## _wp15104

## Source details

**Canonical URL:** [_wp15104](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15104.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15104.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15104.pdf.json)

---

### Introduction and context
- International mutual funds have a high degree of participation in Mexico.
- Mexico was included in the Citigroup’s World Government Bond Index (WGBI) in 2010, attracting new groups of foreign investors.
- Two stress episodes analyzed:
  - the height of the global financial crisis: 2008Q3–2009Q3
  - the tapering announcement: 2013Q2
- Key figures:
  - Between 2013Q1 and 2013Q2, capital inflows (by non-residents) fell by US$24.5 billion, of which US$14 billion was due to a sudden stop in portfolio inflows.
  - Pension funds’ assets have increased by about 18 percent annually over the last 10 years.
- Volatility clustering in Mexico’s 10-year sovereign bond yields: large changes in yields cluster during stress periods, producing persistence in amplitude of yield changes.

### Data, coverage, and construction of cumulative flows
- International mutual funds data: EPFR Global.
  - EPFR Global covers 11,000 equity funds and 4,500 bond funds, with $23.5 trillion in total assets as of March 2014.
  - Sample: 375 foreign mutual funds active in Mexico since 2007: 112 bond funds and 263 equity funds, with $1.43 trillion in total assets as of March 2014.
  - EPFR fund flow data combined with country allocation via unique fund ID.
- Domestic mutual funds data: Econometica.
  - Covers 546 mutual funds and positions across 68 asset types.
  - As of May 2014, total gross assets exceed over $130 billion, approximately half invested in domestic sovereign bonds.
- Sample periods:
  - International mutual funds: monthly January 2007 – March 2014.
  - Domestic mutual funds: monthly January 2011 – May 2014.
- Cumulative flows measure:
  - log difference of the gross cumulative flows of bond and equity funds to Mexico from those in the end of 2010, multiplied by 100 (plotted in Figures 3 and 4).

### Research questions and empirical approach
- Main question: do foreign and domestic mutual fund investors in Mexico behave differently, and how does their behavior affect local market volatility, especially during stress?
- Three-part empirical approach:
  1. Herding measure from Lakonishok et al. (1992) to detect fund herding, especially in stress periods.
  2. Fund-level panel regressions to estimate sensitivities of fund flows to global factors by fund characteristics.
  3. Regress within-month volatility of returns on lagged measures of excess demand to link fund behavior to local volatility.
- Fund characteristics used: investor type (retail vs institutional), redemption structure (open-end vs closed-end), geographic focus (global, emerging markets, Latin America regional, Mexico-dedicated), and fund size (above/below monthly median AUM).

### Descriptive patterns and timing of cumulative flows
- Domestic inflows increased much less than foreign inflows in recent years.
- Foreign mutual funds were net sellers of Mexican assets during stress episodes; domestic mutual funds did not clearly offset these sales.
- Bond flows to Mexico increased faster and were more volatile than equity flows among foreign mutual funds.
- Cumulative flows from Latin America regional bond funds and global bond funds to Mexico rose significantly since early 2011—potentially related to Mexico joining the WGBI in October 2010.
- Selling during stress was more prevalent among:
  - open-end funds,
  - funds with retail investors,
  - Latin America regional funds.
- Example stress episode statistics:
  - At the Lehman collapse in September 2008:
    - around 75 percent of equity funds active in Mexico were selling Mexican assets.
    - around 95 percent of bond funds active in Mexico were selling Mexican assets.
  - During the tapering announcement in 2013:
    - about 50 percent of domestic mutual funds were selling Mexican assets.
    - the number of net sellers among foreign mutual funds exceeded 70 percent.

### Proportion of foreign funds net selling Mexican assets (Table 2)
- All Foreign Funds Investing in Mexico — Percentage of Net Sellers (in Percent):
  - Non-Stress
    - Bond: 48.6
    - Equity: 54.1
  - GFC
    - Bond: 81.6
    - Equity: 69.3
  - Tapering
    - Bond: 65.2
    - Equity: 63.5

### Correlated selling and herding: definitions and high-level findings
- Proportion of funds net sellers computed monthly to give prima facie evidence of correlated selling.
- Herding index: Lakonishok et al. (1992) measure to detect excess co-movement in trading directions.
- Findings:
  - Foreign mutual funds exhibit strong correlated selling during heightened global uncertainties.
  - Correlated selling is much weaker among domestic mutual funds.
  - Herding can destabilize markets, aggravate shocks, and lead to mispricing or bubbles.

### Herding index results (Table 3)
- Herding Indices for All Foreign Funds Investing in Mexico (in Percent):
  - Bond: Non-Stress 2.20; Stress 3.72; Mean Difference 1.52 (standard errors (1.75), (2.65)); p-value [0.03].
  - Equity: Non-Stress 2.71; Stress 3.22; Mean Difference 0.51 (standard errors (1.84), (2.15)); p-value [0.47].
- Interpretation:
  - Foreign bond investors herd more during market stress than foreign equity investors.
  - For bond funds, herding during stress is significantly higher at the 5 percent level; not significant for equity funds.

### Global factors and positive feedback trading — key regression estimates
- Flow model (equation (2)): fund fixed effects, VIX, Stress dummy, VIX*Stress, month fixed effects, clustered SE by fund. Expected VIX coefficient negative; negative interaction implies greater vulnerability during stress.
- Selected coefficients (dependent variable: fund flow in billions of USD; clustered standard errors):
  - Bond funds (Foreign, column (2)):
    - VIX coefficient -0.000955* (0.000498)
    - VIX * Stress -0.00119** (0.000577)
    - Stress -0.00247*** (0.000840)
  - Equity funds (Foreign, column (2)):
    - VIX coefficient -0.000416** (0.000163)
    - VIX * Stress -0.000377* (0.000196)
    - Stress -0.000707** (0.000333)
- Magnitude interpretation:
  - Increase in outflow by foreign bond funds during stress episodes is equivalent to USD 1.19 million for every one percent increase in VIX; for foreign equity funds the increase is USD 0.377 million.

### Momentum (positive feedback) trading — evidence and magnitudes (Table 5)
- Momentum model (equation (3)): lagged returns as predictor; one-month lag used.
- Selected coefficients (dependent variable: fund flow in billions of USD; clustered SE):
  - Foreign bond funds:
    - Lagged return to Mexico 3-month govt bond: 0.00323** (0.00139)
    - Lagged return * Stress: 0.00733** (0.00306)
    - Stress: -0.00151*** (0.000430)
  - Foreign equity funds:
    - Lagged stock market return: 0.0172** (0.00707)
    - Lagged stock market return * Stress: 0.00235*** (0.000624)
    - Stress: -0.000202** (0.0000885)
  - Domestic equity funds:
    - Lagged stock market return: -0.0211** (0.00933) (evidence of negative feedback trading)
- Robustness: main results hold using lagged changes in returns, returns to 10-year government bonds, and dollar-denominated returns (Appendix IV).

### Heterogeneity by fund characteristics — correlated selling, herding, and response magnitudes
- Percentage of Net Sellers among Foreign Funds by Fund Characteristics (selected):
  - Global Emerging Markets — Bond: Non-Crisis 48.4; GFC 80.8; Tapering 66.3.
  - Latin America Regional — Bond: Non-Crisis 53.8; GFC 91.7; Tapering 60.0.
  - Mexico-Dedicated — Bond: Non-Crisis 62.4; Tapering 47.0 (missing data for 2007/03–2009/10).
- Herding indices by characteristic (Table 7, in Percent; selected):
  - Global — Bond: Non-Stress 5.35 (Std Dev (3.27)); Stress 6.44 (4.25).
  - Latin America Regional — Bond: Non-Stress 5.58 (5.05); Stress 6.22 (5.37).
  - Retail — Bond: Non-Stress 2.99 (2.55); Stress 4.21 (2.80).
  - Open redemption — Bond: Non-Stress 2.33 (1.93); Stress 4.13 (3.05).
  - Small funds — Bond: Non-Stress 3.37 (2.62); Stress 5.70 (4.46).
- Three key heterogeneity results:
  1. After Lehman, significantly more mutual funds net sold Mexican assets across all fund types versus non-crisis times.
  2. Bond funds showed more correlated selling than equity funds during the Global Financial Crisis, but the reverse during non-crisis times.
  3. Funds with retail investors, Latin America regional funds, and smaller funds exhibited stronger simultaneous selling than institutional, global, and larger funds.
- Regression extensions (equations (4) and (5)):
  - Regional (equity) funds and open-end (bond) funds respond more strongly to global financial conditions—especially during stress—than global and closed-end counterparts.
  - Regional funds, open-end funds, small funds, and retail-sold funds are more likely to engage in positive feedback trading.
  - Bond funds have a higher tendency to chase returns during stress (three-way interaction positive for bond funds).

### Do foreign flows affect market volatility? — excess demand measures and volatility regressions
- Excess demand measures:
  - dratio (dollar ratio): total dollar increases / (total dollar increases + total dollar decreases) by bond/equity funds in a month.
  - nratio (numbers ratio): number of funds increasing holdings / number of funds changing holdings.
  - One-month lag of excess demand used.
- Selected results (Table 10; dependent variables: within-month volatility of daily returns on Mexican 3-month government bond and Mexican stock market; robust SEs in parentheses):
  - Panel A — Excess Demand by Foreign Mutual Funds:
    - Lagged nratio on bond volatility: 0.0108* (0.00599)
    - Lagged nratio * Stress on bond volatility: -0.0221*** (0.00814)
    - Stress on bond volatility: 0.0110*** (0.00402)
    - Lagged nratio on stock volatility: 2.085*** (0.100)
    - Lagged nratio * Stress on stock volatility: -4.344* (2.245)
    - Stress on stock volatility: 2.597** (0.996)
  - Panel B — Excess Demand by Domestic Mutual Funds:
    - Lagged dratio on bond volatility: 0.00900*** (0.00113)
    - Lagged dratio on stock volatility: 0.607** (0.289)
    - Stress dummy positive and significant for stock market volatility (e.g., 1.817*** (0.529) in comparable foreign funds’ regression)
- Interpretation:
  - In tranquil times, excess purchases by foreign mutual funds tend to amplify local market volatility; in stress times, excess selling is associated with higher volatilities.
  - Excess demand by domestic funds has qualitatively similar effects but quantitatively smaller magnitudes; foreign funds’ excess demand generally has a higher impact on volatility.

### Robustness checks (Appendix IV) — selected statistics
- Robustness check (1): using changes in returns in momentum regressions
  - Foreign Bond Funds (Column (2)): Lagged change in return to Mex 3-month govt bond = 0.0156** (0.00642); Lagged change in return * Stress = 0.0132** (0.00621); Stress = -0.0189** (0.00906); No. of observations = 2851; R-squared = 0.152.
  - Domestic Equity Funds (Column (1)): Lagged change in stock market return = -0.0136* (0.00736); No. of observations = 15834; R-squared = 0.070.
- Robustness check (2): using return on Mexico’s 10-year government bonds
  - Foreign (Bond) Funds (Column (2)): Lagged return = 0.00871*** (0.00256); Lagged bond yield * Stress = 0.00567* (0.00302); Stress = -0.0234** (0.0116); No. of observations = 7287; R-squared = 0.181.
- Robustness check (3): dollar-denominated return on Mexico’s 3-month government bonds
  - Foreign (Bond) Funds (Column (2)): Lagged dollar return = 0.0247* (0.0129); Lagged bond yield * Stress = 0.158*** (0.0470); Stress = -0.0194** (0.00922); No. of observations = 7287; R-squared = 0.177.

### Main conclusions and policy-relevant implications
- Three main contributions:
  1. Foreign and domestic mutual funds behave differently: foreign mutual funds show broad-based evidence of herding and positive feedback trading and are highly responsive to global risks.
  2. Heterogeneity matters: bond funds are substantially more sensitive to global factors and engage in stronger positive feedback trading than equity funds, particularly in stress episodes; small funds, retail-sold funds, open-end funds, and regional funds engage more in momentum trading and herding.
  3. Market impact: foreign funds’ trading behavior is associated with higher local market volatilities, notably during stress; domestic mutual funds play some mitigating role.
- Policy-relevant implication:
  - Changes in the composition of the investor base—e.g., rising share of bond flows to Mexico—may have important consequences for the stability of aggregate flows and for macro-prudential policies in Mexico.
- Caveat:
  - Results apply to mutual funds only and should not be generalized to banks, pension funds, and insurance companies without further analysis.

*Source: _wp15104 (EPFR Global and author’s calculations, as presented in the PDF content unit).*

### 1. Gross Portfolio Inflows and Cumulative Flows of Mutual (Bond) Funds to ............................ 4

### 1. Gross Portfolio Inflows and Cumulative Flows of Mutual (Bond) Funds to ............................ 4

### Introduction and context
- International mutual funds have a high degree of participation in Mexico (see Table 1).
- Mexico was included in the Citigroup’s World Government Bond Index (WGBI) in 2010, attracting new groups of foreign investors.
- Two stress episodes analyzed in the sample:
  - the height of the global financial crisis: 2008Q3–2009Q3
  - the tapering announcement: 2013Q2
- Between the first and second quarter of 2013, capital inflows (by non-residents) fell by US$24.5 billion, of which US$14 billion was due to a sudden stop in portfolio inflows.
- Pension funds’ assets have increased by about 18 percent annually over the last 10 years (as described in the source).
- Volatility clustering is evident in Mexico’s 10-year sovereign bond yields: large changes in yields cluster during stress periods, leading to persistence in amplitude of yield changes.

### Data sources, coverage, and construction
- International mutual funds data: EPFR Global.
  - EPFR Global covers 11,000 equity funds and 4,500 bond funds, with $23.5 trillion in total assets as of March 2014.
  - The paper focuses on 375 foreign mutual funds active in Mexico since 2007: 112 bond funds and 263 equity funds, with $1.43 trillion in total assets as of March 2014.
  - The analysis combines EPFR fund flow data and country allocation data via a unique fund identification number.
- Domestic mutual funds data: Econometica.
  - Covers 546 mutual funds and provides positions across 68 asset types.
  - As of May 2014, total gross assets of these domestic funds exceed over $130 billion, approximately half of which is invested in domestic sovereign bonds.
- Sample periods:
  - International mutual funds: monthly from January 2007 to March 2014.
  - Domestic mutual funds: monthly from January 2011 to May 2014.
- Construction of cumulative flows measures:
  - The measure of cumulative flows plotted in Figures 3 and 4 is the log difference of the gross cumulative flows of bond and equity funds to Mexico from those in the end of 2010, multiplied by 100.

### Key empirical questions and methodology
- Main research question: do foreign and domestic mutual fund investors in Mexico behave differently, and how does their behavior affect local market volatility, especially during stress periods?
- Three-part empirical approach:
  1. Use the herding measure introduced by Lakonishok et al. (1992) to assess which funds exhibit herding behavior, particularly during stress periods.
  2. Estimate fund-level panel regression models to examine how sensitivities of fund flows to global factors differ by fund characteristics.
  3. Investigate the relationship between excess demand generated by foreign mutual funds’ behavior and the volatility of returns to domestic assets by regressing within-month volatility of returns on (lagged) measures of excess demand.
- Fund characteristics exploited for heterogeneity analysis include investor type, redemption structure, geographic focus, and fund size (Appendix I defines fund characteristics).

### Descriptive findings and patterns
- Table 1 (Number of Foreign Mutual Funds Active in Mexico) documents growth in foreign fund representation across fund types and years (2007–2014) for multiple categories (Global, Emerging Markets, Latin America, Regional, Global Mexico-dedicated) for both equity and bond funds.
- Gross portfolio inflows to Mexico contracted sharply during:
  - the global financial crisis (2008–09)
  - after the U.S. Federal Reserve’s tapering announcement in May 2013
- Much of the large decline in inflows during these episodes reflects sharp reductions in foreign mutual funds’ investment in Mexico, especially by small retail funds (EPFR Retail).
- Domestic investors stepped in when portfolio inflows stopped during the global financial crisis by selling foreign assets and repatriating funds.
- Prior behavior documented in the literature motivates hypotheses tested in the paper:
  - mutual funds may herd due to benchmark-based compensation schemes or informational cascades
  - evidence exists of positive feedback (momentum) trading among funds investing in emerging markets

### Scope, limitations, and focus
- The paper focuses exclusively on mutual funds; findings should not be extrapolated to other segments such as banks, pension funds, and insurance companies.
- The EPFR dataset mainly tracks mutual funds and has limited observations on hedge funds and other investor types.
- The paper emphasizes comparisons between foreign and domestic mutual funds and within foreign funds by fund characteristics to assess differential behavior and implications for aggregate flows and volatility.

*Source: _wp15104 - 1. Gross Portfolio Inflows and Cumulative Flows of Mutual (Bond) Funds to ............................ 4*

### Appendix II provides details on the classification of assets.

### Appendix II provides details on the classification of assets.

### Cumulative flows by fund types and timing
- Measure: cumulative flows (plotted on the y-axis) constructed as the log difference of the gross cumulative flows of bond and equity funds to Mexico from those in the end of 2010, multiplied by 100.
- Shaded areas indicate the two stress episodes in the full sample: the height of the global financial crisis and tapering announcement.
- Observations in figures and text:
  - Increase in inflows by domestic mutual funds has been much smaller than the increase in inflows by foreign mutual funds in recent years.
  - Foreign mutual funds were selling domestic assets during stress episodes; there is no clear evidence that domestic mutual funds were buying these assets at the same time.
  - Among foreign mutual funds, bond flows to Mexico have increased much faster and have been more volatile than equity flows.
  - One major difference: cumulative flows from Latin America regional bond funds and global bond funds to Mexico have increased significantly since the beginning of 2011, while equity funds did not experience a similar increase. This could be attributed to Mexico joining the World Government Bond Index (WGBI) in October 2010.
  - Effects of global financial conditions on fund flows differ across types of foreign mutual funds; during stress episodes, selling of Mexican assets was more prevalent among:
    - open-end funds,
    - funds with retail investors,
    - Latin America regional funds.
  - These types of foreign mutual funds seem to have contributed more to capital flow volatility in Mexico than other types of funds during the stress episodes.

### Foreign versus domestic mutual funds during stress episodes
- Example: at the Lehman Brothers collapse in September 2008:
  - around 75 percent of equity funds active in Mexico were selling Mexican assets.
  - around 95 percent of bond funds active in Mexico were selling Mexican assets.
- During the tapering announcement in 2013:
  - about 50 percent of domestic mutual funds were selling Mexican assets.
  - the number of net sellers among foreign mutual funds exceeded 70 percent.

### Proportion of funds net selling Mexican assets (summary)
- Figures 5 and 6 show foreign mutual funds exhibit a strong tendency to sell Mexican assets during periods of heightened global uncertainties.
- Table 2. Percentage of Net Sellers among All Foreign Mutual Funds in Mexico (in Percent)
  - All Foreign Funds Investing in Mexico
    - Non-Stress
      - Bond: 48.6
      - Equity: 54.1
    - GFC
      - Bond: 81.6
      - Equity: 69.3
    - Tapering
      - Bond: 65.2
      - Equity: 63.5
- Notes on Table 2:
  - Columns (1), (2) and (3) report the average percentages of funds that are net sellers of Mexican assets during non-crisis times, after the collapse of Lehman Brothers (2008/09 – 2008/11), and the period around the tapering announcement by the Fed (2013/05 – 2013/07).
  - Source: EPFR Global and author’s calculations.

### Correlated selling and herding behavior
- Definitions and approach:
  - Proportion of all funds active in Mexico (in a particular month) that are net sellers is computed to give prima facie evidence of correlated selling during market stress.
  - Herding index used: the Lakonishok et al. (1992) herding measure is computed to assess which types of funds are most likely to exhibit herding behavior (funds moving in the same direction more often than expected under independent random trading).
- Findings:
  - Foreign mutual funds exhibit strong correlated selling during heightened global uncertainties.
  - Correlated selling is much weaker among domestic mutual funds during the same episodes.
  - Herding: the tendency of funds to move in the same direction (buying or selling) simultaneously, more often than would be expected if funds traded randomly and independently. Such behavior can destabilize markets, aggravate shocks, and lead to mispricing or asset price bubbles.

### Herding index results
- Figure 7 (left) shows:
  - The proportion of equity investors exhibiting herding behavior in the last 10 years is about one half what it was during 2000-2001.
  - There is an upward trend in herding behavior among equity funds in the second half of the 2000s, which might have been caused by the low global interest environment since the mid-2000s.
- Table 3 (summary statement):
  - Herding is prevalent among both bond and equity investors; all herding estimates in columns (1) and (2) are statistically significant.
  - Herding appears to be higher during episodes of market stress than during tranquil times for both bond and equity investors.
  - These findings align with prior studies on mutual fund herding in emerging markets.

*Source: EPFR Global and author’s calculations; figures and tables as presented in the Appendix II text.*

### Appendix III describes the methodology.

### _wp15104 - Appendix III describes the methodology.

### Herding indices and correlated selling
- Herding index computed only for foreign mutual funds due to limited data on domestic mutual funds; focus on comparing foreign bond and equity funds during stress versus non-stress periods.
- Table 3 — Herding Indices for All Foreign Funds Investing in Mexico (in Percent):
  - Bond: Non-Stress 2.20; Stress 3.72; Mean Difference 1.52 (standard errors in parentheses: (1.75), (2.65)); p-value of mean-difference test [0.03].
  - Equity: Non-Stress 2.71; Stress 3.22; Mean Difference 0.51 (standard errors (1.84), (2.15)); p-value [0.47].
- Interpretation and statistical tests:
  - Foreign bond investors herd more during episodes of market stress than foreign equity investors.
  - Proportion of investors with herding behavior during stress times is significantly higher (at 5 percent significance level) than during tranquil times for bond funds, but not for equity funds.
  - Herding index (Lakonishok et al., 1992) measures correlation in trading patterns and can give indicative evidence of “true” herding; positive values may reflect common reactions to public announcements (“spurious” herding), but misclassification is considered unlikely at monthly frequency unless reactions are asymmetric across investor groups.

### Global factors and positive feedback trading — models and key estimates
- Model for fund flow response to global financial conditions (equation (2)):
  - Flow of fund i to Mexico in month t: fund fixed effects, VIX (global volatility index), Stress dummy (global financial crisis and tapering announcement), interaction VIX*Stress, month fixed effects for foreign funds, clustered standard errors by fund.
  - Expected sign: coefficient on VIX negative; negative interaction implies greater vulnerability during stress.
- Table 4 — Effects of Global Financial Conditions (selected findings):
  - Foreign funds respond more strongly to changes in global financial conditions than domestic funds; magnitude of VIX coefficient for foreign funds is more than twice that for domestic funds in some regressions.
  - Increase in outflow by foreign bond funds during stress episodes is equivalent to USD 1.19 million for every one percent increase in VIX; increase for foreign equity funds is USD 0.377 million. This implies foreign bond funds are potentially more destabilizing than equity funds.
- Selected regression coefficient examples (dependent variable: fund flow in billions of USD; clustered standard errors):
  - Bond funds (Foreign, column (2)): VIX coefficient -0.000955* (standard error (0.000498)); VIX * Stress -0.00119** (0.000577); Stress -0.00247*** (0.000840).
  - Equity funds (Foreign, column (2)): VIX coefficient -0.000416** (0.000163); VIX * Stress -0.000377* (0.000196); Stress -0.000707** (0.000333).

### Momentum (positive feedback) trading — evidence and magnitudes
- Model for momentum trading (equation (3)): replace global risk factor with lagged returns (Mexico’s 3-month gov bond returns for bond funds; Mexico stock market returns for equity funds); one-month lag to mitigate endogeneity. Positive coefficient implies positive feedback trading; positive interaction with Stress implies stronger momentum in stress episodes.
- Table 5 — Momentum Trading Behavior (selected coefficients; dependent variable: fund flow in billions of USD; clustered standard errors):
  - Foreign bond funds:
    - Lagged return to Mexico three-month govt bond: 0.00323** (0.00139).
    - Lagged return to bond * Stress: 0.00733** (0.00306).
    - Stress: -0.00151*** (0.000430).
  - Foreign equity funds:
    - Lagged stock market return: 0.0172** (0.00707).
    - Lagged stock market return * Stress: 0.00235*** (0.000624).
    - Stress: -0.000202** (0.0000885).
  - Domestic equity funds exhibit “negative feedback trading”: lagged stock market return -0.0211** (0.00933).
- Robustness:
  - Regression (3) re-estimated using lagged changes in returns, returns to 10-year government bonds, and dollar-denominated returns; main results hold (see Appendix IV referenced in source).

### Effects of fund characteristics
- Correlated selling and herding vary by fund type and characteristics:
  - Table 6 — Percentage of Net Sellers among Foreign Funds by Fund Characteristics (selected entries):
    - Global Emerging Markets — Bond: Non-Crisis 48.4; GFC 80.8; Tapering 66.3.
    - Global Emerging Markets — Equity: Non-Crisis 52.7; GFC 66.1; Tapering 65.2.
    - Latin America Regional — Bond: Non-Crisis 53.8; GFC 91.7; Tapering 60.0.
    - Mexico-Dedicated — Bond: Non-Crisis 62.4; GFC - ; Tapering 47.0 (note: missing fund-level data for Mexico-dedicated funds for 2007/03–2009/10).
    - Institutional vs Retail and Open vs Closed redemption structures and Small vs Large fund-size categories reported with differing levels of net selling across episodes.
  - Three key results from correlated selling statistics:
    - After collapse of Lehman Brothers there were significantly more mutual funds net selling Mexican assets than during non-crisis times across all fund types.
    - Bond funds exhibited more correlated selling than equity funds during the Global Financial Crisis, but the opposite during non-crisis times.
    - Funds with retail investors, Latin America regional funds, and smaller funds exhibit stronger simultaneous selling than funds with institutional investors, global funds, and larger funds.
- Table 7 — Herding Indices for Foreign Funds by Fund Characteristics (in Percent; selected entries):
  - Global — Bond: Non-Stress mean 5.35 (Std Dev (3.27)); Stress mean 6.44 (4.25).
  - Global — Equity: Non-Stress 5.10 (3.58); Stress 5.31 (4.39).
  - Latin America Regional — Bond: Non-Stress 5.58 (5.05); Stress 6.22 (5.37).
  - Retail — Bond: Non-Stress 2.99 (2.55); Stress 4.21 (2.80). Retail — Equity: Non-Stress 4.85 (3.07); Stress 5.15 (3.67).
  - Open redemption — Bond: Non-Stress 2.33 (1.93); Stress 4.13 (3.05).
  - Small funds — Bond: Non-Stress 3.37 (2.62); Stress 5.70 (4.46). Small funds — Equity: Non-Stress 5.06 (4.22); Stress 5.42 (5.34).
  - All herding estimates reported in Table 7 are statistically significant at 5 percent significance level.
- Regression extensions (equations (4) and (5)):
  - Add characteristic dummies and interactions (VIX*Chara, VIX*Chara*Stress, Return*Chara, Return*Chara*Stress) to examine differential sensitivities.
  - Findings from Tables 8 and 9:
    - Regional (equity) funds and open-end (bond) funds respond more strongly to global financial conditions—especially during stress episodes—than global and closed-end counterparts.
    - Regional funds, open-end funds, small funds, and funds sold to retail investors are more likely to engage in positive feedback trading.
    - Bond funds have a higher tendency to chase returns during stress episodes (coefficient on the three-way interaction ߛସ significantly positive for bond funds).

### Do foreign flows affect market volatility? (excess demand measures and volatility regressions)
- Excess demand measures constructed following Lakonishok et al. (1992):
  - dratio (dollar ratio): total dollar increases / (total dollar increases + total dollar decreases) by bond/equity funds in a month.
  - nratio (numbers ratio): number of funds increasing holdings / number of funds changing holdings.
  - One-month lag of excess demand used to mitigate endogeneity.
- Table 10 — Impact of Excess Demand by Foreign & Domestic Mutual Funds on Local Market Volatility (selected coefficients; robust standard errors in parentheses):
  - Panel A — Excess Demand by Foreign Mutual Funds (dependent variables: within-month volatility of daily returns on Mexican 3-month government bond and within-month volatility of daily Mexican stock market returns):
    - Lagged nratio on bond volatility: 0.0108* (0.00599).
    - Lagged nratio * Stress on bond volatility: -0.0221*** (0.00814).
    - Stress on bond volatility: 0.0110*** (0.00402).
    - Lagged nratio on stock volatility: 2.085*** (0.100).
    - Lagged nratio * Stress on stock volatility: -4.344* (2.245).
    - Stress on stock volatility: 2.597** (0.996).
  - Panel B — Excess Demand by Domestic Mutual Funds:
    - Lagged dratio on bond volatility: 0.00900*** (0.00113).
    - Lagged dratio on stock volatility: 0.607** (0.289).
    - Lagged dratio * Stress coefficients smaller or not significant in some specifications.
    - Stress dummy positive and significant for stock market volatility (e.g., 1.817*** (0.529) for foreign funds’ comparable regression).
- Interpretation:
  - In tranquil times, excess purchase of Mexican assets by foreign mutual funds tends to amplify local market volatility; in stress times, excess selling yields higher volatilities.
  - Excess demand by domestic funds has qualitatively similar effects on volatility but quantitatively smaller magnitudes; foreign institutions’ excess demand generally has higher impact on volatility.

### Concluding remarks — key contributions and implications
- Three main contributions:
  1. Comparison of domestic and foreign mutual fund behavior in Mexico: foreign and domestic investors behave differently; foreign mutual funds show broad-based evidence of herding and positive feedback trading and are very responsive to global risks.
  2. Heterogeneity across fund types: bond funds are substantially more sensitive to global factors and engage in positive feedback trading more strongly than equity funds, especially during stress episodes; herding among bond funds is significantly higher in crisis periods. Small funds, funds sold to retail investors, open-end funds, and regional funds tend to engage more in momentum trading and herding.
  3. Impact on market volatility: foreign funds’ trading behavior is associated with higher local market volatilities, notably during market stress; domestic mutual funds play some mitigating role.
- Policy-relevant implication:
  - Changes in the composition of the investor base—e.g., rising share of bond flows to Mexico—may have important consequences for the stability of aggregate flows and for macro-prudential policies in Mexico.
- Caveat and extension:
  - Results should not be taken as a general characterization of all foreign and domestic investor behavior in Mexico; extensions could examine other institutional investors such as Mexican pension funds and insurance companies.

*Source: EPFR Global and author’s calculations, as reported in the PDF content unit.*

### Appendix I. Definition of Fund Characteristics

### Appendix I. Definition of Fund Characteristics

### Fund characteristics (categories and definitions)
- Fund size:
  - Large and small funds are defined as those above and below the median of assets under management (AUM) in each month, respectively.
- Ultimate investor:
  - EPFR Global provides share-level data and many funds report whether fund shares are sold to retail investors or to institutional investors.
  - A fund is identified as sold only to institutional investors or as sold also to retail investors.
  - If a fund does not report types of ultimate investors for all of its shares, the fund is identified as a retail fund when some of its shares are known to be sold to retail investors.
  - Unidentified observations are limited to around 7 percent of bond funds and around 14 percent of equity funds.
- Redemption structure:
  - EPFR Global identifies whether a fund is open-end or closed-end.
  - Open-end funds permit flexible additions and redemptions; closed-end funds do not.
- Geographic focus:
  - Four characteristics are used: global funds, global emerging market funds, Latin America regional funds, and Mexico-dedicated funds.

*Source: EPFR Global and author’s definitions.*

### Appendix II. Asset Classification for Domestic Mutual Funds

### Overview
- Dataset covers positions of 546 across 68 different types of assets.
- Assets are aggregated into 5 classes: bank deposits, equity, private bonds, sovereign bonds, and others (including derivatives).

### (1) Bank deposits (asset items listed verbatim)
- 3-month Savings protection bonds; 6-month Savings protection bonds; Bank notes; Certificate of deposit (Cedes); Certificate of deposit (spot exchange rate); Checking in Dollars; Checking in Euros; Checking in local currency; Savings protection bonds.

### (2) Equity (asset items listed verbatim)
- ADR's (spot exchange rate); Index warrants; Portfolio of warrants; Stock index futures; Stock of foreign co. (spot foreign exchange rate); Stock warrants; Stocks; Stocks of finance, credit and insurance co.; Stocks of foreign co.; Stocks of invest instit equity; Stocks of invest instit Fixed income; Stocks of unlisted companies; Stocks traded in international system; Stocks traded in intl system (spot foreign exchange rate); Tracking stocks (Foreign TRAC'S); Tracking stocks of foreign co. (ETF's spot exchange rate).

### (3) Private bonds (asset items listed verbatim)
- Bank development bonds; Componente repr. cupon segreg. prov. del principal de emisión CBIC; Corporate notes; Corporate notes (spot exchange rate); Corpotrack; Debentures; Debt issued by multilaterals; Debt of foreign issuers; Development bank notes; Eurobonds of private co. (spot exchange rate); Eurobonds of private companies; Foreign debt investment institutions (spot rate); Medium-term notes issued in intl market; Mortgage notes; Notes linked to commercial paper; Notes with detachable coupons; Ordinary participation certificates (CPO, CPOA); Peso denominated notes with detach coupons (princ and coupons); Principal STRIPS inflation-linked bonds; Promissory notes with interest payable at maturity; Subordinated notes.

### (4) Sovereign bonds (asset items listed verbatim)
- Bondes D; BPAG 28 BPA ́s con pago mensual de interés y tasa interés de ref. adic.; BPAG 91 Bonos protecc al ahorro, con pago trim de interés y tasa de ref. adic.; CETES with variant fiscal policy; Fed govern bonds placed internationally (UMS spot exch rate); Government development bonds 20-year (fixed interest); Municipal notes; Notes issued by federal government institutions; Short term government notes; Udibonos.

### (5) Others (including derivatives) (asset items listed verbatim)
- Call; Chicago futures; Chicago options; Dollar futures (minimum initial allocation); Foreign currencies futures; Futures of rates and bonds M3 (Mexder); INPC forward; Swaps; Real estate investment trusts; Separated trading of registered interest and principal securities (STRIPS) of fixed rate (federal government); Separeted trading of registered interest and principal securities (STRIPS) of fixed real rate (bonds); Trac ́s.

### Appendix III. Construction of the Herding Index

### Definition and interpretation
- Herding index R_{c,t} is defined as:
  - R_{c,t} = |p_{c,t} - p_t| - C_{c,t},
    where p_{c,t} is the proportion of all funds active in Mexico and month t that are net buyers, p_t is its expected value, and C_{c,t} is an adjustment factor to ensure R_{c,t} is zero if there is no herding.
- p_t is approximated by the share of funds that are net buyers across all emerging markets and varies over time to capture common trends across countries (e.g., swings in aggregate inflows).
- Sample of emerging market economies used to compute p_t includes:
  - Argentina, Bangladesh, Brazil, Bulgaria, Chile, China, Columbia, Croatia, Czech Republic, Egypt, Hungary, India, Indonesia, Israel, Jordan, Korea, Latvia, Lebanon, Lithuania, Malaysia, Mexico, Morocco, Nigeria, Pakistan, Peru, Philippines, Poland, Romania, Russia, Serbia, South Africa, Sri Lanka, Taiwan Province of China, Thailand, Turkey, Ukraine, Uruguay, and Vietnam.

### Adjustment factor and statistical assumption
- Adjustment factor C_{c,t} is given by:
  - C_{c,t} = E[|p_{c,t} - p_t|].
- The adjustment factor captures the random variation of p_{c,t} around p_t under the null hypothesis of independent trading, assuming the number of buyers follows a binomial distribution with parameter p = p_t.
- As the number of funds active in Mexico becomes larger, then under the null, C_{c,t} will be close to zero.

### Binomial distribution notation (as provided)
- The probability density function of the binomial distribution (with parameter) is given by:
  - f(k; N, p) = (N choose k) p^k (1 - p)^{N-k},
    where N is the number of funds active in Mexico in month t, and k, p correspond to the number of buyers and p_t as defined above.

### Appendix IV. Robustness Checks — Key regression results and statistics

Note: Clustered standard errors (by fund) in parentheses: *p<0.1, **p<0.05, ***p<0.01. The dependent variable is fund flow in billions of USD. Sample period for domestic and foreign funds: 2011 January – 2014 March (stress episode: tapering announcement).

### Robustness check (1) – Using changes in return in momentum trading regressions

- (A) Bond Funds — Selected coefficients and statistics
  - Domestic Funds (Column (1)): Lagged change in return to Mex 3-month govt bond = -0.0895 (0.0672); No. of observations = 18018; R-squared = 0.020.
  - Domestic Funds (Column (2)): Lagged change in return to Mex 3-month govt bond = -0.0784 (0.134); Lagged change in return * Stress = 0.00412 (0.154); Constant = 0.00916*** (0.000171); No. of observations = 18018; R-squared = 0.020.
  - Foreign Funds (Column (1)): Lagged change in return to Mex 3-month govt bond = 0.0228*** (0.00597); No. of observations = 2851; R-squared = 0.143.
  - Foreign Funds (Column (2)): Lagged change in return to Mex 3-month govt bond = 0.0156** (0.00642); Lagged change in return * Stress = 0.0132** (0.00621); Stress = -0.0189** (0.00906); Constant = 0.000802*** (0.000000760); No. of observations = 2851; R-squared = 0.152.

- (B) Equity Funds — Selected coefficients and statistics
  - Domestic Funds (Column (1)): Lagged change in stock market return = -0.0136* (0.00736); No. of observations = 15834; R-squared = 0.070.
  - Domestic Funds (Column (2)): Lagged change in stock market return = -0.00865 (0.00848); Lagged change in return * Stress = -0.0646* (0.0359); Constant = 0.00214*** (0.0000334); No. of observations = 15834; R-squared = 0.070.
  - Foreign Funds (Column (1)): Lagged change in stock market return = 0.00143* (0.000735); No. of observations = 15538; R-squared = 0.147.
  - Foreign Funds (Column (2)): Lagged change in stock market return = 0.00156** (0.000698); Lagged change in return * Stress = -0.000543 (0.00141); Stress = -0.00421* (0.00219); Constant = -0.000147*** (0.000000987); No. of observations = 15538; R-squared = 0.147.

- Source note appended in table: Source: EPFR Global and author’s calculations.

### Robustness check (2) – Using the return on Mexico’s 10-year government bonds

- Domestic (Bond) Funds
  - Lagged return to Mex 10-year govt bond = 0.0370 (0.0264); Lagged bond yield * Stress = -0.115** (0.0506); Constant = 0.00896*** (0.000214); Number of Observations = 20202; R-squared = 0.019.
  - Column (2) reports Lagged return = 0.0718** (0.0317); Constant = 0.00828*** (0.000706); Number of Observations = 20202; R-squared = 0.019.

- Foreign (Bond) Funds
  - Column (1): Lagged return to Mex 10-year govt bond = 0.0110*** (0.00277); Constant = 0.000697*** (0.0000201); Number of Observations = 7287; R-squared = 0.180.
  - Column (2): Lagged return = 0.00871*** (0.00256); Lagged bond yield * Stress = 0.00567* (0.00302); Stress = -0.0234** (0.0116); Constant = 0.000852*** (0.0000581); Number of Observations = 7287; R-squared = 0.181.

- Standard errors (clustered at the fund level) in parentheses. Significance notation: *p<0.10 **p<0.50 ***p<0.01 (as printed).

### Robustness check (3) – Using dollar-denominated return on Mexico’s 3-month government bonds

- Domestic (Bond) Funds
  - Lagged dollar return to Mex 3-month govt bond = -1.176 (0.968); Lagged bond yield * Stress = 0.196 (2.371); Constant = 0.0100*** (0.000637); Number of Observations = 20202; R-squared = 0.019.
  - Column (2) reports Lagged dollar return = -1.367 (1.882); Constant = 0.00995*** (0.00104); Number of Observations = 20202; R-squared = 0.019.

- Foreign (Bond) Funds
  - Column (1): Lagged dollar return to Mex 3-month govt bond = 0.0422** (0.0162); Constant = 0.000728*** (0.0000187); Number of Observations = 7287; R-squared = 0.173.
  - Column (2): Lagged dollar return = 0.0247* (0.0129); Lagged bond yield * Stress = 0.158*** (0.0470); Stress = -0.0194** (0.00922); Constant = 0.000959*** (0.0000629); Number of Observations = 7287; R-squared = 0.177.

### References and data sources cited within appendices
- Source: EPFR Global and author’s calculations.
- Selected bibliographic entries included in the appendix (authors and titles as listed in the source text).

*Appendix I–IV, _wp15104 - Appendix I. Definition of Fund Characteristics_*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15104.pdf_