## On Cross-Border Crypto Flows: Measurements, Drivers, and Policy Implications (wpiea2024261-print-pdf)

## Source details

**Canonical URL:** [On Cross-Border Crypto Flows: Measurements, Drivers, and Policy Implications (wpiea2024261-print-pdf)](https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024261-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2024/english/wpiea2024261-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2024/english/wpiea2024261-print-pdf.pdf.json)

---

### Purpose and scope
- Aim: shed light on three aspects of cross-border crypto flows (CBCFs): their measurement, their drivers, and the related policy implications.
- Motivation: four observations motivating the work:
  - Financial cross-border flows bring benefits and risks; crypto assets can alter the balance via impacts on volumes and volatility.
  - CBCFs can serve as vehicles to circumvent capital flow management measures (CFMs).
  - No official data on CBCFs exist for cross-country analysis—with Brazil a notable exception—because of pseudonymity and cross-border operation of exchanges.
  - Practitioners, academics, and data providers have produced a variety of methods to identify and measure CBCFs, ranging from indirect inferences to direct inference from off- and on-chain data.

### Working definition and practical occurrence
- Working definition: a CBCF is a change in ownership of a crypto asset between a resident and a non-resident in exchange for a flow of resources.
- Practical occurrence:
  - Transactions can occur off the blockchain or on the blockchain.
  - Trades in decentralized peer-to-peer exchanges and in centralized exchanges can be on-chain or off-chain.
  - Transactions between exchanges occur on chain.

### Methodological approach
- Provide a working definition and conceptualization of CBCFs.
- Two case studies for measurement and comparison:
  - Case study 1: Brazil — compare three monthly proxies for CBCFs:
    - peer-to-peer exchanges (Method 1);
    - centralized exchanges (industry: Chainalysis and Crystal, Method 2);
    - official estimates by the Central Bank of Brazil (CBB) using the universe of contracts in the FX market (Method 3).
  - Case study 2: Global CBCFs — aggregate on-chain transaction-level flows across exchanges, excluding transactions estimated to occur within countries.
- Dynamics and drivers:
  - Begin with correlations and second moments using official Brazil data.
  - Build a structural vector autoregressive (SVAR) model to quantify the share of CBCFs’ volatility explained by external (push) factors.
- Policy angle: relate empirical findings to policy recommendations.

### Main findings — Size and measurement
- Increasing volume of CBCFs documented.
- Brazil case:
  - Steady increase in CBC outflows since late 2017.
  - By October 2023, CBC outflows reached a flow that is 25 percent of gross portfolio outflows.
  - CBC outflows are about 70 percent of the cumulative net portfolio flows since 2020. 
- Global estimates:
  - Global CBCFs reach levels of up to 22 percent of total capital flows worldwide when looking at the main 7 crypto assets traded in 2022-23.
  - This number can increase up to 35 percent if more crypto assets and exchanges are added.
- Measurement heterogeneity:
  - Substantial heterogeneity across methodologies.
  - Bilateral CBCFs can be very poorly estimated due to pseudonymity and opacity in tracing residency of market participants.
  - Official estimates following standard BoP procedures may be more reliable but are scarce (Brazil notable exception; El Salvador provides quarterly coverage only from 2021-Q3).
  - Global aggregates that sum across countries may be less imprecise by avoiding sender/receiver bilateral allocation.

### Main findings — Volatility and drivers
- Volatility:
  - CBCFs are as volatile as regular portfolio and foreign direct investment (FDI) flows when measured as deviations from trend.
  - CBCFs do not display any statistically significant correlation with regular (net) capital flows, nor with remittances.
- Comovement with external variables:
  - CBCFs strongly comove with a vector of external variables: the price of Bitcoin, the S&P 500, world industrial production, the VIX index, and a measure of the monetary policy stance in advanced economies.
  - SVAR results: about a third of CBCFs’ variance is associated with this vector of external (push) factors, after correcting for small sample bias.
  - The share of variance explained by external factors for CBCFs is 3 to 6 times the share explained for regular financial flows such as portfolio and FDI.
- Implication: CBCFs are considerably more sensitive to external factors than traditional financial flows.

### Policy implications and recommendations
- Measurement and monitoring:
  - Urgent need for better, more accurate, and comparable measurement and monitoring of CBCFs by country authorities, analogous to regular capital flow statistics.
  - Industry and academic estimates useful for broad trends but limited for precise measurement; crypto exchanges and platforms should:
    - gather residency information; and
    - report bilateral flows across countries to authorities to allow proper accounting of CBCFs.
  - Recommendation 11 of the G20’s Data Gap Initiative (DGI-3) highlighted as relevant for measuring “currency substitution” and “cross-border usage”.
- CFMs and circumvention:
  - Given the increase in size and heightened sensitivity to external factors, and potential circumvention of CFMs, there is a need to reconsider the design of CFMs in a more digitalized world where CBCFs are poorly measured.
  - Four documented circumvention cases discussed; common driver is a premium achievable through transactions that motivates selling cryptos to circumvent CFMs.
- Regulatory and international cooperation:
  - Enhance supervision and adapt regulations to institutional arrangements.
  - Travel-rule–type measures (noting limited global uptake) and international information-exchange schemes recommended to help track CBCFs and reduce loopholes.
  - Extending travel-rule–type regulations to include residency information would aid monitoring and mitigate CBCF impacts on CFMs.

### Structure of the paper
- Section 2: literature review.
- Section 3: working definition of CBCFs and measurement in practice.
- Sections 4 and 5: case studies of Brazil and global CBCFs.
- Section 6: dynamics and drivers analysis.
- Section 7: policy implications.
- Section 8: concluding remarks.
- Appendix: further technical material.

---

### Measurement approaches — conceptual comparison
- Method 1 (P2P platforms; Graph von Luckner et al., 2023; Cerutti et al., 2024):
  - Matches Bitcoin transactions within short time windows (e.g., 5-hour window) to detect cross-border transfers bypassing the financial system.
  - Limitations: may significantly underestimate CBCFs by excluding flows through the local financial system, limited to available P2P platforms, excludes delayed sales and investment holdings.
- Method 2 (industry: Chainalysis; Crystal Intelligence):
  - Attributes on-chain flows between exchanges to countries via exchange-country mapping (Chainalysis uses web traffic; Crystal uses registration country).
  - Limitations: mapping inaccuracies; omission of flows within a single exchange; Chainalysis web-traffic allocation treats all visits equally; Crystal omits exchanges registered in multiple countries or not registered.
- Method 3 (Central Bank of Brazil, CBB):
  - Uses universe of FX contracts with party identification, purpose codes, and free-text fields; classifies crypto-related FX contracts under “acquisition of goods delivered abroad”.
  - Limitations: cannot capture flows bypassing the local financial system; FX payments executed from accounts abroad not captured; relies on self-reporting for classification.

### Case Study 1 — Brazil: key statistics and cross-method comparisons
- CBB official series:
  - Monthly CBCF series from January 2016.
  - Crypto outflows upward trend since end-2017, reaching monthly flows of about $1300 million by end-2023.
  - Crypto inflows peak at $200 million at the end of the sample.
  - Crypto outflows over portfolio outflows reached nearly 25 percent in October 2023.
  - Crypto outflows were 14 percent relative to the sum of portfolio and FDI outflows in October 2023.
  - Cumulative net outflows between January 2020 and October 2023: crypto net outflows represent 82 percent relative to PF net outflows and 40 percent relative to FDI net outflows; crypto net outflows represent 14 percent relative to net exports.
  - Scaled by Brazilian GDP, CBC outflows reach close to one half of a percentage point of annual GDP by the end of the sample in 2022.
- Cross-source magnitude and correlation contrasts:
  - Chainalysis estimates much higher in magnitude: Chainalysis outflows around 200 to 300 percent of portfolio outflows in the last two years analyzed; Chainalysis inflows an order of magnitude larger than other sources.
  - Crystal’s CBC outflows not larger than 1.9 percent (relative to portfolio outflows), while CBB averaged around 25 percent.
  - Correlations:
    - Chainalysis outflows correlate with CBB outflows: 0.6.
    - Chainalysis inflows correlate with CBB inflows: -0.1 (negative, insignificant).
    - Chainalysis inflows and outflows correlation: 0.99.
    - Crystal inflows and outflows correlation: 0.71.
    - CBB inflows and outflows correlation: 0.54.
  - Cumulative flows (2020–2022) and net terms:
    - CBB net outflows represent 63.3 percent of net PF outflows.
    - Chainalysis net outflows represent 0.2 percent of net PF outflows.
    - Crystal net outflows negative and represent 2 percent of net PF outflows.
- P2P (LocalBitcoins/Paxful) vs CBB:
  - From April 2017 to February 2023, CBB recorded CBC inflows of USD228 million and outflows of USD22.276 million; Method 1 registered inflows of USD0.1 million and outflows of USD0.2 million.
  - Correlations between CBB and LocalBitcoins: outflows correlation -0.1 (not significant); inflows correlation 0.2 (negative and statistically significant).
- Main confounding factors across methods:
  - Wallet-to-exchange assignment limitations (e.g., WalletExplorer identified ~5 percent of addresses in one dataset).
  - Crystal restricted to registered exchanges and country registration; transactions by Brazilians on exchanges registered outside Brazil may be omitted.
  - Chainalysis web-traffic allocation treats web visits equally and may misallocate flows.
  - Method 1 misses flows routed through the financial system and is constrained by P2P data availability.

### Case Study 2 — Global CBCFs: data, results, and caveats
- Data providers and coverage:
  - Crystal Intelligence: links transactions to origin and destination registered exchanges; identifies 14 crypto assets; aggregation period 2014–2022; tracks 703 exchanges.
  - Chainalysis: links transactions to registered exchanges; daily data April 2022—March 2023; uses 30 crypto assets; tracks 2253 exchanges.
  - Overlap across providers: 367 exchanges.
- Crystal Intelligence results (time-series 2014–2022):
  - Clear upward trend beginning between 2016 and 2017 with peak in 2021.
  - Peak 2021 scaled values:
    - around 16.2 percent when scaled by portfolio flows;
    - around 6.6 percent when scaled by total capital flows;
    - peak in 2021 around half of a percentage point of world GDP (when scaled by world GDP).
- Chainalysis results (April 2022—March 2023):
  - G-CBCFs amount to about one third (35.1 percent) of all capital flows worldwide registered in the previous 12-month period (denominator = worldwide capital flows in 2021).
  - For the nine overlapping assets (BTC, BUSD, DAI, ETH, LTC, USDT, TUSD, USDC and XRP):
    - Chainalysis: 23.6 percent of total capital flows.
    - Crystal: 2.8 percent of total capital flows for the same 9 assets.
  - Bitcoin-only shares:
    - Chainalysis: Bitcoin alone accounts for 4.2 percent of total capital flows.
    - Crystal: Bitcoin alone accounts for 1.3 percent of total capital flows.
- Backed versus unbacked crypto:
  - Both Crystal and Chainalysis indicate the share of backed stablecoins in G-CBCFs hovers between one third to half of G-CBCFs.
  - Crystal: share of backed stablecoins begins to grow only in latter years of the analysis period.
- Limitations and caveats:
  - Estimates are references, not unbiased measures, due to:
    - Exclusion of within-country transactions potentially biasing results.
    - Wash trades inflating volumes (over 70 percent of reported volume on unregulated exchanges may consist of such trades; some exchanges exaggerate true volume by a factor of 25 to 50; Crystal’s database contains a large share of unregulated exchanges).
    - Denominator constraints: use of 2021 capital flows or GDP for 2022 scaling when 2022 data unavailable.
    - Differences in exchange registration criteria and coverage across data providers.

### Dynamics and drivers — SVAR framework and variance-share results
- SVAR setup:
  - Foreign bloc (X_t, five detrended external variables): VIX index, price of BTC, S&P 500 index, WIP index by Baumeister and Hamilton (2019), and the AEMS.
  - Domestic bloc (Y_t, four detrended Brazilian variables): CBC outflows, monthly economic activity index (IEA), monetary policy rate (MPR), and nominal exchange rate (NER).
  - Key assumptions: Brazil is a small open economy (Y_t does not impact X_t); μ_t orthogonal to ε_t; monthly data January 2018 to March 2023.
  - Small-sample upward bias corrected using a Monte Carlo procedure (N = 1000; sample size T = 63).
- Main variance-share results (small-sample bias corrected):
  - Sequential estimation (single domestic variable): 32 percent of the variance of CBC outflows explained by external variables.
  - Joint estimation (domestic bloc includes IEA, NER, MPR): near 30 percent share of variance of CBC outflows explained by external variables.
  - Comparison: the share for CBC outflows is between 6 and 3 times that explained in PF and PF + FDI outflows.
  - CBC net outflows are around 3 times more sensitive to external factors relative to total net outflows.
- Sensitivity checks:
  - Excluding the price of Bitcoin from the foreign bloc increases the share of variance associated to foreign shocks.
  - Results robust across sequential and joint analyses and alternative foreign-sample start dates where available.

### Remittances, current account, and regressions (appendix findings)
- Remittances:
  - Regression of CBC inflows and outflows on current and four lags of remittance flows (controlling for trend) yields:
    - For CBC outflows: once trends are controlled for, there is no systemic relationship between remittances and CBC outflows.
    - For CBC inflows: a contemporaneous and significantly negative relationship with remittances, suggesting part of the increase in sales of crypto assets might be substituting income from remittances.
  - Robustness: similar findings when using Chainalysis and Crystal flows.
- Current account transactions:
  - Appendix reports correlation coefficients for imports and exports of goods and CBC flows (period January 2018—October 2023); further details in Table A.11 and Figure A.5 (CBB sources).

### Circumvention scenarios documented
- Scenario 1: Exporter abroad purchases crypto with FX earnings, transfers to local exchange to sell; resident buys cryptos locally and transfers overseas to sell—local exchanges act as centralized facilitators.
- Scenario 2: Crypto mining—miners sell newly mined cryptos on local exchange to residents who transfer overseas to sell for FX.
- Additional scenarios:
  - Residents already holding crypto sell on local exchanges to move money abroad when outflow CFMs apply.
  - Exporters retain under-invoiced earnings overseas or sell to residents when foreign accounts are infeasible or surveillance is a concern.
- Policy implication: outflow CFMs cannot effectively coexist with domestic sellers of cryptocurrency (exchanges) unless measures prevent circumvention.

### Concluding remarks — summary of policy-relevant takeaways
- Measurement challenges:
  - Bilateral CBCFs may be very poorly estimated with current ad hoc methods due to pseudonymity and opacity.
  - Methodological imprecision hampers sender/receiver country identification, explaining heterogeneity across estimates.
  - Official series are scarce; Brazil is a notable exception.
  - Global aggregates that sum across countries—without assigning bilateral counterparties—may be less imprecise.
- Sensitivity to external shocks:
  - CBCFs’ higher sensitivity to external push factors implies strong responses to global financial conditions and external shocks.
  - Opacity of crypto flows complicates monitoring and policy responses at the bilateral level and may reduce effectiveness of CFMs.
- Suggested avenues:
  - Multilateral efforts to homogenize national accounting procedures to measure crypto flows across borders (e.g., G20 DGI Recommendation 11).
  - Study CBCF evolution amid increasing cross-border flows through CBDCs.
  - Further research to overcome data challenges and test hypotheses explaining Brazil’s increase in CBCF outflows.

*Source: IMF Working Paper — On Cross-Border Crypto Flows: Measurements, Drivers, and Policy Implications (wpiea2024261-print-pdf).*

### 1. Introduction ........................................................................................................

### 1. Introduction

### Purpose and scope
- Aim: shed light on three aspects of cross-border crypto flows (CBCFs): their measurement, their drivers, and the related policy implications.
- Motivation: four observations motivating the work:
  - Financial cross-border flows bring benefits and risks; crypto assets can alter the balance via impacts on volumes and volatility.
  - CBCFs can serve as vehicles to circumvent capital flow management measures (CFMs).
  - No official data on CBCFs exist for cross-country analysis—with Brazil a notable exception—because of pseudonymity and cross-border operation of exchanges.
  - Practitioners, academics, and data providers have produced a variety of methods to identify and measure CBCFs, ranging from indirect inferences to direct inference from off- and on-chain data.

### Working definition and practical occurrence
- Working definition: a CBCF is a change in ownership of a crypto asset between a resident and a non-resident in exchange for a flow of resources.
- Practical occurrence: such transactions can occur off the blockchain or on the blockchain; trades in decentralized peer-to-peer exchanges and in centralized exchanges can be on-chain or off-chain; transactions between exchanges occur on chain.

### Methodological approach
- Provide a working definition and conceptualization of CBCFs.
- Two case studies for measurement and comparison of methodologies:
  - Case study 1: Brazil — compare three monthly proxies for CBCFs:
    - peer-to-peer exchanges;
    - centralized exchanges;
    - official estimates by the Central Bank of Brazil (CBB) using the universe of contracts in the FX market.
    - Contrast these with monthly official balance of payments (BoP) flows.
  - Case study 2: Global CBCFs — aggregate on-chain transaction-level flows across exchanges, excluding transactions estimated to occur within countries.
- Dynamics and drivers:
  - Begin with correlations and second moments using official Brazil data.
  - Build a structural vector autoregressive (SVAR) model to quantify the share of CBCFs’ volatility explained by external (push) factors.
- Policy angle: relate empirical findings to policy recommendations.

### Main findings — Size and measurement
- Increasing volume of CBCFs documented.
- Brazil case:
  - Steady increase in CBC outflows since late 2017.
  - By October 2023, CBC outflows reached a flow that is 25 percent of gross portfolio outflows.
  - CBC outflows are about 70 percent of the cumulative net portfolio flows since 2020.
- Global estimates:
  - Global CBCFs reach levels of up to 22 percent of total capital flows worldwide when looking at the main 7 crypto assets traded in 2022-23.
  - This number can increase up to 35 percent if more crypto assets and exchanges are added.
- Measurement heterogeneity:
  - Substantial heterogeneity across methodologies.
  - Bilateral CBCFs can be very poorly estimated due to pseudonymity and opacity in tracing residency of market participants.
  - Official estimates following standard BoP procedures may be more reliable but are scarce (Brazil notable exception; El Salvador provides quarterly coverage only from 2021-Q3).
  - Global aggregates that sum across countries may be less imprecise by avoiding sender/receiver bilateral allocation.

### Main findings — Volatility and drivers
- Volatility:
  - CBCFs are as volatile as regular portfolio and foreign direct investment (FDI) flows when measured as deviations from trend.
  - CBCFs do not display any statistically significant correlation with regular (net) capital flows, nor with remittances.
- Comovement with external variables:
  - CBCFs strongly comove with a vector of external variables: the price of Bitcoin, the S&P 500, world industrial production, the VIX index, and a measure of the monetary policy stance in advanced economies.
  - SVAR results: about a third of CBCFs’ variance is associated with this vector of external (push) factors, after correcting for small sample bias.
  - The share of variance explained by external factors for CBCFs is 3 to 6 times the share explained for regular financial flows such as portfolio and FDI.
- Implication: CBCFs are considerably more sensitive to external factors than traditional financial flows.

### Policy implications and recommendations
- Urgent need for better, more accurate, and comparable measurement and monitoring of CBCFs by country authorities, analogous to regular capital flow statistics.
- Industry and academic estimates useful for broad trends but limited for precise measurement; crypto exchanges and platforms should:
  - gather residency information; and
  - report bilateral flows across countries to authorities to allow proper accounting of CBCFs.
- Given the increase in size and heightened sensitivity to external factors, and potential circumvention of CFMs, there is a need to reconsider the design of CFMs in a more digitalized world where CBCFs are poorly measured.
- The paper describes four cases of using crypto assets to circumvent outflows CFMs and briefly discusses how to tackle such circumvention.

### Structure of the paper
- Section 2: literature review.
- Section 3: working definition of CBCFs and measurement in practice.
- Sections 4 and 5: case studies of Brazil and global CBCFs.
- Section 6: dynamics and drivers analysis.
- Section 7: policy implications.
- Section 8: concluding remarks.
- Appendix: further technical material.

*Source: IMF Working Paper — 1. Introduction (wpiea2024261-print-pdf).*

### 2. Measurements of Cross Border Crypto Flows:

### 2. Measurements of Cross Border Crypto Flows: A Literature Review

### Direct measurements
- Direct measurements parse raw transaction data among exchanges or users from anonymous wallet addresses; these initiatives require substantial data or sophisticated algorithms and predominantly rely on platform-provided data.
- Crypto platforms are categorized as on-chain or off-chain and can operate through centralized or decentralized protocols.
- On-chain transactions: treated via the blockchain network, need to be confirmed by validators, usually traded in decentralized platforms.
- Off-chain transactions: take place outside the traditional blockchain, typically traded via peer-to-peer (P2P) platforms or centralized exchanges that provide matchmaking trading.
- Transactions on centralized exchanges often dominate in terms of volume compared to on-chain transactions (Igor and Scholar, 2022).
- Graf Von Luckner et al. (2023):
  - Employ high-frequency transaction data from two decentralized P2P Bitcoin platforms (LocalBitcoins and Paxful).
  - Methodology involves matching Bitcoin transactions within a 5-hour timeframe to identify capital movement across borders.
  - Highlights Bitcoin’s importance as a conduit for remittances and to circumvent capital controls in some emerging markets.
- Cerutti et al. (2024) study cross-border Bitcoin off-chain flows from LocalBitcoins and find off-chain cross-border flows correlated with incentives to avoid capital flow restrictions.
- Hu et al. (2022):
  - Exploits blockchain data to pinpoint cross-border flows evading China's capital controls through cryptocurrencies.
  - Matches wallets to exchanges through Wallet Explorer, which collects exchange wallet data from public sites and internal sources.
  - Finds capital flight’s volume constituted over one-quarter of the total Chinese Bitcoin exchanged and that capital flight from China via Bitcoin is positively related with Chinese economic policy uncertainty and the Bitcoin premium in Chinese RMB.
- Third-party firms (Crystal Intelligence, Chainalysis) collect on-chain crypto transactions between exchanges to capture and assess on-chain crypto flows across countries.
  - Example: Crystal Intelligence (2021) provides analysis across 694 international exchanges from 82 countries, visualizing fund flows from 2013 to 2021.
- BIS Atlas database (launched in 2023, collaboration with central banks of the Netherlands and Germany) combines off-chain exchange data with public blockchain on-chain data to measure cross-border flows (BIS, 2023).
- The authors’ contribution: comparing direct measurements of CBCFs and their underlying assumptions.

### Indirect measurements
- Indirect approaches infer CBCFs using indicators and/or arbitrage theory where direct data are limited.
- Alnasaa et al. (2022): link increased crypto usage with higher perceived corruption and more stringent capital controls using survey-based data and a general-to-specific approach.
- Chen and Sarker (2022): provide evidence consistent with Chinese residents buying Bitcoin in China and selling it for USD in foreign exchanges to evade capital controls (2014–2016 restrictions).
- Cheng and Dai (2020): find evidence of carry trade activities through Bitcoin transactions between CNY and USD platforms and a relationship between crackdowns by Chinese authorities and a weaker response of Bitcoin carry trade returns.
- Ju et al. (2016): propose a Bitcoin-implied exchange rate discount indicator using BTC China and Bitstamp data to detect capital flight from China via Bitcoin before China’s 2013 ban.
- Indirect methods frequently employ arbitrage and market indicators to infer cross-border usage of crypto when direct transaction-level identification is infeasible.

### The Route to Official Statistics
- Limitations exist in both indirect and direct methods; international organizations and central banks are undertaking initiatives to improve statistics and data availability.
- Central Bank of Brazil (CBB):
  - Measures Brazil’s CBCFs relying on FX contracts that FX intermediaries must submit when transferring/receiving funds from abroad for crypto transactions.
  - In 2017, CBB instructed FX operators to classify FX contracts related to crypto assets under the category “acquisition of goods delivered abroad”; CBB relies on a free text field to distinguish crypto assets from other goods (Central Bank of Brazil (2023)).
  - FX contracts include identification of parties (resident and nonresident), a code for purpose, and a free text field. FX contracts do not cover credit or debit card FX transactions; CBB estimates these represent a relatively small share of total CBCFs.
  - Forex transactions for acquisition of virtual assets abroad are monitored by CBB using a Risk-Based Approach tool with parameters defined in a risk matrix; buyers must settle a forex transaction with an institution authorized by the CBB, and authorized institutions must send to the CBB information about each forex transaction (including purpose and names of parties involved).
  - Limitations: cannot capture flows bypassing the local financial system; FX payments executed from accounts abroad are not captured; classification relies on self-reporting by residents (though majority of crypto transactions in Brazil occur through regulated domestic exchanges).
- IMF efforts:
  - Proposed treatment of crypto assets in BPM7 statistics; IMF is working on definitions and changes to treat crypto assets in statistics in proposed BPM7 chapters.
  - Note in source: from June 2024 the IMF, jointly with other international organizations and after a global consultation with compilers of external sector statistics, changed the methodological treatment for the balance of payment statistics regarding crypto assets in the context of the BPM7 implementation: crypto assets without an issuer would be considered as nonfinancial assets and, within this group, as non-produced nonfinancial assets; therefore, these transactions will no longer be included in the current account but in the capital account.
- Data Gaps Initiative (DGI), now in DGI-3:
  - Aims to establish a data collection framework and gather information on digital money (CBDCs, stablecoins, and other crypto assets used for payment).
  - DGI-3 is preparing Recommendation 11 on measuring “currency substitution” and “cross-border usage” in the context of digital money.
  - Emphasizes data collection on both positions and transactions, including cross-border transactions, with two key breakdowns: counterpart economy and institutional sector; and identification of relevant data providers (e.g., e-wallets, service providers, exchanges) and their data provision capabilities considering privacy and other limitations.
- Persistent data constraints limit research scope; authors’ contributions:
  - Document existing measurements through two case studies and compare them to traditional financial flows, highlighting advantages and limitations of ad hoc methods.
  - Explore dynamics and drivers of CBCFs by contrasting volatility and cyclicality of CBCFs against other financial flows.
  - Quantify the role of push vs. pull factors in accounting for CBCF dynamics through structural analysis.
  - Address normative implications by relating findings to policy recommendations.

### Definition and Measurement of CBCFs (Section 3 excerpts)
- Definition:
  - CBCFs defined as a change in ownership of a crypto asset between a resident and a nonresident, following the BoP definition of cross-border flows.
  - Residents may purchase crypto assets from nonresidents for investment (anticipating higher prices), avoiding regulation (capital controls, AML/CTF, taxes), or transactional purposes; such transactions constitute CBCFs.
  - Changes in ownership are likely to imply a flow of FX, thereby impacting the exchange rate, though not always (e.g., if the transaction is conducted abroad, or the flow constitutes a remittance, gift/donation, or swap with another crypto).
  - Accurate measurement requires information on residency of both parties—a key challenge in practice.
- Measurement mechanisms and three methods illustrated conceptually:
  - Method 1 (Graph von Luckner et al., 2023; Cerutti et al., 2024):
    - Captures cases where residents circumvent capital controls/AML/CTF by using P2P platforms: resident buys crypto from a resident (paying local currency; possibly cash) then sells to a nonresident via P2P receiving FX abroad.
    - Identification hinges on matching purchase and sale amounts and short time windows (e.g., 5-hour window).
    - Limitations:
      - May significantly underestimate CBCFs by focusing solely on cases circumventing controls and disregarding flows through the local financial system.
      - Limited to subset of P2P platforms with available Bitcoin transaction data.
      - Excludes transactions where Bitcoin is used as an investment (resident does not immediately sell) and transactions with delay beyond 5 hours.
  - Method 2 (Chainalysis; Crystal Blockchain):
    - Estimates CBCFs through flows between exchanges using on-chain blockchain transactions attributed to exchanges.
    - Crystal assigns exchanges to countries by registration country; exchanges registered in multiple countries are not assigned to any country.
    - Chainalysis distributes flows between exchanges according to countries from which the exchange’s web traffic comes.
    - Potential drawbacks:
      - Mapping exchanges to countries may be inaccurate.
      - Omits CBCFs between residents and nonresidents within a single exchange.
      - Crystal’s method neglects exchanges registered in multiple countries and exchanges not registered in any country.
      - Chainalysis’ web traffic approach treats all website entries equally, regardless of individual wealth, potentially biasing estimated flows; Chainalysis staff indicate their methodology is designed as an indicator of trends rather than as specific measures of individual country-to-country flows.
  - Method 3 (Central Bank of Brazil):
    - Uses universe of FX contracts in Brazil with identification of parties, purpose codes, and free-text fields; CBB instructed classification for crypto-related FX contracts under “acquisition of goods delivered abroad” and uses free text and participant lists to identify crypto transactions.
    - Aligns closely with BoP best practices but cannot capture flows bypassing the local financial system.
    - FX contracts exclude credit or debit card FX transactions; these are estimated by CBB to be a relatively small share of CBCFs.
    - Limitations include non-capture of FX payments executed from accounts abroad and reliance on self-reporting for classification (though many crypto transactions in Brazil occur through regulated domestic exchanges).
- Additional contextual points:
  - Method 1 also captures residents buying cryptos from non-residents as investment and avoiding capital controls, but in the document Method 1 is primarily used to refer to the CBCF measured by Graph von Luckner et al. (2023).
  - A crucial element in Graph von Luckner et al.’s identification method is that the amount purchased and sold must be identical.

*Source: wpiea2024261-print-pdf - 2. Measurements of Cross Border Crypto Flows*

### 4. Case Study 1: Brazil

### 4. Case Study 1: Brazil

### 4.1. Official Data (Method 3)
- The CBB publishes monthly CBCF data starting from January 2016.
- The CBB classifies purchases of crypto assets from nonresidents as imports of crypto assets and sales as exports; imports are referred to as “crypto outflows” (implying an FX outflow) and exports as “crypto inflows” (FX inflow).
- The CBB data analyzed include both unbacked crypto assets and stablecoins (the CBB is not yet separating backed and un-backed crypto assets).
- Key observed trends and magnitudes:
  - Crypto outflows show an upward trend since end-2017, reaching monthly flows of about $1300 million by end-2023.
  - Crypto inflows remain much lower, peaking at $200 million at the end of the sample.
  - Net crypto outflows closely mirror crypto outflows given the much smaller crypto inflows.
  - Crypto outflows over portfolio outflows reached nearly 25 percent in October 2023.
  - Crypto outflows were 14 percent relative to the sum of portfolio and FDI outflows in October 2023.
  - Cumulative net outflows between January 2020 and October 2023: crypto net outflows represent 82 percent relative to PF net outflows and 40 percent relative to FDI net outflows; crypto net outflows represent 14 percent relative to net exports.
  - Scaled by Brazilian GDP, CBC outflows show an upward trend, reaching close to one half of a percentage point of annual GDP by the end of the sample in 2022.
- Potential motives for residents’ acquisition of crypto assets discussed:
  - Portfolio diversification.
  - Convenience for maintaining FX savings given regulations preventing Brazilians from opening FX accounts within the country (despite higher risk exposure).
- Notes and caveats from the source:
  - The CBB data include both backed and un-backed crypto assets; the Appendix presents decomposition from Chainalysis and Crystal showing differing shares across backed vs unbacked cryptos.
  - From June 2024 the IMF changed BPM7 treatment such that crypto assets without an issuer would be considered nonfinancial assets.
  - IOF (tax on financial operations) likely does not explain the upward trend: IOF for FX transfers to accounts abroad is 0.38 percent since 2010 (IOF on transfers to same accountholder’s name is 1.1 percent).

### 4.2. Alternative Methodologies
- Comparison across Method 3 (CBB), Method 2 (industry: Chainalysis and Crystal), and Method 1 (P2P exchange analysis) shows large differences in magnitude and correlation.
- Major empirical comparisons:
  - Chainalysis estimates are much higher in magnitude for both outflows and inflows compared to CBB and Crystal.
    - While CBB crypto outflows reach around 25 percent of portfolio outflows toward end-sample, Chainalysis estimates are around 200 to 300 percent in the last two years analyzed.
    - Chainalysis inflows are an order of magnitude larger than other sources.
  - Crystal vs CBB:
    - Crystal’s CBC outflows have not been larger than 1.9 percent (relative to portfolio outflows), while CBB averaged around 25 percent.
    - Crystal’s CBC inflows relative to portfolio inflows peak at 3.3 percent, while CBB inflows peak at 1.3 percent.
  - Correlations across sources:
    - Chainalysis outflows correlate positively with CBB outflows (0.6) but Chainalysis inflows correlate negatively and insignificantly with CBB inflows (-0.1).
    - Crystal’s outflows and inflows display negative correlation with CBB data.
    - Chainalysis inflows and outflows show near perfect correlation (0.99).
    - Crystal inflows and outflows correlation is 0.71.
    - CBB inflows and outflows correlation is 0.54, similar to regular flows (PF, PF+FDI) correlation of 0.53.
- Cumulative flow comparisons (2020–2022):
  - Chainalysis cumulative gross flows surpass other sources by at least an order of magnitude in both inflows and outflows.
  - Crystal cumulative inflows exceed CBB inflows, but Crystal cumulative outflows are below CBB outflows.
  - Net terms (2020–2022, scaled to PF net outflows):
    - CBB net outflows represent 63.3 percent of net PF outflows.
    - Chainalysis net outflows represent 0.2 percent of net PF outflows.
    - Crystal net outflows are negative and represent 2 percent of net PF outflows.
- Method 1 (P2P LocalBitcoins/Paxful) vs CBB:
  - CBB volumes are several orders of magnitude larger than those registered in LocalBitcoins.
  - From April 2017 to February 2023, CBB recorded CBC inflows of USD228 million and outflows of USD22.276 million; Method 1 registered inflows of USD0.1 million and outflows of USD0.2 million.
  - Correlations between CBB and LocalBitcoins: outflows correlation -0.1 (not significant); inflows correlation 0.2 (negative and statistically significant).
  - Methodological differences explain discrepancies: CBB measures flows through the financial system; P2P studies measure flows bypassing the financial system.
- Methodological sources of discrepancy highlighted:
  - Wallet-to-exchange assignment limitations (e.g., WalletExplorer identified ~5 percent of addresses in one dataset; Liang et al. (2019) found WalletExplorer detects 4.3 percent of addresses in Bitcoin blockchain as of 2018).
  - Crystal only captures transactions among registered exchanges and requires exchanges to be registered in a given country; transactions by Brazilians on exchanges registered outside Brazil may be omitted from Crystal but captured by CBB via FX transfers.
  - Chainalysis allocates CBCFs across countries using web traffic; this may overestimate/underestimate volumes by assuming web visits equal transactions and treating visits equally regardless of wealth (example: Binance).
  - Method 1 misses flows that go through the financial system and is constrained by discretionary data availability from P2P platforms (e.g., LocalBitcoins and Paxful are the only P2P exchanges known to have provided external transaction data).

### 4.3. Taking stock
- Summary of key contrasts across the three methods:
  - Crypto inflows and outflows are orders of magnitude larger in Chainalysis relative to the others.
  - Crypto outflows (inflows) in Crystal are considerably below (above) those reported by the CBB.
  - The correlation between inflows and outflows in Chainalysis and Crystal is much higher than in regular flows.
  - Net flows are near zero in both Chainalysis and Crystal, while CBB net outflows are substantial.
  - Information from P2P (Method 1) is several orders of magnitude lower than all other measures and does not correlate with the official data reported by the CBB.
- Main confounding factors explaining cross-source differences:
  - Incomplete or biased wallet-to-exchange mapping.
  - Crystal’s country assignment limited to transactions among registered exchanges and omission when exchanges are registered outside Brazil.
  - Chainalysis’ web-traffic allocation approach may treat all website entries as transactions and treat entries equally irrespective of individual wealth, potentially misallocating flows (noted example: Binance).
  - Method 1’s reliance on P2P exchange data misses flows through the financial system and is limited by which P2P platforms make transaction data available.

*Source: IMF Working Paper — Chapter 4, "Case Study 1: Brazil."*

### 5. Case Study 2: Global Cross-Border Crypto

### 5. Case Study 2: Global Cross-Border Crypto Flows

### Overview
- Objective: Estimate the volume of total cross-border crypto flows around the globe (G-CBCFs) by summing transaction-level data across countries and crypto assets using Crystal Intelligence and Chainalysis.
- Premise: While bilateral CBCF measurements may be biased, summation across all cross-border transactions (excluding within-country transactions) may offer a less biased reference estimate; bias may remain due to removal of within-country transactions and wash trades.

### Data and methodology
- Crystal Intelligence:
  - Links every transaction to two registered exchanges (origin and destination) and maps exchanges to countries, allowing identification and exclusion of transactions between exchanges registered in the same country.
  - Identifies fourteen different crypto assets (see referenced Table A.1).
  - Aggregation period: years 2014 through 2022.
  - Scales cumulated CBCFs by three alternative denominators: portfolio flows, total capital flows, and GDP.
  - Financial flows and GDP data sourced from IMF’s Financial Flows Analytics (FFA), summed across 188 countries.
  - Note: 2022 shares use 2021 data in the denominator (PF flows, Total Capital flows and GDP), due to availability of data.
- Chainalysis:
  - Links every transaction to two registered exchanges and maps exchanges to countries.
  - Available transactional daily data from April 2022; analysis covers April 2022 to March 2023 (12-month period).
  - Uses 30 crypto assets available in Chainalysis (including eleven overlapping assets with Crystal).
  - G-CBCFs scaled by global total capital flows (denominator uses calendar year 2021 due to data availability).

### G-CBCFs: Results from Crystal Intelligence
- Time-series behavior:
  - Clear upward trend beginning between 2016 and 2017 with considerable volatility.
  - Peak in 2021.
- Peak 2021 scaled values:
  - Around 16.2 percent when scaled by portfolio flows.
  - Around 6.6 percent when scaled by total capital flows.
  - When scaled by world GDP (right vertical axis in Figure 9): peak in 2021 around half of a percentage point of world GDP.

### G-CBCFs: Results from Chainalysis and cross-source comparison (April 2022—March 2023)
- Chainalysis (All 30 assets, April 2022—March 2023):
  - G-CBCFs amount to about one third (35.1 percent) of all capital flows worldwide registered in the previous 12-month period (denominator = worldwide capital flows in 2021).
- Overlap comparisons:
  - Nine assets overlapped between Crystal and Chainalysis: BTC, BUSD, DAI, ETH, LTC, USDT, TUSD, USDC and XRP.
  - Chainalysis: G-CBCFs in these 9 assets account for 23.6 percent of total capital flows.
  - Crystal: G-CBCFs in the same 9 assets account for 2.8 percent of total capital flows for that period.
- Bitcoin-only shares:
  - Chainalysis: Bitcoin alone accounts for 4.2 percent of total capital flows.
  - Crystal: Bitcoin alone accounts for 1.3 percent of total capital flows.
- Interpretation: Chainalysis-based G-CBCFs are substantially larger than Crystal-based estimates even after controlling for overlapping assets.

### Discrepancies and coverage
- Exchange coverage differences:
  - Chainalysis tracks data from 2253 exchanges.
  - Crystal relies on information from only 703 exchanges.
  - Overlap across the two sources: only 367 exchanges.
- Implication: Disparity in number of exchanges covered (and differences in registering criteria across providers) creates a notable disparity in G-CBCF measurement and accounts for much of the discrepancy between Crystal and Chainalysis estimates.

### Backed versus unbacked crypto share in G-CBCFs
- Both Crystal and Chainalysis data indicate the share of backed stablecoins in G-CBCFs is considerable.
- Reported range: share hovers between one third to half of G-CBCFs.
- Crystal-specific timing: the share of backed stablecoins begins to grow only in the latter years of the analysis period (see Appendix Figure A.3 referenced).

### Limitations and caveats
- Estimates should be treated as references and not as unbiased estimates due to:
  - Potential bias from exclusion of within-country transactions if those excluded transactions are systematically biased.
  - Possibility of wash trades inflating measured volumes (wash trading noted as a prominent issue; references to empirical findings: over 70 percent of reported volume on unregulated crypto exchanges may consist of such trades; some exchanges exaggerate true volume by a factor of 25 to 50; Crystal’s database contains a large share of unregulated exchanges).
  - Denominator constraints: use of 2021 capital flows or GDP data for 2022 scaling where 2022 data unavailable.
  - Differences in exchange registration criteria and coverage across data providers.

*Source: IMF staff summary of "5. Case Study 2: Global Cross-Border Crypto Flows" from the supplied PDF content.*

### Appendix for the sake of space (see Tables A.9 and A.10). The systematic correlation between

### wpiea2024261-print-pdf - Appendix for the sake of space (see Tables A.9 and A.10). The systematic correlation between

### Remittances and CBCFs
- Regression of CBC inflows and outflows (separately) on current and four lags of remittance flows, controlling for a trend.
- Findings:
  - For CBC outflows: once trends in remittances and CBCFs are controlled for, there is no systemic relationship between remittances and CBC outflows.
  - For CBC inflows: a contemporaneous and significantly negative relationship with remittances is found, suggesting part of the increase in sales of crypto assets might be substituting income from remittances.
  - The significantly large and upward trend in net crypto purchases is unrelated to remittances.
- Robustness:
  - Regression analysis using flows from Chainalysis and Crystal is largely consistent with these findings.
- Data notes:
  - Imports of crypto assets are referred to as “crypto outflows” (implying an outflow of FX), and exports of crypto as “crypto inflows” (an FX inflow).
  - LocalBitcoins data from Cerutti et.al (2024). Cryptocurrencies included in Crystal and Chainalysis series in Table A.1. Remittances proxied as personal transfers—credits. Source: Central Bank of Brazil, Chainalysis, Crystal Intelligence, Cerutti et.al (2024) and authors’ calculations.
- Appendix pointer:
  - Table A.11 in the Appendix presents further analysis of the positive correlation between CBCFs and import and export of goods.

### SVAR framework and identification
- Objective: quantify share of CBCFs variance explained by external (push) versus domestic (pull) factors for Brazil.
- Model structure:
  - Foreign bloc (vector X_t, five detrended external variables): VIX index, price of BTC, S&P 500 index, WIP index by Baumeister and Hamilton (2019), and the AEMS.
  - X_t follows: X_t = A X_{t−1} + μ_t, where μ_t is i.i.d. mean-zero with variance–covariance Σ_μ; μ_t interpreted as reduced-form global shocks.
  - Domestic bloc (vector Y_t, four detrended Brazilian variables): CBC outflows, monthly economic activity index (IEA), monetary policy rate, and nominal exchange rate.
  - Y_t follows: Y_t = B X_t + C Y_{t−1} + D X_{t−1} + ε_t, where ε_t are reduced-form residuals.
- Key assumptions:
  - Brazil is a small open economy: Y_t does not impact X_t.
  - μ_t is orthogonal to ε_t, so μ_t captures global shocks that affect Brazil via X_t.
  - The analysis estimates the share of variance in Y_t (CBC outflows) attributable to μ_t mediated by X_t; it does not identify structural shocks or derive structural impulse responses.
- Estimation details:
  - Monthly data over the period January 2018 to March 2023.
  - Correct for small-sample upward bias in variance share using Monte Carlo methods; Appendix contains details.
  - Also consider sequential estimation where each variable in Y_t is estimated as the sole domestic variable.
  - Appendix Tables A.3 and A.4 present results with the foreign bloc estimated from as early as 2015; results qualitatively similar.

### Variance share results
- Main quantitative results (small-sample bias corrected):
  - Sequential estimation (single domestic variable):
    - 32 percent of the variance of CBC outflows is explained by external variables (column 1, upper panel).
  - Joint estimation (domestic bloc includes IEA, nominal exchange rate, monetary policy rate):
    - Near 30 percent share of variance of CBC outflows explained by external variables (column 1, lower panel).
  - Comparison to other flows:
    - The share for CBC outflows is between 6 and 3 times that explained in PF and PF + FDI outflows (columns 2 and 3).
  - Robustness:
    - Results robust across sequential and joint analyses.
    - Table 3: CBC net outflows are around 3 times more sensitive to external factors relative to total net outflows.
- Sensitivity:
  - Appendix sensitivity where price of Bitcoin is excluded from the foreign bloc (Tables A.5 and A.6): exclusion increases the share of variance associated to foreign shocks.
  - Interpretation: consistent with literature documenting hedging properties of some crypto assets against stock market volatility.
- Data/sample limitation:
  - Lack of CBCF data prior to 2018 prevents longer time series; foreign bloc could be estimated on longer samples where available.

### Policy implications
- Measurement and monitoring:
  - Need for better, more accurate and comprehensive measurement and monitoring of CBCFs by country authorities, in line with monitoring of other balance-of-payments flows.
  - CBCFs’ growing volume implies potential for increased role as transmitters and propagators of external shocks.
  - Industry and academic estimates provide broad trend indications but can be limited and vary considerably; official statistics comparable across countries are needed.
  - Pioneering methods such as the one developed by the Central Bank of Brazil (CBB) offer useful benchmarks to be explored and enhanced by others.
  - Recommendation 11 of the G20’s Data Gap Initiative seen as promising for closing data gaps; likely entails exchanges and digital platforms systematically gathering residency information and reporting bilateral flows to statistical agencies.
- Capital flow management measures (CFMs) in a digitalized world:
  - Need to rethink optimal design of CFMs given CBCFs and potential circumvention of CFMs via crypto markets.
  - Evidence that CBCF volumes are increasing and may be more sensitive to external factors outside policymakers’ control supports rekindling and adapting CFMs.
- Circumvention scenarios (as documented by Graf von Luckner et al., 2024):
  - Scenario 1: Exporter abroad purchases crypto with FX earnings, transfers to local exchange to sell; resident buys cryptos locally and transfers overseas to sell—local exchanges act as centralized facilitators.
  - Scenario 2: Crypto mining—miners sell newly mined cryptos on local exchange to residents who transfer overseas to sell for FX.
  - Additional scenarios:
    - Residents already holding crypto sell on local exchanges to move money abroad when outflow CFMs apply.
    - Exporters retain under-invoiced earnings overseas or sell to residents when foreign accounts are infeasible or surveillance is a concern.
  - Common driver: premium achievable through the transaction motivates selling cryptos to circumvent CFMs.
  - Implication: outflow CFMs cannot effectively coexist with domestic sellers of cryptocurrency (exchanges) unless measures prevent circumvention.
- Regulatory and international cooperation measures:
  - Enhancing supervision and regulation is challenging and must adapt to institutional arrangements.
  - Case study: Japan—2017 law introduced registration for service providers conducting virtual asset transfers to/from customers in Japan; 2022 regulations aimed at preventing use of crypto to circumvent sanctions and included stablecoins and crypto assets in capital transactions restrictions; introduced “travel rules” consistent with FATF Recommendation 15.
  - Travel rules require notifying counterparty exchange of transaction information (names of customer and recipient), akin to banks.
  - Limited global uptake: number of jurisdictions implementing the travel rule remains limited (FSA, 2024), leaving loopholes in jurisdictions without travel rules.
  - Need for international information-exchange schemes to track CBCFs and prevent jurisdictions becoming loopholes.
  - Extending travel-rule–type regulations to include residency information would aid countries in monitoring CBCFs and mitigating their impact on outflow CFMs.

*Source: Appendix content from IMF Working Paper "On Cross-Border Crypto Flows: Measurements, Drivers, and Policy Implications" (content unit: wpiea2024261-print-pdf).*

### 8. Concluding Remarks

### 8. Concluding Remarks

### Key findings on measurement and scale
- In the Brazil case study, there is a steady increase in cross-border crypto outflows since late 2017, comparable to 25 percent of gross portfolio outflows by the end of 2023.
- Global cross-border crypto flows (G-CBCFs) display a steady increase, reaching levels between 3 percent and 22 percent of total capital flows, increasing up to 35 percent if one adds more crypto assets as alternative sources allow.
- CBCFs are as volatile as regular portfolio and FDI flows.
- CBCFs do not display any statistically significant correlation with regular (net) capital flows, nor with remittances.
- CBCFs comove with a vector of external variables.
- A SVAR model associates about a third of CBCFs’ variance to this vector of external (push) factors.
- The share of variance of CBCFs explained by the external vector is 3 to 6 times the share explained by the same vector for regular financial flows (such as portfolio and FDI), indicating that CBCFs are considerably more sensitive to external factors than financial flows.

### Measurement challenges and data quality
- Bilateral CBCFs may be very poorly estimated using current ad hoc methods because pseudonymity and opacity in crypto markets make it difficult to trace the residency of market participants.
- Methodological imprecision hampers accurate identification of sender and receiver countries, explaining substantial heterogeneity across CBCF estimates from different methodologies.
- Official estimates can be more reliable but are scarce; Brazil is a notable exception with official series available.
- Global measures that sum CBCFs across countries—without assigning sender/receiver country—may be less imprecise.

### Policy implications and sensitivity to external shocks
- CBCFs’ higher sensitivity to external push factors (relative to portfolio and FDI flows) implies that crypto-related cross-border flows can respond strongly to global financial conditions and external shocks.
- The opacity of crypto flows complicates monitoring and policy responses at the bilateral level, potentially limiting the effectiveness of country-level capital flow management and balance of payments surveillance.

### Suggested avenues for future research and data improvements
- Multilateral efforts to homogenize national accounting procedures to measure crypto flows across borders (for example, efforts referenced in G20’s Data Gap Initiatives Recommendation 11) will improve systematic analysis once new data become available; it will be of interest to contrast current findings against such improved data.
- Study how CBCFs evolve amid increasing cross-border flows through central bank digital currencies (CBDCs) as more central banks digitize their currencies and allow digital capital flows across borders.
- Further research should explore ways to overcome data challenges and test the validity of hypotheses explaining the considerable increase in CBCF outflows in Brazil.

*Source: IMF Working Paper — 8. Concluding Remarks*

### section includes also IEA, NER and MPR. Whenever the bias is larger than the non-corrected share, the corrected share sh

### wpiea2024261-print-pdf - section includes also IEA, NER and MPR. Whenever the bias is larger than the non-corrected share, the corrected share sh

### Variance share analysis (outflows and net outflows)
- Tables A.7 and A.8 present variance share decompositions for:
  - CBC outflows (column 1), portfolio outflows (column 2), and portfolio and FDI outflows (column 3) (Table A.7).
  - Net outflows / the variable in the first row explained by global variables (Table A.8).
- Sample periods:
  - Foreign bloc: January 2015 to March 2023.
  - Domestic bloc: January 2018 to March 2023.
- Specification notes:
  - The sequential section includes only the first-row variable in the domestic bloc.
  - The joint section includes IEA (monthly index of economic activity), NER (nominal exchange rate), and MPR (monetary policy rate).
  - Whenever the bias is larger than the non-corrected share, the corrected share shown in the table is 0.
  - All variables are detrended.
  - Imports of crypto assets are referred to as “crypto outflows”, as they imply an outflow of FX; exports of crypto assets are referred to as “crypto inflows”, as an FX inflow occurs.
- Data sources cited: CBB, FRED, Baumeister and Hamilton (2019) and authors’ calculations.

### Remittances — regression framework and specifications (A.5)
- Purpose: assess systematic relationship between remittances and CBC Outflows / CBC Inflows using formal regression analysis with lags.
- CBC Outflows regression specification:
  - CBCOut_t = α + γ t + β0 Rem_t + β1 Rem_{t−1} + β2 Rem_{t−2} + β3 Rem_{t−3} + β4 Rem_{t−4} + ε_t
  - Four lags of remittances are included.
- CBC Inflows regression specification:
  - CBCIn_t = α + γ t + β0 Rem_t + β1 Rem_{t−1} + β2 Rem_{t−2} + β3 Rem_{t−3} + β4 Rem_{t−4} + δ D[Aug−23] + ε_t
  - Includes same four lags plus a dummy for August 2023 due to an unusual increase observed in that year.
- Notes on tables and samples (Tables A.9 and A.10):
  - Standard errors in parenthesis.
  - Significance notation: one star (*) = 10 percent, two stars (**) = 5 percent, three stars (***) = 1 percent.
  - Dependent variable is CBC Outflows in Table A.9 and CBC Inflows in Table A.10.
  - Different sources used across specifications. Sample periods by source:
    - CBB: May 2016—October 2023.
    - Crystal: May 2016—March 2023.
    - Chainalysis: May 2020—December 2022.
    - LocalBitcoins: August 2017—February 2023.
  - Imports of crypto assets referred to as “crypto outflows”; exports referred to as “crypto inflows”.
- Figure A.4:
  - Depicts remittances and CBC inflows from other methodologies.
  - LocalBitcoins data from Cerutti et.al (2024).
  - Cryptocurrencies included in Crystal and Chainalysis series listed in Table A.1 (not reproduced here).
  - Data on remittances into Brazil proxied as personal transfers—credits.
  - Sources: Central Bank of Brazil, Chainalysis, Crystal Intelligence, Cerutti et.al (2024) and authors’ calculations.

### Current account analysis (A.6)
- Objective: explore relationship between current account transactions (imports and exports of goods) and CBC Outflows / CBC Inflows (from Central Bank of Brazil).
- Table A.11:
  - Reports correlation coefficients for current account transactions.
  - Period of analysis: January 2018 through October 2023.
  - Blue colors denote variables related to crypto flows.
  - Imports of crypto assets labeled “crypto outflows”; exports labeled “crypto inflows”.
  - Source: Central Bank of Brazil.
- Figure A.5:
  - Plots current account flows (exports and imports of goods) and CBCFs.
  - Source: Central Bank of Brazil.

### SVAR details and small-sample bias correction (A.7)
- Model setup (foreign and domestic blocs):
  - X_t = A X_{t−1} + μ_t
  - Y_t = B X_t + C Y_{t−1} + D X_{t−1} + ε_t
  - Definitions:
    - F = [A ∅; D A + B C]
    - G = [I ∅; D I]
    - E(μ_t ε_t) = Σ = [Σ_μ ∅; ∅ Σ_ε]
  - Combined form:
    - [X_t; Y_t] = F [X_{t−1}; Y_{t−1}] + G [μ_t; ε_t]
- Small-sample bias correction Monte Carlo procedure (follows Fernández et al. (2017)):
  1. Let F̂, Ĝ, and Σ̂ denote estimates of F, G, and Σ from actual data. Let σ̂ denote the associated estimate of the share of the variance of Y_t explained by μ_t. Use F̂, Ĝ, and Σ̂ to generate artificial time series for Y_t and X_t of a desired length from the SVAR model (the authors use 250 months).
  2. Let T denote the sample size, which is 63 in the sample between January 2018 and March 2023. Use the last T observations of the artificial time series (of length 250 months) to re-estimate both the foreign and domestic blocs of the SVAR (re-estimate A, B, C, D, Σ_μ and Σ_ε).
  3. Steps 1 and 2 yield an estimate of F, G, and Σ from the simulated data. Use this estimate to compute the share of the variance of Y_t explained by μ_t shocks, denoted by σ.
  4. Repeat steps 1–3 N times. The authors set N = 1000. Then compute averages of the resulting estimate of σ and denote it by σ̅.
  5. Define the small-sample bias as σ̅ − σ̂. The corrected estimate of the share of the variance of Y_t explained by μ_t is given by 2 σ̂ − σ̅.

### Definitions, data conventions, and sources
- Crypto flow terminology:
  - “Crypto outflows” = imports of crypto assets (imply an outflow of FX).
  - “Crypto inflows” = exports of crypto assets (imply an inflow of FX).
- Detrending: All variables in the reported analyses are detrended.
- Data and sources referenced across the section: CBB (Central Bank of Brazil), FRED, Baumeister and Hamilton (2019), Chainalysis, Crystal Intelligence, Cerutti et.al (2024), and authors’ calculations.

*Italic: Source — IMF Working Paper section from "On Cross-Border Crypto Flows: Measurements, Drivers, and Policy Implications" (excerpt of PDF content provided).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024261-print-pdf.pdf_