The Queue Nobody Sees Until It Closes
Picture yourself sitting at the treasurer's desk sometime in the third or fourth week of a stress episode. The overnight quotes are still arriving. The rate hasn't moved enough to alarm anyone. The balance sheet, if you pulled it up right now, would look perfectly adequate to anyone glancing over your shoulder. But the quotes are coming in slower than they did last week, and from fewer counterparties, and you have worked these markets long enough to know that when the queue thins, the door is already most of the way shut.
That is how overnight funding access ends for most regional banks. Not with a dramatic rate spike. With silence.
The wholesale interbank lending market, specifically the overnight segment where banks lend surplus reserves to each other for twenty-four hours at a time, is not a neutral exchange where any qualified borrower posts a bid and expects to be filled. It is a network, dense with bilateral relationships, internal credit limits, and reputational memory stretching back years. Understanding which banks lose access first requires understanding how that network is actually constructed, not how the textbook describes it. The textbook, it should be said, is not particularly helpful here.
Tiers, Not Peers
The market organges itself into a rough hierarchy that practitioners call a tiered structure. At the center sit the largest global and national banks: institutions that hold so many correspondent relationships, clear so much payment volume, and maintain such deep balance sheets that other banks need to transact with them regardless of sentiment. Call them core nodes. Around them cluster a second tier of substantial regional players who have established bilateral credit lines with several core nodes and with each other. Further out sit the smaller regionals and community banks that fund themselves primarily through a handful of relationships, often just two or three active counterparties.
The critical mechanic is the bilateral credit limit. Before any overnight loan is extended, the lending bank's credit desk assigns a maximum exposure to each named borrower. These limits are not published anywhere. They are internal decisions, reviewed periodically, and they can be cut unilaterally and silently. A bank can reduce your limit to zero without calling you. You find out when your next request simply is not filled.
Academic work on interbank network topology, notably research using data from the federal funds market and European overnight markets, has consistently shown that this structure is not random. It follows what network theorists call a scale-free distribution: a small number of nodes hold a very large number of connections, and the majority of nodes have very few. The practical consequence is that when stress hits, banks at the periphery lose counterparties rapidly, because each counterparty they have represents a large share of their total access. Lose two out of three, and you have lost most of your market. The geometry of the thing is almost punitive.
The Mechanics of a Quiet Freeze
Consider two banks, both regional, both carrying similar capital ratios and similar loan portfolios. Call them Ridgemont Savings and Lakeview Community Bank. Ridgemont has spent a decade cultivating relationships with eight active overnight counterparties, including two core-node banks and three second-tier regionals spread across different geographic markets. Lakeview has concentrated its relationship-building locally, relying on four counterparties, all of them smaller regionals operating in the same corridor.
A stress signal emerges. Not a catastrophic one: a deteriorating sector exposure reported in the financial press, the kind of item that runs on page six of the business section. Both banks carry modest exposure to that sector. The credit desks at the core-node banks run their internal reviews. For Ridgemont, the review triggers a limit reduction, perhaps from fifty million to thirty million, but the relationship survives because the core-node banks have enough history with Ridgemont's treasurer, and enough diversified exposure, that they are willing to stay in at a lower level. For Lakeview, one of the four local regionals, itself under pressure, pulls its line entirely. A second quietly halves its limit. Lakeview has now lost roughly forty percent of its overnight capacity without a single public event occurring.
Now the feedback loop begins. Lakeview, facing a funding gap, starts bidding slightly above the prevailing rate to attract volume. Counterparties notice. In a market built on relationship pricing, paying up signals desperation as surely as a raised voice in a library signals panic, and the remaining two counterparties flag Lakeview for internal review. One cuts its line as a precaution. Lakeview is now, in practical terms, frozen out of the market it was transacting in seventy-two hours earlier, while its publicly reported capital ratios remain technically adequate.
Ridgemont weathers the same stress event with reduced but functional access. The difference was not the balance sheet. It was the network position.
What People Misread About the Rate Signal
The standard assumption, reasonable enough on the surface, is that rising overnight borrowing rates serve as the early warning sign of a bank losing access. Rates go up, stress becomes visible, regulators and counterparties respond. This is how the mechanism is supposed to work.
It often does not work that way. A bank losing access does not always pay a higher rate. Sometimes it simply stops transacting. If a bank's bilateral limits are cut below the threshold at which it can fund its needs, it exits the market as a borrower before it ever has to post an embarrassing rate. The stress is invisible in the price data. It shows up instead in volume data, specifically in the collapse of a bank's borrowing volume, which is far less scrutinized in real time and sometimes only reconstructed after the fact, when reconstruction is of limited use to anyone.
This is where the literature is admirably candid: rate-based monitoring of overnight markets is a lagging and incomplete signal for peripheral banks. By the time a regional bank's stress appears in the rate, the network has often already made its judgment. The verdict precedes the evidence, at least as the evidence is conventionally measured.
Regulators have known this for some time. Post-crisis reforms in several jurisdictions pushed for greater transaction reporting in overnight markets precisely to capture volume signals alongside price signals. But data aggregation lags, and the bilateral nature of most interbank lending means a complete picture of network position remains difficult to assemble in real time. The architecture of the solution has not quite caught up with the architecture of the problem.
Relationship Capital Is the Actual Reserve
Here is the judgment that the evidence supports, even if it makes treasury departments uncomfortable: the overnight market is not a commodity market where access is determined by your credit rating on any given morning. It is a relationship market where access was determined by decisions made two, five, and ten years ago. Treating it otherwise is not a cost-saving strategy. It is a slow-motion withdrawal from a market you will one day need urgently.
Banks that diversified their counterparty base geographically, that maintained lines with core-node institutions even when cheaper local alternatives were available, and that invested in the unglamorous work of relationship maintenance during calm periods consistently appear in the network literature as more resilient during stress episodes. Not immune. More resilient. The distinction matters, and conflating the two has contributed to a number of avoidable crises at institutions that believed their capital buffers were doing work that only relationships could do.
The banks that lose access first are, almost without exception, the ones that optimized for cost during good times and treated relationship maintenance as overhead. When the credit desks at the core nodes run their stress-period reviews, they are not simply reading balance sheets. They are asking how long they have known this counterparty, how reliably it has behaved, and whether the relationship is worth the risk at the margin. A bank that has been a consistent, professional participant in that market for years gets the benefit of the doubt. A bank that only showed up when it was convenient does not. Can anyone with experience of how institutions actually behave find that surprising?
The overnight market, when you examine it honestly, functions less like a water tap and more like a ledger of favors, written in pencil, kept by people with long memories. The terms are always subject to quiet revision. And the revision almost always arrives first at the banks that assumed the tap would simply keep running because it always had before.