The Payment-Day Plumbing: What CUD Needs from Liquidity
In Colombia’s CUD, the question is not whether a bank intends to pay, but whether it has enough usable balance at the moment the payment is submitted. This chapter examines how that real-time constraint turns liquidity into a timing decision. Once the settlement rule, the available funding channels, and the shape of a CUD business day are clear, the difference between payment volume and payment value becomes much easier to see.
A CUD payment settles only when the balance is there
CUD is Colombia’s large-value payment system, operated by the Central Bank of Colombia. It processes obligations associated with large electronic fund transfers, securities and foreign-exchange trades in local currency, interbank lending, tax payments, and check clearing. Its operating model is real-time gross settlement, or RTGS. “Real-time” means that an eligible payment is processed as it is submitted rather than accumulated for a later batch. “Gross” means that the obligation is settled individually rather than first offset against other obligations. When the payment completes, settlement is immediate and final.
Finality is valuable because it removes uncertainty about whether a completed transfer can later be reversed through the payment system. But finality also makes the balance constraint concrete. A participant cannot make a payment final merely by having a profitable trade, a willing counterparty, or an obligation that is due. The account used for settlement must contain enough funds when the instruction is accepted. In Colombia’s CUD, intraday overdrafts are not permitted. A participant that is short of liquidity must therefore find funds, wait for an incoming payment, or manage the timing of its outgoing instruction.
That rule produces a sequence rather than a single decision. Consider a bank that owes a large payment to another CUD participant. It first has to decide whether sending the payment now is worth using its immediately available balance. If the balance is sufficient, the payment can settle and become final. If it is not, or if using the balance now would leave the bank too exposed to another obligation, the bank has several alternatives: draw on a permitted source of intraday liquidity, obtain repo funding, wait for a payment from another participant, or defer the instruction. Each option changes both the bank’s own position and the liquidity available to its counterparties.
The distinction between these channels matters. A reserve balance is liquidity already held in the central-bank settlement account. Repo funding is a way to obtain liquidity through a financing arrangement. An incoming payment is not a new reserve policy; it is a flow from another participant that can be recycled into the next obligation. These channels may all help a payment settle, but they do not have the same cost, timing, or reliability. A bank that depends on a counterparty’s payment is exposed to that counterparty’s schedule, while a bank with several funding options has more control over when to submit.
The result is strategic interdependence. Paying early can release funds to another participant and help the network settle a chain of obligations. Waiting can conserve liquidity and reveal more about later inflows, but it can also leave the counterparty waiting and reduce the time available to correct an error. A participant is not choosing only between “pay” and “do not pay.” It is choosing how to use scarce intraday balance across obligations that may arrive in an uncertain order.
Imagine three institutions connected through CUD. The first owes the second, and the second owes the third. If the first payment settles early, the second can use the incoming funds to meet its own obligation without finding the full amount elsewhere. If the first institution waits, the second may also wait, even if the second has a payment instruction ready. The delay is not necessarily caused by a lack of liquidity across the system as a whole. It can arise because liquidity is in the wrong account, at the wrong time, or still tied to a decision that another participant has not yet made. This is the practical meaning of payment recycling.
The same chain also explains why finality and timing belong together. Immediate final settlement reduces the risk that a completed obligation remains uncertain, but it requires the system to verify the balance at the moment of settlement. A participant cannot rely on a payment that has merely been promised; it needs an incoming transfer to have actually arrived or needs another funding source to be in place. In a system without intraday overdrafts, the boundary between “will receive” and “has received” is a liquidity boundary. Payment schedules are shaped by that boundary even when the underlying commercial obligations are already known.
Reserve requirements sit in the background of this process. In the Colombian framework described by the study, required reserves are calculated over a biweekly period and held as cash or deposits at the central bank. That maintenance structure allows some intertemporal smoothing: a bank does not have to match every payment to a separate reserve deposit at every instant. Yet the CUD balance constraint still operates minute by minute. A bank can satisfy a reserve requirement over the relevant period and still face a difficult settlement choice at a particular point in the business day.
This difference between a regulatory average and a real-time constraint is easy to miss. A bank may have complied with the reserve rule over the maintenance period and still preserve its current CUD balance because a large payment is due later, because an incoming transfer is expected but not yet settled, or because an alternative use of funds has a better return. Conversely, an institution may have a relatively small reserve position but settle promptly if its obligations are small or its incoming payments arrive reliably. The reserve requirement changes the starting conditions; the payment network determines how those conditions are experienced during the day.
The day has two peaks, but value and volume tell different stories
CUD’s intraday profile gives the balance constraint a visible shape. Payment activity repeatedly surges around the opening of the system, roughly 7:00 to 7:30 a.m., and rises again toward the end of the operating day, around 7:30 to 9:00 p.m. Those two peaks are not simply two versions of the same event. The opening spike is larger when measured by payment value than when measured by transaction count, while smaller and more frequent payments are spread more smoothly after midday.
This difference between value and volume is central. Volume answers, “How many instructions were settled?” Value answers, “How much money moved through the settlement accounts?” Ten small payments and one very large payment may contribute the same transaction count but create very different demands on available liquidity. A system can process a steady stream of instructions while still experiencing a sharp funding requirement when a few high-value obligations arrive close together.
A useful way to read the two measures is to ask what they would imply for a bank’s balance. A rise in volume may reflect many routine transfers whose individual funding demands are modest. A rise in value may reflect a smaller number of transfers that consume a substantial share of immediately available funds. Neither measure is sufficient on its own. Volume helps describe operational workload; value helps describe the scale of the liquidity position that must be managed. The CUD study gives more weight to value because its question is about settlement liquidity and the timing of obligations, not merely the number of messages processed.
The percentile approach follows from the same logic. Suppose, only as an illustration, that a day contains 100 units of payment value. The time at which the first 10 units have settled says little about when the final 10 or 20 units will arrive. A policy change could move the first 10 units earlier while moving the last 10 units later. Looking only at the midpoint would blur that contrast; looking at several cumulative points reveals whether the system is front-loading, back-loading, or spreading value differently across the day.
The drivers of the peaks also changed over the sample. In the earliest period, tax-related payments contributed to the morning activity. In later periods, the early peak was more closely associated with the unwinding of repurchase agreements and the return of interest-bearing deposits connected with central-bank operations. The researchers exclude certain policy-operation and government-related flows from the empirical analysis designed to measure the “intraday game,” but those flows still help explain why the raw CUD day has recurring peaks.
Late-day activity increasingly reflected securities settlement through the DCV. In the period after the new DCV began operating, settlement of DCV securities transactions became a major contributor to the end-of-day peak, alongside smaller contributions from other transfers and deposit operations. This is one reason a payment system cannot be understood by looking at reserve balances in isolation. The obligations created by securities infrastructure can arrive on a schedule that changes how banks use their settlement liquidity.
These peaks should not be read as a fixed timetable imposed on every participant. They are averages across many days and many institutions. A recurring peak can arise because the same type of obligation is commonly due at a particular time, because central-bank operations create a predictable flow, or because institutions coordinate around one another’s expected payments. It can also become steeper when several drivers coincide. The distinction matters for interpretation: a policy change may not move the whole day uniformly; it may change the intensity or composition of an existing peak.
The CUD profile also contains flows that are important for describing the system but not equally useful for measuring inter-institutional liquidity pressure. Transfers between accounts owned by the same institution do not create a new flow between participants. Certain remunerated-deposit operations and government-related transfers likewise reflect institutional operations rather than the ordinary payment decisions the study calls the “intraday game.” The researchers filter such categories in the regression dataset. This improves the connection between the measured outcome and effective liquidity use, while leaving the broader descriptive profile useful for understanding why the day has its visible shape.
CUD’s fee schedule adds another timing signal. From 6:00 a.m. through 17:00:59, a debit entry carries a fee of COP 3,570 per transaction. From 17:01:00 until 5:59:59 the following morning, or until the next opening, the system applies an ad valorem charge of COP 2.5 per million pesos, subject to a minimum fee of COP 5,170. For a COP 5 billion transfer, the paper gives an illustrative comparison: the first window costs COP 3,570, while the second costs COP 12,500, about 3.5 times as much.
If price alone determined timing, that structure would pull large payments before 17:00. The observed data show why price is only one part of the decision. The timing of a large payment may be tied to the money-market transaction that generates it, to a bank’s need to complete its funding adjustment, or to the arrival of liquidity from another participant. The expected return or funding necessity of the underlying transaction can outweigh the settlement fee. A bank may rationally accept a higher fee if moving the payment earlier would require more expensive liquidity or would interfere with another obligation.
For this reason, the paper emphasizes cumulative value percentiles. The 10th percentile is the time by which 10% of the day’s total payment value has settled; the 90th percentile is the time by which 90% has settled. Looking across these points shows whether a change affects the beginning, middle, or tail of the settlement distribution. A single value-weighted average is useful, but it can conceal whether the entire day moved or whether the largest remaining obligations became concentrated late.
The CUD plumbing therefore supplies the necessary interpretation for the policy evidence. A reserve change does not enter an empty system. It enters a real-time network with final settlement, no intraday overdrafts, multiple liquidity sources, recurring payment peaks, securities-related obligations, and prices that vary by time window. The next question is how the system’s timing distribution changed when reserve requirements were reduced—and whether the change was uniform across the day.