Mexico's securities depository didn't ask banks for more cash — it found the trades that could cancel each other out

the problem

When a system runs out of capacity processing transactions one at a time, the default fix is to demand more capital or resources be held in reserve to cover the backlog, rather than asking whether the transactions actually need to be processed one at a time at all

background

what everyone would do

The industry-wide conventional response to settlement gridlock was to require participants to hold larger cash and securities buffers so any single transaction could always clear immediately, locking up capital that then sat idle covering a worst case that rarely actually happened.

what they saw

Indeval, Banco de México and ITAM saw that the actual bottleneck wasn't insufficient capital in the system overall, it was that transactions were being processed one at a time, each requiring its own individually sufficient balance, when many pending trades' obligations would cancel each other out if grouped together, a owes b, b owes c, c owes a. Rather than asking banks to hold more capital in reserve to cover every transaction individually, the fix was building a settlement engine that continuously searched the existing pool of pending obligations for subsets that netted against each other, settling the whole bundle at once instead of each leg separately.

the move

Dalí's core is a linear-programming optimization engine that continuously scans the queue of pending settlement instructions and identifies subsets where obligations net against each other — a owes b, b owes c, c owes a — so the group can settle simultaneously using far less actual cash or securities changing hands than settling each leg individually would require, while still guaranteeing every settlement is final and irreversible.

why it works

Running a linear-programming optimization engine that scanned the queue of pending settlement instructions for mutually canceling obligations meant a group of trades could settle simultaneously using far less actual cash or securities changing hands than clearing each transaction individually would require, while still guaranteeing every settlement remained final and irreversible. Because the fix worked by rearranging and grouping existing pending obligations rather than requiring any participant to deposit new capital, it delivered its entire benefit, a 52% cut in the cash liquidity banks needed and a 26% reduction in required securities collateral, without asking a single bank to add money to the system, making the intervention disproportionately cheap relative to the scale of capital efficiency it unlocked across a system handling over $250 billion in daily transactions. This is why Dalí won the 2010 Franz Edelman Award, the diagnosis correctly identified that gridlock was a sequencing and optimization problem over the transactions themselves, not a capital-shortage problem the industry's standard buffer-holding response had assumed it to be.

the payoff

The netting engine cut the cash liquidity banks needed to keep the settlement system moving by 52% and the securities collateral required by 26%, without any participant depositing additional capital into the system. Dalí won the 2010 Franz Edelman Award from INFORMS, the first Mexican organization and the first financial institution to win the award in its then-38-year history.

where it breaks

The mechanism depends on the pending transaction queue actually containing enough mutually offsetting obligations to net meaningfully — a settlement system where trades don't naturally cluster into cancelable groups, because participants rarely owe each other in overlapping cycles, would find little to optimize and would still need the traditional capital-buffer approach to handle genuinely one-directional flows. It also depends on having the computational and mathematical capability to solve the netting optimization at the speed and scale a real-time settlement system demands, since a naive or slow search for canceling subsets among a large pending queue could itself become a bottleneck, requiring the kind of dedicated optimization engineering Indeval invested in building with academic partners. And netting-based settlement introduces a genuine dependency risk of its own: if the system finds and executes a large netted bundle, a single participant's inability to complete its leg at the last moment can unwind or delay the entire group's settlement, a different failure mode than the isolated, transaction-by-transaction gridlock the traditional buffer approach was originally designed to prevent.

what came after

Dalí is cited in operations-research literature as a landmark real-world application of optimization-based netting to national financial infrastructure, and its approach — treating settlement gridlock as an optimization problem over the existing pool of obligations rather than a capital-shortage problem — has been studied and referenced by other national securities depositories evaluating their own settlement engines.

references

1. [1]Modeling, simulation and analysis of a securities settlement system: the case of Central Securities Depository of MexicoJournal of Economics, Finance and Administrative Science (Elsevier), 2012elsevier.es
2. [2]Mexico's Indeval wins INFORMS Edelman AwardEurekAlert! (AAAS), 2010eurekalert.org

keep it

same kind of clever

• 1991 · MarriottMarriott took the airlines' playbook and priced every room by the night it would die
• 2016 · NHS clinicsNHS clinics clawed back millions of missed appointments with reminder texts
• 1609 · Amsterdam Wisselbank (Bank of Amsterdam)Amsterdam stopped trying to police 800 coin types and just declared what a guilder meant instead