What Changed After Reserves Were Released
The subject is still Colombia’s CUD, but the focus here is narrower: what happened to settlement timing around the reserve-requirement reductions of April 2020 and September 2024? The useful judgment is not whether the reforms were “good” or “bad” in the abstract. It is whether payment value moved earlier, later, or in different directions at different points in the day—and how confidently the data allow us to connect that movement to reserve policy.
Two reductions, two institutional contexts
The BIS study divides its CUD sample into three regimes. The first runs from January 1, 2018, through April 21, 2020. During that period, the reserve requirement was 11% for demand deposits and 4.5% for longer-term deposits. The second begins on April 22, 2020, when those ratios fell to 8% and 3.5%, respectively. The third begins on September 11, 2024, when they fell again, to 7% for demand deposits and 2.5% for longer-term deposits. The sample extends through April 30, 2025.
The dates are more than labels on a chart. They identify different policy environments. The April 2020 reduction was part of the liquidity response to the COVID-19 pandemic. It did not arrive as an isolated change in an otherwise stable market. Other measures were being used to support financial conditions, so any change in CUD timing during that period could reflect a package of interventions, an abrupt change in payment demand, or the behavior of institutions responding to extraordinary uncertainty as well as the reserve ratio itself.
The September 2024 reduction had a different stated rationale. It was associated with the Central Bank of Colombia’s 2022–2025 Strategic Plan and was intended to support financial deepening and reduce intermediation costs. The paper places that decision in a longer institutional transition. As liquidity-risk regulation strengthened and monetary policy moved toward an inflation-targeting framework in which the policy interest rate was the primary instrument, reserve requirements became less central to some of their earlier roles. The 2024 adjustment released an estimated COP 6 trillion, approximately USD 1.5 billion, in bank reserves.
That amount should be read carefully. It describes funds released by a lower regulatory requirement; it does not tell us how much remained in CUD accounts at every moment, how much was placed in another market, or how much was used to settle payments. A reserve reduction changes the constraints and opportunities facing institutions. The settlement outcome depends on what they do with that additional discretion.
The policy background helps explain why the same instrument can have different meanings at different times. Reserve requirements have traditionally served more than one purpose: they can provide a liquidity buffer, influence the growth of deposits and credit, and shape how financial institutions manage their balances. But those roles do not remain equally important as the financial system changes. Stronger liquidity-risk rules can take over some of the safety function, while an inflation-targeting framework can place more weight on the policy interest rate for monetary control. In that setting, lowering the reserve ratio may be less about changing the central bank’s main policy stance than about reducing a structural cost of intermediation. The resulting funds still enter institutions that must decide how to allocate them during the day.
This is also why the word “released” should not be read as “immediately available for every payment.” A bank’s reserve position can improve without its payment schedule becoming mechanically earlier. The funds may support a different asset, satisfy a later liquidity need, or reduce the urgency of obtaining another form of funding. The policy fact is the change in the required ratio; the behavioral question is how that change alters the bank’s sequence of choices.
The two episodes also differ in their relationship to payment infrastructure. The new DCV, which handles custody, clearing, and settlement functions for securities transactions, began operating on April 1, 2024. It therefore falls inside the third reserve regime but predates the September reduction by several months. The latest period cannot be interpreted as a pure test of the reserve change: a major change in securities settlement was already altering the obligations that passed through CUD.
This overlap does not make comparison useless. It changes the claim that comparison can support. The data can show that the timing distribution changed across identifiable regimes and that the change coincided with policy and infrastructure transitions. They can also show whether particular parts of the distribution moved consistently after accounting for other measured conditions. They cannot, without stronger identification, assign every observed minute of delay to the reserve ratio alone.
There are effectively three clocks in the latest comparison. The first is the policy clock: the reserve ratio falls in September 2024. The second is the infrastructure clock: the new DCV begins operating in April 2024 and can change the timing of securities-related obligations. The third is the adaptation clock: institutions may need time to learn a new operating framework and revise their liquidity-management routines. These clocks overlap in the data even though they describe different forces. Treating them as one event would make the result easier to narrate but harder to understand.
The same caution applies to the pandemic period. The April 2020 reduction was introduced in an environment of rapidly changing economic conditions and extraordinary central-bank support. Payment values, institution behavior, and funding markets could all have shifted at once. A before-and-after chart can locate the break in the calendar, but it cannot by itself identify which channel produced the break. The study is most informative when its calendar comparisons are combined with the percentile pattern and the other variables used in the regressions.
The result is a redistribution, not a single speed
The most important evidence appears when the day is divided by cumulative payment value. The study estimates the time at which the 10th, 20th, and subsequent deciles up to the 90th percentile of daily payment value had settled. This approach asks where successive portions of the day’s money flow arrived. It prevents a large late payment from being hidden inside an average and distinguishes an early-day change from a change in the tail.
The comparison between the first and second regimes already contradicts a simple story. After the April 2020 reduction, the 10th percentile settled 32 minutes earlier than before, and the 40th percentile about five minutes earlier. That suggests that part of the day’s value could be discharged more quickly. But the 70th percentile settled 24 minutes later, and the 90th percentile 21 minutes later. The lower and upper portions of the distribution moved in opposite directions.
This is why the headline cannot be “reserve cuts delayed payments” without qualification. Some early value settled sooner. The later portion of the day’s value settled later. Both statements can be true because the policy affected the sequence in which obligations were discharged rather than shifting every transaction by the same amount. A bank can use additional flexibility to meet smaller or more immediate obligations early while preserving liquidity for a larger obligation, a funding trade, or an incoming-payment dependency later.
The latest period shows an even sharper separation between the lower and upper parts of the distribution. Relative to the first regime, the 10th percentile settled three hours and 19 minutes earlier. Relative to the second regime, it settled two hours and 47 minutes earlier. At the 40th percentile, settlement was 16 minutes earlier than in the first regime and 12 minutes earlier than in the second. These are substantial changes at the beginning of the cumulative-value curve.
At the same time, the upper percentiles moved toward later settlement. At the start of the second CUD pricing window, 17:00, the cumulative proportion of settled value in the latest period was below the earlier level, which was closer to the 90th percentile. Put differently, by 17:00 the latest-period system had completed less of the day’s total value, leaving a larger share for the evening window. The shift is concentrated in the high-value tail, not spread evenly across all payments.
The regression analysis gives the descriptive pattern a second test. For each of nine cumulative value percentiles—from the 10th through the 90th—the authors estimate settlement time in seconds from the opening of CUD. They include indicators for the 2020 and 2024 reserve changes and controls for the daily value and number of payments, active participants, the ratio of daily repos to available reserves, the difference between the interbank rate and the policy rate, payment concentration, the new DCV, month-end effects, and a time trend. The standard errors are adjusted for serial correlation and changing variance with the Newey–West method.
This design separates several questions that an aggregate chart would mix together. The daily payment-value control asks whether a busier system is simply slower because more money must move. The transaction-count and active-member controls distinguish scale and participation from the policy regime. Repo use and the interbank-policy-rate differential provide different views of funding pressure and the opportunity cost of holding reserves. The concentration measure asks whether a more concentrated set of participants can coordinate more quickly. The DCV and month-end variables acknowledge that the settlement calendar has its own institutional and seasonal forces.
The important point is not the technical label of the method but the question it permits. If the reserve indicators still have a positive timing association after these measured conditions are taken into account, the evidence is less likely to be a simple reflection of more payments or more participants on a given day. The reported coefficients indicate that the reserve reductions increased the time required for settlement particularly from the 40th through the 90th percentiles. The value-weighted average settlement-time measure also points to later settlement, especially after the 2024 reform.
Yet controls are not a time machine. A regression can account for observed payment value, transaction count, participants, repo conditions, concentration, calendar timing, and the DCV indicator without proving that all relevant confounders have been captured. The 2020 policy change coincided with pandemic measures that are difficult to summarize in one variable. The DCV indicator marks a structural transition, but it may not capture learning, operational redesign, or changes in the sequencing of securities obligations in full.
The study’s robustness analysis strengthens the direction of the finding. When the authors use non-transformed data, the reserve reductions remain linked mainly to delays from the 40th to the 90th percentiles. The patterns for repo reliance, payment concentration, the DCV, month-end effects, and the time trend are broadly similar. This consistency makes it less plausible that the headline result is an artifact of one data transformation. It still does not turn an observational comparison into a randomized experiment.
A disciplined reading therefore has three layers. The first is observation: the CUD value distribution changed, with earlier lower percentiles and later upper percentiles, particularly in the most recent regime. The second is association: reserve-requirement indicators and the value-weighted timing measure point toward later settlement in the upper part of the distribution. The third is interpretation: institutions may have redeployed released liquidity, relied differently on repos and incoming payments, or adjusted to the DCV in ways that delayed large obligations. The third layer is plausible and supported by the structure of the evidence, but it remains an explanation to test rather than a direct recording of every bank’s motive.
That separation matters for policy judgment. If the observed result were a uniform system-wide delay, an aggregate average might be enough to describe it. Because the result is distributional, the relevant warning signal is where the unsettled value is located. A system that settles its first 40% of value earlier but leaves more of its final 20% for late afternoon may feel faster in ordinary activity while becoming more dependent on successful end-of-day coordination for its largest flows.
The Colombian evidence thus establishes a precise empirical starting point: less required reserves coincided with a reallocation of CUD settlement timing, not a universal acceleration or universal delay. The timing pattern is real enough to demand a mechanism, but the overlapping pandemic response and DCV transition make it unsafe to tell a one-variable story. The next step is to examine how participants can rationally use newly available liquidity in ways that produce exactly this split between earlier lower-value settlement and later large-value settlement.