Maethel

Essays find a mechanism, test whether it runs elsewhere, and say what would prove it does not.

Most Bad Outcomes Are Rational

How individually sensible decisions can produce a collective failure, and what would make that explanation wrong.

Tufi Sele ·

On 9 March 2023, depositors withdrew over US$40 billion from Silicon Valley Bank. Another US$100 billion was expected to leave the next day, according to the Federal Reserve’s review.

For a depositor worried about getting their money back, withdrawing was sensible. On my reading, the withdrawals also made staying more dangerous.

That is the problem I want to pull at. A decision can make sense for the person making it while helping produce a failure for the group. Understanding that gap might tell us more than deciding everyone involved was greedy or stupid.

The biggest assumption is that the decision really was sensible given what the person knew and could do. A bad result alone cannot establish that.

What the outcome tells us

When a system produces a bad result, it is tempting to work backwards. Someone must have wanted it. Or someone failed to understand what they were doing.

Both explanations deserve a hearing. But they carry an assumption: the collective outcome reflects somebody’s intention, or somebody’s mistake.

What if the outcome is assembled from decisions that make sense separately?

The bank run makes the distinction easier to see. Wanting to protect a deposit is different from wanting a bank to fail. Those intentions can diverge even when the same action contributes to both outcomes.

Agreement that failure would be bad does not settle the next decision. A depositor still has to decide whether to leave their money there while others withdraw.

How the decisions connect

My working explanation is a short chain.

A depositor fears that others will withdraw. Getting out looks safer than waiting. Their withdrawal adds to the outflow. That gives the remaining depositors another reason to worry, which makes leaving more attractive.

Each decision changes the conditions for the next one.

A bank run feeds itself: fear leads to withdrawal, withdrawal adds to the outflow, and the outflow gives the remaining depositors another reason to worry. A loop of three steps, read from the top. Fear that others will withdraw makes getting out look safer than waiting. Withdrawal adds to the outflow. The Federal Reserve review reports an outflow of over US$40 billion on 9 March 2023 and another US$100 billion expected the next day. The outflow gives the remaining depositors another reason to worry, and the loop starts again. The loop is the author's working interpretation; only the two figures are the Federal Reserve's. One proposed test: the account is wrong if someone gains from the outcome and can preserve the rules that produce it. The chain paragraph above the figure says the same in words. HOW A RUN FEEDS ITSELF Fear that others will withdraw makes getting out look safer than waiting Withdrawal adds to Outflow gives the remaining depositors another reason to worry Over US$40bn out on 9 March 2023 US$100bn more expected next day Fed review, pp. 7, 18 WRONG IF TEST 1 OF 4 someone gains from the outcome and can preserve the rules that produce it. Proposed test. No forecast logged yet.
The loop is my reading.

The Federal Reserve’s review describes social media fuelling the run among a tight network of depositors. That fits the chain: information about other people’s behaviour can alter the decision in front of you.

It does not establish that this chain explains every part of the bank’s failure. The run is the narrow problem here. Explaining why the bank was vulnerable would require a fuller account.

This is what I mean by locally rational: a choice that makes sense within someone’s information, incentives and constraints. It is not a claim that the choice is fair, harmless or correct with hindsight.

The question becomes: what makes a damaging action the sensible option, and who bears the consequence?

Agreement and the ability to act

This problem is older than a bank run.

Hume used the example of draining a meadow. Two neighbours can arrange to do it together; coordinating a thousand people is a different problem. The example separates seeing the benefit from organising the action.

That distinction matters. If I assume agreement is enough, continued failure looks like evidence that people do not care. If agreement cannot organise the work, more persuasion may leave the obstacle untouched.

I would want to ask who must act, what acting costs them, and what assurance they have that others will follow.

There is existing work to learn from here. Ostrom’s work on governing shared resources belongs beside an account of collective failure.

The tragedy of the commons describes how individually sensible use can exhaust a shared resource. But that outcome is not inevitable: an open-access resource is not the same thing as a governed commons.

So a shared resource should start an investigation into its rules. It should not end the investigation with a prediction of collapse.

Where the mechanism travels: four relations from the bank run, checked against Hume's meadow and Ostrom's commons. A grid. Rows are four relations in the author's reading of the bank run: each act adds to the shared loss; the loss gives others reason to act; an outcome none of them chose; rules change the outcome. Columns are the three cases the essay names: the bank run, Hume's meadow and Ostrom's commons. A dashed cell holds the author's reading in the essay's words; a square names a cited source (Hume, Treatise 3.2.7; Cox 1985). A cell reading 'Not stated in the essay' is a relation the essay does not state for that case: the return (the loss gives others reason to act) is stated only for the bank run; rules changing the outcome is stated only for the meadow and the commons. One proposed test: the account is wrong if new people under the same rules produce a different outcome. WHERE THE MECHANISM TRAVELS Bank run Hume's meadow Ostrom's commons Each act adds to the shared loss Adds to the outflow Not stated in the essay Use draws the resource down The loss gives others reason to act Others have another reason to worry Not stated in the essay Not stated in the essay An outcome none of them chose Staying becomes more dangerous The work does not get organised The resource is exhausted Rules change the outcome Not stated in the essay Two neighbours arrange it Hume A governed commons is not open access Cox 1985 WRONG IF new people under the same rules produce a different outcome. Then the people mattered more than this account allowed. Test 4 of 4. Proposed. No forecast logged yet.
Three cases share one sensible choice producing group failure; the return is stated only for the bank run. Dashed cells are my reading, a square names the source, and a cell reading Not stated is where the transfer breaks. All four tests.

[ What would prove this wrong ]

First, I need to look for beneficiaries with power. If someone gains from the outcome and can preserve the rules producing it, calling the result unwanted would hide part of the explanation. This is the strongest rival to my account. I should ask who wanted it before assuming nobody did.

Second, I need to separate incentives from error. If people acted on mistaken information, correcting that information may change behaviour without changing the reward. That would weaken the incentive account.

Third, “rational” must not become a word that explains any behaviour after it happens. I would write down the predicted response to a specified change in costs or rewards beforehand. If behaviour repeatedly goes the other way, the model needs revising. I would also need to check whether the change weakens a norm that was helping people cooperate.

Finally, replacement is a useful question, but not a complete experiment. If new people produce different outcomes under the same rules, the people mattered more than this account allowed. If the outcome persists, I still need to distinguish incentives from shared errors or other constraints.

These are proposed tests. I have not logged a forecast for this yet. I would rather keep that gap visible than turn an explanation into evidence for itself.

Where I land

I am keeping the narrower idea: individually sensible decisions can point towards a bad collective outcome.

The title is the claim I am testing. The evidence here supports that narrower possibility in bank runs and shared-resource problems. It does not establish that this explains most bad outcomes.

For now, the useful question is what would keep producing the outcome if the people changed. That gives me somewhere to look: the rewards, constraints and expectations attached to a decision.

It also gives the explanation a job to do. If it identifies the mechanism, it should help predict what happens when that mechanism changes. Otherwise I have a description that fits the past, with no reason yet to trust it about the next case.

Sources and claim map

  • Federal Reserve SVB review, pp. 7, 18 and 38: opening figures and social-media account; the withdrawal chain is my working interpretation.
  • Hume, Treatise 3.2.7: the meadow example.
  • Ostrom’s Nobel lecture: the governing-commons research direction.
  • Cox (1985), No Tragedy on the Commons, pp. 53-56: the English commons had restricted access and rules governing use, unlike an open-access resource.
  • Feeny et al. (1990), The Tragedy of the Commons: Twenty-Two Years Later, pp. 4-5 and 12-13: the open-access distinction and rejection of inevitable failure across shared-resource arrangements.
  • Local decisions, collective outcomes and the assumptions examined here are my working account. Beneficiaries, error, norms, replacement and payoff changes are rivals and proposed checks, not claimed findings or completed forecasts. The title’s “most” is the claim under test, not a finding established here.

This idea was developed through an ongoing dialogue with AI. AI was used to challenge the argument, locate related research and test alternative explanations. The conclusions are mine unless otherwise stated.