## A New Dataset of High-Frequency Monetary Policy Shocks — Introduction and Annex II (wpiea2024224-print-pdf)

## Source details

**Canonical URL:** [A New Dataset of High-Frequency Monetary Policy Shocks — Introduction and Annex II (wpiea2024224-print-pdf)](https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024224-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2024/english/wpiea2024224-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2024/english/wpiea2024224-print-pdf.pdf.json)

---

### Purpose and contributions
- Introduces a newly-constructed cross-country database of monetary policy shocks using the high-frequency method for 29 countries from 2000 to 2022.
- Key contributions:
  - Constructs monetary policy shocks at a daily frequency for 29 countries corresponding to 20 central banks (12 AE central banks and 8 EM central banks).
  - Provides the first cross-country database with standardized monetary policy shocks using a uniform high-frequency method (one-year interest rate swaps, IRS).
  - Introduces a simple framework that splits high-frequency monetary policy surprises into an exogenous monetary policy shock and an endogenous central bank information effect.
  - Uses the shocks in panel local projections to estimate effects on asset prices and to study high-frequency spillovers across countries.
  - Documents a novel empirical finding that monetary policy events of small open economy central banks have substantial spillovers to interest rates in other countries, and finds evidence consistent with an exchange rate puzzle in EMs (contractionary shocks leading to small exchange rate depreciations).

### Coverage and data summary
- Sample period: 2000 to 2022.
- Scope:
  - 29 countries (21 advanced economies (AEs) and 8 emerging markets (EMs)).
  - 20 central banks.
  - 3,545 monetary policy events collected.
  - 60 percent of events are from AEs and 40 percent from EMs.
- For the euro area, financial data are collected for 10 countries (bringing total country count to 29).
- Announcements sourced from the Bloomberg Economic Calendar; with the exception of China, all announcements are associated with a central bank meeting.
- Meeting frequency varies by country (from four to twelve times per year). Regular meeting spacing is one to three months.
- Emergency meetings are classified as meetings taking place outside the regular schedule; emergency meetings are relatively rare and concentrated around major crises (9/11 aftermath, Global Financial Crisis, European sovereign debt crisis, 2013-2014 Taper Tantrum, Covid-19 pandemic).
- Average collection intensity: about 40 meetings per quarter, of which 20 are for AEs and 20 for EMs.

### Standardization of surprises (financial instrument choice)
- Surprises are measured consistently across countries using one-year interest rate swaps (IRS) where the index rate has the shortest maturity possible (for most countries the index rate is the overnight rate).
- Rationale for one-year IRS:
  - Short-maturity instruments capture effects related to the level of the policy rate relative to the effective lower bound (citing Brennan et al., 2024).
  - The one-year IRS surprise equals a weighted average of the “target” factor (unexpected change in the policy rate) and the “path” factor (unexpected change in the future path of policy), capturing forward guidance and large-scale asset purchase features.

### Construction of high-frequency monetary policy surprises
- Raw monetary policy surprise definition:
  - mps_n t ≡ E_n t i_n t+j − E_n t−1 i_n t+j (horizon j equals one year).
- Timing rules for measuring surprises:
  - For announcements during or before market open: change in closing price compared to closing price the day before the announcement.
  - For announcements after market close: difference between the closing price on the day after the announcement and the day of the announcement.
  - For weekend announcements: difference between the closing price on Monday and the previous Friday.

### Identification framework and decomposition (main equations and sources)
- Central bank reaction function:
  - i_n t+j = f_n (X_n t+k) + μ_n t+k
  - μ_n t+k is an exogenous monetary policy “shock” (horizon j equals one year).
- Market expectation the day before the event:
  - E_n t−1 i_n t+j = g_n t−1 (E_n t−1 X_n t+k) + E_n t−1 μ_n t+k
- Monetary policy surprise ex-post predictable components:
  - mps_n t ≡ ΔE_n t μ_n t+k + [ g_n t (E_n t X_n t+k) − g_n t (E_n t−1 X_n t+k) ] + Δg_n t (E_n t−1 X_n t+k)
- The three sources of a monetary policy surprise:
  1. An exogenous monetary policy shock.
  2. A “central bank response to news” effect.
  3. A “central bank information” (CBI) effect.
- Imposed structure on beliefs and dynamics:
  - g_n t (E_n t X_n t+k) = (c_n t−1 + φ_n t) E_n t X_n t+k
  - X_n t+k = γ_n X_n t + γ̃_n t+k
  - X_n t = ζ_n X_n t−1 + η̃_n t
  - γ̃_n t+k and η̃_n t are vectors of exogenous i.i.d. shocks; c_n t−1 is the prior of the private sector; φ_n t is an update to private-sector beliefs.
- Predictive regression for surprises:
  - mps_n t = α_n + β_n X_n t−1 + ε̃_n t
  - β_n ≡ γ_n ζ_n φ̅_n and ε̃_n t ≡ ΔE_n t μ_n t+k + (c_n t−1 + φ_n t) η̃_n t.
- Orthogonalized surprise:
  - mps_n t^o = mps_n t − β̂_n X_n t−1 (uncorrelated with X_n t−1).

### Decomposition into monetary policy shock and central bank information shock
- Orthogonalized surprise still contains central bank information component (c_n t−1 + φ_n t) η̃_n t.
- Decomposition:
  - mps_n t^o ≡ ε_n t^MP + ε_n t^CBI
  - ε_n t^CBI is the central bank information shock.
- Identification strategy:
  - Uses the response of stock prices to mps_n t^o:
    - If a positive surprise reflects better macro fundamentals (CBI), stock prices will increase (at rate s_n t) despite higher discount rates.
    - If the surprise is a true tightening MP shock, stock prices should decline.
- Preferred classification: “poor man’s sign restrictions” using co-movement of mps_n t^o and stock price change to separate ε_n t^MP and ε_n t^CBI.

### Implementation: predictors and estimation
- Equation (8) estimated using the elastic net operator with predictors:
  - Stock prices: the growth rate, 65 trading days before the central bank event to the day before the event.
  - Exchange rate: growth rate of NEER from three months (65 trading days) before the event to the day before.
  - Sovereign bonds: changes in 1-year and 10-year yields from three months (65 trading days) before the event to the day before.
  - Sovereign yield curve slope: change between the 10-year yield and the 1-year yield over three months (65 trading days).
  - Commodity prices: growth rate of the Bloomberg Commodity Spot Price index from three months (65 trading days) to the day before.
  - Financial market volatility: change in the CBOE Volatility Index (VIX) from three months (65 trading days) to the day before.
  - Expected macro fundamentals: one-year ahead mean forecast of the 3-month interest rate, year-on-year percentage change in real GDP, and year-on-year percentage change of CPI from Consensus Forecasts.
  - Forecast errors of the above one-year ahead mean forecasts from Consensus Forecasts.
- Only data available on the day before the central bank meeting are used; procedure applied country by country.
- The elastic net reduces overfitting and selects predictors that help predict ex-post surprises out of sample.
- For 60 percent of countries in the sample, the elastic net selects no variables (ability to predict ex-post surprises is low).
- Reported R-squared across countries: ranges from 0 to 0.22, with a mean of 0.04 and a median of 0.

### Orthogonalization and shock classification (operational definitions)
- Poor man’s sign restrictions:
  - ε_n t^MP,1 = mps_n t^o if mps_n t^o × s_n t ≤ 0, otherwise 0.  (Equation 11)
  - ε_n t^CBI,1 = 0 if mps_n t^o × s_n t ≤ 0, otherwise mps_n t^o.  (Equation 12)
- Alternative rotational sign restrictions (Jarociński, 2022) noted but not used.

### Summary statistics of monetary policy surprises and shocks (basis points)
- Event days — Advanced economies (Obs. = 2,216):
  - mps_n t: Mean -0.2; Median 0.0; Std. dev. 9.6; Min -85.0; Max 306.5
  - mps_n t^o: Mean 0.0; Median 0.0; Std. dev. 9.5; Min -73.3; Max 305.2
  - ε_n t^MP,1: Mean 0.0; Median 0.0; Std. dev. 8.4; Min -73.3; Max 305.2
  - ε_n t^CBI,1: Mean 0.0; Median 0.0; Std. dev. 4.3; Min -46.0; Max 27.5
- Event days — Emerging markets (Obs. = 1,329):
  - mps_n t: Mean 0.3; Median 0.0; Std. dev. 11.5; Min -93.0; Max 137.0
  - mps_n t^o: Mean 0.0; Median -0.2; Std. dev. 11.2; Min -89.6; Max 137.2
  - ε_n t^MP,1: Mean -0.1; Median 0.0; Std. dev. 9.0; Min -89.6; Max 137.2
  - ε_n t^CBI,1: Mean 0.1; Median 0.0; Std. dev. 6.6; Min -57.6; Max 75.6
- Other days — Advanced economies (Obs. = 61,237):
  - mps_n t: Mean 0.0; Median 0.0; Std. dev. 4.0; Min -374.5; Max 121.0
  - mps_n t^o: Mean 0.0; Median 0.0; Std. dev. 4.0; Min -371.0; Max 120.7
  - ε_n t^MP,1: Mean 0.0; Median 0.0; Std. dev. 2.9; Min -371.0; Max 120.7
  - ε_n t^CBI,1: Mean 0.0; Median 0.0; Std. dev. 2.7; Min -124.3; Max 103.7
- Other days — Emerging markets (Obs. = 37,344):
  - mps_n t: Mean 0.0; Median 0.0; Std. dev. 6.8; Min -123.0; Max 167.0
  - mps_n t^o: Mean 0.0; Median 0.0; Std. dev. 6.8; Min -125.1; Max 165.4
  - ε_n t^MP,1: Mean 0.0; Median 0.0; Std. dev. 5.3; Min -120.4; Max 165.4
  - ε_n t^CBI,1: Mean 0.0; Median 0.0; Std. dev. 4.3; Min -125.1; Max 127.6
- Monetary policy surprises center around zero for both AEs and EMs; EMs display more dispersion and fatter tails.
- For the ECB, the authors’ daily surprises correlate 0.84 with intra-day one-year OIS surprises from Altavilla et al. (2019).

### Frequency and liquidity
- Bid-ask spreads (IRS rates) suggest liquidity:
  - Advanced economies average bid-ask spread about 3 basis points between 2015 and 2021.
  - Within AEs: 2 basis points in Australia, Canada, the eurozone and Japan; 5 basis points in New Zealand and Norway.
  - Emerging markets average bid-ask spread about 5 basis points; some EMs (including Mexico and Thailand) have spreads similar to AEs.

### Empirical approach — same-day effects and specification
- Same-day specification:
  - Δy_n t = α_n + β_m n t + δ_t + ε_n t  (equation 14)
  - Δy_n t is the daily change in an interest rate or asset price in country n; m_n t is the monetary policy surprise or shock.
  - α_n is a country-specific fixed effect; δ_t is a year-month fixed effect.
  - Dependent variables include sovereign bond yields at different maturities, FX-denominated sovereign bond spreads, nominal effective and bilateral exchange rates, and stock market indices.
  - Variables are winsorized at the 1 percent tails; standard errors clustered at the country level.

### Sovereign bond yield same-day effects (impact per 100 basis points surprise; basis points)
- Advanced Economies — impact of a 100 basis point surprise:
  - 3-month yield: 40; SE 7.6; R2 0.25
  - 1-year yield: 68; SE 4.6; R2 0.48
  - 2-year yield: 77; SE 4.3; R2 0.54
  - 5-year yield: 71; SE 6.1; R2 0.46
  - 10-year yield: 45; SE 4.8; R2 0.36
- Emerging Markets — impact of a 100 basis point surprise:
  - 3-month yield: 40; SE 7.8; R2 0.29
  - 1-year yield: 38; SE 5.5; R2 0.36
  - 2-year yield: 52; SE 8.1; R2 0.45
  - 5-year yield: 50; SE 4.9; R2 0.37
  - 10-year yield: 46; SE 5.7; R2 0.38
- Three main results:
  - Stronger effects on the middle of the yield curve (1-year, 2-year, 5-year).
  - Pass-through to medium-term sovereign bond yields is higher in AEs than in EMs (e.g., 1-year: 68 in AEs vs. 38 in EMs; 2-year: 77 in AEs vs. 52 in EMs; 5-year: 71 in AEs vs. 50 in EMs).
  - Transmission of orthogonalized surprises (mps_n t^o) to sovereign yields is almost identical to transmission of raw surprises (mps_n t).

### Other same-day asset responses (key coefficients and statistics)
- Stock market indices: associated with a 16 percent same-day decline in stock market indices on average in both groups of countries.
- NEER and ER vs. USD same-day responses (selected Table 4 coefficients):
  - NEER (Advanced economies): 2.1; SE 0.7; R2 0.16 (example entry)
  - ER vs. USD (Advanced economies): 2.2; SE 0.8; R2 0.31 (example entry)
  - NEER (Emerging markets): -0.2; SE 0.4; R2 0.15 (example entry)
  - ER vs. USD (Emerging markets): -0.4; SE 0.7; R2 0.17 (example entry)
- Sovereign bond spread (Table 4 examples):
  - Advanced economies: 1.1; SE 3.2; R2 0.31 (example)
  - Emerging markets: 1.9; SE 1.8; R2 0.24 (example)
- Note: Table 4 reports multiple columns; above are representative exact numeric entries as provided in the source excerpt.

### Dynamic impact (local projections; up to 65 trading days)
- Impulse responses estimated from:
  - y_n,t+h − y_n,t−1 = δ_n,h + δ_t+h + β_h m_n,t + γ′ x_n,t−1 + u_n,t  (equation 15)
- Controls include three-month growth rates of stock index, NEER, bilateral USD exchange rate, global commodity prices, Brent crude oil prices, three-month changes in 3-month, 1-year, and 10-year government bond yields, and the VIX.
- Horizon examined up to 65 trading days (approximately three months).
- Impulse responses for orthogonalized surprises and central bank information shocks reported in Annex III (not reproduced here).

### Sovereign bond yields: magnitude and persistence (dynamic results)
- 3-month yields:
  - Cumulative, persistent impacts of about 150 basis points after three months in AEs (for a 100 basis points surprise).
  - Cumulative, persistent impacts of about 125 basis points after three months in EMs.
- 1-year yields:
  - After same-day impact of a 100 basis points surprise:
    - Impact doubles in AEs and triples in EMs after six weeks, rising to about 150 basis points.
    - Persists at least three months in both AEs and EMs.
- 2-year yields:
  - Transmission patterns similar to the 1-year yield.
- 10-year yields:
  - Same-day impact similar to shorter maturities, but effects increase less over time.
  - Flattening around 60 basis points higher after three months for monetary policy surprises.
  - Monetary policy shocks: impact rises gradually to about 90 basis points after three months.

### Exchange rates: divergent AE vs EM responses and the "exchange rate puzzle"
- Advanced economies (AEs):
  - Following a 100 basis points monetary policy shock, NEERs in AEs:
    - Appreciate quickly, with a peak impact of 6.3 percent appreciation after 20 trading days.
    - Remain at 3.3 percent after three months.
- Emerging markets (EMs):
  - Following a 100 basis points monetary policy shock:
    - NEER shows a slight depreciation of about 0.8 percent after 20 trading days.
    - Bilateral exchange rates vis a vis the US dollar depreciate (header statistic: the bilateral depreciation against the US dollar is 2.1 percent on average).
    - Effects disappear after three months.
- Puzzle interpretation:
  - Finding for EMs is "unexpected from standard open-economic macroeconomic models".
  - Proposed explanation: fiscal dominance and increases in real interest rates can increase default risks in EMs; contractionary monetary policy can lead to nominal depreciation if changes in the risk premium dominate the interest rate differential.

### Sovereign bond spreads: asymmetric AE vs EM responses (dynamic)
- Following a 100 basis points monetary policy surprise:
  - EMs: sovereign spreads increase by 31 basis points 20 trading days after the surprise.
  - AEs: sovereign spreads fall by 16 basis points 20 trading days after the surprise.
  - After three months:
    - EM spreads are 38 basis points higher.
    - AE spreads have fallen by 29 basis points.
- For monetary policy shocks (orthogonalized):
  - EMs: 100 basis points shock → spreads are 63 basis points higher after three months.
  - AEs: 100 basis points shock → spreads are -20 basis points after three months.

### High-frequency spillovers (international spillovers from AE source countries)
- Specification pools observations for all countries n (AEs and EMs) other than source country m and estimates effects over a three trading day horizon to account for time zone differences (equation 16).
- Dependent variables include IRS rates and 3-month, 1-year, 2-year, and 10-year sovereign yields.
- Controls z_t include three-day change in VIX, three-day growth rate of Brent oil price, and three-day change in the one-year US Treasury yield for non-US source countries.
- Key spillover findings (impacts per 100 basis points surprise or shock; selected exact entries from Table 5):
  - United States: Swap rate impacts 19 (surprise) and 21 (shock); 3-month 4 and 15; 1-year 13 and 17; 2-year 0 and 25; 10-year -21 and 3.
  - Eurozone (ECB): Swap rate impacts 39 and 41.
  - Japan (BoJ): Swap rate impacts 38 and 71.
  - Australia: Swap rate impacts 24 and 27.
  - Canada: Swap rate impacts 14 and 26.
  - Sweden: Swap rate impacts 30 and 36.
  - United Kingdom (BoE): Swap rate impacts 28 and 21.
- Summary statistics for non-US source countries (Table 5):
  - Mean non-US: Swap rate impacts 24 and 28; Median non-US: Swap rate impacts 24 and 26.
- Persistence and maturities:
  - These spillover impacts persist after 10 trading days (Annex Table A.7).
  - Median pass-through of a 100 basis points monetary policy shock by non-US central banks to foreign bond yields:
    - 1-year: 18 basis points
    - 2-year: 24 basis points
    - 10-year: 29 basis points
  - Magnitudes are similar but slightly smaller for central bank information shocks.

### Interpretation of spillovers and channels
- Large central banks (Fed, ECB, BoJ, BoE) generate sizeable spillovers to foreign IRS and sovereign yields.
- Small open economies (Australia, Canada, Sweden) also generate substantial spillovers.
- Possible channels:
  - Trade and financial flows (standard channels).
  - Central bank information effects: when macro shocks are correlated across countries, foreign monetary policy decisions update private-sector estimates of global fundamentals and domestic expectations.
- Findings suggest central bank information effects may explain part of spillovers from small open economies.

### Conclusions and research implications
- Dataset introduced:
  - Cross-country dataset of monetary policy shocks for 21 advanced economies and 8 emerging markets from 2000 to 2022.
  - High-frequency identification via changes in interest rate swap rates around central bank announcements.
  - Procedure: remove "central bank response to news" effect and decompose remaining surprises into exogenous monetary policy shocks and central bank information shocks.
- Two notable empirical findings challenging conventional theory:
  - Exchange rate puzzle in EMs: contractionary monetary policy shocks lead to small exchange rate depreciations rather than appreciations.
  - Significant spillovers from small open-economy central banks to interest rates abroad, implicating information channels in addition to trade and financial channels.
- Research avenues:
  - Further investigation into drivers of exchange rate dynamics in EMs in response to monetary policy shocks.
  - Deeper exploration of mechanisms underlying transmission of monetary policy spillovers.
  - Use the dataset as a starting point for future research.

### Annex II — Framework (definitions and derivation)
- Definition of the monetary policy surprise (Equation (A.1)):
  - mps_n t ≡ ΔE_n t μ_n t+k + [ g_n t (E_n t X_n t+k) − g_n t−1 (E_n t−1 X_n t+k) ].
- Derivation using equation (5) (equation (A.2)):
  - mps_n t = ΔE_n t μ_n t+k + (c_n t−1 + φ_n t) E_n t X_n t+k − c_n t−1 E_n t−1 X_n t+k.
- Expectations from equations (6) and (7):
  - E_n t X_n t+k = γ_n (ζ_n X_n t−1 + η̃_n t)
  - E_n t−1 X_n t+k = γ_n ζ_n X_n t−1
- Inserting into (A.1)/(A.2) gives (A.3):
  - mps_n t = ΔE_n t μ_n t+k + γ_n ζ_n φ_n t X_n t−1 + (c_n t−1 + φ_n t) η̃_n t.
- Mapping to main text (equation (8)):
  - β_n ≡ γ_n ζ_n φ̅_n
  - ε̃_n t ≡ ΔE_n t μ_n t+k + (c_n t−1 + φ_n t) η̃_n t

_Italic source attribution: IMF Working Paper — "A New Dataset of High-Frequency Monetary Policy Shocks", Introduction and Section II (Data and methodology) excerpts from wpiea2024224-print-pdf._

### Introduction ...........................................................................................................

### Introduction

### Purpose and contributions
- Introduces a newly-constructed cross-country database of monetary policy shocks using the high-frequency method for 29 countries from 2000 to 2022.
- Key contributions:
  - Constructs monetary policy shocks at a daily frequency for 29 countries corresponding to 20 central banks (12 AE central banks and 8 EM central banks).
  - Provides the first cross-country database with standardized monetary policy shocks using a uniform high-frequency method (one-year interest rate swaps, IRS).
  - Introduces a simple framework that splits high-frequency monetary policy surprises into an exogenous monetary policy shock and an endogenous central bank information effect.
  - Uses the shocks in panel local projections to estimate effects on asset prices and to study high-frequency spillovers across countries.
  - Documents a novel empirical finding that monetary policy events of small open economy central banks have substantial spillovers to interest rates in other countries, and finds evidence consistent with an exchange rate puzzle in EMs (contractionary shocks leading to small exchange rate depreciations).

### Coverage and data summary
- Sample period: 2000 to 2022.
- Scope:
  - 29 countries (21 advanced economies (AEs) and 8 emerging markets (EMs)).
  - 20 central banks.
  - 3,545 monetary policy events collected.
  - 60 percent of events are from AEs and 40 percent from EMs.
- For the euro area, financial data are collected for 10 countries (bringing total country count to 29).
- Announcements sourced from the Bloomberg Economic Calendar; with the exception of China, all announcements are associated with a central bank meeting.
- Meeting frequency varies by country (from four to twelve times per year). Regular meeting spacing is one to three months.
- Emergency meetings are classified as meetings taking place outside the regular schedule; emergency meetings are relatively rare and concentrated around major crises (9/11 aftermath, Global Financial Crisis, European sovereign debt crisis, 2013-2014 Taper Tantrum, Covid-19 pandemic).
- Average collection intensity: about 40 meetings per quarter, of which 20 are for AEs and 20 for EMs.

### Standardization of surprises (financial instrument choice)
- Surprises are measured consistently across countries using one-year interest rate swaps (IRS) where the index rate has the shortest maturity possible (for most countries the index rate is the overnight rate).
- Rationale for one-year IRS:
  - Short-maturity instruments capture effects related to the level of the policy rate relative to the effective lower bound (citing Brennan et al., 2024).
  - The one-year IRS surprise equals a weighted average of the “target” factor (unexpected change in the policy rate) and the “path” factor (unexpected change in the future path of policy), capturing forward guidance and large-scale asset purchase features.

### Construction of high-frequency monetary policy surprises
- Raw monetary policy surprise definition (equation (3) in source):
  - mps_n t ≡ E_n t i_n t+j − E_n t−1 i_n t+j
  - where the horizon j equals one year in the dataset.
- Timing rules for measuring surprises:
  - For announcements during or before market open: change in closing price compared to closing price the day before the announcement.
  - For announcements after market close: difference between the closing price on the day after the announcement and the day of the announcement.
  - For weekend announcements: difference between the closing price on Monday and the previous Friday.

### Identification framework and decomposition
- Central bank reaction function (notation from source):
  - i_n t+j = f_n (X_n t+k) + μ_n t+k    (equation (1))
  - μ_n t+k is an exogenous monetary policy “shock” (horizon j equals one year).
- Market expectation the day before the event:
  - E_n t−1 i_n t+j = g_n t−1 (E_n t−1 X_n t+k) + E_n t−1 μ_n t+k    (equation (2))
- Monetary policy surprise ex-post predictable components (equation (4) in source):
  - mps_n t ≡ ΔE_n t μ_n t+k + [ g_n t (E_n t X_n t+k) − g_n t (E_n t−1 X_n t+k) ] + Δg_n t (E_n t−1 X_n t+k)
- The three sources of a monetary policy surprise (as summarized in the source):
  1. An exogenous monetary policy shock.
  2. A “central bank response to news” effect (when the event changes the private sector’s estimate of the central bank reaction function).
  3. A “central bank information” (CBI) effect (when the central bank’s observation of the state of the economy differs from the previous expectation of the private sector).
- Imposed structure on beliefs and dynamics (equations (5)–(7) in source):
  - g_n t (E_n t X_n t+k) = (c_n t−1 + φ_n t) E_n t X_n t+k   (equation (5))
  - X_n t+k = γ_n X_n t + γ̃_n t+k   (equation (6))
  - X_n t = ζ_n X_n t−1 + η̃_n t   (equation (7))
  - γ̃_n t+k and η̃_n t are vectors of exogenous i.i.d. shocks; c_n t−1 is the prior of the private sector; φ_n t is an update to private-sector beliefs.
- Predictive regression for surprises (equation (8) in source):
  - mps_n t = α_n + β_n X_n t−1 + ε̃_n t
  - where β_n ≡ γ_n ζ_n φ̅_n and ε̃_n t ≡ ΔE_n t μ_n t+k + (c_n t−1 + φ_n t) η̃_n t.
- Orthogonalized surprise (equation (9) in source):
  - mps_n t^o = mps_n t − β̂_n X_n t−1
  - by construction mps_n t^o is uncorrelated with macroeconomic data X_n t−1.

### Decomposition into monetary policy shock and central bank information shock
- Orthogonalized surprise still contains central bank information component (c_n t−1 + φ_n t) η̃_n t.
- Decomposition (equation (10) in source):
  - mps_n t^o ≡ ε_n t^MP + ε_n t^CBI
  - ε_n t^CBI is the central bank information shock.
- Identification strategy to disentangle ε_n t^MP and ε_n t^CBI:
  - Uses the response of stock prices to mps_n t^o: if a positive surprise reflects better macro fundamentals (CBI), stock prices will increase (at rate s_n t) despite higher discount rates; if the surprise is a true tightening MP shock, stock prices should decline.

### Relations to existing literature and advantages
- Relates to high-frequency approaches dating to Kuttner (2001), Bernanke and Kuttner (2005), Nakamura and Steinsson (2018), and to studies across the eurozone, UK, China, and a variety of applications to financial, macroeconomic, and firm-level outcomes.
- Advantages of this database over other cross-country datasets:
  - Careful identification of shocks and breadth of coverage.
  - Standardized measurement using the same financial instrument (one-year IRS) to ensure comparability.
  - Panel dimension increases statistical power and permits study of state-dependence under milder exogeneity assumptions.
  - Daily surprises enable study of high-frequency spillovers across countries.

### Empirical findings highlighted in the Introduction
- Confirms results previously found in the literature on asset price responses using an expanded cross-country sample.
- Finds evidence consistent with an exchange rate puzzle in EMs: contractionary monetary policy shocks lead to small exchange rate depreciations in emerging markets.
- Documents that monetary policy events of small open economy central banks have substantial spillovers to interest rates in other countries, suggesting spillovers can manifest through information effects as well as traditional trade and financial channels.

_Italic source attribution: IMF Working Paper — "A New Dataset of High-Frequency Monetary Policy Shocks", Introduction and Section II (Data and methodology) excerpts from wpiea2024224-print-pdf._

### Annex II contains a derivation.

### wpiea2024224-print-pdf - Annex II contains a derivation

### Implementation: construction and predictors of monetary policy surprises
- Monetary policy surprises (푚푝푠푛푡) are constructed using the identity in equation (3) and data on swap rates and central bank announcements.
- Equation (8) is estimated using the elastic net operator with predictors focusing on macroeconomic news and financial variables previously found by Bauer and Swanson (2023a, b) to predict monetary policy surprises:
  - Stock prices: the growth rate, 65 trading days before the central bank event to the day before the event.
  - Exchange rate: the growth rate of a country’s nominal effective exchange rate (NEER) from three months (65 trading days) before the central bank event to the day before the event.
  - Sovereign bonds: the changes in the yields of sovereign bonds with 1-year and 10-year maturities from three months (65 trading days) before the central bank event to the day before the event.
  - Sovereign yield curve: the change in the slope of the sovereign yield curve, measured as the difference between the 10-year yield and the 1-year yield, from three months (65 trading days) before the central bank event to the day before the event.
  - Commodity prices: the growth rate of the Bloomberg Commodity Spot Price index from three months (65 trading days) before the central bank event to the day before the event.
  - Financial market volatility: the change in the Chicago Board Options Exchange's CBOE Volatility Index (VIX) from three months (65 trading days) before the central bank event to the day before the event.
  - Expected macroeconomic fundamentals: the one-year ahead mean forecast of the 3-month interest rate, the year-on-year percentage change in real GDP, and the year-on-year percentage change of the consumer price index (CPI) from Consensus Forecasts.
  - Forecast errors of macroeconomic fundamentals: the forecast errors of the one-year ahead mean forecast of the 3-month interest rate, the year-on-year percentage change in real GDP, and the year-on-year percentage change of the consumer price index (CPI) from Consensus Forecasts.
- Only data available on the day before the central bank meeting are used; procedure applied country by country.
- The elastic net reduces overfitting and selects predictors that help predict ex-post surprises out of sample.
- For 60 percent of countries in the sample, the elastic net selects no variables (ability to predict ex-post surprises is low).
- Reported R-squared across countries: ranges from 0 to 0.22, with a mean of 0.04 and a median of 0.

### Orthogonalization and shock classification
- “Poor man’s sign restrictions” are used to categorize each monetary policy surprise as either a monetary policy shock (휖푛푡푀푃,1) or a central bank information shock (휖푛푡퐶퐵퐼,1) depending on the co-movement between the policy rate surprise and stock price change:
  - 휖푛푡푀푃,1 = 푚푝푠푛푡표 if 푚푝푠푛푡표 × 푠푛푡 ≤ 0, otherwise 0.  (Equation 11)
  - 휖푛푡퐶퐵퐼,1 = 0 if 푚푝푠푛푡표 × 푠푛푡 ≤ 0, otherwise 푚푝푠푛푡표.  (Equation 12)
- The paper notes an alternative “rotational sign restrictions” approach (Jarociński, 2022) but prefers the transparent poor man’s sign restriction approach.

### Summary statistics of monetary policy surprises and shocks
- Table 2: Summary statistics (basis points) for event days and other days, grouped by Advanced Economies (AEs) and Emerging Markets (EMs).
- Event days — Advanced economies (Obs. = 2,216):
  - 푚푝푠푛푡: Mean -0.2; Median 0.0; Std. dev. 9.6; Min -85.0; Max 306.5
  - 푚푝푠푛푡표: Mean 0.0; Median 0.0; Std. dev. 9.5; Min -73.3; Max 305.2
  - 휖푛푡푀푃,1: Mean 0.0; Median 0.0; Std. dev. 8.4; Min -73.3; Max 305.2
  - 휖푛푡퐶퐵퐼,1: Mean 0.0; Median 0.0; Std. dev. 4.3; Min -46.0; Max 27.5
- Event days — Emerging markets (Obs. = 1,329):
  - 푚푝푠푛푡: Mean 0.3; Median 0.0; Std. dev. 11.5; Min -93.0; Max 137.0
  - 푚푝푠푛푡표: Mean 0.0; Median -0.2; Std. dev. 11.2; Min -89.6; Max 137.2
  - 휖푛푡푀푃,1: Mean -0.1; Median 0.0; Std. dev. 9.0; Min -89.6; Max 137.2
  - 휖푛푡퐶퐵퐼,1: Mean 0.1; Median 0.0; Std. dev. 6.6; Min -57.6; Max 75.6
- Other days — Advanced economies (Obs. = 61,237):
  - 푚푝푠푛푡: Mean 0.0; Median 0.0; Std. dev. 4.0; Min -374.5; Max 121.0
  - 푚푝푠푛푡표: Mean 0.0; Median 0.0; Std. dev. 4.0; Min -371.0; Max 120.7
  - 휖푛푡푀푃,1: Mean 0.0; Median 0.0; Std. dev. 2.9; Min -371.0; Max 120.7
  - 휖푛푡퐶퐵퐼,1: Mean 0.0; Median 0.0; Std. dev. 2.7; Min -124.3; Max 103.7
- Other days — Emerging markets (Obs. = 37,344):
  - 푚푝푠푛푡: Mean 0.0; Median 0.0; Std. dev. 6.8; Min -123.0; Max 167.0
  - 푚푝푠푛푡표: Mean 0.0; Median 0.0; Std. dev. 6.8; Min -125.1; Max 165.4
  - 휖푛푡푀푃,1: Mean 0.0; Median 0.0; Std. dev. 5.3; Min -120.4; Max 165.4
  - 휖푛푡퐶퐵퐼,1: Mean 0.0; Median 0.0; Std. dev. 4.3; Min -125.1; Max 127.6
- Monetary policy surprises center around zero for both AEs and EMs; EMs display more dispersion and fatter tails.
- For the ECB, the authors’ daily surprises correlate 0.84 with intra-day one-year OIS surprises from Altavilla et al. (2019).

### Frequency and liquidity
- Correlation between the paper’s daily ECB surprises and intra-day surprises from Altavilla et al. (2019): 0.84.
- Bid-ask spreads (IRS rates) suggest liquidity:
  - Advanced economies average bid-ask spread about 3 basis points between 2015 and 2021.
  - Within AEs: 2 basis points in Australia, Canada, the eurozone and Japan; 5 basis points in New Zealand and Norway.
  - Emerging markets average bid-ask spread about 5 basis points; some EMs (including Mexico and Thailand) have spreads similar to AEs.

### The high-frequency impact of monetary policy shocks — empirical approach
- Empirical specification for same-day effects:
  - Δ푦푛푡 = 훼푛 + 훽푚푛푡 + 훿푡 + 휖푛푡  (equation 14)
  - Δ푦푛푡 is the daily change in an interest rate or asset price in country n; 푚푛푡 is the monetary policy surprise or shock.
  - 훼푛 is a country-specific fixed effect; 훿푡 is a year-month fixed effect.
  - Dependent variables include sovereign bond yields at different maturities, FX-denominated sovereign bond spreads, nominal effective and bilateral exchange rates, and stock market indices.
  - Variables are winsorized at the 1 percent tails; standard errors clustered at the country level.

### Sovereign bond yield responses — same-day effects (Table 3)
- Coefficients represent impact of a 100 basis point increase in the surprise on the corresponding bond yield (basis points). Country and year-month fixed effects included. Unbalanced sample from 2000 to 2022.
- Advanced Economies (columns 1–4) — impact of a 100 basis point surprise:
  - 3-month yield: 40; SE 7.6; R2 0.25
  - 1-year yield: 68; SE 4.6; R2 0.48
  - 2-year yield: 77; SE 4.3; R2 0.54
  - 5-year yield: 71; SE 6.1; R2 0.46
  - 10-year yield: 45; SE 4.8; R2 0.36
- Emerging Markets (columns 5–8) — impact of a 100 basis point surprise:
  - 3-month yield: 40; SE 7.8; R2 0.29
  - 1-year yield: 38; SE 5.5; R2 0.36
  - 2-year yield: 52; SE 8.1; R2 0.45
  - 5-year yield: 50; SE 4.9; R2 0.37
  - 10-year yield: 46; SE 5.7; R2 0.38
- Three main results highlighted:
  - Monetary policy surprises have a stronger effect on sovereign bond yields in the middle of the yield curve (large impacts on 1-year, 2-year, and 5-year yields relative to 3-month or 10-year yields).
  - Pass-through to medium-term sovereign bond yields is higher in AEs than in EMs. Example comparisons for a 100 basis point surprise:
    - 1-year yields: 68 in AEs vs. 38 in EMs (44 percent lower in EMs).
    - 2-year yields: 77 in AEs vs. 52 in EMs (32 percent lower in EMs).
    - 5-year yields: 71 in AEs vs. 50 in EMs (29 percent lower in EMs).
  - Transmission of orthogonalized monetary policy surprises (푚푝푠푛푡표) to sovereign bond yields is almost identical to transmission of raw surprises (푚푝푠푛푡); point estimates, standard errors, and R-squared are very similar.
  - Monetary policy shocks (휖푛푡푀푃,1 and 휖푛푡퐶퐵퐼,1) have effects larger in magnitude than raw monetary policy surprises for both AEs and EMs.

### Other asset prices — summary of key findings
- Sovereign bond spreads:
  - Measured as yield of USD- or Euro-denominated bonds for each country minus yield of similar-profile US Treasury bonds or German government bonds.
  - Effects of monetary policy on sovereign bond spreads are negligible in AEs.
  - In EMs, effects are larger and statistically significant: a 100 basis point positive surprise and contractionary monetary policy shock push up spreads by 1.9 and 5.5 basis points respectively (impacts for AEs: 1.1 and 3.7 basis points, not significant).
- Exchange rates:
  - Advanced Economies: a 100 basis points surprise is associated with a 2.1 and 2.2 percent same-day appreciation of the NEER and the nominal exchange rate vis-à-vis the US dollar, respectively.
  - Emerging Markets: estimated responses to monetary policy surprises are negative (average depreciation of the NEER of 0.2 percent and of the bilateral rate of 0.4 percent), but these impacts are not significant.
  - For contractionary monetary policy shocks, the effect is statistically significant in EMs, indicating that the NEER depreciates by [text ends at provided excerpt].

*IMF Working Papers — A New Dataset of High-Frequency Monetary Policy Shocks.*

### 1.2 percentage points on average, while the bilateral depreciation against the US dollar is 2.1

### wpiea2024224-print-pdf - 1.2 percentage points on average, while the bilateral depreciation against the US dollar is 2.1

### Stock market and other same-day asset responses
- Stock market indices: "associated with a 16 percent same-day decline in stock market indices on average in both groups of countries."
- NEER and ER vs. USD same-day responses (Table 4 coefficients, Advanced economies / Emerging markets):
  - NEER: 2.1; 2.0; 2.0; 2.6; -0.2; -0.2; -1.2; 0.6
  - SE (NEER): 0.7; 0.7; 0.9; 0.8; 0.4; 0.4; 0.8; 0.4
  - R2 (NEER): 0.16; 0.15; 0.16; 0.16; 0.15; 0.15; 0.16; 0.16
  - ER vs. USD: 2.2; 2.1; 1.3; 3.3; -0.4; -0.5; -2.1; 1.0
  - SE (ER vs. USD): 0.8; 0.8; 1.0; 1.0; 0.7; 0.6; 1.2; 0.3
  - R2 (ER vs. USD): 0.31; 0.31; 0.31; 0.31; 0.17; 0.17; 0.19; 0.19
- Stock market (Table 4): 0.0; -0.1; -16.2; 16.0; -2.2; -2.4; -15.8; 12.7
  - SE (Stock market): 0.7; 0.7; 1.1; 1.1; 0.9; 0.9; 1.7; 1.5
  - R2 (Stock market): 0.34; 0.34; 0.56; 0.56; 0.19; 0.19; 0.50; 0.50
- Sovereign bond spread (Table 4): 1.1; 0.6; 3.7; -5.4; 1.9; 2.4; 5.5; -1.2
  - SE (Sovereign bond spread): 3.2; 3.4; 6.1; 3.7; 1.8; 1.5; 2.6; 3.4
  - R2 (Sovereign bond spread): 0.31; 0.31; 0.31; 0.31; 0.24; 0.24; 0.24; 0.24

### Dynamic impact (local projections; horizon up to 65 trading days)
- Methodology:
  - Impulse responses estimated from equation (15): yn,t+h − yn,t−1 = δn,h + δt+h + βh mn,t + γ′ x n,t−1 + un,t.
  - Country- and time-fixed effects included.
  - Controls: three-month growth rates of country stock index, nominal effective exchange rate, bilateral USD exchange rate, global commodity prices, Brent crude oil prices, three-month changes in 3-month, 1-year, and 10-year government bond yields, and the CBOE Volatility Index (VIX).
  - Impulse responses examined up to 65 trading days (approximately three months).
  - Impulse responses for orthogonalized surprises and central bank information shocks reported in Annex III.

### Sovereign bond yields: magnitude and persistence
- Short-term yields (3-month):
  - Effect of a "100 basis points monetary policy surprise" on 3-month yields:
    - Cumulative, persistent impacts of about 150 basis points after three months in AEs.
    - Cumulative, persistent impacts of about 125 basis points after three months in EMs.
  - Transmission similar for monetary policy shocks (right panel referenced).
- 1-year yields:
  - Following same-day impact of a 100 basis points monetary policy surprise:
    - Impact doubles in AEs and triples in EMs after six weeks, rising to about 150 basis points.
    - Persists at least three months in both AEs and EMs.
  - Monetary policy shocks show similar impacts.
- 2-year yields:
  - Transmission patterns similar to the 1-year yield.
- 10-year yields:
  - Same-day impact of a 100 basis points monetary policy surprise is similar to shorter maturities, but effects increase by less over time.
  - Flattening around 60 basis points higher after three months for monetary policy surprises.
  - Monetary policy shocks: impact rises gradually to about 90 basis points after three months.
- Note: 5-year bond yield impulse responses are reported in Annex III.

### Exchange rates: divergent AE vs EM responses and the "exchange rate puzzle"
- Advanced economies (AEs):
  - Following a 100 basis points monetary policy shock, NEERs in AEs:
    - Appreciate quickly, with a peak impact of 6.3 percent appreciation after 20 trading days.
    - Remain at 3.3 percent after three months.
- Emerging markets (EMs):
  - Following a 100 basis points monetary policy shock:
    - NEER shows a slight depreciation of about 0.8 percent after 20 trading days.
    - Bilateral exchange rates vis a vis the US dollar depreciate (exact bilateral statistic in header: "the bilateral depreciation against the US dollar is 2.1 percent on average").
    - Effects disappear after three months.
- Puzzle and literature:
  - Finding for EMs is "unexpected from standard open-economic macroeconomic models".
  - Prior studies cited documenting similar puzzles (Kohlscheen 2014; Blanchard 2005; Hnatkovska et al. 2016; Alberola et al. 2022; Dominguez and Foschi 2024).
  - Proposed explanation: fiscal dominance and increases in real interest rates can increase default risks in EMs; contractionary monetary policy can lead to nominal depreciation if changes in the risk premium dominate the interest rate differential.

### Sovereign bond spreads: asymmetric AE vs EM responses
- Following a 100 basis points monetary policy surprise:
  - Emerging markets (EMs): sovereign spreads increase by 31 basis points 20 trading days after the surprise.
  - Advanced economies (AEs): sovereign spreads fall by 16 basis points 20 trading days after the surprise.
  - After three months:
    - EM spreads are 38 basis points higher.
    - AE spreads have fallen by 29 basis points.
- For monetary policy shocks (orthogonalized):
  - EMs: 100 basis points shock → spreads are 63 basis points higher after three months.
  - AEs: 100 basis points shock → spreads are -20 basis points after three months.

### High-frequency spillovers (international spillovers from AE source countries; equation (16))
- Specification:
  - Pool observations for all countries n (AEs and EMs) other than source country m; estimate yn,t+h − yn,t−1 = δn,h + βh m m≠n,t + γ′ x n,t−1 + ρ′ z t + un,t.
  - Horizon of three trading days for main specification to account for time zone differences.
  - Dependent variables: IRS rates and 3-month, 1-year, 2-year, and 10-year sovereign yields.
  - Controls zt include three-day change in VIX, three-day growth rate of Brent oil price, and three-day change in the one-year US Treasury yield for non-US source countries.
- Key spillover findings (Table 5; impacts per 100 basis points surprise or shock by source country):
  - United States: Swap rate impacts 19 (surprise) and 21 (shock); 3-month 4 and 15; 1-year 13 and 17; 2-year 0 and 25; 10-year -21 and 3.
    - SEs (United States): 2.4; 10.0; 3.5; 8.2; 4.8; 6.5; 3.9; 8.5; 5.5; 6.0
    - R2s reported for each column (United States): 0.08; 0.08; 0.03; 0.03; 0.05; 0.05; 0.05; 0.05; 0.06; 0.06
  - Australia: Swap rate impacts 24 and 27; other maturities reported in Table 5.
    - SEs (Australia): 2.8; 5.3; 3.7; 6.6; 2.9; 4.6; 2.7; 3.7; 2.7; 4.4
    - R2s (Australia): 0.18; 0.18; 0.07; 0.07; 0.11; 0.12; 0.13; 0.13; 0.09; 0.09
  - Canada: Swap rate impacts 14 and 26.
    - SEs (Canada): 3.0; 4.8; 3.0; 5.1; 3.2; 5.4; 3.3; 5.2; 3.5; 5.9
    - R2s (Canada): 0.15; 0.15; 0.07; 0.07; 0.09; 0.09; 0.12; 0.13; 0.07; 0.07
  - Eurozone (ECB): Swap rate impacts 39 and 41.
    - SEs (Eurozone): 5.7; 6.9; 6.3; 10.9; 5.7; 8.1; 6.7; 11.3; 8.1; 10.6
    - R2s (Eurozone): 0.16; 0.16; 0.06; 0.06; 0.11; 0.11; 0.18; 0.18; 0.12; 0.12
  - Japan (BoJ): Swap rate impacts 38 and 71.
    - SEs (Japan): 11.5; 17.2; 10.7; 16.4; 14.1; 17.7; 14.7; 19.7; 15.3; 21.1
    - R2s (Japan): 0.09; 0.09; 0.03; 0.03; 0.04; 0.05; 0.06; 0.07; 0.08; 0.09
  - Norway: Swap rate impacts 6 and 0.
    - SEs (Norway): 2.2; 3.1; 3.3; 2.7; 2.8; 4.8; 4.2; 5.7; 3.3; 4.8
    - R2s (Norway): 0.11; 0.11; 0.04; 0.04; 0.07; 0.07; 0.09; 0.09; 0.04; 0.04
  - New Zealand: Swap rate impacts 8 and 2.
    - SEs (New Zealand): 2.5; 3.1; 2.2; 4.3; 1.9; 3.2; 2.3; 2.7; 1.8; 4.1
    - R2s (New Zealand): 0.08; 0.09; 0.04; 0.04; 0.07; 0.07; 0.08; 0.08; 0.06; 0.06
  - Sweden: Swap rate impacts 30 and 36.
    - SEs (Sweden): 4.0; 5.6; 3.7; 5.8; 4.2; 6.4; 5.2; 9.2; 3.7; 7.9
    - R2s (Sweden): 0.11; 0.11; 0.06; 0.06; 0.09; 0.09; 0.12; 0.13; 0.10; 0.11
  - United Kingdom (BoE): Swap rate impacts 28 and 21.
    - SEs (United Kingdom): 4.9; 6.5; 2.8; 5.3; 5.7; 6.3; 6.7; 8.0; 6.6; 7.0
    - R2s (United Kingdom): 0.11; 0.11; 0.04; 0.04; 0.09; 0.09; 0.11; 0.11; 0.06; 0.07
- Summary statistics for non-US source countries (Table 5):
  - Mean non-US: Swap rate impacts 24 and 28; other maturities reported across columns.
  - Median non-US: Swap rate impacts 24 and 26; other medians reported across columns.
- Persistence and maturities:
  - These spillover impacts persist after 10 trading days (Annex Table A.7).
  - Median pass-through of a 100 basis points monetary policy shock by non-US central banks to foreign bond yields:
    - 1-year: 18 basis points
    - 2-year: 24 basis points
    - 10-year: 29 basis points
  - Magnitudes are similar but slightly smaller for central bank information shocks.

### Interpretation of spillovers and channels
- Large central banks (Fed, ECB, BoJ, BoE) generate sizeable spillovers to foreign IRS and sovereign yields.
- Small open economies (Australia, Canada, Sweden) also generate substantial spillovers.
- Possible channels:
  - Trade and financial flows (standard channels).
  - Central bank information effects: when macro shocks are correlated across countries, foreign monetary policy decisions update private-sector estimates of global fundamentals and domestic expectations.
- The findings suggest central bank information effects may explain part of spillovers from small open economies.

### Conclusions and research implications
- Dataset introduced:
  - Cross-country dataset of monetary policy shocks for 21 advanced economies and 8 emerging markets from 2000 to 2022.
  - High-frequency identification via changes in interest rate swap rates around central bank announcements.
  - Procedure:
    - Remove "central bank response to news" effect (Bauer and Swanson, 2023).
    - Decompose remaining surprises into exogenous monetary policy shocks and central bank information shocks (Jarociński and Karadi, 2020).
- Two notable empirical findings challenging conventional theory:
  - Exchange rate puzzle in EMs: contractionary monetary policy shocks lead to small exchange rate depreciations rather than appreciations.
  - Significant spillovers from small open-economy central banks to interest rates abroad, implicating information channels in addition to trade and financial channels.
- Research avenues:
  - Further investigation into drivers of exchange rate dynamics in EMs in response to monetary policy shocks.
  - Deeper exploration of mechanisms underlying transmission of monetary policy spillovers.
  - Use the dataset as a starting point for future research.

*Source: IMF WORKING PAPERS — A New Dataset of High-Frequency Monetary Policy Shocks (excerpts from the provided PDF content).*

### References

### References (wpiea2024224-print-pdf)

### Core bibliographic sources
- Contains a comprehensive list of cited works on high-frequency identification of monetary policy, monetary policy surprises, information effects, international spillovers, and monetary transmission. Key authors and works include: Albagli et al. (2019); Altavilla et al. (2019); Bauer & Swanson (2023); Bernanke & Kuttner (2005); Blanchard (2005); Brand, Buncic & Turunen (2010); Braun, Miranda-Agrippino & Saha (2023); Brandao‑Marques et al. (2020); Brennan et al. (2024); Bu, Rogers & Wu (2021); Camara (2021); Campbell et al. (2012, 2017); Cesa‑Bianchi et al. (2020); Choi, Willems & Yoo (2024); Cieslak (2018); Cieslak & Schrimpf (2019); Cloyne et al. (2023); Corbet et al. (2021); D’Amico & Farka (2011); Daniel (2001, 2010); Das & Song (2023); Deb et al. (2023); Di Giovanni & Hale (2022); Dominguez & Foschi (2024); Farmer, Nakamura & Steinsson (2024); Fratzscher, Lo Duca & Straub (2016); Georgiadis (2016); Gertler & Karadi (2015); Gilchrist, López‑Salido & Zakrajšek (2015); Gürkaynak et al. (2005, 2021, 2022); Hansen, McMahon & Tong (2019); Hanson & Stein (2015); Hnatkovska, Lahiri & Vegh (2016); IMF, October 2023, Regional Economic Outlook Asia and Pacific, Box 1.2; Jarociński (2022); Jarociński & Karadi (2020); Jeenas (2023); Jordà, Schularick & Taylor (2020); Kalemli‑Özcan (2019); Kerssenfischer (2022); Kohlscheen (2014); Kuttner (2001); Lloyd (2021); Miranda‑Agrippino & Rey (2020); Miranda‑Agrippino & Ricco (2021); Nakamura & Steinsson (2018); Ottonello & Winberry (2020); Pinchetti & Szczepaniak (2023); Romer & Romer (1994, 2000); Rogers, Scotti & Wright (2014); Swanson (2021); Tenreyro & Thwaites (2016); Willems (2020); Witheridge (2024); Wright (2012); Zhang (2022).

### Annex I. Data — A. Sample and variables
- Sample construction:
  - Starting sample: 28 central banks for which interest rate swaps (IRS) are available in Refinitiv Datastream.
  - Dropped countries: Colombia, Iceland, Indonesia, Malaysia, Philippines, Russia, South Africa, and Turkey (reasons: missing one-year government bond yields, IRS maturity too short, or IRS not liquid enough).
  - Remaining central banks: 20.
  - Euro area: financial data collected for 10 countries (Austria, Belgium, Germany, Spain, Finland, France, Ireland, Italy, Netherland, and Portugal).
  - Total countries in dataset: 29.
  - Classification: 22 countries are advanced economies (AEs) and 8 are emerging markets (EMs).
- Surprise variable construction:
  - Use IRS rates to construct the surprise variable. Data extracted from Refinitiv DataStream.
  - Maturities of interest swap series: one year.
  - If available, use overnight index swap (OIS) rate; if OIS not available, use an IRS with the shortest possible floating-tenor.
- Central bank announcement data:
  - Source: Bloomberg Economic Calendar (dates, time of announcements, current and previous monetary policy rate).
  - For most countries, announcements occur before financial market close; exceptions: Brazil, Chile and Philippines announce policy rates after market close.
  - Construction of monetary surprise:
    - For most countries: difference in the swap rate at t and t-1.
    - For Brazil, Chile, and Philippines: difference of the swap rates at t+1 and t.
    - For announcements during weekends: difference of swap rates on Monday and previous Friday.
- Other variables and data handling:
  - Other variables listed in Table A.3 and categorized by frequency into daily, monthly, and quarterly data.
  - Interpolate missing observations for bond yields with maturities ranging from three months to 10 years.
  - Data sources for government bond yields: Haver and the Global Financial Data (GFD), whichever has best coverage.

### Annex I. Data — Table A.1 (interest rate derivative rates)
- Maturities and series:
  - Tenor for all listed swap series: 1Y (one year), except where noted by (*) because one-year swap series are not available.
  - If OIS available, the floating rate is the overnight interbank rate (examples: Australia AONIA; Eurozone EONIA to €STR; United States EFFR).
  - Table entries (country — name of swap series — code — unit — tenor — type of floating rate) include, for example:
    - Australia: ICAP AUD 1Y OIS — IA$OI1Y — A$ — Overnight Interbank rate (AONIA)
    - Brazil: BRAZIL DI-PRE FIXED FLOAT IRS 1Y — BRDPR1Y — C — Overnight Interbank rate (DI)
    - Canada: ICAP CAD 1Y OIS — IC$OI1Y — C$ — Overnight Repo rate (CORRA)
    - China: RFV CNY QM A/365 7D REPO IRS 1Y — CNQMR1Y — CH — 7 days Repo rate
    - Mexico: RFV MXN 28D BND/28D TIIE IRS 13M* — ICMX13M — MP — 28 days Repo rate (TIIE) (*)
    - United Kingdom: RFV GBP AM A/365 3M LIBOR IRS 1Y — TRUK31Y — £ — 3 months Interbank rate (LIBOR)
    - United States: US DOLLAR 1 YEAR OIS — OIUSD1Y — U$ — Overnight Interbank rate (EFFR)
  - Notes:
    - 1/ The series with (*) are selected because one-year swap series are not available in those countries.
    - 2/ Repo market is the secured segment under interbank borrowing market.
  - Source: Refinitiv DataStream.

### Annex I. Data — Table A.2 (central bank meeting frequency)
- Meeting frequency (per year) by country:
  - Australia: 11
  - Brazil: 8
  - Canada: 8
  - Chile: 8
  - China: N/A (see note)
  - Euro Area: 8
  - Hungary: 12
  - India: 6
  - Israel: 8
  - Japan: 8
  - Korea: 8
  - Mexico: 8
  - New Zealand: 7
  - Norway: 8
  - Poland: 11
  - Sweden: 5
  - Switzerland: 4
  - Thailand: 6
  - United Kingdom: 8
  - United States: 8
- Notes:
  - China: series from Das et al. (2023); policy events include changes to reserve requirement ratio (RRR), PBC’s 7-day reverse repo rate, benchmark deposit and lending rates (LDR), and the rate on the PBC’s medium-term lending facility (MLF).
  - ECB: main rates include MRO, deposit facility (DFR), and marginal lending facility (MLF); announcements regarding MRO rate are considered.
  - Japan: main rates include interest rate applied to the Policy-Rate Balance in current account and JGB yields; announcements regarding the policy balance rate are considered.
  - Source: Central bank websites, BIS.

### Annex I. Data — Table A.3 (dependent variables and frequencies)
- Daily frequency variables and units/sources:
  - Government bond yield (3m-10y) — % — Haver; GFD
  - Stock market index — index — Haver
  - Nominal effective exchange rate (NEER) — index — Haver
  - Exchange rate per US$ — LCU — Haver
  - Commodity price index — index — Bloomberg
  - Inflation data — index — State Street
  - Monetary policy rate — % — Haver
  - Sovereign spread (FX denominated bonds) – weighted average — index — IMF
  - United States CBOE Volatility Index — index — Haver
  - BofA Merrill Lynch Option Volatility Estimation Index (1,3, and 6 months) — index — Haver
- Note: GFD refers to Global Financial Data.

### Annex I. Data — Figure A.1 liquidity measures
- Bid-ask spread defined as average daily bid minus average daily ask.
- Figure plots annual means by country, and for EMs and AEs as group (unweighted means).
- Series plotted include EM average and AE average and country lines (examples listed): Chile, China, Hungary, India, Mexico, Thailand (EM); Australia, Canada, Eurozone, Japan, New Zealand, Norway, Sweden, Switzerland, United Kingdom, United States (AE).

### Annex I. Data — B. Surprise series (Table A.5: summary statistics)
- Table A.5 reports summary statistics of monetary policy surprises and shocks. Columns include: Obs, Mean, Median, S.D. for variables labeled 푚푝푠푛푡, 푚푝푠푛푡표, 휖푛푡푀푃,1, and 휖푛푡퐶퐵퐼,1 (see text for definitions).
- Selected country-level entries (exact values as reported):
  - Australia: Obs. 230; Mean -0.3; Median 0.0; S.D. 9.9 (columns correspond to 푚푝푠푛푡 series), and Obs. 230; Mean 0.0; Median -0.2; S.D. 9.2 (푚푝푠푛푡표), Obs. 230; Mean -0.4; Median 0.0; S.D. 7.5 (휖푛푡푀푃,1), Obs. 230; Mean 0.4; Median 0.0; S.D. 5.3 (휖푛푡퐶퐵퐼,1).
  - Brazil: Obs. 110; Mean -0.6; Median 0.0; S.D. 14.2; Obs. 110; Mean 0.0; Median 0.6; S.D. 14.2; Obs. 110; Mean -0.3; Median 0.0; S.D. 11.4; Obs. 110; Mean 0.3; Median 0.0; S.D. 8.5.
  - Canada: Obs. 175; Mean -0.1; Median 0.0; S.D. 7.8; Obs. 175; Mean 0.0; Median 0.1; S.D. 7.7; Obs. 175; Mean 0.1; Median 0.0; S.D. 6.9; Obs. 175; Mean -0.1; Median 0.0; S.D. 3.3.
  - Chile: Obs. 136; Mean 1.8; Median 0.0; S.D. 14.7; Obs. 136; Mean 0.0; Median 0.0; S.D. 13.0; Obs. 136; Mean 0.4; Median 0.0; S.D. 11.7; Obs. 136; Mean -0.4; Median 0.0; S.D. 5.7.
  - China: Obs. 166; Mean -0.8; Median 0.0; S.D. 8.5; Obs. 166; Mean 0.0; Median 0.1; S.D. 7.8; Obs. 166; Mean -0.2; Median 0.0; S.D. 6.0; Obs. 166; Mean 0.2; Median 0.0; S.D. 5.0.
  - Eurozone: Obs. 267; Mean 0.2; Median -0.1; S.D. 5.9; Obs. 267; Mean 0.0; Median -0.2; S.D. 5.9; Obs. 267; Mean 0.0; Median 0.0; S.D. 4.4; Obs. 267; Mean 0.0; Median 0.0; S.D. 4.0.
  - Hungary: Obs. 237; Mean -0.2; Median 0.0; S.D. 12.9; Obs. 237; Mean 0.0; Median 0.2; S.D. 12.9; Obs. 237; Mean 0.0; Median 0.0; S.D. 11.9; Obs. 237; Mean 0.0; Median 0.0; S.D. 5.1.
  - India: Obs. 125; Mean 2.3; Median 1.0; S.D. 15.1; Obs. 125; Mean 0.0; Median -1.3; S.D. 15.1; Obs. 125; Mean -0.2; Median 0.0; S.D. 12.1; Obs. 125; Mean 0.2; Median 0.0; S.D. 9.0.
  - Israel: Obs. 177; Mean 1.3; Median 0.0; S.D. 24.2; Obs. 177; Mean 0.0; Median -1.3; S.D. 24.2; Obs. 177; Mean 0.5; Median 0.0; S.D. 23.9; Obs. 177; Mean -0.5; Median 0.0; S.D. 3.7.
  - Japan: Obs. 142; Mean 0.2; Median 0.0; S.D. 1.2; Obs. 142; Mean 0.0; Median -0.1; S.D. 1.2; Obs. 142; Mean -0.1; Median 0.0; S.D. 0.9; Obs. 142; Mean 0.1; Median 0.0; S.D. 0.7.
  - Korea: Obs. 253; Mean -0.4; Median 0.0; S.D. 7.5; Obs. 253; Mean 0.0; Median 0.0; S.D. 7.2; Obs. 253; Mean -0.3; Median 0.0; S.D. 5.6; Obs. 253; Mean 0.3; Median 0.0; S.D. 4.6.
  - New Zealand: Obs. 179; Mean -0.6; Median 0.0; S.D. 8.5; Obs. 179; Mean 0.0; Median 0.6; S.D. 8.5; Obs. 179; Mean -0.3; Median 0.0; S.D. 6.8; Obs. 179; Mean 0.3; Median 0.0; S.D. 5.1.
  - Norway: Obs. 183; Mean -0.6; Median 0.0; S.D. 7.6; Obs. 183; Mean 0.0; Median 0.6; S.D. 7.6; Obs. 183; Mean 0.4; Median 0.0; S.D. 5.6; Obs. 183; Mean -0.4; Median 0.0; S.D. 5.2.
  - Poland: Obs. 209; Mean 0.6; Median 0.0; S.D. 7.8; Obs. 209; Mean 0.0; Median -0.7; S.D. 7.6; Obs. 209; Mean -0.3; Median 0.0; S.D. 4.8; Obs. 209; Mean 0.3; Median 0.0; S.D. 5.9.
  - Sweden: Obs. 135; Mean -1.0; Median -0.5; S.D. 7.6; Obs. 135; Mean 0.0; Median 0.2; S.D. 6.7; Obs. 135; Mean -0.1; Median 0.0; S.D. 4.5; Obs. 135; Mean 0.1; Median 0.0; S.D. 5.0.
  - Switzerland: Obs. 48; Mean -0.6; Median 0.6; S.D. 8.1; Obs. 48; Mean 0.0; Median 1.1; S.D. 8.1; Obs. 48; Mean 0.4; Median 0.0; S.D. 2.4; Obs. 48; Mean -0.4; Median 0.0; S.D. 7.7.
  - Thailand: Obs. 157; Mean 0.0; Median 0.0; S.D. 5.0; Obs. 157; Mean 0.0; Median 0.0; S.D. 5.0; Obs. 157; Mean -0.1; Median 0.0; S.D. 2.9; Obs. 157; Mean 0.1; Median 0.0; S.D. 4.0.
  - United Kingdom: Obs. 253; Mean -0.4; Median -0.1; S.D. 4.8; Obs. 253; Mean 0.0; Median 0.3; S.D. 4.8; Obs. 253; Mean 0.1; Median 0.0; S.D. 3.3; Obs. 253; Mean -0.1; Median 0.0; S.D. 3.5.
  - United States: Obs. 174; Mean -0.4; Median 0.0; S.D. 5.2; Obs. 174; Mean 0.0; Median 0.4; S.D. 5.2; Obs. 174; Mean 0.0; Median 0.0; S.D. 2.7; Obs. 174; Mean 0.0; Median 0.0; S.D. 4.5.
- Notes: See text for definitions and details. Sources: See Annex I.

*Content and numeric values reproduced exactly from the source PDF section "References" and Annex I. Data.*

### Annex II. Framework

### Annex II. Framework

### Definition of the monetary policy surprise
- Equation (A.1):
  - 푚푝푠푛푡 ≡ Δ퐸푛푡휇푛푡+푘 + [푔푛푡(퐸푛푡푿풏풕+풌) − 푔푛푡−1(퐸푛푡−1푿풏풕+풌)].

### Derivation using equation (5)
- Inserting equation (5) yields equation (A.2):
  - 푚푝푠푛푡 = Δ퐸푛푡휇푛푡+푘 + (푐푛푡−1 + 휙푛푡) 퐸푛푡푿풏풕+풌 − 푐푛푡−1 퐸푛푡−1푿풏풕+풌.

### Expressions for expectations from equations (6) and (7)
- Equations (6) and (7) imply:
  - 퐸푛푡푿풏풕+풌 = 훾푛(휁푛푿풏풕−ퟏ + 휼̃풏풕)
  - 퐸푛푡−1푿풏풕+풌 = 훾푛휁푛푿풏풕−ퟏ.

- Inserting these into (A.1) and (A.2) gives equation (A.3):
  - 푚푝푠푛푡 = Δ퐸푛푡휇푛푡+푘 + 훾푛휁푛 휙푛푡 푿풏풕−ퟏ + (푐푛푡−1 + 휙푛푡) 휂̃푛푡.

### Mapping to the main text (equation (8))
- Equation (A.3) corresponds to equation (8) in the main text, with the following definitions preserved exactly:
  - 훽푛 ≡ 훾푛휁푛 휙̅푛
  - 휖̃푛푡 ≡ Δ퐸푛푡휇푛푡푘 + (푐푛푡−1 + 휙푛푡) 휂̃푛푡

*Source: Annex II. Framework (wpiea2024224-print-pdf).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024224-print-pdf.pdf_