## 1. Stress in Financial Markets, January–July 2020

## Source details

**Canonical URL:** [1. Stress in Financial Markets, January–July 2020](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021014-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021014-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021014-print-pdf.pdf.json)

---

### Introduction and key context
- Even before their own COVID-19 outbreaks, EMDEs were affected by global investor risk aversion, lower commodity prices, rising local currency bond yields, capital outflows, and sharp currency depreciations in March 2020.
- Many EMDEs introduced their first unconventional monetary policy (UMP) measures in the form of asset purchase programs (APPs).
- The constructed database covers COVID19-related UMP measures for 27 EMDEs and 8 small AEs and focuses on the period January through August of 2020.

### Timeline, sample coverage, and measurement
- 50 asset purchase programs announced between March and August 2020 across 27 EMDEs and 8 small advanced economies.
- Analysis period: January through August of 2020.
- A market is defined as highly stressed when the 1-day price change is in the tail of the distribution (2.58 times the historical standard deviation above the historical average, computed over 2017–2019).
- Thresholds for mid and low stress are equal to 1.96 and 1.64, respectively.

### Pre-crisis policy and fiscal context for EMDEs
- Many EMDEs began announcing UMP measures while still having room for policy rate cuts and relatively low public debt levels.
- All countries faced inflation close to or above target except for Thailand.
- The same countries except India, Hungary, and South Africa entered the crisis with relatively low public debt levels and announced fiscal easing measures in response to COVID-19.

### Main objectives observed for APPs in EMDEs
- Majority aimed at:
  - affecting the sovereign yield curve,
  - easing stress, and
  - bolstering liquidity of targeted financial markets.
- Stabilizing sovereign secondary bond markets and easing broader financial conditions improved funding conditions for governments.
- About a quarter of the EMDE programs (and less than a fifth of small open AE programs) announced support for pandemic-related fiscal needs as a main objective.

### Modalities and scope of APPs observed
- Interventions mostly targeted government bond markets; subset targeted corporate or bank bond markets (BEAC, Brazil, Chile, Ethiopia, Hungary, Israel, Korea, Mauritius, and Norway).
- Egypt is the only country in the sample that purchased equities.
- About two thirds of the programs were quantity-based (fixed or maximum amount of purchases).
  - Examples of quantity-based programs: Bolivia, Ethiopia, Ghana, Iceland, India, Mauritius, Mexico, New Zealand, Norway, and Thailand.
- Other programs were quantity-based but flexible, with purchase amounts calibrated to market conditions and/or the economic and inflation outlook (Angola, Canada, Colombia, Costa Rica, and Hungary).
- Except for Chile (price-based APP targeting bank bonds), none of the programs were price-based.
- Dataset includes primary- and secondary-market purchases, twist operations (Colombia, India, and Mexico), special purpose vehicles/funds (Korea, Mauritius, Norway, Thailand), direct monetary financing (one-off contribution by Bank of Mauritius), and purchases of loans to SMEs (People’s Bank of China).

### Novelty and comparison with AEs
- In contrast to some small open AEs during the GFC, most EMDEs in this sample had policy rates above zero and faced external pressures when they launched APPs.
- Size of programs in EMDEs was comparable to that of AEs in the sample.
- Most APPs for EMs were newly announced during the COVID-19 crisis, except Indonesia where it was an expansion of a pre-existing program.

### Database construction and contents
- Primary sources: central bank press releases; supplemented by other announcements, news (for Egypt and Ethiopia), and legal/official communications when relevant.
- The database consolidates:
  - initial program announcements and all subsequent announcements;
  - implementation dates and related published information;
  - information on communication style, joint announcements with other authorities, and whether programs were part of policy packages;
  - realized transaction data when publicly available and clearly linked to the APP announcement.
- The database covers all central bank purchases (or sales) of private and/or public securities on primary or secondary markets, and measures that affect the yield curve even without expanding central bank balance sheets.

### Empirical strategy overview
- Analyses performed:
  - Event-study style investigation of sovereign bond yields and exchange rates, and of equity prices, corporate bond yields, and EMBI spreads for each country.
  - Effects estimated one, two, and three days following announcements.
  - Country-specific regressions testing whether concurrent policy announcements, external factors, and COVID-19’s impact alter APP announcement effects.
  - Panel regression controlling for various factors to test robustness.
- Comparisons made with conventional monetary policy (CMP) actions.

### Key empirical findings (summary)
- APP announcements in EMDEs:
  - Confirm a negative and statistically significant multi-day effect on bond yields.
  - Magnitude and persistence depend on program focus: quantity- vs. price-based, and whether purchases involved government bonds, private securities, or both.
  - Results hold when excluding APP announcements that coincide with policy rate cuts, when controls are added, and in panel regression settings.
- Conventional monetary policy (policy rate cut announcements):
  - Have a negative and statistically significant multi-day effect on bond yields across maturities but of smaller magnitude than that of APP announcements.
- For small AEs, results are broadly consistent with existing literature findings.

### Theoretical transmission channels highlighted
- Five mechanisms through which APPs impact economic and financial environment:
  - Direct liquidity channel: change in liquidity available on the targeted market leads to price adjustments.
  - Portfolio-rebalancing channel: reallocation of liquidity across asset portfolios, spilling over to other markets.
  - Signaling channel: APP announcements signal central bank objectives and future interventions, coordinating expectations about future short-term interest rates.
  - Liquidity-to-credit channel: increased liquidity relaxes banks’ constraints and can increase credit supply.
  - Exchange rate channel: changes in interest differentials or purchases of foreign-currency assets can lead to exchange rate depreciation, which can counteract funding cost reductions but can stimulate exports and affect outlook and inflation.

### Findings from related literature (selected quantitative points)
- US studies: decrease in the 10-year bond yield of about 0.2 percent after UMP announcements.
- Hartley and Rebucci (2020): average response of -0.28 to -0.43 percent of the 10-year bond yield in EMs; -0.11 to -0.14 percent for developed markets.
- Arslan et al (2020): central bank bond purchases on average reduced benchmark long-term bond yields and interrupted depreciation trends; APPs appear to have shored up the exchange rate on average.
- Sever et al. (2020): APP announcements had significant impact on bond yields and helped turn around sentiment but did not lead to depreciation of emerging market currencies.

### Box 1 — A Tale of Two Central Banks: India and South Africa (comparative operational findings)
- Commonalities:
  - Both RBI and SARB purchased government securities in secondary markets, used quantity-based programs, and made no joint announcements with other national authorities.
  - Objective: shore up market confidence and tackle market dysfunctionality.
- SARB:
  - Single press release on March 25th; Q&A on March 26th clarified objectives and that the program was not direct financing, debt monetization, or QE.
  - Market outcome: bond yields across all maturities were lower after the announcement.
- RBI:
  - Numerous daily press releases focused on practicalities; implementation was discretionary and preannounced two days ahead.
  - Key dated actions (selected):
    - March 18th: announced a first open market operation involving government dated securities.
    - March 20th: quantity-based program implemented two days after announcement.
    - March 24th and March 30th: two 1.5-bigger tranche purchases covering longer maturities.
    - April 23rd: announced a new “twist program” to be implemented two days later.
    - June 29th: announced another “twist”-like intervention to be made two days later.
  - Market outcome: APP announcements had only a slight effect on bond yields at medium- and long-term maturities.
- RBI maturity targeting and operations:
  - Initial purchases: 2 to 5 years bonds.
  - Later target changes: 2 to 9 years.
  - Twist operations details preserved as in source.

### Cross-country communication variation and information collected
- Central banks vary in the extent of information released; Canada provided information on all 27 variables collected, Romania provided information on 12 variables.
- Among EMDEs, Colombia, India, and Mexico rank first, second, and third in extent of information provided.
- South Africa and Costa Rica identified as having detailed communication strategies.

### Cross-country implementation and selected program sizes (selected entries from Table 4)
- Australia: First announcement 19-Mar-20; First implementation 20-Mar-20; Announced USD 35 bn / LCU 53 bn / 2.8 percent of GDP; Assets: Sovereign bonds, local government bonds.
- Canada: First announcement 24-Mar-20; First implementation 25-Mar-20; Announced USD 50 bn / 2.4 percent of GDP; Assets: Sovereign; Primary and Second.
- Iceland: First announcement 23-Mar-20; First implementation May 2020; Announced USD 1 bn / LCU 150 bn / 9.2 percent of GDP (represents an upper bound); Realized USD 0 bn / LCU 1 bn / 0.1 percent of GDP.
- Israel (sovereign): First announcement 15-Mar-20; First implementation March 2020; Announced USD 14 bn / LCU 50 bn / 3.7 percent of GDP; Realized USD 34 bn / LCU 118 bn / 8.8 percent of GDP.
- Korea (sovereign): First announcement 19-Mar-20; First implementation 20-Mar-20; Announced USD 3 bn / LCU 3000 bn / 0.2 percent of GDP; Realized USD 3 bn / LCU 3000 bn / 0.2 percent of GDP; Maturities: 3-, 5-, and 10-year.
- New Zealand: First announcement 20-Mar-20; First implementation 25-Mar-20; Announced USD 38 bn / LCU 60 bn / 22.4 percent of GDP; Realized USD 15 bn / LCU 24 bn / 9.4 percent of GDP.
- Norway: First announcement 16-Mar-20; Announced USD 5 bn / LCU 50 bn / 1.4 percent of GDP; Assets: Corporate bonds; Primary and Second.
- Colombia (private securities): First announcement 23-Mar-20; First implementation 24-Mar-20; Announced USD 10 bn / LCU 38015 bn / 3.7 percent of GDP; Realized USD 9 bn / LCU 33186 bn / 3.8 percent of GDP; Maturity: Up to 3 years.
- India: First announcement 18-Mar-20; First implementation 20-Mar-20; Announced USD 8 bn / LCU 60000 bn / 0.4 percent of GDP; Realized USD 8 bn / LCU 60000 bn / 0.4 percent of GDP; Maturity: 2 to 5 years.
- Thailand (two entries): 17-Mar-20 announced USD 3 bn / LCU 93 bn / 0.8 percent of GDP; realized USD 5 bn / LCU 152 bn / 1.4 percent of GDP (sovereign). 22-Mar-20 announced USD 13 bn / LCU 400 bn / 3.6 percent of GDP (private securities).
- Notes: Realized size refers to realized transactions until August 2020, if data available. Percentages are percent of 2019 GDP.

### Empirical assessment methodology (sample and variables)
- Sample: 15 EMs and 8 small AEs (constrained by availability of financial variables).
- Event study analyzing 1-, 2-, and 3-day impacts of APP announcements; also examine announcements that coincide with policy rate cuts.
- Variables of interest:
  - Yield curve: 10-year, 5-year, 2-year, 6-month bond yields.
  - Exchange rate: local currency per US dollar.
  - EMBI spread.
  - Equity index and corporate bond yields.
- Data: Daily data from Reuters; standard deviations computed using daily data from January 1, 2017 through the day prior to the announcement (about 1,000 observations).

### Main empirical findings (detailed)
- Bond yields:
  - APP announcements reduce bond yields; negative and statistically significant multi-day effects across maturities between 2 and 5 years as well as the 10-year yield.
  - Quantity-based programs and programs focusing on government securities appear more effective in reducing bond yields.
  - APP announcements that coincided with FX intervention were not effective in lowering borrowing costs; they had a positive and statistically significant effect on the 10-year bond yield (Table III.1).
  - Excluding 9 APP announcements that coincide with policy rate cuts does not change the bond yield findings.
- Exchange rate:
  - When all APP announcements are considered, results are predominantly positive and statistically significant, indicating exchange rate depreciation following many APP announcements.
  - Excluding APP announcements close to policy rate cuts yields inconclusive results for the exchange rate, suggesting a weaker exchange rate channel.
- EMBI / external borrowing costs:
  - APP announcements predominantly have a positive and statistically significant effect on the EMBI spread, with significant heterogeneity across the sample.
  - The EMBI finding holds even when excluding APP announcements close to policy rate cuts.
- Equity index and corporate bond yields:
  - Significant heterogeneity and an inconclusive average median effect.
- Implementation dates:
  - Event study on implementation dates yields broadly consistent results for limited samples with implementation dates available (5 small AEs: Sweden, Canada, Australia, New Zealand, Korea; 6 EMs: Colombia, India, Thailand, Hungary, Croatia, Poland).

### Selected econometric magnitudes (examples from Annex II and Annex IV)
- Country 1-, 2-, and 3-day effects on 10-year government bond yield (selected):
  - Colombia 23-Mar-20: -0.375***, -0.5***, -1.351***.
  - South Africa 25-Mar-20: -0.66***, -0.9***, -0.71***.
  - India 20-Mar-20: -0.151***, -0.031, -0.106.
  - Poland 17-Mar-20: 0.227***, -0.223***, -0.052.
- Country 1-, 2-, and 3-day effects on nominal exchange rate (selected):
  - Indonesia 19-Mar-20: 690***, 737***, 1352***.
  - Mexico 12-Mar-20: 0.5564***, 0.5392***, 1.4967***.
  - India 20-Mar-20: 0.1651, 1.245***, 1.0462**.
- Country 1-, 2-, and 3-day effects on EMBI spread (selected):
  - India 18-Mar-20: 44***, 80***, 120***.
  - Mexico 12-Mar-20: 53***, 486***.
  - Colombia 23-Mar-20: 23***, -8, -85***.

### Annex IV — event-study with controls and panel regression (key results)
- Country-specific OLS regressions (Jan 1, 2020–end-August) specification included ANN (announcement dummy), VIX, Δ policy rate domestic, Δ FED, Oxford COVID-19 Government Response Tracker, Google mobility index.
- Country regression coefficients (announcement dummy, 10Y BY selected):
  - Brazil: 0.0384
  - Colombia: -0.173
  - India: -0.0583**
  - Poland: -0.172***
  - Romania: -0.308***
  - South Africa: -0.841***
  - Thailand: -0.0685*
- Panel aggregates from country coefficients:
  - Panel (AEs+EMDEs) 10Y BY: -0.0862***
  - Panel (EMDEs) 10Y BY: -0.135***
- Panel regression (EMDE sample) announcement dummy coefficients:
  - 10Y BY: -0.135***
  - 5Y BY: -0.0921***
  - 2Y BY: -0.0732**
  - 6M BY: -0.0216
  - EMBI: -2.327
  - Equities: 75.70
  - FX: -2.966
- Key controls (EMDE panel examples):
  - VIX on 10Y BY: 0.00179***.
  - Δ policy rate USA on 10Y BY: 0.319***.
  - Google Index on 10Y BY: 0.00199***.
  - Oxford Index on 10Y BY: 0.000537*.
- Panel sample and fit (EMDE panel):
  - Observations: 10Y BY: 1,317; 5Y BY: 1,317; 2Y BY: 968; 6M BY: 790; EMBI: 1,056; Equities: 1,202; FX: 1,290.
  - R-squared: 10Y BY: 0.072; 5Y BY: 0.057; 2Y BY: 0.035; 6M BY: 0.015; EMBI: 0.087; Equities: 0.022; FX: 0.023.
  - Number of countrycode: 10Y BY: 15; 5Y BY: 15; 2Y BY: 11; 6M BY: 9; EMBI: 12; Equities: 14; FX: 15.

### Heterogeneity and interaction findings (from Annex IV)
- Central bank credibility:
  - APP announcements by credible central banks are more effective in lowering bond yields (example: CB credibility on 10-year Bond: -0.193, p-value .033).
- Announced program size (share of GDP):
  - Larger announced size associated with smaller effects on government 10-year bond markets (Announced size on 10-year Bond: .02, p-value .002).
- Price-based vs quantity-based:
  - Price-based program (Chile) had opposite effect on sovereign bond markets relative to quantity-based programs (Price-based on 10-year Bond: .306, p-value .001).
- Multiple announcements and low monetary policy space:
  - Multiple announcements associated with smaller effects (Multiple announcements on 10-year Bond: .249, p-value .028).
  - Low monetary policy space associated with smaller effects (Low monetary policy space on 10-year Bond: .139, p-value .072).
- Central bank transparency:
  - No robust role in full sample; when restricting to first announcements only, central bank transparency matters (first-announcement sample: CB transparency on 10-year Bond: -0.048, p-value .022).
- EMDEs vs AEs (panel comparisons):
  - Panel a (all announcements): EMDEs 10-year Bond: -0.122 (p-value 0); AEs 10-year Bond: .014 (p-value .378).
  - Panel b (first announcements): EMDEs 10-year Bond: -0.067 (p-value .16); AEs 10-year Bond: .103 (p-value .004).
  - Overall: stronger and more statistically significant APP impacts on bond yields for EMDEs than for AEs in the sample.
- Indonesia excluded from some panels as a large outlier.

### Robustness, limitations, and suggestions for future work
- Including an implementation-date dummy did not yield statistically significant results where data is available.
- Data limitations restrict exploration of APP effects on stress indicators (liquidity risk), tail risks, and market liquidity indicators (bid/ask spreads, volatility, market turnover) for most EMDEs.
- The metric counting number of announcement variables does not capture communication quality or market impact.
- Limitations imply future work should:
  - Evaluate effectiveness of APP programs in supporting market functioning (stated objective for most programs).
  - Analyze longer-term impact of these measures.
  - Extend observations beyond current crisis to understand underpinnings and channels of UMP effectiveness in EMDEs.

*Source: wpiea2021014-print-pdf, IMF.*

### 1. Stress in Financial Markets, January–July 2020 __________________________________8

### 1. Stress in Financial Markets, January–July 2020

### Introduction and key context
- Even before their own COVID-19 outbreaks, EMDEs were affected by global investor risk aversion, lower commodity prices, rising local currency bond yields, capital outflows, and sharp currency depreciations in March 2020.
- EMDEs responded with conventional and unconventional measures; in particular, many introduced their first unconventional monetary policy (UMP) measures in the form of asset purchase programs (APPs).
- The database constructed covers recent COVID19-related UMP measures for 27 EMDEs and 8 small AEs and focuses on the period January through August of 2020.

### Timeline and sample coverage
- 50 asset purchase programs announced between March and August 2020 across 27 EMDEs and 8 small advanced economies.
- The analysis period: January through August of 2020.
- Empirical analysis constrained at times to smaller sub-samples based on data availability.

### Stress measurement (as used in the chapter)
- A market is defined as highly stressed when the 1-day price change is in the tail of the distribution (2.58 times the historical standard deviation above the historical average, computed over 2017–2019).
- Thresholds for mid and low stress are equal to 1.96 and 1.64, respectively.

### Pre-crisis policy and fiscal context for EMDEs
- Many EMDEs began announcing UMP measures while still having room for policy rate cuts and relatively low public debt levels.
- Figure 2 (referenced) shows policy rates for 24 EMDEs at the beginning of March 2020; countries that announced APPs are represented in orange.
- All countries faced inflation close to or above target except for Thailand.
- The same countries except India, Hungary, and South Africa entered the crisis with relatively low public debt levels and announced fiscal easing measures in response to COVID-19.

### Main objectives observed for APPs in EMDEs
- Majority aimed at affecting the sovereign yield curve, easing stress, and bolstering liquidity of targeted financial markets.
- Stabilizing sovereign secondary bond markets and easing broader financial conditions improved funding conditions for governments.
- Only about a quarter of the EMDE programs (and less than a fifth of small open AE programs) announced support for pandemic-related fiscal needs as a main objective.

### Modalities and scope of APPs observed
- Interventions mostly targeted government bond markets; a subset targeted corporate or bank bond markets (BEAC, Brazil, Chile, Ethiopia, Hungary, Israel, Korea, Mauritius, and Norway).
- Egypt is the only country in the sample that purchased equities.
- About two thirds of the programs were quantity-based (fixed or maximum amount of purchases).
  - Examples of quantity-based programs: Bolivia, Ethiopia, Ghana, Iceland, India, Mauritius, Mexico, New Zealand, Norway, and Thailand.
- Other programs were quantity-based but flexible, with purchase amounts calibrated to market conditions and/or the economic and inflation outlook (Angola, Canada, Colombia, Costa Rica, and Hungary).
- Except for Chile (price-based APP targeting bank bonds), none of the programs were price-based.
- Dataset includes primary- and secondary-market purchases, twist operations (Colombia, India, and Mexico), special purpose vehicles/funds (Korea, Mauritius, Norway, Thailand), direct monetary financing (one-off contribution by Bank of Mauritius), and purchases of loans to SMEs (People’s Bank of China).

### Novelty and comparison with AEs
- In contrast to some small open AEs during the GFC, most EMDEs in this sample had policy rates above zero and faced external pressures when they launched APPs.
- Size of programs in EMDEs was comparable to that of AEs in the sample.
- Most APPs for EMs were newly announced during the COVID-19 crisis, except Indonesia where it was an expansion of a pre-existing program.

### Database construction and contents
- Primary sources: central bank press releases; supplemented by other announcements, news (for Egypt and Ethiopia), and legal/official communications when relevant.
- The database consolidates dispersed information and includes:
  - initial program announcements and all subsequent announcements;
  - implementation dates and related published information;
  - information on communication style, joint announcements with other authorities, and whether programs were part of policy packages;
  - realized transaction data when publicly available and clearly linked to the APP announcement.
- The database covers all central bank purchases (or sales) of private and/or public securities on primary or secondary markets, and measures that affect the yield curve even without expanding central bank balance sheets.

### Empirical strategy overview
- Analyses performed:
  - Event-study style investigation of the distribution of sovereign bond yields and exchange rates, and of equity prices, corporate bond yields, and EMBI spreads for each country;
  - Effects estimated one, two, and three days following announcements;
  - Country-specific regressions to test whether concurrent policy announcements, external factors, and COVID-19’s impact on activity alter APP announcement effects;
  - Panel regression controlling for various factors to test robustness.
- Comparisons made with conventional monetary policy (CMP) actions.

### Empirical findings (summary)
- APP announcements in EMDEs:
  - Confirm a negative and statistically significant multi-day effect on bond yields.
  - The magnitude and persistence depend on program focus: quantity- vs. price-based, and whether purchases involved government bonds, private securities, or both.
  - Results hold when excluding APP announcements that coincide with policy rate cuts, when controls are added, and in a panel regression setting.
- Conventional monetary policy measures (policy rate cut announcements):
  - Have a negative and statistically significant multi-day effect on bond yields across maturities but of a smaller magnitude than that of APP announcements.
- For small AEs, results are broadly consistent with existing literature findings.

### Theoretical transmission channels highlighted
- Five main mechanisms (one direct and four indirect) through which APPs impact the economic and financial environment:
  - Direct liquidity channel: change in liquidity available on the targeted market leads to price adjustments.
  - Portfolio-rebalancing channel: beneficiaries of intervention reallocate liquidity across asset portfolios, potentially spilling over to other markets.
  - Signaling channel: APP announcements signal central bank objectives and future interventions, coordinating expectations about future short-term interest rates.
  - Liquidity-to-credit channel: increased liquidity relaxes banks’ constraints and can increase credit supply.
  - Exchange rate channel: changes in interest differentials or purchases of foreign-currency assets can lead to exchange rate depreciation, which can in turn affect effectiveness (weaker local currency can counteract funding cost reductions, but can stimulate exports and have offsetting positive effects on outlook and inflation).

### Findings from the literature (selected quantitative points)
- US studies: decrease in the 10-year bond yield of about 0.2 percent after UMP announcements.
- Briciu and Lis (2015): a cumulative two-day effect on the 10-year bond yield ranging from 33 to -17 bps for seven ECB balance sheet policies between 2008 and 2015.
- Hartley and Rebucci (2020): average response of -0.28 to -0.43 percent of the 10-year bond yield in EMs, higher than for developed markets (-0.11 to -0.14 percent).
- Arslan et al (2020): central bank bond purchases on average reduced benchmark long-term bond yields in a significant and persistent manner and interrupted depreciation trends; APPs appear to have shored up the exchange rate on average, though results vary by country.
- Sever et al. (2020): APP announcements had a significant impact on bond yields and helped turn around sentiment but did not lead to depreciation of emerging market currencies.

### Database utility and research contributions
- Provides granular, operational analysis of APP schemes, their modalities, communication aspects, and implementation—enabling assessment of whether these features affect APP impact.
- Consolidates existing trackers and extends coverage to more countries and a longer period.
- Includes both announcement and implementation information to allow separation of announcement effects from implementation effects.

*Source: wpiea2021014-print-pdf, IMF.*

### Box 1. A Tale of Two Central Banks: India and South Africa

### Box 1. A Tale of Two Central Banks: India and South Africa

### Comparative overview
- Both central banks (Reserve Bank of India, RBI; South African Reserve Bank, SARB) purchased government securities in secondary markets, used quantity-based programs, and made no joint announcements with other national authorities.
- Both faced high sovereign bond yields in mid-March: in India yields reached a historic high; in South Africa a 3 percent upward shift of the yield curve coincided with the onset of the pandemic.
- Objective in both cases: shore up market confidence and tackle market dysfunctionality.

### SARB approach and timeline
- Single press release on March 25th; press release gave no implementation details (size, frequency, preannouncement, target maturities).
- March 26th: SARB published a Q&A detailing APP objectives, relation to other policies, and clarifying that the program was neither direct financing, debt monetization, nor quantitative easing, and did not represent a shift in the CB’s mandate.
- Market outcome: bond yields across all maturities were lower after the announcement.

### RBI approach and timeline
- Numerous daily press releases related to UMP, focused on practicalities; did not detail how the APP fit with other policies.
- Implementation was discretionary and preannounced two days ahead.
- Key dated actions:
  - March 18th: RBI announced a first open market operation involving government dated securities.
  - March 20th: quantity-based program implemented two days after announcement; received a “positive response”.
  - March 24th and March 30th: RBI announced two 1.5-bigger tranche purchases covering longer maturities.
  - March 23rd: second purchase (announced March 30th) later advanced by four days (to March 26th).
  - April 23rd: RBI announced a new “twist program” to be implemented two days later (simultaneous purchase of long-term securities and sale of short-term maturities).
  - June 29th: RBI announced another “twist”-like intervention to be made two days later in a similar amount, targeting slightly longer maturities.
- Market outcome: APP announcements had only a slight effect on bond yields at medium- and long-term maturities.

### RBI maturity targeting and operations
- Initial purchases: 2 to 5 years bonds.
- Later target changes: 2 to 9 years.
- Twist program operations:
  - First twist: sold securities with 2 months to 1 year maturity and purchased those with 6 to 10 years maturity.
  - Subsequent twist: sold securities with 3 to 10 months maturity and purchased those with 7 to 13 years maturity.
- Implementation details: repeated changes in target maturities; tranche size adjustments (two 1.5-bigger tranche purchases noted).

### Key comparative findings and relevant operational aspects
- Diversity of country approaches to UMP highlighted: frequency of announcements; whether UMP was part of a policy package; level of detail provided (size, maturity, pre-announced frequency and dates, counterparties); commentary on rationale and implications for CB mandate and monetary policy stance; innovativeness (e.g., swaps); and whether programs were expansions of existing programs or new programs.
- Communication mattered: SARB emphasized rationale and limits of APP in Q&A and clarified non-monetization; RBI emphasized operational details and frequent preannouncement.
- Implementation mechanics differed: SARB provided limited initial implementation detail then detailed objectives; RBI implemented quantity-based operations with announced short notice and later introduced twist operations.

*Source: Box 1. A Tale of Two Central Banks: India and South Africa, wpiea2021014-print-pdf.*

### introduction of the APP on March 31st was the result of a Presidential decree, and it was

### wpiea2021014-print-pdf - introduction of the APP on March 31st was the result of a Presidential decree, and it was

### Variation in APP announcements and central bank communication
- Central banks (CBs) vary in the extent of information released when introducing an APP; advanced economies (AEs) generally provide more information than emerging market and developing economies (EMDEs).
- For each announcement the authors collected up to 27 variables describing the APP at announcement (objective, size, implementation timeframe, etc.). The number of variables collected is used as a rough indicator of the CB communication strategy.
- Canada provided information on all 27 variables collected.
- Romania provided information on 12 variables.
- Among EMDEs, Colombia, India, and Mexico rank respectively first, second, and third in extent of information provided.
- South Africa and Costa Rica are also identified as having detailed communication strategies, with long and detailed CB announcements.
- Note: The metric counts quantity of information and may not capture quality or impact of communication.

### Cross-country implementation and program characteristics (selected summarized data from Table 4)
- Countries are listed more than once if announced programs had different maturity targets; reported information reflects the most updated information from the central bank.
- Table 4 (selected entries):
  - Australia: First announcement 19-Mar-20; First implementation 20-Mar-20; Announced USD 35 bn / LCU 53 bn / 2.8 percent of GDP; Assets: Sovereign bonds, local government bonds; Type: Outright; Objective: Boost confidence and tackle market dysfunctionalities.
  - Canada: First announcement 24-Mar-20; First implementation 25-Mar-20; Announced USD 50 bn / 2.4 percent of GDP; Assets: Sovereign; Type: Outright; Primary and Second.; Objective: Boost confidence and tackle market dysfunctionalities.
  - Iceland: First announcement 23-Mar-20; First implementation May 2020; Announced USD 1 bn / LCU 150 bn / 9.2 percent of GDP (represents an upper bound); Realized USD 0 bn / LCU 1 bn / 0.1 percent of GDP.
  - Israel (sovereign): First announcement 15-Mar-20; First implementation March 2020; Announced USD 14 bn / LCU 50 bn / 3.7 percent of GDP; Realized USD 34 bn / LCU 118 bn / 8.8 percent of GDP.
  - Korea (sovereign): First announcement 19-Mar-20; First implementation 20-Mar-20; Announced USD 3 bn / LCU 3000 bn / 0.2 percent of GDP; Realized USD 3 bn / LCU 3000 bn / 0.2 percent of GDP; Maturities: 3-, 5-, and 10-year.
  - New Zealand: First announcement 20-Mar-20; First implementation 25-Mar-20; Announced USD 38 bn / LCU 60 bn / 22.4 percent of GDP; Realized USD 15 bn / LCU 24 bn / 9.4 percent of GDP.
  - Norway: First announcement 16-Mar-20; Announced USD 5 bn / LCU 50 bn / 1.4 percent of GDP; Assets: Corporate bonds; Type: Outright; Primary and Second.; Objective: Boost confidence and tackle market dysfunctionalities.
  - Colombia (private securities): First announcement 23-Mar-20; First implementation 24-Mar-20; Announced USD 10 bn / LCU 38015 bn / 3.7 percent of GDP; Realized USD 9 bn / LCU 33186 bn / 3.8 percent of GDP; Maturity: Up to 3 years.
  - India: First announcement 18-Mar-20; First implementation 20-Mar-20; Announced USD 8 bn / LCU 60000 bn / 0.4 percent of GDP; Realized USD 8 bn / LCU 60000 bn / 0.4 percent of GDP; Maturity: 2 to 5 years.
  - Thailand (two entries): 17-Mar-20 announced USD 3 bn / LCU 93 bn / 0.8 percent of GDP; realized USD 5 bn / LCU 152 bn / 1.4 percent of GDP (sovereign). 22-Mar-20 announced USD 13 bn / LCU 400 bn / 3.6 percent of GDP (private securities).
- Notes from the table:
  - Realized size refers to realized transactions until August 2020, if data available.
  - Percentages are percent of 2019 GDP.
  - Some program details: quantity- vs price-based, outright purchases, swaps, repos, primary and secondary market participation, maturity targets, objectives (e.g., boost confidence, provide monetary stimulus, support fiscal needs).

### Empirical assessment: methodology for effectiveness of UMP measures in EMs
- Sample: 15 EMs and 8 small AEs (constrained by availability of financial variables).
- Methodology: Event study analyzing 1-, 2-, and 3-day impacts of APP announcements; also examine announcements that coincide with policy rate cuts.
- Variables of interest:
  - Yield curve: 10-year, 5-year, 2-year, 6-month bond yields.
  - Exchange rate: local currency per US dollar.
  - EMBI spread to capture second-round effects on external funding costs.
  - Also considered equity index and corporate bond yields.
- Data: Daily data from Reuters, omitting weekends and official holidays; standard deviations computed using daily data from January 1, 2017 through the day prior to the announcement (about 1,000 observations).
- Approach details:
  - Compute 1-, 2-, and 3-day change starting on the day prior to announcement and divide by the unconditional standard deviations.
  - Test null hypothesis that APP announcements have no significant impact.
  - Consider multi-day windows recognizing asset prices may not react instantaneously.
  - Also conduct a second event study and panel regressions (Annex IV) to control for other policy announcements and shocks.

### Main empirical findings
- Bond yields:
  - APP announcements reduce bond yields; the effect is statistically significant on average and consistent with previous literature (Hartley and Rebucci (2020), Arslan et al. (2020), Sever et al. (2020)).
  - Negative and statistically significant multi-day effects are found across maturities between 2 and 5 years as well as the 10-year yield.
  - Quantity-based programs and programs focusing on government securities appear more effective in reducing bond yields.
  - APP announcements that coincided with FX intervention were not effective in lowering borrowing costs; they had a positive and statistically significant effect on the 10-year bond yield (Table III.1).
  - Heterogeneity: In some cases the first APP announcement had a significant negative effect on the 10-year yield; in other cases subsequent announcements were more effective or first announcements were ineffective—conclusion is inconclusive on a consistent “surprise effect.”
  - Excluding 9 APP announcements that coincide with policy rate cuts does not change the bond yield findings.
- Exchange rate:
  - Results depend on proximity to policy rate cuts. When all APP announcements are considered, results are predominantly positive and statistically significant, indicating exchange rate depreciation following many APP announcements.
  - Several central banks announced policy rate cuts the day before or the same day as APP announcements (Chile, Indonesia, Mexico, Poland, Thailand), creating likely spillovers from rate cuts.
  - Excluding APP announcements close to policy rate cuts yields inconclusive results for the exchange rate, suggesting a weaker exchange rate channel.
- EMBI / external borrowing costs:
  - APP announcements predominantly have a positive and statistically significant effect on the EMBI spread, with significant heterogeneity across the sample.
  - This indicates announcements were often not sufficient to calm international investor perceptions and reduce the cost of external borrowing.
  - The EMBI finding holds even when excluding APP announcements close to policy rate cuts.
- Equity index and corporate bond yields:
  - Results show significant heterogeneity and an inconclusive average median effect.
- Implementation dates:
  - Applying the event study to implementation (rather than announcement) dates yields broadly consistent results for the more limited sample with available implementation dates (5 small AEs: Sweden, Canada, Australia, New Zealand, Korea; 6 EMs: Colombia, India, Thailand, Hungary, Croatia, Poland).

### Additional notes and limitations highlighted by the authors
- The metric counting number of announcement variables does not capture communication quality or market impact.
- Data limitations restrict exploration of APP effects on stress indicators (liquidity risk), tail risks, and market liquidity indicators (bid/ask spreads, volatility, market turnover) for most EMDEs.
- The EM sample is constrained by availability of financial variables.
- Some country-specific program types and details vary (e.g., swaps, repos, grants, primary vs secondary market participation).
- The authors attempt to control for other factors (pandemic effects, macroeconomic fundamentals) in the panel regressions presented in Annex IV.

*wpiea2021014-print-pdf - introduction of the APP on March 31st was the result of a Presidential decree, and it was*

### Annex IV.

### Annex IV.

### Event-study and panel-regression confirmation
- The event study with control variables as well as the panel regression, presented in Annex IV, broadly confirm the findings above (Annex IV).
- The results of panel regressions find no significant effect of first announcements on the 10-year bond yield (Table IV.6, panel b).

### Effectiveness of APP announcements
- APP announcements were effective in reducing bond yields across different maturities; the implementation effects were similar in impact to the announcement effects.
- APPs reduced bond yields to a greater degree than conventional monetary policy (CMP) implemented in mid-March–end-April (see Table III.3).
- Programs that coincided with FX intervention had the opposite expected effect on bond yields (positive and statistically significant).
- Some program characteristics appeared to worsen outcomes:
  - Announcements made jointly with other authorities.
  - Multiple announcements.
  - Size of program (larger programs, as measured by the announced size as a share of GDP, associated with smaller effects on the government 10-year bond markets).
- Panel regressions reveal that APP announcements made by credible central banks are more effective (in terms of their impact on bond yields).
- No evidence was found for an effect of:
  - Central bank transparency.
  - Non-residential investment share.
  - The monetary regime.
  - The exchange rate regime.

### Country factors and monetary space
- Larger programs (announced size as a share of GDP) are associated with smaller effects on government 10-year bond markets.
- Programs announced in countries with low monetary space are associated with smaller effects on government 10-year bond markets.
- A few country-specific factors seemed to improve APP outcomes:
  - Central Bank credibility.
  - High monetary policy space.
  - Low share of non-resident holdings of government bonds.
- Other factors that did not seem to have an impact:
  - Central Bank transparency.
  - Monetary or exchange rate regime.
  - Foreign investment share.

### Conventional policy rate cut announcements
- Policy rate cut announcements had negative and statistically significant multi-day effects on bond yields across the maturity curve, mostly over the period mid-March to end-April.
- Overall, the impact of conventional monetary policy transmission to bond yields is slightly less than that of APPs (Table III.3).
- Second-round effects from policy rate cut announcements:
  - Reduced the cost of external borrowing.
  - Had the expected depreciating effect on the exchange rate.

### Limitations and suggestions for future work
- This analysis does not evaluate the effectiveness of APP programs in EMDEs related to the support of market functioning; such an analysis is suitable for future work as this was the stated objective for most of the programs in the database.
- Important future work would include an analysis of the longer-term impact of these measures.
- Additional observations beyond the current crisis and further work are needed to understand the underpinnings and channels of the effectiveness of unconventional monetary policy (UMP) in EMDEs to assess their usefulness in normal times.

### Context and sample
- The paper builds a comprehensive database of APP announcements and implementations by a sample of 27 EMDEs and 8 small AEs from the onset of the crisis in March until August 2020.
- Figures referenced in the analysis include: Figure 15 (10-Year Bond Yield and Exchange Rate Results), Figure 16 (Other Bond Yields and Financial Variables Results), and Figure 17 (Results for the 10-Year Bond Yield by Type and Number of Announcements).

*Source: Annex IV.*

### Annex II. Details on Taxonomy of Objectives (continued)

### Annex II. Details on Taxonomy of Objectives (concluded)

### Annex II. Details on Taxonomy of Objectives (concluded)

### Country Objectives (selected entries)
- Thailand (07/04/2020): to stabilize the corporate bond market by providing liquidity backstop to ensure its continued functioning; to provide bridge financing to high-quality firms with bonds maturing during 2020-2021, at higher-than-market ‘penalty’ rates. Indicators in table: Y (Objective 1), Y (Objective 2), Y (Objective 3). URL provided in source.
- Turkey (31/03/2020): enhance the effectiveness of the monetary transmission mechanism via increasing the market depth, enabling sound asset pricing and providing banks with flexibility in liquidity management. Indicators: Y (Objective 1), Y (Objective 2), Y (Objective 3). URL provided in source.
- Turkey (17/04/2020): to maintain market depth, strengthen the monetary policy transmission mechanism and support the Primary Dealership system. Indicators: Y (Objective 1), Y (Objective 2).
- Uganda (06/04/2020): to ease [Microfinance Deposit taking Institutions (MDIs) and Credit Institutions (CIs)] liquidity distress whenever it arises. Indicator: Y (Objective 1).

### Econometric results — 1-, 2-, and 3-day effects on the 10-year government bond yield following APP announcement dates (selected entries)
- Brazil 26-Jun-20: 0.035, -0.075, -0.155; FXI indicator: ●
- Brazil 21-Jul-20: 0.06, 0.07, 0.295; FXI indicator: ●
- Turkey 31-Mar-20: -0.01, 0.29, 0.75; FXI indicator: ●
- Turkey 17-Apr-20: -0.46, -2.37***, -1.94***; FXI indicator: ●
- Hungary 16-Mar-20: 0.45***, 0.10, 0.51***; Gov. Securities indicator: ●●●●
- Colombia 23-Mar-20: -0.375***, -0.5***, -1.351***; Gov. Securities indicator: ●●●
- Poland 16-Mar-20: 0.227***, -0.223***, -0.052
- South Africa 25-Mar-20: -0.66***, -0.9***, -0.71***; Gov. Securities indicator: ●●
- Mexico 12-Mar-20: 0.43***, 0.37***, 0.52***; Gov. Securities indicator: ●●●
- India 20-Mar-20: -0.151***, -0.031, -0.106; Gov. Securities indicator: ●
- Uganda 6-Apr-20: 0.10, -0.35; 3-day: n/a
- Romania 20-Mar-20: 0.2**, -0.8***, -1***; Gov. Securities indicator: ●

Notes from table: Countries above the first bold line announced quantity-based programs and countries between the two bold lines announced price-based programs. One dot indicates whether the APP focused on government securities, the CB is considered credible, and the APP was announced together with FXI. Two dots indicated that the APP focused both on government and private securities. For the March 22nd announcement for Thailand (Sunday), March 20th was used instead.

### Econometric results — 1-, 2-, and 3-day effects on the nominal exchange rate following APP announcement dates (selected entries)
- Brazil 26-Jun-20: 0.1232***, 0.0418, 0.1049; FX regime: NoFloating
- Brazil 21-Jul-20: 0.1602***, -0.2151***, -0.1193*
- Chile 16-Mar-20: 16.6***, 10.4329, 9.52***; FX regime: YesFree floating
- Colombia 23-Mar-20: 0, 11.69, -12.43; FX regime: NoFloating
- Croatia 13-Mar-20: 0.0197, -0.0193, 0.0971**; FX regime: YesStabilized arrangement
- Hungary 16-Mar-20: 2.96**, 10.97***, 19.06***; FX regime: YesFloating
- India 20-Mar-20: 0.1651, 1.245***, 1.0462**; FX regime: NoFloating
- Indonesia 19-Mar-20: 690***, 737***, 1352***; FX regime: NoFloating
- Mexico 12-Mar-20: 0.5564***, 0.5392***, 1.4967***; FX regime: YesFree floating
- Poland 17-Mar-20: 0.0881***, 0.1467***, 0.2943***; FX regime: YesFree floating
- Thailand 17-Mar-20: 0.18**, 0.364***, 0.462***; FX regime: NoFloating
- Uganda 6-Apr-20: 6.897, 7.555***; FX regime: YesFloating

Note: One dot indicates whether the APP announcement was around the time of a policy rate cut. The Philippines announced a Php 300 billion purchase of government securities under repo on March 22nd, not included in the database.

### Econometric results — 1-, 2-, and 3-day effects on the nominal exchange rate following policy rate cut announcement dates (selected entries)
- Brazil 17-Mar-20: 0.0088, 0.1085**, 0.0957*
- Brazil 16-Jun-20: 0.0881**, 0.072, 0.2217***
- Chile 16-Mar-20: 16.6***, 10.4329, 9.52***
- India 26-Mar-20: -0.9212***, -1.1562***, -0.6599*
- Indonesia 18-Mar-20: 50740***, 787***
- Mexico 19-Mar-20: 0.3272**, 0.7147***, 1.6546***
- Poland 17-Mar-20: 0.0881***, 0.1467***, 0.2943***
- South Africa 19-Mar-20: 0.3715***, 0.5145***, 0.7415***
- Thailand 20-Mar-20: 0.05, 0.39***, 0.25

### Econometric results — 1-, 2-, and 3-day effects on the EMBI spread following APP announcement dates (selected entries)
- Brazil 21-Jul-20: 9, -12, -3
- Turkey 31-Mar-20: 440**, 57***
- Turkey 17-Apr-20: 6, -35*, -29
- Hungary 16-Mar-20: 20***, -115**
- Colombia 23-Mar-20: 23***, -8, -85***
- Colombia 14-Apr-20: 919*, 34**
- Mexico 12-Mar-20: 53***, 486***
- Mexico 20-Mar-20: 39***, 8, -36***
- India 18-Mar-20: 44***, 80***, 120***
- India 20-Mar-20: 40***, 56***, 52***
- Romania 20-Mar-20: 12.41**, 22.61***, 11.8
- Chile 16-Mar-20: 44***, 42***, 82***
- Croatia 13-Mar-20: 15***, 56***, 67***
- Uganda 6-Apr-20: n/a, n/a, n/a

Note: For the March 22nd announcement for Thailand, which is on Sunday, March 20th was used instead, as data is not available on weekends. The Philippines’ March 22nd repo purchase is not included in the database.

*Source: wpiea2021014-print-pdf - Annex II. Details on Taxonomy of Objectives (concluded); canonical URL: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021014-print-pdf.pdf*

### Annex IV. Event Study with Controls and a Panel Regression

### Annex IV. Event Study with Controls and a Panel Regression

### Methodology
- Country-specific OLS regressions estimated using daily data from January 1, 2020 until the end of August.
- Regression specification:
  - ∆Y_iit = α + β1 ANN_iit + β2 VIX_iit + β3 ∆I_iit + β4 ∆FED_iit + β5 Oxford_iit + β6 Google_iit + ε_i
  - Where ANN is the dummy for APP announcement dates; I is the domestic policy rate change; FED is the Fed’s policy rate change; VIX is the volatility index; Oxford is the Oxford COVID-19 Government Response Tracker; Google is the Google mobility index.
- Also run panel regressions with fixed effects using the same dependent and independent variable specifications; robustness checks include lagged dependent variables (not materially different).

### Country-level event study results (announcement dummy)
- Country regressions (coefficients on the announcement dummy) confirm:
  - APP announcements have a statistically significant and negative effect on bond yields across many countries.
  - Little effect on the exchange rate (FX), plausibly because some announcements moved FX positively while others moved FX negatively, washing out significance.
  - Using a dummy for implementation dates did not yield statistically significant results where data is available.
- Selected country coefficients (announcement dummy, 10Y BY unless otherwise noted):
  - Brazil: 0.0384
  - Chile: -0.0580
  - Colombia: -0.173
  - Croatia: 0.0691***
  - Hungary: 0.0471
  - India: -0.0583**
  - Indonesia: 0.0355
  - Mexico: -0.0281
  - Poland: -0.172***
  - Romania: -0.308***
  - South Africa: -0.841***
  - Thailand: -0.0685*
  - Turkey: -0.244
  - Uganda: 0.364
- Panel aggregates from country coefficients:
  - Panel (AEs+EMDEs) 10Y BY: -0.0862***
  - Panel (EMDEs) 10Y BY: -0.135***
- Note on significance notation: *** p<0.01, ** p<0.05, * p<0.1.

### Panel regression results (EMDE sample) — key coefficients and statistics
- Announcement dummy (Announce) coefficients (Table IV.3, EMDE panel):
  - 10Y BY: -0.135***
  - 5Y BY: -0.0921***
  - 2Y BY: -0.0732**
  - 6M BY: -0.0216
  - EMBI: -2.327
  - Equities: 75.70
  - FX: -2.966
  - Standard errors reported in parentheses below coefficients in source.
- Key control variable coefficients (EMDE panel):
  - VIX:
    - 10Y BY: 0.00179***
    - 5Y BY: 0.00148***
    - 2Y BY: 0.000602
    - EMBI: 0.237***
    - Equities: -5.868***
    - FX: 0.0736***
  - Δ policy rate USA:
    - 10Y BY: 0.319***
    - 5Y BY: 0.318***
    - 2Y BY: 0.343***
    - 6M BY: 0.150
    - EMBI: 36.07***
    - Equities: -749.70
    - FX: 0.859
  - Google Index:
    - 10Y BY: 0.00199***
    - 5Y BY: 0.00213***
    - 2Y BY: 0.00129***
    - 6M BY: 0.000946
    - EMBI: 0.155***
    - Equities: -1.686
    - FX: 0.0785***
  - Oxford Index:
    - 10Y BY: 0.000537*
    - 5Y BY: 0.000755**
- Panel regression sample and fit (EMDE panel):
  - Observations: 10Y BY: 1,317; 5Y BY: 1,317; 2Y BY: 968; 6M BY: 790; EMBI: 1,056; Equities: 1,202; FX: 1,290.
  - R-squared: 10Y BY: 0.072; 5Y BY: 0.057; 2Y BY: 0.035; 6M BY: 0.015; EMBI: 0.087; Equities: 0.022; FX: 0.023.
  - Number of countrycode: 10Y BY: 15; 5Y BY: 15; 2Y BY: 11; 6M BY: 9; EMBI: 12; Equities: 14; FX: 15.

### Robustness and implementation-date findings
- Including a dummy for implementation dates did not yield statistically significant results for countries where data is available; likely due to limited variability in the data.
- Indonesia excluded from the full and EMDE panel samples as a large outlier in some specifications.

### Heterogeneity analysis — interactions with structural characteristics (EMDEs)
- Specification: announcement dummy interacted with structural country characteristics; each interaction estimated in separate regressions.
- Main patterns (Table IV.4 and Table IV.5 summaries):
  - Central bank credibility:
    - APP announcements by credible central banks are more effective in lowering bond yields.
    - Example coefficients (EMDE interactions, Panel): CB credibility on 10-year Bond: -0.193 (p-value .033).
  - Central bank transparency:
    - No robust role in full-sample interactions; when restricting to first announcements only, central bank transparency matters for effectiveness (Table IV.5: CB transparency on 10-year Bond: -0.048, p-value .022).
  - Announced size (APP size as share of GDP):
    - Larger announced size associated with smaller effects on government 10-year bond markets (interpreted as markets not fully reacting or able to absorb novel policy information; pandemic-driven uncertainty may contribute).
    - Example coefficients: Announced size (Panel IV.4) on 10-year Bond: .02 (p-value .002).
  - Price-based versus quantity-based programs:
    - Announcements of the Chile price-based program targeting bank bonds had the opposite effect on sovereign bond markets compared to quantity-based programs.
    - Table IV.4 Price-based on 10-year Bond: .306 (p-value .001).
  - Multiple announcements and low monetary policy space:
    - Programs with multiple announcements and those in countries with low monetary space are associated with smaller effects on government 10-year bond markets.
    - Multiple announcements on 10-year Bond: .249 (p-value .028).
    - Low monetary policy space on 10-year Bond: .139 (p-value .072).
  - No robust evidence of the role of:
    - Central bank transparency (in full sample), non-residential investment share, monetary regime (inflation targeting vs other), or exchange rate regime on APP effectiveness — though lack of heterogeneity (most countries floating and inflation-targeting) limits inference.
- First-announcement-only sample (Table IV.5) — surprise effect:
  - Broadly similar results to full-sample interactions.
  - Notable: central bank transparency matters when restricting to first announcement.
  - Examples from first-announcement interactions:
    - CB credibility on 10-year Bond: -.289 (p-value .042).
    - Multiple announcements on 10-year Bond: .349 (p-value .004).
- EMDEs versus AEs (Table IV.6):
  - Panel a (all announcements):
    - EMDEs 10-year Bond: -.122 (p-value 0)
    - AEs 10-year Bond: .014 (p-value .378)
    - EMDEs and AEs combined 10-year Bond: -.076 (p-value 0)
  - Panel b (first announcements):
    - EMDEs 10-year Bond: -.067 (p-value .16)
    - AEs 10-year Bond: .103 (p-value .004)
    - EMDEs and AEs combined 10-year Bond: -.03 (p-value .401)
  - Overall: stronger and more statistically significant APP impacts on bond yields for EMDEs than for AEs in the sample; implementation-date effects generally not significant.
  - Indonesia excluded from these panels.

### Summary of substantive findings and interpretation
- APP announcements:
  - Statistically significant and negative effects on bond yields across maturities (especially in EMDEs).
  - Little or no consistent effect on exchange rates in the aggregate regressions.
- Controls:
  - VIX and Fed policy rate changes often have statistically significant coefficients (VIX typically positive; Fed policy rate changes often positive in bond regressions), affecting FX and other variables and sometimes washing out announcement effects on FX.
- Program characteristics and institutional features:
  - Central bank credibility increases APP effectiveness on bond yields.
  - Larger announced program sizes associated with smaller immediate bond-market effects.
  - Price-based programs may affect sovereign bond markets differently (opposite) relative to quantity-based programs.
  - Multiple announcements and low monetary policy space associated with smaller bond-market effects.
- Data and inference caveats:
  - Implementation-date dummy results are not significant, likely due to limited data variability.
  - Some null results (e.g., on exchange rate regime or monetary regime) may reflect little heterogeneity in the sample rather than genuine absence of effects.
  - Robustness checks exist but are not shown in full in the source.

*Source: Annex IV. Event Study with Controls and a Panel Regression, wpiea2021014-print-pdf*

### Annex V. Literature Review (continued)

### Annex V. Literature Review (continued

### Evidence on ECB Unconventional Monetary Policy (UMP) Spillovers
- Studies: "Cross Country Report on Spillovers" (IMF Country Report No. 16/212); country-level VAR; Global VAR.
- Methodology: Event study method; Country VAR; Global VAR.
- Main findings:
  - Event study: Spillovers from ECB UMP occurred via sovereign bond yields, with the exchange rate channel becoming significant more recently. Effects on indicators of capital flows were less significant. Emerging markets experienced larger financial spillovers than advanced economies. The ongoing APP exerted much larger financial spillover effects compared to earlier ECB UMP. Spillover effects were partly counteracted by market expectations of tighter U.S. monetary policy.
  - Country VAR: Confirms spillovers through sovereign bond markets; shocks to euro area term spreads impact currencies; no clear evidence of real spillovers. Shocks to euro area term spreads spill over to domestic term spreads and policy rates even after controls. Limited real-sector spillovers are consistent with short sample and literature that finds real effects mainly in the medium term (e.g., after 18 months).
  - Global VAR: ECB UMP not found to have had a statistically significant impact on CESEE and Nordic economies; exchange rate is closest to a significant response.
- Variables examined:
  - Government bond yields, exchange rates.
  - 10-year bond yield, nominal exchange rate, GDP growth, inflation, short term rate, domestic term spread.
  - GDP growth, inflation, nominal exchange rate, policy rate.
- Sample and period:
  - Czech Republic, Hungary, Poland, Sweden, Denmark; 2008-2015 (for event study and VAR analyses).

### Spillovers from U.S. Monetary Policy and Large-Scale Asset Purchases
- Studies: "Spillovers from United States Monetary Policy on Emerging Markets: Different This Time?" (Chen, Mancini-Griffoli, Sahay); "Large-Scale Asset Purchases by the Federal Reserve: Did They Work?" (Gagnon et al.); Krishnamurthy and Vissing-Jorgensen.
- Methodology: Event study method; novel decomposition approach to extract market and signaling factors.
- Main findings:
  - Chen et al.: Larger spillovers stem more from structural factors, such as the use of new instruments (asset purchases). Developed a methodology to extract, separate, and interpret U.S. monetary policy shocks.
  - Gagnon et al.: Purchases led to economically meaningful and long-lasting reductions in longer-term interest rates across a range of securities, including securities not included in purchase programs. Reductions primarily reflect lower risk premiums, including term premiums, rather than lower expectations of future short-term rates.
  - Krishnamurthy & Vissing-Jorgensen: Evidence for a signaling channel, a unique demand for long-term safe assets, and an inflation channel for both QE1 and QE2; an MBS prepayment channel and corporate bond default risk channel for QE1 only. Effects depend critically on which assets are purchased: MBS purchases in QE1 crucial for lowering MBS yields and corporate yields; Treasury-only purchases in QE2 disproportionately affected Treasuries and agency bonds.
- Variables examined:
  - Asset prices and capital market flows; market and signaling factors.
  - 2-year and 10-year Treasury yields, 10-year agency and debt yields, current-coupon 30-year agency MBS yields, the 10-year Treasury term premium, the 10-year swap rate, BAA corporate bond index yields.
  - Treasury yields, agency (Fannie Mae), agency MBS yields.
- Sample and period:
  - Chen et al.: US and 21 EMs; January 2000-July 2007.
  - Gagnon et al.: USA; 2008-2009.
  - Krishnamurthy & Vissing-Jorgensen: USA; 2008-2009.

### Operation Twist and Event-Study Estimates of Announcement Effects
- Study: "Let’s Twist Again: A High-Frequency Event-Study Analysis of Operation Twist and Its Implications for QE2" (Eric Swanson).
- Methodology: High-frequency event study.
- Main findings:
  - Operation Twist and QE2 are similar in magnitude.
  - Identified six significant, discrete announcements during Operation Twist; four had statistically significant effects.
  - The cumulative effect of the six announcements on longer-term Treasury yields is about 15 basis points (bp), highly statistically significant but moderate.
  - Effects on long-term agency and corporate bond yields are smaller: about 13 bp for agency securities and 2 to 4 bp for corporates.
  - Effects diminish moving from Treasury securities toward private sector credit instruments.
- Variables examined:
  - Treasury yields (3-month, 1-year, 2-year, 5-year, 10-year, 30-year), agency and corporate bonds.
- Sample and period:
  - USA; 2008-2009.

### Macroeconomic Effects of Asset Purchases
- Studies: "What are the macroeconomic effects of asset purchases" (Weale & Wieladek); event-study analysis of ECB balance sheet policies since October 2008 (Briciu & Lisi, EC Economic Brief 001 | JULY 2015).
- Methodology: B-VAR; event study.
- Main findings:
  - Weale & Wieladek (B-VAR): An asset purchase announcement of 1% of GDP leads to a statistically significant rise of 0.58% (0.25%) and 0.62% (0.32%) rise in real GDP and CPI for the US (UK). Transmission channels differ across countries.
  - Briciu & Lisi (ECB brief): The set of ECB balance sheet policies announced in 2014 had the broadest immediate impact on euro area financial conditions; the expanded asset purchase programme (EAPP) had the strongest impact on the exchange rate and significantly lowered longer-term government bond yields. Effects possibly augmented by parallel conventional monetary policy announcements.
- Variables examined:
  - Real GDP and CPI, government bonds.
  - 2-year bond yields, 10-year bond yields, exchange rate, equity market indices.
- Sample and period:
  - UK and US (Weale & Wieladek).
  - Euro Area; Oct 2008-Jan 2015 (Briciu & Lisi).

*IMF Working Paper — Annex V. Literature Review (continued).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021014-print-pdf.pdf_