What the Data Show—and What They Do Not
The question now is not merely how a payment could be delayed after reaching the beneficiary bank, but what the available data show about whether financial-account restrictions are associated with that delay. An IMF working paper studies that relationship across countries by combining Swift payment records with an index derived from the IMF’s data on exchange arrangements and exchange restrictions. The useful outcome for the reader is a disciplined interpretation: understand the estimate in its proper unit, see where the pattern is stronger or weaker, and know why a statistically clear association is not yet a causal account of an individual payment.
That distinction matters because the last-mile process is easy to describe more confidently than it can currently be measured. A bank may have a reason to hold a payment, but a country-level average cannot tell us which check held a particular transfer. The study brings the hidden interval into quantitative view while also showing how much remains outside the measurement.
Measuring the interval and the restriction index
The study uses two linked but different kinds of information. The first comes from Swift transaction-level data. Swift’s Global Payments Innovation system uses a Unique End-to-End Transaction Reference to track a payment through its stages. For the study, those observations are anonymized and aggregated to the country-year level. The outcome is called Beneficiary Elapsed Time: the time from when the beneficiary bank receives a payment to when the end-customer’s account is credited.
That definition is narrower than the total time a payment takes from sender to recipient. It excludes the time before the originating bank submits the instruction and the time while the payment is travelling through the international network. It also is not a measure of how long a bank spends on one named compliance check. It is an elapsed-time measure for completed payments, calculated over transactions received in a country during a year.
The second data source is the IMF’s Annual Report on Exchange Arrangements and Exchange Restrictions, or AREAER. From that material, the researchers construct the Financial Account Restriction Index, known as FARI. The inflow version is the relevant measure because the beneficiary leg concerns money entering the recipient’s country. FARI ranges from 0 to 1 and averages binary indicators across 29 financial-account components. A higher value means that more of the covered types of restriction are present; it does not mean that one particular rule is necessarily more severe.
That last distinction is fundamental. An index built from whether restrictions exist can compare the breadth of a country’s framework, but it cannot by itself say how intensely a rule is enforced, how often an exception is granted, how much discretion a bank has, or how efficiently the bank processes the required evidence. Two countries could have similar index values while applying their rules very differently. Conversely, a country could tighten an existing restriction without the binary indicator changing. FARI is therefore a proxy for the regulatory environment, not a direct meter of the work performed on every incoming payment.
The study covers 176 countries over 2020–2022 and contains 528 country-year observations. Its summary statistics provide a sense of the outcome’s scale. Average Beneficiary Elapsed Time is 30.72 hours, the median is 27.84 hours, and the observed values range from 2.50 to 120 hours. Those figures describe country-year aggregates, not a typical customer’s guaranteed wait. They also show why a few very long waits can matter to an average and why the median is a useful complementary measure.
The transaction sample is restricted to completed single-customer credit transfers. The researchers keep payments whose beneficiary account is confirmed as credited and remove future-value-date payments, because those transactions may have reached the beneficiary bank while the funds are deliberately unavailable until a later value date. Domestic payments are excluded, since domestic transfers are not the direct object of a cross-border capital-control question. Specialized interbank, securities, trade-finance, and cash-management transfers are also excluded.
These choices improve the match between the outcome and the question, but they narrow the population. The data cannot identify which payments were actually subject to a capital-control check. Retail payments and remittances remain in the sample even though they may be less affected by financial-account restrictions than some business or investment-related transfers. That mixture can make the measured relationship less direct. Countries subject to Swift access restrictions during the period are excluded as well. A result for this selected set of completed customer transfers should not automatically be read as a result for every cross-border payment, every corridor, or every type of recipient.
The baseline statistical exercise is a pooled panel regression. It relates average Beneficiary Elapsed Time to the inflow FARI, first on its own and then alongside a parsimonious set of macrofinancial, financial-access, and digital-readiness controls. The controls matter because the same underlying conditions could affect both a country’s use of financial-account restrictions and the speed of its payment systems. A country with weaker infrastructure, different trade patterns, or less developed financial access might both process payments more slowly and maintain a different regulatory framework.
The univariate relationship is positive. The reported FARI coefficient is 34.87 hours per index unit. The standard deviation of the inflow FARI in the sample is 0.22, so a one-standard-deviation increase corresponds to nearly eight additional hours in the unadjusted association. When the control variables are added, the FARI coefficient falls to 19.54 hours per index unit. Applied to the same 0.22 standard deviation, that is slightly more than four hours. The paper’s abstract therefore describes a statistically significant association of roughly four to eight hours at the beneficiary leg, depending on the specification.
The translation is more important than the headline. “A one-standard-deviation increase” is a comparison within the study’s distribution, not a policy instruction to add or remove a fixed number of rules. “Associated with” is also doing essential work. The estimate says that country-years with higher inflow FARI tend to have longer measured beneficiary elapsed times after accounting for the variables in the stated model. It does not say that changing one rule will reduce every payment’s wait by four hours, or that capital controls caused the full difference between two countries.
A simplified view of the study is:
The diagram separates the construction of the variables from the interpretation of the result. It also places the non-causality boundary next to the estimate rather than in a footnote. The study measures a relationship between a proxy for the extent of restrictions and an aggregated outcome. It does not observe a treatment being assigned to otherwise identical payments, nor does it observe the full implementation path between a rule and a customer credit.
Interpreting heterogeneity without overclaiming
The additional results are useful because they show that the relationship is not uniform. They are also easy to misuse. A difference between groups can reveal where the association is stronger, but it does not automatically reveal why it is stronger.
The income-group comparison separates Advanced Economies from Emerging and Developing Economies. For Advanced Economies, the estimated FARI effect is not statistically distinguishable from zero in the interaction model. For EMDEs, the interaction is positive but only marginally significant. The paper translates the difference as approximately nine hours of additional delay for an EMDE relative to an advanced economy for a one-standard-deviation increase in FARI. The same model shows a substantial baseline gap: after accounting for capital controls, payments in EMDEs are about 13 hours slower than in advanced economies.
Those are two different statements. The baseline gap says that the groups differ in overall measured payment speed. The interaction concerns how the association with FARI changes across groups. A country can have slow payments for many reasons even if the estimated FARI slope is not especially large. Treating the whole baseline gap as the effect of capital controls would confuse an intercept difference with a slope difference and would exceed what the model identifies.
Digital readiness produces a similar lesson. The study finds no statistically significant difference in the FARI slope between countries with higher and lower digital readiness. That does not mean digital capability has no relationship with payment speed. In the reported model, countries with lower digital readiness have payments about 22 hours slower overall than countries with higher readiness, after accounting for capital controls. The cautious interpretation is that digital readiness appears relevant to the level of speed in this comparison, while the study does not establish that it changes the relationship between FARI and elapsed time.
Regional results are also heterogeneous. Europe and Central Asia serve as the reference region, with a reported FARI association of about 13 hours for a one-standard-deviation increase. Several other regions have faster payments relative to that reference as FARI rises, while the FARI effect in South Asia is not statistically different from zero in the stated specification. The regional intercepts differ as well; some regions have much slower baseline payments apart from the estimated FARI relationship. These results tell us that geography is carrying information about different systems and contexts. They do not identify a particular national rule, bank practice, or infrastructure deficit as the explanation.
The authors test the result in ways that address some, but not all, concerns. They repeat the baseline analysis with median rather than average Beneficiary Elapsed Time, which reduces the influence of outliers. They also examine whether a squared FARI term is needed, rather than assuming that the relationship must be linear across the full index. The reported signs and broad significance pattern support the main association, and the non-linearity exercise does not show that the result is simply an artifact of the linear specification.
Robustness is not the same as identification. Replacing the mean with the median tests sensitivity to the distribution of delays; it does not reveal which rule produced them. Adding a squared term tests one shape of the relationship; it does not account for every omitted institutional factor. Adding macrofinancial, access, and digital controls attenuates the coefficient, which is informative because it shows that the variables overlap, but the controls are still proxies. An unmeasured feature of governance, infrastructure, payment composition, or bank practice could influence both FARI and elapsed time.
The study’s own limitations make the boundary clearer. FARI captures the extensive margin—the presence or absence of covered restrictions—not the intensive margin of how restrictive a rule is. It does not measure enforcement or implementation efficiency. Current-account restrictions, including those related to imports and invisible transactions, are outside the index even though they may matter to some payments. The Swift data cannot tell whether a particular observed transfer was held for a capital-control purpose, an AML/CFT review, sanctions screening, a tax or reporting requirement, a fraud check, an operating window, or some combination.
The time period matters too. The sample covers 2020–2022, the COVID period, when lockdowns and changing policy environments may have affected payment speed and capital flows. Large cross-country variation makes the association informative, but it does not remove the possibility that unusual conditions shaped the result. A longer series would help distinguish slow-moving differences between countries from changes within a country when its rules, systems, or operating conditions shift.
What evidence would change the conclusion? The strongest next step would join transaction-level timing with detailed information about the applicable AREAER components and their intensity. If researchers could observe when a specific control changed and compare affected and plausibly unaffected payments before and after that change, they could use more credible causal designs. They would also need to separate payment types, identify the action that created the wait, and observe implementation quality rather than only the existence of a rule.
For now, the defensible conclusion is substantial but bounded. Countries with more extensive measured financial-account restrictions tend to have longer beneficiary-leg elapsed times in the study’s selected sample. The association survives a set of controls and appears stronger in some income groups and regions, while digital readiness affects baseline speed without a statistically different FARI slope. The evidence is consistent with the process mechanism in which local requirements and administrative work can delay crediting. It does not prove that capital controls alone cause the delay, that every country responds in the same way, or that removing a named restriction would produce a predictable customer-level gain.
That calibrated result points beyond statistics. If the relationship reflects a real last-mile friction, the next question is why it persists across systems and why domestic payment networks have often overcome similar problems more effectively. Answering that requires looking at the correspondent institutions, network effects, and interoperability arrangements that determine whether information and obligations can travel with the payment as easily as the message does.