Where the internal price stops, a bank's business areas can decide separately
Evidence markers: [V] verified · [D] derived · [E] estimate with an explicit assumption · [H] unproven hypothesis
A universal bank makes many decisions, and a few of them compete for the same resources, consume the same constraints, and change the state on which the others are made. When each one is resolved within its own area and against its own target, the bank can end up with decisions that are locally correct and jointly inconsistent [H]. This Perspective proposes a map of those decisions—1 core that issues prices, 7 systems that consume them, and 1 second-order system that sets what the core optimizes—and the rule that distinguishes a legitimate boundary between 2 systems from a single system split in 2. The mechanism that holds the map together is this: in its matched-maturity form, transfer pricing assigns each transaction an internal rate that incorporates the bank's funding cost and matches the maturity structure of its cash flows; because that rate is fixed at origination, the business margin is protected from the interest rate and funding risk that the price transfers [H]. Where the internal price leaves a resource out, or prices it with average and static parameters, the areas can go back to deciding separately [H]. The piece is a reasoned derivation from banking economics and operations; it describes no institution, and no industry-wide census supports it.
0 · Structure of this note
Sections 1 and 2 explain the balance-sheet interdependencies and the limits of the internal price. Sections 3 and 4 present the core and the coupled systems. Sections 5 to 7 specify the boundaries, the capacity costs, and the fraud policy. Sections 8 to 10 describe targets, signals, and institutional capability. Sections 11 and 12 gather the open boundaries and the questions for the senior executive. The note distinguishes verified facts, derivations, estimates, and hypotheses.
1 · Bank decisions that share resources can, when resolved separately, produce results that are correct for each area and inconsistent for the bank
Three chains are enough to show it.
Liabilities. The deposit price influences the volume raised and its composition. Stability also depends on the segment and on the features of the product [H]. That stability determines the liquidity buffer and the duration that can be taken. The duration taken determines the sensitivity of net interest income and of economic value. Capital has to absorb, above all, the sensitivity of economic value [H]. The link to liquidity has a regulatory basis. In its 30-day stress scenario, the Basel III standard distinguishes categories of retail deposits [V]. It normally assigns stable deposits a run-off (outflow) rate of 5% [V]. Less stable deposits receive a minimum of 10%, depending on the categories and conditions of the standard [V]. In Mexico, the Coeficiente de Cobertura de Liquidez (CCL, Liquidity Coverage Ratio) of commercial banks has a required minimum of 100% [V]. The Comisión Nacional Bancaria y de Valores (CNBV, National Banking and Securities Commission), the banking supervisor, sets out that minimum [V]. If the CCL weights outflows as the Basel standard does, a change in the deposit mix can change its weighted outflows. The change can occur within the same month in which the price was set [H].
Credit. What the bank charges influences who accepts the offer. Price and origination must be resolved together when the price changes the selection of borrowers and the composition of risk [H]. The authorized limit conditions future exposure. That exposure also depends on how much is drawn, on amortization, and on the use of undrawn lines [H]. This quarter's origination also conditions the potential collections workload of the following months. The peak depends on the product. Available collections capacity should bound the origination appetite [H].
Fraud. Tightening the authorization threshold can reduce the fraud that gets through the filter. It can also increase false positives, with their friction for the customer, and change the demand for investigation. When each of those 3 magnitudes is the target of one area, they are optimized separately [H]. Total economic cost can worsen while fraud loss improves [H].
In all 3 chains, each link can be the decision of a different area with its own target. If its internal price leaves out what it consumes from the others, the optima can diverge. The bank's optimum can stop coinciding with the sum of the areas' optima [H].
2 · Fragmentation can arise from the perimeter of the internal price and from area-level targets, and adding models within each area leaves it standing
In its matched-maturity form, transfer pricing assigns each transaction an internal rate. The rate is charged to loans and credited to deposits. It is built from the risk-free curve, the bank's own funding premium, and maturity matching. The matching follows the structure of the transaction's cash flows. The rate is fixed at origination for the term of each cash flow. The business margin is protected from subsequent movements in market rates and in the funding premium. That risk passes to the unit that manages the bank's assets and liabilities, which handles it centrally [H]. The areas can go back to deciding separately where the internal price leaves resources out or prices them with static averages. The price can leave out collections capacity and the value of the customer's whole relationship. It can price capital consumed, marginal credit risk, and cost to serve with average and static parameters [H].
Transfer pricing is the case in which coordination by price already works. Origination and deposit-taking incorporate funding cost and interest rate risk without resolving each transaction jointly with that unit. The unit manages the risk it receives centrally [H]. If an area consumes a resource whose cost already comes in its internal price, the area can decide alone and decide well for the bank [H]. If the resource is left out of the price, the area can treat it as free or assign it an implicit value of its own. Another area has to reconcile afterward [H]. Fragmentation thus has 2 origins. The internal price leaves resources out or charges for them with static averages. Targets are set by area [H]. Adding independent models within each area, by itself, leaves the 2 origins standing [H].
A volume target already contains a pricing decision. Setting the volume first and then looking for the rate that achieves it solves the problem backward. The price becomes the consequence of a figure that nobody priced [H].
3 · Liability pricing, the liquidity corridor, duration, and balance-sheet capacity form a single core that issues prices to the rest
The 4 decisions share resources, constraints, and a central assumption about liability behavior. That assumption covers the pass-through beta and the average life of non-maturity deposits. It also takes in the assumptions contributed by other liabilities, assets, and maturities [H]. The size of the buffer can change the marginal cost of liquidity. If the CCL weights outflows as the Basel standard does, the deposit mix can change its weighted outflows [H]. When the decisions are resolved in sequence, a later one can make an earlier one obsolete within the same month. Someone would have to reconcile them by hand [H]. There is reason to resolve those decisions together when coordination through prices, limits, and states loses effectiveness. The condition is that such coordination stops preserving the quality of the decision within its horizon [H]. That is why this Perspective reads the balance sheet as 1 core.
Duration rests on a behavioral assumption: how long non-maturity deposits remain. In April 2016, the Basel Committee on Banking Supervision published standards on interest rate risk in the banking book [V]. The standards ask banks to consider key behavioral and modeling assumptions when they measure that risk [V]. An assumption subject to a standard becomes a constraint. How much duration to take remains a choice under uncertainty [H].
The binding constraint indicates when to reopen the core. This Perspective proposes 4 regimes [H]. In normal conditions, the binding constraint is the volume target that system 9 sets. Near the internal liquidity limit, liquidity binds, and cost stops being the criterion. In a rate-cutting cycle, the sensitivity of net interest income binds, and the decision becomes one of duration. With ample slack, profitability binds. The same price can be right in the first regime and wrong in the other 3 [H].
The core issues 3 prices that the other systems consume. The transfer curve already includes the liquidity premium. The charge for the regulatory liquidity buffer and the capital charge are added to it [H]. After the binding constraint, the second signal for reopening the core is a performance gap. The realized return by allocation is compared with the expected return. If that comparison is not kept, nothing triggers a review of how the balance sheet is allocated between year-end closes [H]. The capitalization framework and its constraints are set in the second-order system (§8). The core allocates the marginal balance-sheet capacity available within that framework [H].
4 · Seven systems consume those prices and decide within them, connected by price, by limit, and by state
# | System | What is decided | How it is coupled | Cadence |
1 | Balance-sheet allocation and structure | Liability mix and pricing, liquidity corridor, duration, allocation of balance-sheet capacity | Core that issues prices | Monthly and upon a material event |
2 | Credit pricing, limits, and offer | To whom to originate, with what amount, price, and limit | By price | Per transaction |
3 | Commercial, customer service, operational, and technology capacity | What capacity to build and in what order | By limit and by price | Quarterly or annual |
4 | Markets: inventory, pricing, and hedging | What inventory to take to serve the customer, at what price, and with what hedge | By price and by limit | Continuous and intraday |
5 | Recovery of impaired loans | What intervention to make on each account, with what intensity, and with which provider | By price and by state | Per event and per account |
6 | Intraday liquidity, collateral, and payments | How the position moves during the day and what collateral is mobilized | By limit, by price, and by state | Minutes |
7 | Allocation of customer service capacity and of the offer | Through which channel, how often, and with what offer to reach each customer | By limit, by price, and by state | Continuous, with a monthly policy |
8 | Fraud, authorization, and intervention | Where the thresholds are set | By limit and by state | Monthly policy over real-time execution |
9 | Targets, budget, and allocation of ambition | Appetite, allocation of ambition across businesses, and plan assumptions | Second order: produces the objective function of the core | Annual, reopened on performance |
Source: InsightIA derivation [H].
The systems connect without merging, and that avoids replacing many small silos with 9 new silos [H]. By price: one system consumes another's resource and internalizes its price at its own cadence. With the capital charge as a number, the credit system prices loans without resolving each transaction jointly with the core. By limit: one system receives a boundary within which it optimizes freely. The intraday system receives the corridor, and the fraud thresholds receive the investigation capacity. By state: one system returns a realized state that changes the next solution of the other. The originated vintage changes the recovery problem. By several paths at once: recovery and intraday liquidity return states and marginal costs that other systems have to incorporate [H].
Four of the coupled systems deserve a line of their own.
Credit (2). Setting the limit after the price amounts to setting it on a portfolio whose composition the price has already helped to determine [H]. The month's origination can materially change marginal capital consumption or the binding constraint. If it does, the capital charge used to price the loans of the third week may need to be recalculated [H].
Capacity (3). It brings together 4 resources that cost money, take time to build, and hand the core a short-term constraint [H]. When they are decided as a portfolio of projects against a budget, on the judgment of each sponsor, the order is set by whoever best defends a project [H]. With capital and liquidity slack, capacity can become the binding constraint while the budget discussion is still about capital [H].
Markets (4). Article 46 of the Ley de Instituciones de Crédito (Credit Institutions Law) covers deposit-taking and lending [V]. It also covers foreign exchange and derivatives transactions, subject to the applicable provisions [V]. That is why the map includes markets. Taking inventory without the current capital charge amounts to serving the customer at a price that omits what the position consumes [H].
Customer service and offer (7). A natural objection to this system is that, without binding shared constraints, it degenerates into a recommender. Installed customer service capacity can be the shared constraint that was missing, contested by several products at once. Contact limits and customer response also bound the allocation of offers [H]. How much capacity to sustain belongs to system 3. How to allocate it belongs to the customer service and offer system. This Perspective proposes producing cost to serve as a marginal cost per customer and channel. It would replace the average allocated by product and channel [H].
5 · A boundary between systems holds as long as an update trigger can keep up with the effect of one system on the prices of the other
It is normal for a coupled system to alter the state of the core. Every price-based decomposition works that way [H]. The boundary needs an update trigger if that effect makes prices or limits inadequate before the core resolves. If even with that trigger the prices cannot keep up with the state, there is reason to resolve the 2 systems together [H].
The credit example illustrates the rule. Suppose the core issues a capital charge calculated for the month. If that month's origination materially changes marginal capital consumption, the charge can become inadequate by the third week [H]. If the capital charge is set annually in the budget, a material change in marginal consumption can make it inadequate. This can happen before its next update [H]. The problem can lie in the update cadence, in the methodology of the charge, or in both [H]. This section discusses cadence. Under that assumption, what is missing is a trigger that reopens the calculation when origination deviates. Merging the 2 systems would be the wrong answer [H].
The rule has 2 further conditions. Observability: the state that the coupled system returns must be observable and attributable to a segment and a vintage. Only then can the boundary be governed. If that condition is missing, the bank has to instrument that return or reconsider the separation [H]. Object: the boundary runs through the object of the decision. The coupled system optimizes its local decision under prices, limits, and signals. Those signals represent the constraints and externalities that matter for the whole. The core decides the composition of the whole, and the credit system decides a transaction within it [H].
6 · When a capacity can be bought at a known cost, part of the coupling is expressed as that cost
Many capacities that look like hard constraints can be elastic to a price [H]. Two cases with a public basis illustrate it.
Collections. In Mexico, financial institutions can hire external collection agencies [V]. The Comisión Nacional para la Protección y Defensa de los Usuarios de Servicios Financieros (Condusef, National Commission for the Protection and Defense of Financial Services Users) is the authority that protects users of financial services [V]. On October 7, 2014, it published provisions that set obligations for institutions on how their agencies deal with users [V]. It administers the Registro de Despachos de Cobranza (REDECO, Registry of Collection Agencies) and states that institutions pay for that activity [V]. If collections capacity is contracted at a price, the marginal recovery cost by profile can be passed on to origination. That passes on part of the interdependence, and the boundary between the 2 systems can hold. The condition is to observe recovery results, effective capacity, and customer-treatment constraints as well [H]. The circularity between origination and collections changes place. It becomes the tension that the price has to reflect. If a bad vintage makes recovery more expensive and that cost does not reach pricing, the interdependence remains. Origination stops internalizing it [H]. That cost can vary with the cycle and with recovery itself. A bad vintage can coincide with a credit deterioration across the banking system that saturates the same agencies. Condusef states that compensation is usually a commission tied to the amount recovered [V]. That is why the cost can depend on the recovery rate itself [H].
Intraday liquidity. Banco de México provides intraday liquidity to banks and brokerage firms so that they can settle transactions in the payment systems [V]. To banks, it grants overdrafts in their accounts up to a limit that depends on the capital of each bank, always collateralized [V]. To banks and authorized brokerage firms, it grants intraday repos against eligible securities [V]. These are securities of the Federal Government, of the Instituto para la Protección al Ahorro Bancario (IPAB, Institute for the Protection of Bank Savings), the deposit insurer, or of Banco de México itself [V]. That liquidity carries no cost within the day [V]. If it extends beyond the close, it costs 2 times the Tasa Ponderada de Fondeo Bancario (weighted bank funding rate) [V]. With a provider of that kind, a poor decision on the intraday position produces cost and delay. The coupling with the core combines the opportunity cost of liquidity and collateral with availability limits and intraday states [H].
When additional capacity is available at a known cost, part of the scarcity is expressed as marginal cost. Availability and execution limits remain [H].
7 · In fraud, knowing where the right threshold lies has a cost, and paying that cost is a policy decision
The decision on each transaction is contained in policy and model. The system sits one level above. There, loss appetite, the commercial cost of false positives, and investigation capacity compete. One lever, the threshold, moves all 3 magnitudes at once [H].
In fraud, obtaining part of the evidence needed to calibrate the threshold carries an exploration cost [H]. A blocked transaction does not directly reveal what its outcome would have been had it been authorized [H]. One way to obtain that evidence is to deliberately authorize a controlled sample of transactions that would otherwise be blocked. Another is to subject them to step-up authentication. Both routes have costs and limits on what they let the bank observe [H]. Choosing the route and how much to pay for the evidence is a policy decision with an explicit cost. Someone with authority has to approve it [H].
8 · A planning process that allocates an already decided figure can leave poorly framed the systems that resolve well internally
A system whose output is the objective function of another forms a second-order system. Its authority and horizon differ from those of the decisions that execute it [H]. If it set its own criterion, the definition would be circular. That is why targets, the budget, the allocation of ambition, and the capitalization framework sit outside the core. There, the objectives and constraints for the period are set. The core operates within them and returns signals that can trigger their review [H].
The plan assumptions can arrive fixed from above, while the areas contribute inputs that carry no force as constraints. In that case, the exercise allocates an already decided figure instead of discovering it [H]. If every perspective takes part and no real constraint can move the result, the process is integrated in the organizational sense. It lacks decision capability. Taking part differs from deciding [H]. Its reopening trigger is realized performance against projected performance. If the budget is set once and only tracked for 12 months, nothing reopens it [H].
Resolved poorly, this system can leave the other 8 poorly framed even when each one resolves well internally [H].
9 · The map needs 7 signals; 2 of them are marginal costs that the internal price can leave out, and 1 is the evidence that reopens the core
If 2 systems consume the same magnitude and each has its own version, a shared representation is missing. What remains is a pending reconciliation that someone would have to do by hand, every time [H]. If the magnitude is not produced, the system that needs it works on an assumption [H].
Signal | What it is | Who produces it and who consumes it |
Behavior of non-maturity deposits | Pass-through beta and average life | Internal to the core |
Transfer curve | Internal rate of each cash flow: base rate at its repricing term plus funding premium at its liquidity term; for liabilities, at the behavioral term | Produced by the core; consumed by 2 and 4 |
Capital charge and liquidity buffer charge | Cost of consuming balance sheet | Issued by the core; consumed by 2 and 4 |
Probability of default and loss severity | Estimated by borrower, transaction terms, collateral, and scenario; their composition in the portfolio changes with the offer | Produced by the risk models, with independent validation, for system 2; consumed by pricing, the limit, and the core |
Marginal recovery cost by profile | Recovery component identified separately, without duplicating what severity already includes | Produced by 5; needed by 2 |
Marginal cost to serve by customer and channel | Incremental cost of attending to and serving the customer by channel, for the decision at hand | Proposed to be produced by 7; needed by 2 and by every measure of customer profitability |
Realized versus expected return, by allocation | Gap between what was expected and what happened | Produced by the decision and outcome log (§10); received by the core, and triggers the review when the deviation exceeds a defined threshold |
Source: InsightIA derivation [H].
The last 3 could matter more than adding systems [H]. The first 2 of those 3 are the marginal recovery cost and the marginal cost to serve. They can enter the prices on which decisions are made [H]. As an international regulatory reference, Article 5 of the European capital requirements regulation (Regulation (EU) 575/2013) defines economic loss [V]. For capital requirements for credit risk, the definition includes material direct and indirect costs associated with collection [V]. That reference does not establish the treatment applicable in Mexico. This Perspective proposes that the recovery component reach pricing disaggregated by profile. Pricing must avoid duplicating what severity already includes. The proposal rests on the fact that the component is produced in recovery and varies with whom the bank originated to [H]. Realized versus expected return belongs to another category. It is performance evidence that triggers a review of the core when what was realized departs from what was assumed beyond a threshold [H].
10 · Integrating areas organizes who takes part; decision capability is what remains when the participants change
Identifying the systems is the first step. The next is knowing what has to be installed for decision capability to remain across cycles and across people.
If the people who do the analysis today disappeared, what would have to remain for the bank to keep making that decision?
There are 6 things:
The formulation of the decision: an objective function and constraints that are explicit, written down, and alive
The shared representations: with an owner, a version, and periodic calibration
The joint resolution mechanism: the procedure that produces the answer, with its assumptions declared
The decision and outcome log: what was decided, under what state of the world, and what happened afterward
Governance: who can change what, with what evidence, and what is reserved for judgment
The reopening triggers: declared thresholds on monitored variables, so that the decision reopens when the state changes and stops depending on someone remembering
A bank can bring together areas, data, models, and a committee, and have a perfectly institutionalized process, without having built a decision capability [H]. Integration organizes who takes part. Decision capability determines what remains when the participants change [H].
11 · Three boundaries remain open, and the piece leaves them stated
Should the 4 resources of system 3 be resolved together? What §4 says unites them: all of them cost money, take time to build, and deliver a short-term constraint. Whether they must be resolved jointly remains unproven.
Does merchant acquiring require its own system? It remains a candidate, pending application of the boundary criterion.
Is there a tenth? No derivation exercise proves that the list is complete.
The boundaries of systems 3 and 9 hold by argument. They still have to be tested decision by decision. The targets system could carry the greatest consequence if it is resolved poorly, because it produces the objective functions of the others [H].
12 · A senior executive can check on Monday, with 3 questions, whether the map applies to the bank
This piece describes an architecture and leaves unanswered how any particular bank decides. A senior executive can check within the institution whether the map applies:
What does each area consume—capital, customer service capacity, collections capacity—that its internal price leaves out or charges for with an average that does not update with actual use?
If one month's origination deviates materially from plan, who checks whether the deviation changed the marginal cost of capital and, when warranted, triggers its update?
Where is the realized return by allocation recorded against the projected return, and what decision changes when they diverge?
Which analytical family suits each system—optimization, causality, pattern learning, rules—follows from the problem and comes later. Technology arrives last, as a consequence of those 3 answers.
Provenance
Authorship. Prepared by InsightIA, with comments from Ramón Sanz
What is verified. The 8 sourced claims, each with its marker in the passage where it appears, have a public source in the supporting evidence book.
What this piece leaves out. It does not diagnose any institution, does not state how any bank decides today or how many banks have which systems, and is not a list of artificial intelligence use cases.
