## wpiea2020192-print-pdf

## Source details

**Canonical URL:** [wpiea2020192-print-pdf](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020192-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020192-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020192-print-pdf.pdf.json)

---

### Rise of Benchmark-Driven Investments (BDI) in EM Bond Markets
- Definition: An investment is benchmark-driven to the extent that its portfolio allocation across countries is guided by the country weights in a benchmark index.
- Empirical evidence: Raddatz and others (2017) find that 70 percent of country allocations of investment mutual funds are influenced by benchmark indices.
- Drivers:
  - Rising assets under management of passive funds.
  - Increasing number of active funds “hugging” the index.
  - Growing number of countries represented in benchmark indices.
- Consequence: Portfolio flows to EM local bond markets have been more synchronized for countries included in benchmark indices (e.g., J.P. Morgan GBI-EM Global) than for non-included countries.

### Size and Measurement of Benchmark-Driven Investors
- Estimated size: around $300 billion as of end 2019 (pre-COVID-19).
- Multiple estimation approaches:
  - JP Morgan survey estimates (local bond funds tracking JP Morgan indices close to $230 billion as of Q3:2019).
  - Regression-based empirical estimates (constrained least squares).
  - Event-study estimates (red crosses in Figure 2 denote event-study-based estimates).
- Regression-based estimate: pool of benchmark-driven investments in EM local currency debt markets was $330 billion at end-2019 (Table 2, years 2014–2019).
- Event-study implied estimates:
  - Uruguay (Jul–Sep 2017): implied BDI ≈ 373 (in billions of U.S. dollar).
  - Dominican Republic (Apr–Jun 2018): implied BDI ≈ 339 (in billions of U.S. dollar).
- Overall statement: As of end-2019, the size of benchmark-driven investments in EM local currency sovereign bonds is estimated between $300 billion, about 40 percent of the foreign investor base.

### Sensitivity to Global Factors and Correlation of Flows
- Benchmark-driven investments emphasize EMs as an asset class and focus on factors common to EMs rather than country-specific developments.
- Empirical findings:
  - Benchmark-driven flows are about three to five times more sensitive to global risk factors than balance-of-payments measures of portfolio flows.
  - A one standard deviation increase in the VIX on average reduces invested assets of benchmark-driven EM investors by 2 percent, compared with ½ percent for total portfolio investment.
  - A one standard deviation increase in U.S. 10-year Treasury yields reduces invested assets by 1½ percent, compared with about ¼ percent for total portfolio investment.
  - Sensitivity of benchmark-driven flows to external factors has increased in recent years.
  - ETFs are significantly more sensitive to interest rate and risk aversion shocks than mutual funds.
- Correlation and common-factor evidence:
  - For 10-year local bond yields of Chile, Colombia and Czech Republic, 3-month correlation with the overall J.P. Morgan GBI-EM yield increases from 16 percent to 53 percent after inclusion events.
  - Variation explained by the first principal component increases from 64 percent to 78 percent post inclusion.
- During COVID-19, heightened sensitivity and growing assets under management contributed to record outflows from emerging markets (IMF 2020); portfolio outflows from local currency bond markets were strongly correlated with the weight of the country in the GBI-EM benchmark index (Figure 4).

### Country Heterogeneity and Transmission of Shocks
- Different investor types (benchmark-driven vs. unconstrained) exhibit fundamentally different allocation behaviors.
- Countries exhibit widely different investor bases and country weights in benchmark indices.
- Same external or domestic shock can prompt very different capital flow responses across countries, affecting shock transmission to the domestic economy.
- Financial accelerator framework and implications:
  - Feedback loop A: deteriorating domestic fundamentals → higher credit spreads → tighter financial conditions → slows economy.
  - Feedback loop C: external shocks (e.g., global risk aversion) tighten external conditions → worsen domestic financial conditions.
  - Low share of benchmark-driven investments: foreign portfolio flows highly responsive to domestic factors (A strong), less to external developments (C weak). Example: Argentina.
  - High share of benchmark-driven investors: flows less responsive to domestic factors (A weak), highly responsive to external developments (C strong). Example: Colombia.
- Cross-country heterogeneity examples:
  - Turkey: high share of benchmark-driven investments as percent of total local currency foreign holdings but relatively small as percent of GDP.
  - Hungary: systemic importance of benchmark-driven investors is much higher.

### Key Index Construction and Inclusion Effects
- Major EM bond benchmarks:
  - J.P. Morgan EMBIG for dollar-denominated bonds.
  - J.P. Morgan GBI-EM for local currency bonds.
- Since 2007, EMBIG doubled to more than 70 countries; GBI-EM increased from 12 to 18 (text).
- Diversified weighting:
  - Many EM benchmarks use a "diversified" weighting method that reduces the weight of larger issuers and redistributes excess to smaller countries.
  - For local currency government bonds, diversified indices limit the maximum weight to 10 percent.
  - Example: Brazil’s weight is capped 8 percentage points lower than under market capitalization weights used in global benchmarks.
  - Smaller issuers such as Colombia, Hungary and Peru see increases in their weights by 1 to 2 percent.
  - Given an estimated index-tracked amount of $300 billion, a 2 percentage points higher weight would mean $6 billion additional benchmark-driven investments due to index rules.
- Inclusion and liquidity effects:
  - Index inclusion decisions and liquidity criteria can materially change country or security weights and trigger mechanical rebalancing by benchmark-following funds.
  - Example: Malaysia’s weight reduction in GBI-EM GD from 10 percent on February 2016 to 6 percent on August 2017; part of reduction attributed to changes in the NDF market and perceived MGS liquidity impacts.
  - Inclusion of China in the GBI-EM Global index will boost flows to China and reduce index weights for other countries during the transition period, prompting rebalancing by benchmark-driven funds.

### How Benchmarks Affect Fund Behavior and Evidence of “Closet Indexing”
- Active share measure: sum of absolute deviations of a fund’s country weights from its benchmark.
- Miyajima and Shim (2014): median active share among active EM local-currency bond funds tracked by EPFR Global is 17 percent, implying an 83 percent overlap in country weights with benchmarks.
- Funds classification:
  - Active share 0–10 percent: "closet index" funds.
  - Active share 10–20 percent: "weakly active."
- Nearly 70 percent of actively managed EM bond funds tracked by EPFR Global are either "closet index" or "weakly active"—i.e., benchmark driven.
- Active share of EM local currency bond funds tracked by EPFR has been declining steadily since 2008.
- Trend partly explained by underperformance of active funds over the last decade and retail investors’ shift to low-cost funds.
- Average tracking error metrics corroborate the trend of active funds becoming more passive.

### Regression Analysis: Methodology and Results (constrained least squares)
- Model partitions total foreign holdings F_i,t into two pools: benchmark-driven (a_t) and unconstrained (b_t):
  - F_it = a_t w_i,t + b_t W_i,t + ε_i,t
  - Constraint: a_t + b_t = ∑_{i=1}^N F_it
  - Definitions:
    - a_t: pool of benchmark-driven investments at time t.
    - b_t: pool of unconstrained investors at time t.
    - w_i,t: weight of country i in the J.P. Morgan GBI-EM Global Diversified index at time t.
    - W_i,t: market-cap weight of country i's bond market at time t.
    - F_i,t: nominal amount of foreign holdings of country i's bonds at time t (U.S. dollars).
    - ε_i,t: over-/under-weight of portfolio managers for country i at time t.
- Estimation details:
  - Constrained least squares (CLS); sample January 2010 to June 2020.
  - 17 countries (13 “benchmark” countries; 4 “off-benchmark”: Czech Republic, India, Israel, Korea).
- Regression results (Table 2; benchmark-driven pool a_t and unconstrained pool b_t, with t-stats and Prob>F all reported as 0.00):
  - a_t (benchmark-driven pool) by year:
    - 2014: 259.6 (t-stat 4.32, P>|t-stat| 0.00)
    - 2015: 247.1 (t-stat 4.98, P>|t-stat| 0.00)
    - 2016: 258.1 (t-stat 4.67, P>|t-stat| 0.00)
    - 2017: 359.0 (t-stat 8.64, P>|t-stat| 0.00)
    - 2018: 311.1 (t-stat 8.62, P>|t-stat| 0.00)
    - 2019: 327.0 (t-stat 7.23, P>|t-stat| 0.00)
  - b_t (unconstrained pool) by year:
    - 2014: 367.3 (t-stat 6.11, P>|t-stat| 0.00)
    - 2015: 318.4 (t-stat 6.41, P>|t-stat| 0.00)
    - 2016: 339.6 (t-stat 6.14, P>|t-stat| 0.00)
    - 2017: 365.8 (t-stat 8.80, P>|t-stat| 0.00)
    - 2018: 362.5 (t-stat 10.05, P>|t-stat| 0.00)
    - 2019: 400.9 (t-stat 8.86, P>|t-stat| 0.00)
  - F-stat by year:
    - 2014: 37.36 (Prob>F 0.00)
    - 2015: 41.13 (Prob>F 0.00)
    - 2016: 37.68 (Prob>F 0.00)
    - 2017: 77.43 (Prob>F 0.00)
    - 2018: 100.97 (Prob>F 0.00)
    - 2019: 78.49 (Prob>F 0.00)
- Time variation and episodes: benchmark-driven flows were less sticky during the mid-2013 Fed taper scare, April–August 2018 EM sell-off, and March 2020 COVID-19 shock.

### Event-Study Methodology and Results
- Decomposition of change in index weight into exogenous (inclusion) and buy-and-hold (valuation) components:
  - w_it+1 − w_it = (w_it+1 − w_it * R_it / R_bt) + (w_it * R_it / R_bt − w_it)
  - Exogenous component used to estimate benchmark-driven investor base:
    - B_t = f_it / (w_it+1 − w_it * R_it / R_bt)
    - B_t in U.S. dollars; f_it is net foreign purchases between t and t+1 in U.S. dollars.
- Uruguay (July–September 2017 inclusion):
  - Country weight: 0.00 (End-June 2017) → 0.29 (End-Sep 2017).
  - Exogenous component (a): 0.29.
  - Net foreign purchases (b): 1.08 (In billions of U.S. dollar).
  - Estimated B_t (c = b / a): 373 (In billions of U.S. dollar).
  - Observed net foreign inflows into Uruguay’s government bonds were $1.1 billion over Jul–Sep 2017.
- Dominican Republic (April–June 2018 inclusion):
  - Country weight: 0.00 (End-Mar 2018) → 0.10 (End-Jun 2018).
  - Exogenous component (a): 0.10.
  - Net foreign purchases (b): 0.34 (In billions of U.S. dollar).
  - Estimated B_t (c = b / a): 339 (In billions of U.S. dollar).
  - Observed net foreign inflows into Dominican Republic’s government bonds were $340 million over Apr–Jun 2018.
- Event-study pooled estimates:
  - Pool of benchmark-driven investments around $370 billion during Q3:2017.
  - Pool around $340 billion during Q2:2018.

### Policy-Relevant Implications and Recommendations
- Benefits of index inclusion:
  - Provides access to a larger and more diverse pool of external financing.
  - Capital flows become less sensitive to domestic economic developments, potentially reducing outflows in response to domestic shocks.
- Risks from rising passive and benchmark-driven investments:
  - Benchmark-driven flows amplify sensitivity to external/global financial conditions.
  - Some smaller and medium-sized EMs are disproportionately exposed (examples: Colombia, Hungary, Peru).
  - Index construction can introduce fragmentation and concentration risks; partial or premature inclusion of debt instruments can exacerbate risks.
- Recommended actions for stakeholders:
  - Enhance dialogue among index providers, the investment community, and regulators.
  - Index providers should increase transparency on eligibility criteria and provide advance communication of forthcoming index changes to promote consistency and reduce flow volatility.
  - Issuers should pursue index inclusion prudently and avoid fragmentation and concentration risks from premature or partial inclusion.
  - Authorities should monitor risks related to foreign ownership of local currency bonds, especially when a large share is held by benchmark-driven investors.
  - Consider potential effects of policy actions on index eligibility (e.g., capital flow management measures).
  - Reduce external vulnerabilities and strengthen buffers by:
    - Reducing excessive external liabilities.
    - Reducing reliance on short-term debt.
    - Maintaining adequate fiscal buffers and foreign exchange reserves.

### ANNEX I: EMERGING MARKET AND GLOBAL BOND INDICES — Key Points
- J.P. Morgan GBI-EM:
  - Launched in 2005; three versions (GBI-EM Broad, GBI-EM Global, GBI-EM) with diversified overlays.
  - Diversified version places a 10 percent cap for each country.
  - Main entry requirement: market accessibility; no minimum-rating requirements.
  - Bills and inflation-indexed bonds are not eligible.
  - As of end-April 2020, 19 countries included: Brazil, Chile, China, Colombia, Czech Republic, Dominican Republic, Hungary, Indonesia, Malaysia, Mexico, Peru, Philippines, Poland, Romania, Russia, South Africa, Thailand, Turkey and Uruguay.
- Bloomberg-Barclays Emerging Markets Local Currency Government Index:
  - Launched in 2008; requires market accessibility and minimum market capitalization of $5 billion.
  - As of end-April 2020, broadly same countries as GBI-EM Global with minor differences.
- FTSE EMGBI:
  - Launched in 2013; requires market accessibility, minimum market capitalization of $10 billion, and minimum rating of C (S&P) / Ca (Moody’s).
- Major global indices (April 2020 snapshots):
  - Bloomberg-Barclays Global Aggregate Index: market capitalization of about $60 trillion.
    - As of April 2020, 22 EMs have sovereign bonds included with country weights listed in the source (examples: Indonesia (0.5 percent), Mexico (0.5 percent), Thailand (0.4 percent), etc.).
  - FTSE WGBI: market capitalization of $23 trillion; as of April 2020 only three EMs included: Malaysia (0.4 percent), Mexico (0.8 percent) and Poland (0.5 percent).
- China’s index inclusions:
  - China added to Bloomberg-Barclays Global AGG in April 2019 with a 20-month transition period; estimated $2.0−2.5 trillion tracking the Global AGG and a final country weight of about 6 percent imply expected capital inflows of $120−150 billion to China by end-2020.
  - From April 2019 to April 2020, foreign holdings of Chinese government and policy bank bonds have risen by $69 billion.
  - J.P. Morgan GBI-EM Global included China in February 2020 with a 10-month transition period and a maximum cap of 10 percent; as of April 2020, China has a 2 percent weight in the index.
  - Estimates suggest China’s inclusion may lead to a reduction in EM-dedicated fund allocations of around $1−3 billion for other countries in the index due to mechanical rebalancing.

*Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020192-print-pdf.pdf*

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

### wpiea2020192-print-pdf - References

### Rise of Benchmark-Driven Investments (BDI) in EM Bond Markets
- An investment is benchmark-driven to the extent that its portfolio allocation across countries is guided by the country weights in a benchmark index.
- Empirical evidence: Raddatz and others (2017) find that 70 percent of country allocations of investment mutual funds are influenced by benchmark indices.
- Drivers of the rise of BDI:
  - Rising assets under management of passive funds.
  - Increasing number of active funds “hugging” the index.
  - Growing number of countries represented in benchmark indices.
- Consequence: Portfolio flows to EM local bond markets have been more synchronized for countries included in benchmark indices (e.g., J.P. Morgan GBI-EM Global) than for non-included countries.

### Size and Measurement of Benchmark-Driven Investors
- Estimated size of benchmark-driven investments in local EM bond markets: around $300 billion as of end 2019 (pre-COVID-19).
- Multiple estimation approaches are used:
  - JP Morgan survey estimates.
  - Empirical estimates.
  - Estimates from country-level event studies (red crosses in Figure 2 denote event-study-based estimates).
- Visual evidence: Figures referenced include correlation patterns and estimates of benchmark-driven investor size (Figures 1–6 and 9–16).

### Sensitivity to Global Factors and Correlation of Flows
- Benchmark-driven investments emphasize EMs as an asset class and focus on factors common to EMs rather than country-specific developments.
- As a result, benchmark-driven portfolio flows are more sensitive to common/global factors and more correlated across countries.
- Empirical results in the paper demonstrate heightened sensitivity of benchmark-driven investments to external factors, yielding elevated correlation of such flows across countries.
- During the COVID-19 pandemic, this heightened sensitivity and growing assets under management contributed to record outflows of foreign portfolio capital from emerging markets (IMF 2020).
  - Portfolio outflows from local currency bond markets during COVID-19 were strongly correlated with the weight of the country in the GBI-EM benchmark index (Figure 4).

### Country Heterogeneity and Transmission of Shocks
- Different investor types (benchmark-driven vs. unconstrained) exhibit fundamentally different allocation behaviors.
- Different countries exhibit widely different investor bases and country weights in benchmark indices.
- The same external or domestic shock can prompt very different capital flow responses across countries, affecting how shocks are transmitted to the domestic economy.

### Evidence and Empirical Tools Referenced
- Figures and tables used to support findings include (by number):
  - Figures 1–16 (correlation coefficients, estimates of benchmark-driven investors, portfolio flows proxies, event-study illustrations, sensitivity analyses, index inclusion effects, holdings breakdowns).
  - Tables 1–3 and Annexes I–III (sample of countries, regression results, country event studies, index construction notes, foreign holdings breakdown).
- Specific empirical approaches noted:
  - Regression analysis (e.g., constrained least squares estimation).
  - Country-level event studies (examples include Uruguay and Dominican Republic event studies).
  - Sensitivity regressions to global factors (reported in regression tables).

### Key Statistics and Exact Figures from the Source
- 70 percent: share of country allocations of investment mutual funds influenced by benchmark indices (Raddatz and others (2017)).
- $300 billion: estimated size of benchmark-driven investments in local EM bond markets as of end 2019 (before the COVID-19 shock).
- Time series horizon shown in portfolio flow proxies figure: Jan-19 through Jul-20.
- Visual data sources listed in figures: Bloomberg L.P.; IIF; EPFR; HaverAnalytics; and JP Morgan.

### Policy-Relevant Implications (from the source analysis)
- Inclusion in major benchmark indices provides access to a larger and more diverse pool of external financing.
- However, higher shares of BDI financing raise countries’ sensitivity to external/push factors and to common shocks that affect EMs as a group.
- Monitoring the composition of foreign investor bases and country weights in major indices is important for assessing vulnerability to synchronous outflows.
- Understanding investor-type heterogeneity (benchmark-driven vs. unconstrained) is critical for designing policies to manage capital flow volatility and shock transmission.

*Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020192-print-pdf.pdf*

### Section VI of this paper, building  on Arslanalp and Tsuda (2015).

### wpiea2020192-print-pdf - Section VI of this paper, building  on Arslanalp and Tsuda (2015).

### What are benchmark-driven investments and implications for capital flows
- Benchmark-driven investments arise mainly via investment funds that use benchmark indices to guide portfolio allocation.
- Passive funds explicitly replicate benchmarks; active funds deviate to varying degrees; unconstrained funds choose allocations irrespective of benchmarks.
- Retail investors generally have limited access to unconstrained funds.
- Benchmark-driven behavior increases similarity in country allocations and can amplify cross-country correlation of capital flows.

### Key EM bond benchmarks and index construction
- Most EM bond funds are benchmarked to a small number of indices; the most widely followed are:
  - J.P. Morgan Emerging Market Bond Index (EMBIG) for dollar-denominated bonds.
  - J.P. Morgan Government Bond Index–Emerging Markets (GBI-EM) for local currency bonds.
- Since 2007, the number of countries in the EMBIG has doubled to more than 70.
- The number of countries in the GBI-EM increased from 12 to 18.
- Global investment grade bond indices have been relatively stable in country representation due to more demanding investability criteria.
- Global bond benchmarks tend to be rating-sensitive and can produce large selling upon loss of investment grade status (examples noted: Pemex and South Africa in April 2020; Petrobras in September 2015).
- EM bond benchmarks commonly used are typically not rating sensitive; bonds remain eligible if liquidity criteria are satisfied.

### Index weighting and allocation distortions
- Many EM benchmarks use a "diversified" weighting method that reduces the weight of larger issuers and redistributes excess to smaller countries.
- For local currency government bonds, diversified indices limit the maximum weight to 10 percent.
- Example: Brazil’s weight is capped 8 percentage points lower than under market capitalization weights used in global benchmarks.
- Smaller issuers such as Colombia, Hungary and Peru see increases in their weights by 1 to 2 percent.
- Given an estimated index-tracked amount of $300 billion, a 2 percentage points higher weight would mean $6 billion additional benchmark-driven investments due to index rules.

### Index inclusion and security-level liquidity effects
- Index inclusion decisions and liquidity criteria can materially change country or security weights and trigger mechanical rebalancing by benchmark-following funds.
- Example: Malaysia’s weight reduction in GBI-EM GD from 10 percent on February 2016 to 6 percent on August 2017; part of reduction attributed to changes in the NDF market and perceived MGS liquidity impacts.
- Inclusion of China in the GBI-EM Global index will boost flows to China and reduce index weights for other countries during the transition period, prompting rebalancing by benchmark-driven funds.

### How benchmark indices affect fund behavior
- Active share measure: the sum of absolute deviations of a fund’s country weights from its benchmark.
- Miyajima and Shim (2014): median active share among active EM local-currency bond funds tracked by EPFR Global is 17 percent, implying an 83 percent overlap in country weights with benchmarks.
- Funds with active share 0–10 percent labeled "closet index" funds; 10–20 percent labeled "weakly active."
- Nearly 70 percent of actively managed EM bond funds tracked by EPFR Global are either "closet index" or "weakly active"—i.e., benchmark driven.
- Active share of EM local currency bond funds tracked by EPFR has been declining steadily since 2008.
- Trend toward more passive behavior partly explained by underperformance of active funds over the last decade and retail investors’ shift to low-cost funds.
- Average tracking error metrics corroborate the trend of active funds becoming more passive.

### Potential implications for capital flows
- Benchmarks can explain around 70 percent of country allocations after controlling for macro, industry, and country-specific effects (Raddatz and others 2017).
- Cerutti and others (2019): EMs relying more on global mutual funds are more sensitive to external factors for gross equity and bond inflows; recipient market liquidity and index inclusion increase sensitivities.
- Pandolfi and Williams (2020): index inclusion events lead to significant declines in government bond yields and currency appreciation.
- Mutual funds can propagate shocks via (i) direct holdings, (ii) overlapping ownership, and (iii) fire sales.
- Two contrasting policy implications:
  - Vulnerability: rising share of benchmark-driven investments increases exposure to external shocks as flows become more sensitive to external conditions.
  - Resilience: rising share reduces exposure to domestic shocks as flows become less sensitive to deteriorating country fundamentals.
- Financial accelerator framework:
  - Feedback loop A: deteriorating domestic fundamentals → higher credit spreads → tighter financial conditions → slows economy.
  - Feedback loop C: external shocks (e.g., global risk aversion) tighten external conditions → worsen domestic financial conditions.
  - Low share of benchmark-driven investments: foreign portfolio flows highly responsive to domestic factors (A strong), less to external developments (C weak). Example: Argentina.
  - High share of benchmark-driven investors: flows less responsive to domestic factors (A weak), highly responsive to external developments (C strong). Example: Colombia.

### How large are benchmark-driven investments in EM bond markets
- Foreign investors doubled holdings of EM government debt to more than $1.5 trillion over the past decade; more than 60 percent of this increase came from foreign asset managers.
- Useful concept: treat benchmark-driven investments as a pool allocated purely according to index weights; a percentage change in index weight × dollar amount of benchmark-driven investments = expected reallocation amount.
- EPFR and other data limitations: EPFR covers mutual funds and ETFs that disclose holdings (often retail-focused); institutional investor holdings data are limited.
- JPMorgan investor survey: as of Q3:2019, local bond funds tracking JP Morgan indices were close to $230 billion.
- Two estimation approaches used: regression-based analysis (Ballston and Melin (2013) / Arslanalp and Tsuda (2015) approach) and event studies.
- Both approaches yield broadly consistent estimates suggesting benchmark-driven investments stood around $300 billion at end-2019.

### Regression analysis: methodology and results
- Model partitions total foreign holdings into two pools: benchmarked to GBI-EM Global Diversified and unconstrained (market-cap-weighted) investors.
- Equations:
  - F_it = a_t w_i,t + b_t W_i,t + ε_i,t
  - Subject to: a_t + b_t = ∑_{i=1}^N F_it
  - Definitions:
    - a_t: pool of benchmark-driven investments at time t.
    - b_t: pool of unconstrained investors at time t.
    - w_i,t: weight of country i in the J.P. Morgan GBI-EM Global Diversified index at time t.
    - W_i,t: market-cap weight of country i's bond market at time t.
    - F_i,t: nominal amount of foreign holdings of country i's bonds at time t (U.S. dollars).
    - ε_i,t: over-/under-weight of portfolio managers for country i at time t.
- Estimation: constrained least squares (CLS); sample January 2010 to June 2020; 17 countries (13 “benchmark” countries; 4 “off-benchmark”: Czech Republic, India, Israel, Korea).
- Table 2 regression results (years 2014–2019):
  - a_t (benchmark-driven pool) by year:
    - 2014: 259.6 (t-stat 4.32, P>|t-stat| 0.00)
    - 2015: 247.1 (t-stat 4.98, P>|t-stat| 0.00)
    - 2016: 258.1 (t-stat 4.67, P>|t-stat| 0.00)
    - 2017: 359.0 (t-stat 8.64, P>|t-stat| 0.00)
    - 2018: 311.1 (t-stat 8.62, P>|t-stat| 0.00)
    - 2019: 327.0 (t-stat 7.23, P>|t-stat| 0.00)
  - b_t (unconstrained pool) by year:
    - 2014: 367.3 (t-stat 6.11, P>|t-stat| 0.00)
    - 2015: 318.4 (t-stat 6.41, P>|t-stat| 0.00)
    - 2016: 339.6 (t-stat 6.14, P>|t-stat| 0.00)
    - 2017: 365.8 (t-stat 8.80, P>|t-stat| 0.00)
    - 2018: 362.5 (t-stat 10.05, P>|t-stat| 0.00)
    - 2019: 400.9 (t-stat 8.86, P>|t-stat| 0.00)
  - F-stat by year:
    - 2014: 37.36 (Prob>F 0.00)
    - 2015: 41.13 (Prob>F 0.00)
    - 2016: 37.68 (Prob>F 0.00)
    - 2017: 77.43 (Prob>F 0.00)
    - 2018: 100.97 (Prob>F 0.00)
    - 2019: 78.49 (Prob>F 0.00)
- Regression-based estimate: pool of benchmark-driven investments in EM local currency debt markets was $330 billion at end-2019.

### Time variation, episodes, and cross-country heterogeneity
- Composition of foreign investors changed during 2010–2019; benchmark-driven portfolio flows were less sticky during three recent episodes of significant capital flow reversals:
  - Mid-2013 prospect of faster-than-anticipated Federal Reserve normalization triggered large portfolio outflows by EM-dedicated investment funds.
  - April–August 2018 EM sell-off saw sharp portfolio outflows by EM benchmark-driven funds after almost two years of buildup.
  - March 2020 COVID-19 shock caused similar outflows.
- Cross-country estimates vary significantly:
  - Turkey: high share of benchmark-driven investments as percent of total local currency foreign holdings but relatively small as percent of GDP.
  - Hungary: systemic importance of benchmark-driven investors is much higher.

### Event studies and supporting evidence
- Two event study sets (country inclusion and security inclusion events) used to estimate benchmark-driven pool; approach follows Arslanalp and Tsuda (2015).
- Event study results:
  - Pool of benchmark-driven investments around $370 billion during Q3:2017.
  - Pool around $340 billion during Q2:2018.
- Event study logic: index inclusion events (country or security inclusion) mechanically raise a country’s index weight and should trigger inflows by benchmark-driven investors but not by unconstrained investors.

*Source: Section VI of wpiea2020192-print-pdf, building on Arslanalp and Tsuda (2015).*

### 4.7 percent between March 2014 and end-March 2015, mainly due to the sharp drop in the

### wpiea2020192-print-pdf - 4.7 percent between March 2014 and end-March 2015, mainly due to the sharp drop in the

### Methodology: separating valuation effects and exogenous inclusion effects
- Change in a country’s index weight (w_it) is decomposed into two components:
  - Exogenous component (inclusion events).
  - Buy-and-hold component (valuation effects).
- Decomposition formula (as presented):
  - w_it+1 − w_it = (w_it+1 − w_it * R_it / R_bt) + (w_it * R_it / R_bt − w_it)
  - First term labeled Exogenous component; second term labeled Buy-and-hold component.
- Notation:
  - w_it is the weight of country i in the J.P. Morgan GBI-EM Global Diversified index at time t.
  - R_it and R_bt are the total gross returns of country i's bonds and the benchmark, respectively, from time t to t+1.
- Estimate of the benchmark-driven investor base (B_t) using only the exogenous component:
  - B_t = f_it / (w_it+1 − w_it * R_it / R_bt)
  - B_t is the benchmark-driven investor base at time t in U.S. dollars.
  - f_it is the net foreign purchases of country i's government bonds between t and t+1 in U.S. dollars.

### Event studies: Uruguay and Dominican Republic (2017−18)
- Uruguay (July-September 2017 inclusion)
  - J.P. Morgan phased inclusion over July-September 2017.
  - Country weight (GBI-EM global diversified): 0.00 (End-June 2017) → 0.29 (End-Sep 2017).
  - Change in country weight: 0.29 (In percent).
  - Buy-and-hold component: 0.00.
  - Exogenous component (a): 0.29.
  - Net foreign purchases during Jul-Sep 2017 (b): 1.08 (In billions of U.S. dollar).
  - Estimated benchmark-driven investor base (c = b / a): 373 (In billions of U.S. dollar).
  - Observed: net foreign inflows into Uruguay’s government bonds were $1.1. billion over July-September 2017—the highest quarterly figure recorded in official balance of payments statistics between 2000−17.
  - Interpretation: implied benchmark-driven investor base of around $370 billion at the time of the event.
- Dominican Republic (April-June 2018 inclusion)
  - J.P. Morgan phased inclusion over April-June 2018.
  - Country weight (GBI-EM global diversified): 0.00 (End-Mar 2018) → 0.10 (End-Jun 2018).
  - Change in country weight: 0.10 (In percent).
  - Buy-and-hold component: 0.00.
  - Exogenous component (a): 0.10.
  - Net foreign purchases during Apr-Jun 2018 (b): 0.34 (In billions of U.S. dollar).
  - Estimated benchmark-driven investor base (c = b / a): 339 (In billions of U.S. dollar).
  - Observed: net foreign inflows into Dominican Republic’s government bonds were $340 million over April-June 2018—the highest quarterly figure recorded in official balance of payments statistics between 2011−18.
  - Interpretation: implied benchmark-driven investor base of around $340 billion at the time of the event.

### Sensitivity of benchmark-driven investments to global factors
- Empirical setup
  - Monthly data from January 2010 to December 2018.
  - Dependent variable: EPFR Global flows into EM dedicated bond funds used as a proxy for benchmark-driven bond flows to EMs.
  - Regression specification:
    - Flows_t = α_0 + α_1 ∙ Flows_{t−1} + β ∙ Risk_t + γ ∙ Rates_t + δ ∙ Domestic_t + ε_t
  - Variables:
    - Risk_t: proxy for global risk aversion measured by the U.S. corporate BBB spread over U.S. Treasury securities (alternative: VIX).
    - Rates_t: 10-year U.S. Treasury yield (alternative: three-year-ahead expected federal funds rate in the Eurodollar market).
    - Domestic_t: EM economic surprise index compiled by Citigroup (pull variable).
  - Data sources include EPFR for fund flows and IIF for overall bond flows to EMs.
- Key empirical findings
  - Benchmark-driven flows are about three to five times more sensitive to global risk factors than balance-of-payments measures of portfolio flows.
  - Example magnitudes (reported comparisons):
    - A one standard deviation increase in the VIX on average reduces invested assets of benchmark-driven EM investors by 2 percent, compared with ½ percent for total portfolio investment.
    - A one standard deviation increase in U.S. 10-year Treasury yields reduces invested assets by 1½ percent, compared with about ¼ percent for total portfolio investment.
  - Sensitivity of benchmark-driven flows to external factors has increased in recent years.
  - ETFs are significantly more sensitive to interest rate and risk aversion shocks than mutual funds.
  - Table 5 and Figure 15 present regression results and graphical evidence (referenced in text).

### Asset correlations and common-factor importance after index inclusion
- Analysis of 10-year local bond yields for Chile, Colombia and Czech Republic over the last five years:
  - 3-month correlation of country yields with the overall J.P. Morgan GBI-EM yield increases from 16 percent to 53 percent after inclusion events.
  - Variation explained by the first principal component increases from 64 percent to 78 percent post inclusion.
  - Conclusion: importance of common factors rises as a result of a country’s inclusion in a benchmark index.
  - Note: 6-month correlation results corroborate findings, though to a lesser extent.

### Overall findings, risks, and policy implications
- Market-size estimates and investor base
  - As of end-2019, the size of benchmark-driven investments in EM local currency sovereign bonds is estimated between $300 billion, about 40 percent of the foreign investor base.
  - Event-study implied benchmark-driven investor base estimates: 373 (Uruguay, in billions of U.S. dollar) and 339 (Dominican Republic, in billions of U.S. dollar).
- Benefits of index inclusion
  - Provides access to a larger and more diverse pool of external financing.
  - Capital flows become less sensitive to domestic economic developments, potentially reducing outflows in response to domestic shocks.
- Risks from rising passive and benchmark-driven investments
  - Benchmark-driven flows amplify sensitivity to external/global financial conditions.
  - Some smaller and medium-sized EMs are disproportionately exposed (examples: Colombia, Hungary, Peru).
  - Index construction can introduce fragmentation and concentration risks; partial or premature inclusion of debt instruments can exacerbate risks.
- Policy recommendations and actions for stakeholders
  - Enhance dialogue among index providers, the investment community, and regulators.
  - Index providers should increase transparency on eligibility criteria and provide advance communication of forthcoming index changes to promote consistency and reduce flow volatility.
  - Issuers should pursue index inclusion prudently and avoid fragmentation and concentration risks from premature or partial inclusion.
  - Authorities should monitor risks related to foreign ownership of local currency bonds, especially when a large share is held by benchmark-driven investors.
  - Consider potential effects of policy actions on index eligibility (e.g., capital flow management measures).
  - Reduce external vulnerabilities and strengthen buffers by:
    - Reducing excessive external liabilities.
    - Reducing reliance on short-term debt.
    - Maintaining adequate fiscal buffers and foreign exchange reserves.

*Source: IMF staff calculations and analysis in the provided wpiea2020192-print-pdf content.*

### ANNEX I. EMERGING MARKET AND GLOBAL BOND INDICES

### ANNEX I. EMERGING MARKET AND GLOBAL BOND INDICES

### Overview
- The J.P. Morgan GBI-EM index is the most widely used index for investing in EM local government bonds.
- Global bond indices by Bloomberg-Barclays and FTSE include several EMs, but the weight of EMs remains relatively small: less than one percent at the country level and four percent at the aggregate level, as of April 2020.

### Major EM bond indices
- J.P. Morgan Government Bond Index–Emerging Markets (GBI-EM)
  - Launched in 2005.
  - Three versions: GBI-EM Broad, GBI-EM Global, and GBI-EM; each version has a diversified overlay.
  - The diversified version places a 10 percent cap for each country to limit concentration risk.
  - J.P. Morgan surveys report GBI-EM Global Diversified accounts for more than 90 percent of assets benchmarked to all six versions of the index.
  - Main entry requirement: market accessibility. No minimum-rating requirements or explicit market size limits.
  - Bills and inflation-indexed bonds are not eligible (only fixed-rate nominal bonds).
  - As of end-April 2020, 19 countries were included in the GBI-EM Global Diversified index: Brazil, Chile, China, Colombia, Czech Republic, Dominican Republic, Hungary, Indonesia, Malaysia, Mexico, Peru, Philippines, Poland, Romania, Russia, South Africa, Thailand, Turkey and Uruguay.

- Bloomberg-Barclays Emerging Markets Local Currency Government Index
  - Launched in 2008.
  - Main entry requirements: market accessibility and a minimum market capitalization of $5 billion.
  - No minimum-rating requirements.
  - Eligible: nominal fixed-rate bonds and bills; inflation-indexed bonds are not eligible.
  - Large overlap with the J.P. Morgan GBI-EM Global index; as of end-April 2020 it covered broadly the same countries, including two additional countries (Israel and Korea) and excluding two countries (Dominican Republic and Uruguay).

- FTSE Emerging Markets Government Bond Index (EMGBI)
  - Launched in 2013.
  - Main entry requirements: market accessibility, minimum market capitalization of $10 billion, and minimum rating requirements of C by Standard & Poor’s and Ca by Moody’s.
  - Bills and inflation-indexed bonds are not eligible (only fixed-rate nominal bonds).
  - Large overlap with the J.P. Morgan GBI-EM Global index.
  - As of end-April 2020, the index includes all but three countries in the GBI-EM Global index (Czech Republic, Dominican Republic and Uruguay), all with small weights on the GBI-EM Global index.

### Major global bond indices
- Bloomberg-Barclays Global Aggregate Index (Global AGG)
  - Tracks fixed-rate and investment-grade bonds of both developed and emerging markets.
  - Market capitalization of about $60 trillion, as of April 2020.
  - Created in 1992; historical data available from January 1987.
  - The Treasury sector tracks central government bonds issued by 37 countries in 24 currency markets, representing a total market capitalization of $32 trillion, as of April 2020.
  - Main entry requirements:
    - countries must have investment-grade status, based on the middle rating of Fitch, Moody’s, and S&P (or the lower rating when not all ratings are available);
    - “a freely traded, convertible currency with a liquid forward market that allows investors to hedge their currency exposure.”
  - As of April 2020, 22 EMs have sovereign bonds included in the Bloomberg-Barclays Aggregate index (with country weights shown in parentheses):
    - Indonesia (0.5 percent), Mexico (0.5 percent), Thailand (0.4 percent), Russia (0.3 percent), Malaysia (0.3 percent), Poland (0.3 percent), Saudi Arabia (0.2 percent), Chile (0.2 percent), Qatar (0.1 percent), Czech Republic (0.1 percent), Hungary (0.1 percent), India (0.1 percent), Colombia (0.1 percent), Philippines (0.1 percent), Peru (0.1 percent), Romania (0.1 percent), and Brazil, Morocco, Kazakhstan, Kuwait, South Africa, Uruguay (< 0.05 percent each).

- FTSE World Government Bond Index (WGBI)
  - Tracks fixed-rate and investment-grade sovereign bonds of both developed and emerging markets.
  - Market capitalization of $23 trillion as of April 2020.
  - Created in 1986; historical data available from December 1984.
  - Main entry requirements:
    - minimum market capitalization of $50 billion;
    - a domestic long-term credit rating of A-/A3 by S&P/Moody's;
    - no barriers to entry as reflected in policies that “actively encourage foreign investor participation.”
  - As of April 2020, there are only three EMs included in the WGBI (with country weights shown in parentheses): Malaysia (0.4 percent), Mexico (0.8 percent) and Poland (0.5 percent).

### China’s inclusion in benchmark bond indices (Annex II summary)
- China’s renminbi-denominated government and policy bank bonds were added to the Bloomberg-Barclays Global Aggregate Bond index (Global AGG) in April 2019 with a 20-month transition period.
  - Estimated $2.0−2.5 trillion tracking the Global AGG index and a final country weight of about 6 percent (after full inclusion) imply expected capital inflows of $120−150 billion to China by end-2020.
  - From April 2019 to April 2020, foreign holdings of Chinese government and policy bank bonds have risen by $69 billion.
- Inclusion in other bond indices could eventually lead to inflows around $300 billion (IMF, 2019).
- J.P. Morgan’s GBI-EM Global index included China in February 2020 with a 10-month transition period and a maximum cap of 10 percent.
  - As of April 2020, China has a 2 percent weight in the index.
- Estimates suggest China’s inclusion may lead to a reduction in EM-dedicated fund allocations of around $1−3 billion for the other countries in the index due to mechanical rebalancing of index weights; these outflows may be significant for EMs with a relatively high presence of benchmark-driven investors in their investor base.

*Source: ANNEX I. EMERGING MARKET AND GLOBAL BOND INDICES (from wpiea2020192-print-pdf).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020192-print-pdf.pdf_