I ran on one bank. Why did three fail?

Kay and Jo run a depositor run on one bank, then sell its loans in a hurry. The strip is below, and the steps to repeat it in Konjugate are under it.

Transcript

Four panels. Kay, a round teal character, and Jo, a dry blue one, are joined by a dotted line. Kay asks: A bank only fails when it runs out of money, right? Jo answers: Or when everyone wants theirs back at once. Step 2, Scenario. A card reads: run on one bank, 8,615 million lost, nobody fails. Kay says: Phew. Jo says: Now sell its loans in a hurry, at half price. Step 3, Results. A card reads: the same run with half-price sales, 3 of 8 banks fail, 49,339 million lost. Kay says: Three? I only ran on one. Jo says: Losses travel along the edges. Step 4, Explore. A scale of the forced-sale discount, green below 18.5% and red above it, marks where the first bank fails. Kay asks: How sure are you? Jo says: Halve my reserve target and nobody fails. It is a guess.

Image for sharing

What just happened

The sample data in Konjugate's Fintech toolbox has eight banks. A depositor run on Alder Bank, withdrawing for eight days, costs the system 8,615 million of equity and nobody fails. Run it again with a forced-sale discount of 50% and 3 of 8 banks fail (Alder, Holly and Fir), with 49,339 million lost. Losses spread because a bank that fails passes its shortfall on to the banks it owes, and because forced sales push prices down for everyone.

Then Explore asks how fragile that is. Alder Bank first fails at a forced-sale discount of about 18.5%, and one assumption in the model moves the answer more than any other: the reserve target, the cash a bank wants to hold. Halve it and the damage falls by 86% with nobody failing. The reserve target is an assumption of the model, not a measurement, which is why Jo calls it a guess.

Repeat it in Konjugate

1. Install the Konjugate Fintech Toolbox from the Discover tab of Extensions ("Installing add-ons and plugins" shows how). Press Fintech in the toolstrip and open Interbank Stress. It needs the desktop app, because add-ons are not in the browser edition.
2. On Your data, use the sample data. On Scenario, choose Depositor run and Alder Bank, and run it. The result is 8,615 million lost with nobody failing.

3. Go back to Scenario and choose Depositor run with a severe asset-price shock, the same run with forced sales fetching half of book value. On Results, 3 of 8 institutions fail and 49,339 million is lost.

4. Open Explore, then Find the breaking point, and press Find it. It searches the forced-sale discount for the point where Alder Bank first fails.

5. Still in Explore, open Which assumptions matter and press Check every assumption. Each assumption is halved and raised by half, one at a time, and the bars show how much the answer moves.

What these numbers do and do not say

The figures come from the toolbox's sample data and were produced by the run described above on 2026-10-16. They are illustrative. The model is not calibrated to any real bank, and the interbank losses spread in a stylized way, not through a solved clearing process. The useful question is not "how many banks fail" but "how much does the answer move when an assumption does", and Explore exists to answer that.

Go deeper

• "Introducing the FinTech toolbox" explains what the toolbox models.
• "Forking a run to try a what-if" shows how to branch one run and compare a change in Konjugate itself.
• "Installing and updating Konjugate" covers getting the desktop app.