bps using most dovish forecast.
- Magnitude and evolution:
  - Monthly mean absolute surprise for mean (median) forecast: about 6 bps (5 bps).
  - Mean (median) forecast absolute annual surprise: 48 (52) bps; excluding 2009: 28 (31) bps.
  - Correlation between monetary policy surprises and inflation expectations: 0.46.
  - Period 2008–2010 shows highest forecast disagreement (175 bps between most dovish and hawkish forecast).
  - Mean absolute surprise excluding 2009 declined from about 2 bps (2003–2008) to less than 1 bps (2010–2017).
- Miscommunication episodes:
  - Jan 2004 and July 2007: central bank surprised markets by cutting rates (by 50 and 25 bps, respectively) when Bloomberg survey expected no cuts and showed no forecast uncertainty.
- Conclusion on predictability: CBC has relatively high predictability which on average has improved over time.

### Efficacy: fixed-income markets (yield curve and forward guidance)
- Impact of surprises vs anticipated changes:
  - Anticipated policy changes: no significant impact on long-term or short-term rates (e.g., 3-month swap).
  - Surprise policy changes: strong and highly significant impact on swap rates across maturities.
- Magnitudes (from regressions):
  - A 100 bps surprise increases the 3-month swap rate by about 53 bps and explains about 64 percent of its variation.
  - Impact declines monotonically with maturity; impact on 10-year rate: 7 bps (Table II).
  - If surprise reflected only timing uncertainty (temporary 1 month): implied impact at most 33 bps for 3-month rate and 0.8 bps for 10-year rate (about 1/10th of observed).
  - Removing 2009Q1 increases impact on 3-month (1-year) rate to 76 (58) bps but reduces impact on 10-year rate (Table VIII-A and B).
- Forward guidance via statements:
  - Even absent a policy rate surprise, statements (words) have significant effects on yield curve.
  - Regression conditioning on whether there was a surprise shows:
    - Impact coefficient of 3-month swap rate on yields in absence of surprises is >1 up to the 5-year yield and 0.9 for the 10-year yield (strongly significant).
  - Interpretation: forward guidance from statements shifts the entire yield curve; surprise actions tend to tilt the curve.
- IPoM and minutes:
  - IPoM releases (policy rate unchanged on release day): 3-month swap rate movements used.
  - Statements generally had stronger impact than IPoM and minutes historically; more recently, impact of statements similar in magnitude to IPoM and minutes.
  - IPoM: swap rates strongly co-move on IPoM release day; regressions of longer-maturity swap rates on 3-month swap rate show very high and significant coefficients (close to one at short horizons, declining with maturity).
    - IPoM impact stronger than typical reaction over sample and stronger than statements, suggesting IPoM plays important role for forward guidance.
  - Minutes: do not provide similar forward guidance; policy information content of minutes substantially less relevant for markets (Table IV).

### Efficacy: exchange rate and stock market impacts
- Typical impacts of a 100 bps tightening (statement) surprise:
  - Exchange rate: appreciates at impact by about 1.5 percent.
  - Equity prices: reaction has correct sign but not significant (Table V).
- Robustness notes:
  - If monetary policy surprise proxied by movement in 3-month swap rate: impact on exchange rate no longer significant; fall in equity prices of about 1.6 percent becomes significant.
  - No significant equity or FX market reaction found for IPoM (IPoM shocks proxied by 3-m swap rate movement on IPoM dates), possibly due to short sample.
- Minutes and equities:
  - Minutes have a strong negative, mildly significant, impact on equity prices but only conditional on being “hawkish” (minutes that increase 3-month swap rate on release date).
  - For dovish minutes (reduce 3-month swap rate), the impact flips sign: more dovish minutes associated with lower equity prices.
  - Interpretation: asymmetric result suggests dovish minutes may convey information about state of the economy rather than Board members’ preferences for looser policy.

### Monetary policy communication and inflation forecast dispersion
- Objective: assess whether monetary policy decisions or communication affect dispersion of inflation forecasts at 12-month horizon (measured as standard deviation across forecasts).
- Regression specification: regress dispersion on its lag, consensus forecast inflation level, and constructed surprise series (3-month swap rate movement from IPoM and minutes releases, median- and mean-forecast surprise from statements), with surprises split into hawkish (>0) and dovish (<0).
- Sample statistics for dispersion of inflation forecasts:
  - Average dispersion: 0.28 point.
  - Standard deviation of dispersion: 0.10 point.
  - Maximum dispersion: 0.68 point.
- Empirical findings (summary):
  - Lagged dispersion and level of inflation both have positive coefficients and are highly significant (Table VII A and B).

*Source: wp18156 - Section 2*

### Section 3

### wp18156 - Section 3

### Effects of hawkish vs dovish surprises on inflation forecast dispersion
- Hawkish surprises: not significant.
- Dovish surprises: reduce inflation dispersion significantly.
- A -25 bps surprise from IPoM increases inflation dispersion by 0.15 points.
- A -25 bps statement surprise increases the interquartile range by about 0.03 points.
- Constructed surprise series autocorrelations: from 0.4 to 0.6; removing sizeable consecutive surprising policy cuts in Q1 2009 yields an autocorrelation around 0.2 that is not significant for the mean-forecast surprise.

### Identification strategy and VAR specification for macroeconomic impact
- Identification approach: blend recursive VAR timing assumption with constructed monetary policy surprises used as an instrument for changes in the policy rate (proxy VAR a la Gertler and Karadi (2015)).
- Constructed surprise series are designed to be orthogonal to output and inflation forecasts and thus used to infer causal effects of raising interest rates on economic activity and prices.
- Monthly VAR variables (all in logs): index of economic activity (IMACEC), consumer price index (CPI), constructed surprise series, monetary policy rate (TPM), end-of-period stock market index (IGPA), end-of-period peso-USD exchange rate.
- Alternative specification in log-difference gives similar results.

### Impulse responses and key quantitative findings from a 100 bps tightening surprise
- Instrumented shock: a 100 bps monetary policy tightening shock (shock to the monetary policy surprise series).
- Policy rate response:
  - Hump-shaped reaction peaking at 30 bps.
  - Returns to its original level after about 1 year.
- Economic activity (IMACEC):
  - Peak decline of about 2 pp. after 1 year.
- CPI response:
  - Initial price puzzle: prices increase initially.
  - Prices revert after about 2 months.
  - End result: CPI ends up 1 percent lower after 2-3 years.
- Financial markets:
  - Stock market declines by about 0.8 percent after 3 months.
  - Exchange rate reaction: not significant (in VAR IRFs).
- Additional reported results (not robust across specifications):
  - A 100 bps tightening surprise typically results in an appreciation of the exchange rate at impact by about 1.5 percent.
  - Reaction of equity prices: not significant, though with the correct sign.

### Yield curve, statements, minutes, and forward guidance
- Monetary policy surprises affect the medium- and long-end of the yield curve more than other shocks.
- Forward guidance from statements seems to shift the entire yield curve.
- Surprise actions tend to tilt rather than shift the curve.
- Minutes deliver a noisier signal.
- IPoM provides substantial forward guidance for the policy rate, inducing changes in medium- and long-term yields superior to those induced by “surprise” meetings.

### Summary of conclusions
- CBC predictability: generally high (except during the GFC) and has increased over time as:
  - the number of surprise meetings has declined, and
  - the overall number of meetings with a change in the policy rate has declined,
  - even though macroeconomic volatility has not changed much.
- Efficacy of statements at times of monetary policy surprises is high.
- Inflation-forecast dispersion increases when markets are surprised by the CBC’s dovish tone.
- Proxy-VAR results reaffirm that a 100 bps tightening shock induces a hump-shaped policy rate response and a peak IMACEC decline of about 2 pp. after 1 year; CPI initially rises then falls to 1 percent lower after 2-3 years.

*Source: wp18156 - Section 3*

### Section 4

### wp18156 - Section 4

### Effects of monetary policy minutes surprises on stock and FX markets (Table VI)
- Regressions (1) and (2): stock market index changes (IGPA) and Peso-USD changes on the 3-month swap rate surprise at minutes release.
- Coefficients:
  - 3-month swap rate (+): -4.470* (standard error 2.264) for IGPA-3m swap; -2.857 (standard error 1.727) for FX-3m swap
  - 3-month swap rate (-): 7.390*** (standard error 2.671) for IGPA-3m swap; 5.959*** (standard error 2.037) for FX-3m swap
  - Constant: 0.224*** (standard error 0.0811) for IGPA-3m swap; 0.112* (standard error 0.0619) for FX-3m swap
- Sample and fit:
  - Observations: 140 (both regressions)
  - R-squared: 0.064 (IGPA), 0.065 (FX)
- Significance notation reported: *** p<0.01, ** p<0.05, * p<0.1

### Dispersion of inflation expectations — interquartile range (Table VII-A)
- Dependent variable: Inf. disp. = interquartile range of 12-month ahead inflation expectations in the next month. Sample period 2003-2017.
- Common covariates with coefficients (standard errors):
  - Inflation dispersion(t-1): coefficients across specifications: 0.390*** (0.0656); 0.380*** (0.0672); 0.392*** (0.0664); 0.378*** (0.0678); 0.447*** (0.0742); 0.450*** (0.0744); 0.447*** (0.0755); 0.447*** (0.0760)
  - Inflation expectation: 0.0543*** (0.0147); 0.0539*** (0.0147); 0.0531*** (0.0147); 0.0521*** (0.0147); 0.0510*** (0.0156); 0.0512*** (0.0156); 0.0507*** (0.0156); 0.0507*** (0.0170)
- Selected surprise coefficients:
  - Median f. surprise: -0.106** (0.0534) in specification (3)
  - Median f. surprise (+): 0.0149 (0.180) in specification (4)
  - Median f. surprise (-): -0.121** (0.0576) in specification (4)
  - Mean f. surprise: -0.0978* (0.0587) in specification (5)
  - Mean f. surprise (+): 0.0901 (0.208) in specification (6)
  - Mean f. surprise (-): -0.122* (0.0642) in specification (6)
  - 3-m rate IPoM surpri se: -0.126 (0.258) in specification (7)
  - 3-m rate minutes surprise and plus/minus splits show small and statistically insignificant coefficients in shown specs.
- Constants: range from 0.0174 (0.0462) to 0.0160 (0.0526)
- Sample sizes and fit:
  - Observations: 192 for specs (1)-(4); 141 for specs (5)-(6); 140 for specs (7)-(8)
  - R-squared: 0.276; 0.278; 0.272; 0.275; 0.324; 0.327; 0.320; 0.320
- Standard errors in parentheses; significance: *** p<0.01, ** p<0.05, * p<0.1

### Dispersion of inflation expectations — standard deviation (Table VII-B)
- Dependent variable: Inf. disp. = standard deviation of 12-month ahead inflation expectations in the next month. Sample period 2003-2017.
- Common covariates with coefficients (standard errors):
  - Inflation dispersion(t-1): 0.443*** (0.0629); 0.436*** (0.0651); 0.441*** (0.0635); 0.434*** (0.0651); 0.470*** (0.0696); 0.465*** (0.0693); 0.469*** (0.0714); 0.470*** (0.0717)
  - Inflation expectation: 0.0439*** (0.00966); 0.0439*** (0.00969); 0.0436*** (0.00963); 0.0434*** (0.00966); 0.0424*** (0.0105); 0.0421*** (0.0105); 0.0410*** (0.0106); 0.0417*** (0.0116)
- Selected surprise coefficients:
  - Median f. surprise: -0.0571 (0.0352) in spec (3)
  - Median f. surprise (-): -0.0632 (0.0384) in spec (4)
  - Mean f. surprise: -0.0617 (0.0386) in spec (5)
  - Mean f. surprise (-): -0.0696 (0.0425) in spec (6)
  - 3-m rate IPoM surprise: -0.358** (0.169) in spec (7)
  - 3-m rate IPoM surpr. (-): -0.591*** (0.225) in spec (8)
  - 3-m rate minutes surprise coefficients are positive but not statistically significant in reported specs.
- Constants: range from 0.0164 (0.0302) to 0.0182 (0.0349)
- Sample sizes and fit:
  - Observations: 192 for specs (1)-(4); 141 for specs (5)-(6); 140 for specs (7)-(8)
  - R-squared: 0.370; 0.371; 0.370; 0.371; 0.432; 0.442; 0.419; 0.419
- Standard errors in parentheses; significance: *** p<0.01, ** p<0.05, * p<0.1

### Robustness: Effects of monetary policy surprises on swap rates since 2010 (Table VIII-A)
- Regressions over sub-sample 2010-2017.
- Anticipated surprises coefficients (standard errors):
  - 3 month swap: 0.0767** (0.0371)
  - 6 month swap: 0.0931*** (0.0331)
  - 1 year swap: 0.0367 (0.0363)
  - 5 year swap: 0.0135 (0.0300)
  - 10 year swap: 0.00541 (0.0249)
- Unanticipated surprises coefficients (standard errors):
  - 3 month swap: 0.762*** (0.0816)
  - 6 month swap: 0.713*** (0.0728)
  - 1 year swap: 0.579*** (0.0799)
  - 5 year swap: 0.0961 (0.0661)
  - 10 year swap: 0.0484 (0.0548)
- Constants and model fit:
  - Constants: 0.00625 (0.00534); 0.00615 (0.00476); 0.00809 (0.00523); -0.00309 (0.00433); -0.00324 (0.00359)
  - Observations: 95 (all columns)
  - R-squared: 0.539; 0.576; 0.398; 0.030; 0.011
- Standard errors in parentheses; significance: *** p<0.01, ** p<0.05, * p<0.1

### Robustness: Effects of monetary policy surprises on swap rates excluding 2009Q1 (Table VIII-B)
- Regressions over full sample but exclude 2009Q1.
- Anticipated surprises coefficients (standard errors):
  - 3 month swap: 0.111*** (0.0332)
  - 6 month swap: 0.0551* (0.0279)
  - 1 year swap: 0.0700** (0.0307)
  - 5 year swap: 0.109*** (0.0393)
  - 10 year swap: 0.0206 (0.0270)
- Unanticipated surprises coefficients (standard errors):
  - 3 month swap: 0.709*** (0.0776)
  - 6 month swap: 0.765*** (0.0653)
  - 1 year swap: 0.354*** (0.0534)
  - 5 year swap: 0.165** (0.0683)
  - 10 year swap: 0.0882 (0.0560)
- Constants and model fit:
  - Constants: 0.00577 (0.00534); 0.00299 (0.00449); 0.00532 (0.00496); -0.00762 (0.00634); -0.00457 (0.00454)
  - Observations: 135; 135; 171; 171; 156
  - R-squared: 0.474; 0.556; 0.247; 0.086; 0.022
- Note: Tables A1 and A2 regress swap rates on anticipated and un-anticipated policy rate decisions based on mean-forecast monetary policy surprises over different subsamples. Regressions in Table A1 are over 2010-2017; regressions in Table A2 are over the full sample but exclude 2009Q1.
- Standard errors in parentheses; significance: *** p<0.01, ** p<0.05, * p<0.1

### Robustness: Forward guidance effects of statements (Table IX)
- Surprises constructed using the mean-forecast surprise. Surprise dummy = 1 if there was at least some surprise; 0 otherwise. "3-mXdummy" = interaction between the surprise dummy and the 3-month swap rate. Of 138 observations, 79 had a surprise meeting.
- Variables and key coefficients (standard errors):
  - 3-m swap (direct) as regressor:
    - Column (2) 6 month swap: 1.138*** (0.205)
    - Column (3) 1 year swap: 1.525*** (0.283)
    - Column (4) 2 year swap: 1.511*** (0.242)
    - Column (5) 5 year swap: 1.208*** (0.232)
    - Column (6) 10 year swap: 0.881*** (0.207)
  - surprise dummy: coefficients range from 0.0170 (0.0128) to -0.00463 (0.00874); none reported as significant.
  - 3-mXdummy (interaction): -0.767*** (0.205); -1.331*** (0.284); -1.345*** (0.243); -1.102*** (0.233); -0.721*** (0.208) across relevant columns
  - Unanticipated surprise: 0.535*** (0.0449); 0.234*** (0.0432); 0.142** (0.0597); 0.0970* (0.0510); 0.0160 (0.0489); -0.000344 (0.0436)
  - Anticipated surprise: 0.0316 (0.0336); -0.00886 (0.0227); -0.0136 (0.0313); -0.0353 (0.0268); 0.00898 (0.0257); -0.0512** (0.0229)
- Constants and fit:
  - Constants: 0.000259 (0.00960); 0.00288 (0.00644); 0.000224 (0.00891); -0.00600 (0.00761); -0.00306 (0.00730); -0.00234 (0.00651)
  - Observations: 138 (all columns)
  - R-squared: 0.648; 0.755; 0.412; 0.393; 0.238; 0.204
- Standard errors in parentheses; significance: *** p<0.01, ** p<0.05, * p<0.1

*Source: wp18156 - Section 4*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2018/wp18156.pdf_