## 1. The Rise of Portfolio Flow Proxies: What Accounts for the Differences between EPFR, IIF and BoP Data?

## Source details

**Canonical URL:** [1. The Rise of Portfolio Flow Proxies: What Accounts for the Differences between EPFR, IIF and BoP Data?](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020171-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020171-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020171-print-pdf.pdf.json)

---

### Introduction and motivation
- In early 2020, the COVID-19 pandemic triggered one of the sharpest reversals in portfolio flows to emerging markets (EMs) on record; timely high-frequency proxies signaled the shock well before quarterly BoP releases.
- High-frequency proxies (monthly, weekly, daily) from EPFR and IIF showed in March 2020 that the reversal was extremely severe in speed and magnitude.
- Data availability on international capital flows has improved dramatically over the past two decades, especially for high-frequency portfolio flow proxies.
- Conceptual and measurement issues, and limited methodological documentation from private providers, complicate comparison across datasets and with standard balance of payments (BoP) accounting.

### Four main contributions of the paper
- Overview of widely used datasets, strengths and weaknesses, with emphasis on high-frequency portfolio/fund flow measures and common misconceptions about BoP accounting.
- Meta-study on data use in the empirical literature, showing how data choice may have shaped empirical results (e.g., possible over-emphasis on external “push” factors due to prevalence of fund flow data and benchmark effects).
- Provision of a free online monthly EM portfolio flow dataset tailored for academic research to help close gaps in BoP-consistent, high-frequency data availability.
- Quantitative assessment of how well portfolio flow proxies track BoP-based portfolio flows in real time; nowcasting results show substantial predictive content.

### Key empirical findings and statistics
- From 2010:Q1–2019:Q2 average quarterly fund flows to EMs as a group:
  - $11 billion according to EPFR fund flow data
  - $68 billion for the IIF Portfolio Flows Tracker
  - $71 billion for IMF BoP portfolio flow data
- Meta-study coverage:
  - Academic dataset: 88 studies; IMF balance of payments data used in 39 percent of studies, BIS in 20 percent, EPFR in 14 percent.
  - Policy dataset: 111 studies; high-frequency portfolio flow proxies used in about 50 percent of studies since 2010; EPFR accounts for about 67 percent of those, IIF for 21 percent (full sample).
- Nowcasting horse race:
  - IIF and EPFR data reduce forecast errors by 80-90 percent relative to an autoregressive benchmark.
  - Daily and weekly proxies outperform monthly data in the first half of the current quarter.
  - IIF data generally outperform EPFR in predictive performance for BoP-based portfolio flows (equity and debt RMSFE results).

### Conceptual and measurement issues highlighted
- Private-sector datasets often lack detailed public methodological documentation and differ conceptually from BoP accounting.
- Important BoP accounting principles:
  - Residency: flows arise from acquisitions/disposals between residents of different countries.
  - Quadruple entry bookkeeping: transactions recorded in both countries, reflecting source and use of funds.
  - Transactions at market value: flows recorded using market value at time of transaction; valuation effects are separate.
- The term “flow” differs across literatures: capital flow literature uses “flow” for transactions; macro statistics “flow” can include transactions, valuation changes, and other flows (BPM6 ¶2.2).

### Overview of common data sources
- IMF Balance of Payments Statistics (BOPS)
  - Frequency and lag: quarterly and annual; typically released with a lag of two to four months.
  - Coverage: almost all EMs; comprehensive; data available gross and net.
  - Advantages: internationally recognized BPM6 standards; comparable across countries and time.
  - Caveats: statistical break in 2005–2008 due to BPM5→BPM6 shift.
- EPFR (Emerging Portfolio Fund Research)
  - Frequency and lag: daily/weekly/monthly; short release lag (1-5 days).
  - Coverage: almost all EMs, some frontier markets; inflows into EM-dedicated investment funds (equity, debt).
  - Advantages: high frequency, short lag, granular breakdowns.
  - Caveats: conceptually different from BoP; sample of reporting funds; top-down country estimates; institutional investors underrepresented; some debt flows limited to local-currency and/or sovereign bonds; recent months often revised.
- IIF (Institute of International Finance) Portfolio Flows Trackers and Capital Flows Databases
  - Frequency and lag: daily/weekly/monthly trackers and quarterly/annual databases; typical release lag 1–4 months depending on product.
  - Coverage: limited country sample relative to IMF BOPS.
  - Advantages: short release lag for trackers; includes forecasts in some products.
  - Caveats: subscription-based; limited country sample relative to IMF BOPS.
- BIS and other sources
  - BIS Locational Banking Statistics: quarterly; comprehensive cross-border banking activity; flows constructed from stocks and adjusted for FX valuation effects; “FX and break-adjusted changes” are technically not “flows” in a narrow sense.

### Meta-study insights on dataset usage and implications
- Academic literature (88 papers since 1993 sample): IMF BOPS most used (39 percent), followed by BIS (20 percent), EPFR (14 percent).
- Policy reports (220 reviewed since 2010, sample of 111 shown): high-frequency proxies increasingly used; EPFR is dominant among those proxies.
- Use of fund-flow datasets may bias empirical inference toward external “push” drivers because fund flows are subject to benchmark effects common across EMs.
- Underuse of BoP-consistent portfolio flow data at high frequencies in academic research, partly due to subscription barriers and sample construction challenges.

### KP Monthly Portfolio Flow Dataset (BoP-consistent) — construction and coverage
- Purpose: provide monthly EM portfolio flows broadly consistent with balance of payments (BoP) accounting and geared toward academic use.
- Availability: dataset posted online with the paper and periodically updated.
- Coverage:
  - Constructed from national sources for a set of 18 EMs.
  - Data for total portfolio flows and debt and equity portfolio flows.
  - Data availability begins in 2010 for most countries.
- KP definition: Transaction-based between residents and non-residents, not subject to valuation effects, reflect transactions at market prices.
- Countries included (18): Brazil, Bulgaria, Chile, Czech Republic, Hungary, India, Lebanon, Mexico, Pakistan, Philippines, Poland, Romania, South Africa, Korea, Sri Lanka, Thailand, Turkey, Ukraine.
  - Note: Czech Republic and Korea are not part of the IMF’s classification of emerging markets but are included in private-sector EM classifications.
- Proxy data used for some countries: government debt flows for Mexico; non-government debt flows for Thailand; transactions on national stock exchanges by non-residents for South Africa and Sri Lanka; portfolio inflows by non-residents for India.
- KP dataset coverage (2010:Q1 to 2019:Q2): around 46 percent of total flows in the BOPS aggregate.
  - Equity coverage: 53 percent.
  - Debt coverage: 44 percent.

### KP dataset metadata (selected entries preserved exactly)
- Brazil: Central Bank of Brazil, 1-2 months, USD, correlation 1.00
- Bulgaria: Eurostat, 2-3 months, EUR, correlation 1.00
- Chile: Central Bank of Chile, 2-3 months, USD, correlation 1.00
- Czech Republic*: Czech National Bank, 2-3 months, EUR, correlation 1.00
- Hungary: Eurostat, 2-3 months, EUR, correlation 1.00
- India**: Securities & Exchange Board of India, 0-1 months, INR, Proxy Y, correlation 0.92
- Korea*: Bank of Korea, 1-2 months, USD, correlation 1.00
- Lebanon: Bank of Lebanon, 10-11 months, USD, correlation 0.97
- Mexico: Bank of Mexico, 3-4 months, USD, Proxy Y, correlation 0.86
- Pakistan: State Bank of Pakistan, 1-2 months, USD, correlation 0.98
- Philippines: Central Bank of the Philippines, 4-5 months, USD, correlation 1.00
- Poland: National Bank of Poland, 2-3 months, EUR, correlation 1.00
- Romania: Eurostat, 2-3 months, EUR, correlation 1.00
- South Africa: Johannesburg Stock Exchange, 2-3 months, ZAF, Proxy Y, correlation 0.64
- Sri Lanka: Colombo Stock Exchange, 2-3 months, USD, Proxy Y, correlation 0.30
- Thailand: Bank of Thailand, 3-4 months, USD, Proxy Y, correlation 0.92
- Turkey: Central Bank of Turkey, 2-3 months, USD, correlation 1.00
- Ukraine: National Bank of Ukraine, 2-3 months, USD, correlation 1.00
- Notes:
  - * Czech Republic and Korea are included despite IMF classification.
  - ** India’s portfolio flow data are recorded on a reporting day basis rather than on a trading day basis, which may contribute to a lower correlation between data used in the KP dataset and Balance of Payments data.

### Correlations with Balance of Payments portfolio flows (2010:Q1–2019:Q2) — selected entries
- All EMs: KP Equity 0.85; KP Debt 0.64.
- IIF Monthly: Equity 0.86; Debt 0.83.
- EPFR Monthly: Equity 0.67; Debt 0.68.
- India: KP Equity 0.94; KP Debt 0.93.
- South Africa: KP Equity 0.74; KP Debt 0.64.
- Thailand: KP Equity 1.00; KP Debt 0.65.
- Turkey: KP Equity 1.00; KP Debt 1.00.

### Comparisons of variance and means with BoP
- Average quarterly flows in 18 country cases (except South Africa) are equal or close to BoP averages.
- Standard deviation: majority show KP co-movement roughly in line with BoP; exceptions include Mexico, Sri Lanka, South Africa, and Thailand.
- Compared to EPFR, the KP dataset aligns more closely with standard BoP definitions.

### Real-Time Tracking of Portfolio Flows — Nowcasting horse race
- Proxies evaluated:
  - Weekly EPFR (published with a lag of about 7 days).
  - Monthly EPFR (published with a lag of about 15 days).
  - Daily IIF (published with a lag of 1-3 days).
  - Monthly IIF (published with a lag of about 1-5 days).
  - KP Monthly dataset (underlying data typically available after about 2-3 months).
- Sample period: 2010:Q1 until 2019:Q2.
- Baseline model: aggregate flows for EM countries where all proxies are available, scaled by combined GDP; pseudo-out-of-sample rolling regression/nowcast producing repeated i-step nowcasts within each quarter.
- Evaluation metric: Root-mean squared forecast error (RMSFE), computed for sums of high-frequency proxies incrementally through each quarter.
- Sample composition for nowcasting:
  - Equity flows sample: Brazil, India, Pakistan, Philippines, South Africa, Korea, Sri Lanka, Thailand, and Turkey.
  - Debt flows sample: Hungary, India, Poland, South Africa, Thailand, and Turkey.

### Nowcasting results — key findings and quantitative performance
- General patterns:
  - Forecast errors for all proxies decline over the course of the quarter as more data accumulate.
  - Daily IIF and weekly EPFR are available earliest (as early as 2 and 7 days into quarter respectively); monthly proxies become available around 33 (IIF) and 45 (EPFR) days into the quarter.
  - RMSFEs for the first few observations of daily and weekly proxies are relatively high; daily IIF performance improves notably after 20–30 days of data.
- Relative performance:
  - IIF proxies generally outperform EPFR proxies.
  - Higher-frequency proxies generally outperform lower-frequency proxies, especially in the first half of the current quarter.
  - IIF’s monthly tracker consistently performs better than either EPFR dataset, but only outperforms IIF daily flows after the second data release in the quarter on average.
  - EPFR monthly mostly underperforms EPFR weekly, with the exception of debt flows toward the end of the quarter.
- Economic significance:
  - RMSFEs of debt and equity flows based on daily and monthly IIF data outperform weekly and monthly EPFR data by around 20 percent early in the quarter and 50 percent at the end of the quarter (expressed as a share of absolute average quarterly flows).
  - All proxies outperform a benchmark autoregressive regression with one lag and a constant term by at least 80 percent (RMSFEs at least 80 percent lower).
- KP dataset performance:
  - KP monthly dataset has similar performance to IIF monthly tracker, but KP data are subject to a greater release lag within the current quarter.
  - Caveat: IIF monthly tracker and KP dataset observations for the latest 1-2 months are revised with each monthly release for the EM aggregate and for many country-level flows, which may bias results toward smaller forecast errors for these datasets. Other sources generally do not get revised.

### Robustness checks and extensions
- Extending sample back to 2005:Q1 did not substantially change average forecast errors.
- Including quarterly averages of global financial variables (VIX, US Treasury yields) as additional predictors did not consistently reduce RMSFEs nor change relative magnitudes of forecast errors substantively.
- Nowcasting exercises run for aggregate EM flows and for individual EM countries where data from all sources are available.

### BIS Locational and Consolidated Banking Statistics (overview)
- LBS:
  - Frequency and history: LBS data are available on a quarterly basis since December 1977. Instrument breakdown available from December 1995.
  - Relation to BoP: LBS most closely related to the “other investment” component in the BoP; LBS also capture banks’ holdings of debt securities (classified as portfolio investment in BoP).
  - Adjustments: BIS derive “FX and break-adjusted changes”; these are technically not “flows” in a narrow sense.
- CBS:
  - Treatment of intra-group positions: intra-banking group claims are excluded in CBS, in contrast to LBS.
  - Frequency and history: CBS on an immediate counterparty basis semi-annual from December 1983 for 16 years and quarterly from December 1999; quarterly CBS on guarantor basis publicly available since early 2005.
  - Counterparty basis: CBS provide immediate counterparty and guarantor basis data (guarantor basis records flows to country of guarantor residence).

### Conclusion and policy implications
- IMF balance of payments statistics remain a key benchmark; high-frequency proxies have improved real-time information provision.
- Fund flow data have been used by a sizeable share of empirical studies (around 14 percent) over the past decade and allow insights into rapid investor behavior shifts.
- Conceptual and empirical differences between fund flow data and BoP portfolio flows imply empirical findings are likely shaped by unique properties of fund flow data.
- Greater focus on high-frequency data consistent with BoP accounting seems warranted but is hampered by subscription requirements and dataset limitations.
- The KP monthly dataset covering 18 EMs may facilitate academic research and help assess the relative importance of external drivers versus domestic factors in portfolio flow dynamics.
- Policymakers and researchers should understand the nuances of capital flow datasets to choose the best-suited data for the question at hand.

*Source: wpiea2020171-print-pdf - 1. The Rise of Portfolio Flow Proxies: What Accounts for the Differences between EPFR, IIF and BoP Data?*

### 1. The Rise of Portfolio Flow Proxies: What Accounts for the Differences between EPFR,

### 1. The Rise of Portfolio Flow Proxies: What Accounts for the Differences between EPFR, IIF and BoP Data?

### Introduction and motivation
- In early 2020, the COVID-19 pandemic triggered one of the sharpest reversals in portfolio flows to emerging markets (EMs) on record; timely high-frequency proxies signaled the shock well before quarterly BoP releases.
- High-frequency proxies (monthly, weekly, daily) from EPFR and IIF showed in March 2020 that the reversal was extremely severe in speed and magnitude.
- Data availability on international capital flows has improved dramatically over the past two decades, especially for high-frequency portfolio flow proxies (Figure 1).
- However, conceptual and measurement issues, and limited methodological documentation from private providers, complicate comparison across datasets and with standard balance of payments (BoP) accounting.

### Four main contributions of the paper
- Overview of widely used datasets, strengths and weaknesses, with emphasis on high-frequency portfolio/fund flow measures and common misconceptions about BoP accounting.
- Meta-study on data use in the empirical literature, showing how data choice may have shaped empirical results (e.g., possible over-emphasis on external “push” factors due to prevalence of fund flow data and benchmark effects).
- Provision of a free online monthly EM portfolio flow dataset tailored for academic research to help close gaps in BoP-consistent, high-frequency data availability.
- Quantitative assessment of how well portfolio flow proxies track BoP-based portfolio flows in real time; nowcasting results show substantial predictive content.

### Key empirical findings and statistics
- From 2010:Q1–2019:Q2 average quarterly fund flows to EMs as a group:
  - $11 billion according to EPFR fund flow data
  - $68 billion for the IIF Portfolio Flows Tracker
  - $71 billion for IMF BoP portfolio flow data
- Meta-study coverage:
  - Academic dataset: 88 studies; IMF balance of payments data used in 39 percent of studies, BIS in 20 percent, EPFR in 14 percent.
  - Policy dataset: 111 studies; high-frequency portfolio flow proxies used in about 50 percent of studies since 2010; EPFR accounts for about 67 percent of those, IIF for 21 percent (full sample).
- Nowcasting horse race:
  - IIF and EPFR data reduce forecast errors by 80-90 percent relative to an autoregressive benchmark (Figure 3).
  - Daily and weekly proxies outperform monthly data in the first half of the current quarter.
  - IIF data generally outperform EPFR in predictive performance for BoP-based portfolio flows (equity and debt RMSFE results shown in Figure 3).

### Conceptual and measurement issues highlighted
- Private-sector datasets often lack detailed public methodological documentation and differ conceptually from BoP accounting.
- Important BoP accounting principles to anchor analysis:
  - Residency: flows arise from acquisitions/disposals between residents of different countries.
  - Quadruple entry bookkeeping: transactions recorded in both countries, reflecting source and use of funds.
  - Transactions at market value: flows recorded using market value at time of transaction; valuation effects are separate.
- The term “flow” differs across literatures: in capital flow literature it refers to transactions; in macro statistics “flow” can include transactions, valuation changes, and other flows (BPM6 ¶2.2).

### Overview of common data sources (high level)
- IMF Balance of Payments Statistics (BOPS)
  - Frequency and lag: quarterly and annual; typically released with a lag of two to four months.
  - Coverage: almost all EMs; comprehensive coverage of cross-border transactions; data available both gross and net for major components.
  - Advantages: internationally recognized BPM6 standards; comparable across countries and time.
  - Caveats: statistical break in 2005–2008 due to BPM5→BPM6 shift (IMF BOPS publishes historical data in BPM6 format).
- EPFR (Emerging Portfolio Fund Research)
  - Frequency and lag: daily/weekly/monthly; short release lag (1-5 days).
  - Coverage: almost all EMs, some frontier markets; inflows into EM-dedicated investment funds (equity, debt).
  - Advantages: high frequency, short lag, insights into ultimate investor behavior, granular breakdowns.
  - Caveats: conceptually different from BoP; sample of reporting funds; country-level flows estimated top-down; institutional investors underrepresented; some debt flows limited to local-currency and/or sovereign bonds; recent months often revised.
- IIF (Institute of International Finance) Portfolio Flows Trackers and Capital Flows Databases
  - Frequency and lag: daily/weekly/monthly trackers and quarterly/annual databases; typical release lag 1–4 months depending on product.
  - Coverage: limited country sample relative to IMF BOPS (e.g., 25 EMs in some products).
  - Advantages: short release lag for trackers; includes forecasts in some products.
  - Caveats: subscription-based; limited country sample relative to IMF BOPS.
- BIS and other sources (high level)
  - BIS Locational Banking Statistics: quarterly; comprehensive cross-border banking activity; no direct mapping to standard BoP flow components because flows are constructed from stock data and adjusted for FX valuation effects.

### Meta-study insights on dataset usage and implications
- Academic literature (88 papers since 1993 sample): IMF BOPS most used (39 percent), followed by BIS (20 percent), EPFR (14 percent); some sources grouped under "Other" if used less than 5 times.
- Policy reports (220 reviewed since 2010, sample of 111 shown): high-frequency proxies increasingly used; EPFR is dominant among those proxies.
- Use of fund-flow datasets may bias empirical inference toward external “push” drivers because fund flows are subject to benchmark effects common across EMs.
- There is underuse of BoP-consistent portfolio flow data at high frequencies in academic research, partly due to subscription barriers and sample construction challenges.

### Nowcasting and practical performance for policymakers
- High-frequency proxies (IIF, EPFR) provide substantial real-time information on BoP-based portfolio flows and can be used to reduce uncertainty before official BoP releases.
- Relative performance:
  - IIF trackers generally outperform EPFR in nowcasting BoP portfolio flows.
  - Higher-frequency data (daily/weekly) provide better information within the quarter than monthly proxies.
- The paper supplies a free monthly EM portfolio flow dataset aimed at academic users to facilitate BoP-consistent analysis.

### Structure of the paper (for navigation)
- Section 2: Overview of capital flow data sources and misconceptions.
- Section 3: Meta-study on data usage in empirical literature and introduction of the new monthly dataset.
- Section 4: Assessment of predictive content of several portfolio flow proxies (nowcasting exercises).
- Section 5: Conclusion.
- Annexes include bilateral (“from-whom-to-whom”) data, daily IIF data description, list of papers in the academic dataset, and portfolio flow nowcasting results by country.

*Source: wpiea2020171-print-pdf - 1. The Rise of Portfolio Flow Proxies: What Accounts for the Differences between EPFR, IIF and BoP Data?*

### 2. BIS Locational and Consolidated Banking Statistics

### 2. BIS Locational and Consolidated Banking Statistics

### Overview of BIS cross-border banking data
- The BIS provides comprehensive coverage of international banking flows and positions, including currency composition, instrument type, and sector and residency of counterparty (BIS 2019).
- Data are reported to the BIS by national central banks and compiled following methods broadly consistent with BoP accounting principles.

### Locational Banking Statistics (LBS)
- Frequency and history:
  - LBS data are available on a quarterly basis since December 1977.
  - Instrument breakdown available from December 1995.
- Relation to balance of payments:
  - LBS data are most closely related to the “other investment” component in the BoP, which includes cross-border bank loans and deposits.
  - LBS also capture banks’ holdings of debt securities, which in the BoP are classified as portfolio investment.
- Adjustments and interpretation:
  - The reported currency breakdown and break-in-series allow the BIS to derive “FX and break-adjusted changes.”
  - Unlike BoP flows, the BIS “FX and break-adjusted changes” are technically not “flows” in a narrow sense.

### Consolidated Banking Statistics (CBS)
- Treatment of intra-group positions:
  - In the CBS, intra-banking group claims are excluded, in contrast to the LBS.
- Frequency and historical availability:
  - CBS data on an immediate counterparty basis are available at the semi-annual frequency from December 1983 for 16 years and on a quarterly basis from December 1999 onwards.
  - Quarterly CBS data on a guarantor basis are publicly available since early 2005.
- Counterparty basis definitions:
  - CBS provide data both on an immediate counterparty basis and on a guarantor basis, where the guarantor is the entity assuming contractual responsibilities if the immediate counterparty defaults (BIS 2019).
  - Guarantor basis data record a cross-border banking flow between country A and the country of residence of the counterparty’s guarantor, rather than the country of residence of the counterparty itself.

*Source: wpiea2020171-print-pdf - 2. BIS Locational and Consolidated Banking Statistics*

### 1.3 trillion, or around 7 percent, while outward investment stocks roughly double from

### wpiea2020171-print-pdf - 1.3 trillion, or around 7 percent, while outward investment stocks roughly double from

### III.A Meta-Study of Data Sources Used in the Empirical Literature
- Sample: 88 studies published over the last 27 years (selection begins in 1993 with Calvo, Leiderman and Reinhart); preference to studies with 20+ citations and top-two-quartile journals per SCImago.
- Timing: Sixty-three studies (about 70 percent) were published after 2007.
- Data frequency evolution:
  - 1993–2007: around 50 percent of studies used annual frequency.
  - 2008–2019: annual-only studies account for around 20 percent; quarterly around 50 percent; monthly or higher around 25 percent.
- Capital flow component usage by frequency: increase in portfolio flow usage with higher-frequency data; bank/FDI data less available at high frequency.
- Concern: Dominance of fund flow data (notably EPFR) in high-frequency studies may bias findings toward external (“push”) factors because fund flows are more affected by push factors and investment benchmarks.

### Data Frequency, Push vs. Pull, and Study Focus
- Finding: High-frequency data studies tend to emphasize external (push) factors; lower-frequency studies tend to emphasize domestic (pull) factors.
- Example evidence: Ananchotikul and Zhang (2014) find push factors dominate at short horizons while pull factors gain importance at longer horizons.
- Implication: Data frequency choice can affect estimated statistical significance and economic importance of push versus pull factors.

### III.B Construction of KP Monthly Portfolio Flow Dataset (BoP-consistent)
- Purpose: Provide monthly emerging market (EM) portfolio flows broadly consistent with balance of payments (BoP) accounting and geared toward academic use.
- Availability: Dataset posted online with the paper and periodically updated.
- Coverage:
  - Constructed from national sources for a set of 18 EMs.
  - Data for total portfolio flows and debt and equity portfolio flows.
  - Data availability begins in 2010 for most countries.
- KP definition: Transaction-based between residents and non-residents, not subject to valuation effects, reflect transactions at market prices.
- Countries included (18): Brazil, Bulgaria, Chile, Czech Republic, Hungary, India, Lebanon, Mexico, Pakistan, Philippines, Poland, Romania, South Africa, Korea, Sri Lanka, Thailand, Turkey, Ukraine.
  - Note: Czech Republic and Korea are not part of the IMF’s classification of emerging markets but are included in private-sector EM classifications.
- Proxy data used for some countries: government debt flows for Mexico; non-government debt flows for Thailand; transactions on national stock exchanges by non-residents for South Africa and Sri Lanka; portfolio inflows by non-residents for India.

### Table 3 (KP dataset metadata — key entries preserved)
- Release lag (approx.), Currency, Proxy, Correlation with total BoP flows (selected entries):
  - Brazil: Central Bank of Brazil, 1-2 months, USD, correlation 1.00
  - Bulgaria: Eurostat, 2-3 months, EUR, correlation 1.00
  - Chile: Central Bank of Chile, 2-3 months, USD, correlation 1.00
  - Czech Republic*: Czech National Bank, 2-3 months, EUR, correlation 1.00
  - Hungary: Eurostat, 2-3 months, EUR, correlation 1.00
  - India**: Securities & Exchange Board of India, 0-1 months, INR, Proxy Y, correlation 0.92
  - Korea*: Bank of Korea, 1-2 months, USD, correlation 1.00
  - Lebanon: Bank of Lebanon, 10-11 months, USD, correlation 0.97
  - Mexico: Bank of Mexico, 3-4 months, USD, Proxy Y, correlation 0.86
  - Pakistan: State Bank of Pakistan, 1-2 months, USD, correlation 0.98
  - Philippines: Central Bank of the Philippines, 4-5 months, USD, correlation 1.00
  - Poland: National Bank of Poland, 2-3 months, EUR, correlation 1.00
  - Romania: Eurostat, 2-3 months, EUR, correlation 1.00
  - South Africa: Johannesburg Stock Exchange, 2-3 months, ZAF, Proxy Y, correlation 0.64
  - Sri Lanka: Colombo Stock Exchange, 2-3 months, USD, Proxy Y, correlation 0.30
  - Thailand: Bank of Thailand, 3-4 months, USD, Proxy Y, correlation 0.92
  - Turkey: Central Bank of Turkey, 2-3 months, USD, correlation 1.00
  - Ukraine: National Bank of Ukraine, 2-3 months, USD, correlation 1.00
- Notes:
  - * Czech Republic and Korea are included despite IMF classification.
  - ** India’s portfolio flow data are recorded on a reporting day basis rather than on a trading day basis, which may contribute to a lower correlation between data used in the KP dataset and Balance of Payments data.

### KP Dataset Coverage and Correlations
- Coverage of KP dataset (2010:Q1 to 2019:Q2): around 46 percent of total flows in the BOPS aggregate.
  - Equity coverage: 53 percent.
  - Debt coverage: 44 percent.
- Correlation with Balance of Payments Portfolio Flows (2010:Q1–2019:Q2) — selected entries:
  - All EMs: KP Equity 0.85; KP Debt 0.64.
  - IIF Monthly: Equity 0.86; Debt 0.83.
  - EPFR Monthly: Equity 0.67; Debt 0.68.
  - India: KP Equity 0.94; KP Debt 0.93.
  - South Africa: KP Equity 0.74; KP Debt 0.64.
  - Thailand: KP Equity 1.00; KP Debt 0.65.
  - Turkey: KP Equity 1.00; KP Debt 1.00.
  - IIF Daily and IIF Monthly generally show high correlations for several country cases; EPFR Monthly shows lower correlations in several country comparisons.

### Comparisons of Variance and Means with BoP
- Average quarterly flows in 18 country cases (except South Africa) are equal or close to BoP averages.
- Standard deviation: For majority, KP co-movement shows standard deviation roughly in line with BoP; exceptions include Mexico, Sri Lanka, South Africa, and Thailand.
- Compared to EPFR, the KP dataset aligns more closely with standard BoP definitions.

### IV. Real-Time Tracking of Portfolio Flows — Nowcasting Horse Race
- Motivation: Official BoP data have large release lags; policymakers need timely information for monetary policy and FX intervention.
- Proxies evaluated:
  - Weekly EPFR (published with a lag of about 7 days).
  - Monthly EPFR (published with a lag of about 15 days).
  - Daily IIF (published with a lag of 1-3 days).
  - Monthly IIF (published with a lag of about 1-5 days).
  - KP Monthly dataset (underlying data typically available after about 2-3 months).
- Sample period: 2010:Q1 until 2019:Q2.
- Baseline model: aggregate flows for EM countries where all proxies are available, scaled by combined GDP; pseudo-out-of-sample rolling regression/nowcast producing repeated i-step nowcasts within each quarter.
- Evaluation metric: Root-mean squared forecast error (RMSFE), computed for sums of high-frequency proxies incrementally through each quarter.
- Sample composition for nowcasting (by component):
  - Equity flows sample contains Brazil, India, Pakistan, Philippines, South Africa, Korea, Sri Lanka, Thailand, and Turkey.
  - Debt flows sample contains Hungary, India, Poland, South Africa, Thailand, and Turkey.

### Nowcasting Results — Key Findings
- General patterns:
  - Forecast errors for all proxies decline over the course of the quarter as more data accumulate.
  - Daily IIF and weekly EPFR are available earliest (as early as 2 and 7 days into quarter respectively); monthly proxies become available around 33 (IIF) and 45 (EPFR) days into the quarter.
  - RMSFEs for the first few observations of daily and weekly proxies are relatively high; daily IIF performance improves notably after 20–30 days of data.
- Relative performance:
  - IIF proxies generally outperform EPFR proxies.
  - Higher-frequency proxies generally outperform lower-frequency proxies, especially in the first half of the current quarter.
  - IIF’s monthly tracker consistently performs better than either EPFR dataset, but only outperforms IIF daily flows after the second data release in the quarter on average.
  - EPFR monthly mostly underperforms EPFR weekly, with the exception of debt flows toward the end of the quarter.
- Economic significance:
  - RMSFEs of debt and equity flows based on daily and monthly IIF data outperform weekly and monthly EPFR data by around 20 percent early in the quarter and 50 percent at the end of the quarter (expressed as a share of absolute average quarterly flows).
  - All proxies outperform a benchmark autoregressive regression with one lag and a constant term by at least 80 percent (RMSFEs at least 80 percent lower).
- KP dataset performance:
  - KP monthly dataset has similar performance to IIF monthly tracker, but KP data are subject to a greater release lag within the current quarter.
  - Caveat: IIF monthly tracker and KP dataset observations for the latest 1-2 months are revised with each monthly release for the EM aggregate and for many country-level flows, which may bias results toward smaller forecast errors for these datasets. Other sources generally do not get revised.

### Robustness and Extensions
- Robustness checks:
  - Extending sample back to 2005:Q1 did not substantially change average forecast errors.
  - Including quarterly averages of global financial variables (VIX, US Treasury yields) as additional predictors did not consistently reduce RMSFEs nor change relative magnitudes of forecast errors substantively.
  - Nowcasting exercises run for aggregate EM flows and for individual EM countries where data from all sources are available (results discussed but not reproduced here).

*Source: IMF staff calculations and KP dataset as presented in the supplied content.*

### Annex IV. For the EM aggregate exercise, the monthly proxy data often perform better than

### Annex IV. For the EM aggregate exercise, the monthly proxy data often perform better than

### V. CONCLUSION
- Overview:
  - The paper provides an overview of the main capital flow data sources used in the empirical literature and among policymakers.
  - IMF balance of payments statistics remain a key data source and serve as an important benchmark for assessing the properties of other datasets.
  - Data availability has improved significantly in recent years, especially with respect to high frequency proxies for portfolio flows at the daily, weekly, and monthly frequencies.
- Usage of fund flow data:
  - Fund flow data have been used by a sizeable share of empirical studies (around 14 percent) over the past decade.
  - Fund flow data allow researchers to gain deeper insight into the drivers of rapid shifts in investor behavior.
- Conceptual and empirical differences:
  - The paper highlights conceptual and empirical differences between fund flow data and portfolio flows as measured in the IMF’s balance of payments statistics.
  - These differences imply that findings in the empirical literature are likely shaped by unique properties of fund flow data.
  - A greater focus on high frequency data that are consistent with balance of payments accounting principles seems warranted but may have been hampered by data subscription requirements and limitations of existing datasets.
- New dataset provided:
  - The authors provide a monthly portfolio flow dataset designed specifically for academic use.
  - The dataset covers 18 of the largest emerging market economies and tracks the quarterly balance of payments data of these countries closely.
  - The dataset may be particularly useful for analyzing rapid shifts in portfolio flows, such as those witnessed during the COVID-19 pandemic.
  - The dataset may help facilitate future research investigating whether the role of external drivers (relative to domestic factors) is as important as suggested by papers using fund flow data (which are by construction subject to common factors).
- Nowcasting assessment:
  - The paper assesses the predictive content of various portfolio flow proxies in a nowcasting “horse race.”
  - Results:
    - All portfolio flow proxies have significant predictive content for balance of payments portfolio flows.
    - Among predictors, IIF portfolio flow trackers generally outperform EPFR fund flow data.
    - Higher frequency proxies generally outperform lower-frequency data, especially in the first half of the current quarter.
- Policy and research implications:
  - Researchers and policymakers should understand in detail the capital flow data and proxies they rely upon.
  - Understanding data nuances will help economists utilize the best-suited dataset depending on the question asked, and will help inform how the answers are framed.
  - The growing number of capital flow datasets provide valuable opportunities for advancing economic research and informing policy decisions in real time.

### Annex I. Bilateral Capital Flows Data (“From-Whom-to-Whom” Data)
- Motivation:
  - The global financial crisis exposed the need for detailed statistics of cross-border financial linkages (FSB and IMF 2009).
  - Research showed that global banking flows contributed to the spread of cross-border funding shocks in the global financial crisis (Cetorelli and Goldberg 2011).
  - Specific recommendations by the IMF and FSB called on the G20 members to strengthen participation in the Coordinated Portfolio Investment Survey and Coordinated Direct Investment Survey (CPIS, CDIS) and in the International Banking Statistics maintained by the Bank for International Settlements.
- CPIS and CDIS coverage and timing:
  - CPIS:
    - Contains annual data from 2001 and semi-annual data from 2013.
    - The data are published with a 9-month lag.
    - In 2018, all G20 members reported semi-annual data on bilateral portfolio investment stocks through the CPIS with overall coverage at 73 reporting countries and territories.
    - Portfolio investment liability positions for participating and non-participating countries and territories are derived from portfolio investment asset positions reported by counterpart countries and territories.
  - CDIS:
    - Begins with annual data in 2009.
    - Data are released with a lag of approximately 12 months.
    - 110 countries and territories reported data at the end of 2018.
    - For non-participating countries, the CDIS derives data using all participating countries’ outward direct investment positions in country A as a proxy for country A’s inward direct investment position.1
- BIS International Banking Statistics:
  - Provide the most comprehensive bilateral stock and flow data of cross-border banking flows.
  - Quarterly data are released with a relatively short lag of 4 months or less.
  - With around 48 central banks or monetary authorities reporting data, the data capture around 95 percent of global banking flows (BIS 2019).
- National and frequency coverage:
  - Some national statistics providers and central banks publish bilateral capital flow and stock data at higher frequencies.
  - Examples:
    - The United States publishes monthly portfolio transactions data.
    - Japan, Germany, and Canada publish quarterly data.
  - Most G7 or G20 economies do not publicly provide data on bilateral portfolio flows.
  - While data availability for bilateral direct investment stocks or flows is more common at quarterly frequencies, more than half of G20 members do not publicly share this data.

1. Country A’s outward direct investment position is also derived using all participating countries’ inward direct investment positions from country A as a proxy for country A’s outward direct investment position.

*Source: wpiea2020171-print-pdf - Annex IV. For the EM aggregate exercise, the monthly proxy data often perform better than*

### 4.      Alberola, Enrique, Aitor Erce, and José Maria Serena, 2016, “International Reserves

### 4. Alberola, Enrique, Aitor Erce, and José Maria Serena, 2016, “International Reserves and Gross Capital Flows Dynamics.” — Bibliography excerpt and Annex IV

### Bibliography entries (items 4–88)
- Lists citations 4 through 88 covering empirical and theoretical studies on capital flows, cross‑border banking, portfolio flows, FDI, global liquidity, monetary policy spillovers, and related topics.
- Each entry preserves original bibliographic detail format: author(s), year, title, outlet (journal / working paper series / institution), volume/issue/pages or report identifiers, month where shown, and DOI or internet availability where provided.
- Representative topics and study types (as labelled in the source):
  - International reserves and gross capital flows dynamics (item 4).
  - World market integration and foreign direct investors (item 5).
  - Supply- and demand-side factors in global banking (item 6).
  - Portfolio flows, global risk aversion, and asset prices in emerging markets (item 7).
  - Determinants and resilience of bond flows to LDCs, 1990–1995 (item 8).
  - The shifting drivers of global liquidity (item 9).
  - Push versus pull factors in capital flows and global financial cycle (multiple entries, e.g., items 19, 29, 46).
  - Capital flow episodes: surges, stops, flight, retrenchment, and nowcasting (multiple entries, e.g., items 44, 63, 78).
- Preservation of numeric and bibliographic precision as presented (no rounding or reinterpretation).

### Annex IV — Portfolio Flow Nowcasting Results by Country (figures and notes)
- Figure A4.1 title: Root mean squared forecasting error (RMSFE) of total flows and broken down in equity and debt flows (in % of GDP).
- Figure A4.1 sources: BOPS, EPFR, IIF, IMF.
- Visual elements and exact numeric labels shown in Figure A4.1 (preserved exactly as in source):
  - Vertical axis tick labels (examples shown): 0.0%, 0.1%, 0.2%, 0.3%, 0.4%, 0.5%, 0.6%, 0.7%, 0.8%, 0.05%, 0.10%, 0.15%, 0.20%.
  - Horizontal time markers (days after begin of a quarter (w/o weekends)): 1 day, 7 days, 13 days, 19 days, 25 days, 31 days, 37 days, 43 days, 49 days, 55 days, 61 days, 67 days, 73 days, 79 days, 1 month2 months3 months4 months.
  - Legends / series labels (as shown): Daily IIF, Weekly EPFR, Monthly EPFR, Monthly tracker IIF, KP Monthly, Previous quarter (rhs).
  - Panel headings preserved: Total Flows to all EMDEs; Equity Flows to all EMDEs; Debt Flows to all EMDEs.
- Figure A4.2 title: Average root mean squared forecasting error (RMSFE) of equity and debt flows by country (in % of GDP).
- Figure A4.2 sources: BOPS, EPFR, IIF, IMF.
- Figure A4.2 sample‑period notes (preserved verbatim): 1/ Sample period starting in 2005. 2/ Sample period starting in 2010.
- Visual elements and exact numeric labels shown in Figure A4.2 (preserved exactly as in source):
  - Vertical axis tick labels (examples shown): 0.00%, 0.20%, 0.40%, 0.60%, 0.80%, 1.00%, 1.20%, 1.40%, 1.60%, 1.50%, 1.80%, 2.10%, 2.40%, 2.70%, 3.00%, 0.50%, 1.00%, 1.50%, 2.00%, 2.50%, 3.00%, 3.50%.
  - Horizontal time markers (days after beginning of a quarter (w/o weekends)): 1 day, 9 days, 17 days, 25 days, 33 days, 41 days, 49 days, 57 days, 65 days, 73 days, 81 days, 1 month2 months3 months4 months.
  - Legends / series labels (as shown): Daily IIF, Weekly EPFR, Monthly EPFR, Monthly tracker IIF, Previous quarter, KP Monthly.
  - Country panels explicitly labelled in the figure: Turkey 1/, South Africa 1/, India 2/, Thailand 2/.
- Preservation of figure captions, axes labels, series names, country labelling, and all numeric annotations exactly as they appear in the source.

*Source: wpiea2020171-print-pdf (Annex IV and bibliography excerpt as provided).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020171-print-pdf.pdf_