When More Liquidity Makes the Day End Later
The subject is Colombia’s Sistema de Cuentas de Depósito, or CUD, the country’s large-value payment system operated by the Central Bank of Colombia. The question is deceptively simple: if banks are required to hold fewer reserves, why might more of the system’s largest payments arrive later in the day? The answer matters because a payment system can look liquid in aggregate while still concentrating its most consequential obligations near closing time. This series uses the CUD evidence to build a practical way to distinguish the amount of liquidity from its timing, allocation, and ability to move through an interconnected network.
A reserve release did not move every CUD payment earlier
The BIS study covering 2018 to 2025 finds that reductions in Colombia’s reserve requirements were associated with a redistribution of settlement timing. That is different from saying that every payment became slower. After the April 2020 reduction, the lower end of the cumulative payment-value distribution moved earlier: settlement at the 10th percentile occurred 32 minutes earlier than in the preceding regime, and the 40th percentile about five minutes earlier. Higher-value portions moved in the opposite direction. The 70th percentile settled 24 minutes later and the 90th percentile 21 minutes later.
The pattern became more pronounced in the latest period, which followed the September 2024 reserve reduction and overlapped with the introduction of a new Central Securities Depository, or DCV, in April 2024. Relative to the first period, the 10th percentile settled three hours and 19 minutes earlier, while the 40th percentile settled 16 minutes earlier. Yet the upper part of the distribution remained more concentrated late in the operating day. At 17:00, the cumulative share of value settled in the latest period was below the earlier level, which was near the 90th percentile, leaving a larger share for the evening window.
This is a distributional result. It describes where the day’s payment value accumulated, not the fate of every individual transfer. A system may clear many smaller obligations earlier while postponing a smaller number of very large obligations. Counting transactions alone could therefore make the change look modest. The amount settled in each interval is more revealing because large-value transfers consume or release more intraday liquidity and can influence the ability of other participants to pay.
The practical concern is not that late settlement is automatically unsafe. Payment systems are designed to accommodate activity across the operating day, and the CUD study does not report a disruption caused by the observed pattern. The concern is concentration: when a greater share of high-value obligations is left for late afternoon or evening, a technical failure, funding shock, or coordination problem has less time to be absorbed before the system closes. Timing changes the room for recovery even when the total value processed during the day continues to rise.
Percentiles make that concentration visible without pretending that there is one fixed category called a “large payment.” The 90th percentile, for example, is the point at which 90% of the day’s cumulative payment value has been settled. It is a position in the day’s value distribution, not a legal threshold or a statement about the size of one particular transfer. If that point moves later, the evidence says that the final portion of the day’s value is arriving later in aggregate. The change can be economically important even when most smaller obligations have already cleared.
There is also a crucial wording distinction in the question. A lower required reserve ratio is not the same as a policy decision to remove an equal amount of reserves from the banking system. It lowers the regulatory floor for balances associated with deposits and gives institutions more discretion over funds that had been tied to that requirement. Whether those funds remain available for settlement, move into another market, or are used to meet a later obligation is a decision made inside the payment and funding network. The CUD result is about that allocation of discretion.
Settlement liquidity is about timing, not only quantity
The phrase “more liquidity” can hide several different facts. A reserve requirement determines how much of certain deposit liabilities a credit institution must hold as cash or as deposits at the central bank. Reducing that requirement can release funds from a low-yield or otherwise constrained use. But released funds do not have to remain in the account used for settlement. An institution can redeploy them, hold them against another risk, use them in money-market activity, or wait for incoming payments that make outgoing payments cheaper to fund.
Settlement liquidity is therefore an intraday condition. It depends on how much is available at the moment a payment must be sent, how expensive it is to obtain or retain, and whether other participants are sending funds into the network. The same total reserve balance can support different payment schedules. If banks submit obligations early, liquidity is consumed early but payment chains can recycle funds sooner. If they wait, they preserve liquidity and gain information about incoming cash, funding costs, and counterparties, but they also leave more obligations outstanding and make the network more dependent on late-day coordination.
A simple hypothetical makes the distinction concrete. Suppose a bank has enough immediately usable funds to send a payment, but it also expects an incoming transfer later in the day and can earn a return by placing part of its cash elsewhere. Sending now removes uncertainty for the counterparty and releases liquidity into the network. Waiting preserves optionality and may be financially attractive, provided the payment can still be completed before the system closes. The choice is not between “liquid” and “illiquid” in the abstract. It is between different uses of liquidity across time, each with a different exposure to funding, settlement, and coordination risk.
That distinction explains why a policy intended to lower intermediation costs need not produce an across-the-board acceleration in CUD settlement. The policy changes the feasible choices; it does not dictate the time at which each institution will use its released funds. The study’s interpretation is that some participants may allocate additional liquidity to short-term money-market operations before executing other payments. In that case, the reserve release improves the institution’s overall liquidity position while changing the order in which its obligations are discharged.
The same choice can look different for a small institution and a large one. A smaller participant may need a particular incoming payment before it can send its own obligation, so an earlier payment by a major counterparty can materially improve its options. A large institution may be able to fund itself through several channels and decide that preserving or redeploying liquidity is worth accepting a later submission. Aggregate balances can rise while the network’s dependence on a few timing decisions becomes more pronounced.
The network makes the timing relational. One institution’s outgoing payment can be another institution’s source of funds. A bank that relies heavily on incoming payments may delay its own transfers until those inflows arrive. A large participant’s decision can alter the liquidity available to smaller counterparties, while a smaller participant may have little ability to absorb a missed or delayed inflow. “System liquidity” is consequently not just the sum of balances on individual accounts. It also includes the timing and predictability of flows between those accounts.
The CUD data show why value and volume should be read separately. Payment activity has a recurring spike around the opening of the system, roughly 7:00 to 7:30 a.m., and a pronounced rise toward the end of the operating day. The opening peak is larger in value than in transaction count, indicating that relatively high-value obligations are concentrated there. After midday, smaller and more numerous payments make the volume profile smoother. A change in the timing of high-value payments can therefore matter for liquidity management even if the number of transactions changes little.
What this case can—and cannot—tell us
The reserve requirement changes provide the CUD study with a useful before-and-after structure, but not a perfectly isolated experiment. Before April 2020, the applicable ratios were 11% for demand deposits and 4.5% for longer-term deposits. On April 22, 2020, they fell to 8% and 3.5% as part of pandemic-era liquidity measures. On September 11, 2024, they fell again to 7% and 2.5%, a policy change the paper links to financial deepening and lower intermediation costs. The authors estimate that the 2024 adjustment released about COP 6 trillion, or approximately USD 1.5 billion, in bank reserves.
The 2020 intervention coincided with other measures taken during the pandemic, so its settlement effect cannot be assigned to reserve requirements alone with complete confidence. The latest period also includes the new DCV, whose operating changes affected securities settlement and could alter the timing of CUD flows independently of reserve policy. The authors address the problem through descriptive comparisons, regressions across cumulative value percentiles, controls for payment activity and liquidity conditions, participant-size analysis, and a robustness exercise using non-transformed data. Those methods make the reported pattern more credible, but they do not erase the institutional overlap.
That qualification does not make the evidence uninformative. It tells us what kind of question the study can answer well. The paper can show that the timing distribution changed around identifiable policy regimes, that the change is concentrated in particular parts of the value distribution, and that related liquidity and institutional variables move in coherent directions. It is less able to say that a precisely measured number of minutes was caused by one reserve decision alone. In payment systems, policy and infrastructure often move together because a central bank changes several operating conditions in response to the same economic problem.
This is why the upper tail is a more useful object of attention than a single average settlement time. An average could rise because every payment moved slightly later, because a few very large payments moved much later, or because the composition of payments changed. The percentile profile separates those possibilities. It also links the statistical result to an operational question: which obligations remain unsettled when the system has the least time left to absorb a problem?
The strongest conclusion is therefore conditional. In this Colombian RTGS system, lower required reserves were associated with later settlement in the upper part of the payment-value distribution, while lower percentiles could move earlier. The result is consistent with a mechanism in which released funds change opportunity costs, repo reliance, payment recycling, and the timing strategies of participants. It is not a universal rule that reserve reductions always delay large payments, nor does it establish that the same threshold or response would appear in another payment system.
What the case does establish is a better question for reading liquidity policy. Instead of asking only whether banks hold more or fewer reserves, ask when those reserves are usable for settlement, what competing uses they have, which participants depend on incoming flows, and where the largest obligations sit in the day’s distribution. Those questions keep CUD—the actual payment system and its participants—at the center while treating settlement liquidity as the lens that makes the evidence intelligible.
The rest of this series develops that lens from the payment-day plumbing upward. The essential point is already visible: liquidity policy changes the conditions of settlement, but institutions and networks decide how those conditions become a timetable. When that timetable shifts, the day can end later even when the system begins with more room to maneuver.