The Credibility Line: Why Rules Survive Only Where Overrides Are Seen

Part 24 of the Fat Tail Notes series — the price of breaking a promise.

The hook

Part 23 priced the policy loophole and found the cliff: a 10% exemption expectation eats 64% of the backstop's damage. But that model treated the exemption rate p as something the market just expects. This part asks the question the series has been walking toward since Part 18: why is p small at all? Why does a policymaker not override the rule?

The answer is that exemption is not free. It has an expected cost — the probability that it is seen, times the punishment when it is, times how much the policymaker values the credibility it destroys. Change the price of exemption and you change the exemption rate. Commitment is not a matter of will. It is a matter of price.

This part makes p endogenous and finds three things. There is a credibility line — a threshold below which exemption is free and the promise is empty — and institutions move that line. The spiral is self-reinforcing: once credibility dips, exemption cheapens itself. And the vaccine is not punishment but visibility: a severe punishment that never fires protects nothing, because the exemption was never seen.

The machine

The policymaker compares the private benefit of exemption b (political temptation: saving the day, keeping a position, avoiding short-run pain) against its expected cost:

p* = min(1, b / (v × s × C))

• v = visibility: the probability that an exemption is seen by the market;
• s = punishment: the credibility loss when it is seen;
• C = current credibility. The expected cost of one exemption is v·s·C — proportional to the credibility it destroys. High credibility makes exemption expensive; low credibility makes it nearly free.

Effective capacity is then C_eff = C × (1 − p*), and the market prices tail losses at C_eff (the P4/P6 machinery, 2,000 paths per level, seed 20260921).

Two structural facts fall out immediately, before any numbers:

The credibility line. C_crit = b/(v·s). Below C_crit, p* = 1: exemption is free, there is no reason left to keep the promise. The line is the institutional position of Part 23's cliff.

The spiral is endogenous. p* rises as C falls — the cost of exemption falls with the credibility it destroys. A wounded reputation invites more exemptions, which wound it further.

The anchor: reputation and institutions are the same price

Kreps & Wilson (1982). The chain-store paradox said a monopolist can never credibly promise to fight entry — backward induction kills it. The paradox dissolved when Kreps and Wilson added a tiny uncertainty about the monopolist's "type": if there is even a small probability that he is irrational enough to fight, then it is rational for a normal monopolist to fight early to build a reputation — because the reputation has future value. Reputation makes commitments credible under imperfect information, and the mechanism is price: fighting is expensive today and cheapens tomorrow's entry.

North (1993 Nobel) and North-Weingast (1989). Institutions are devices that make sovereign commitments credible. After the Glorious Revolution, the English crown bound itself — Parliament controlled taxation, the king could not unilaterally renege — and the result was a fall in borrowing costs and the birth of a standing public debt market. The sovereign did not become more honest; the price of reneging went up. Credibility is an institutional property, not a personal one.

The rhetorical override. Draghi's "whatever it takes" (2012) was followed by OMT — conditional, auditable, never activated. The promise was credible because the override would have been seen: visibility is what made the price of exemption higher than its benefit.

Finding 1 — the credibility line is institutional

Critical threshold C_crit = b/(v·s), with b = 0.10:

v·s	0.10	0.25	0.50	1.00	2.00
C_crit	1.00	0.40	0.20	0.10	0.05

The line is set by the institutional environment, not by the policymaker's character. At v·s = 0.10 (opaque environment, weak consequences) the credibility line sits at 1.00 — the entire credibility space is free-exemption territory. At v·s = 2.00 (transparent, audited, consequential) the line drops to 0.05 — even a nearly-broken reputation still holds its promise. The same policymaker, the same temptation, different institutions: one keeps his word, the other does not. You do not trust the person. You trust the line.

Finding 2 — the spiral is self-reinforcing

Endogenous exemption rate p* per discipline level:

C	v·s=0.10	v·s=0.25	v·s=0.50	v·s=1.00	v·s=2.00
1.00	1.00	0.40	0.20	0.10	0.05
0.75	1.00	0.53	0.27	0.13	0.07
0.50	1.00	0.80	0.40	0.20	0.10
0.25	1.00	1.00	0.80	0.40	0.20
0.00	1.00	1.00	1.00	1.00	1.00

Read the v·s=0.25 column. At C=1.00 the exemption rate is 0.40. At C=0.50 it is 0.80. At C=0.25 it is 1.00 — the promise is empty. Exemption cheapens itself: every bit of credibility destroyed makes the next exemption cheaper, which destroys more credibility. This is the institutional reason behind Part 23's "cliff in the first mouthful": the first exemption brings C near the credibility line, and below the line the remaining exemptions arrive on their own. There is no halfway healing — once the spiral starts, it runs to the line.

Finding 3 — visibility is the vaccine, punishment is only half of it

Effective capacity C_eff = C × (1 − p*):

C	v·s=0.10	v·s=0.25	v·s=0.50	v·s=1.00	v·s=2.00
1.00	0.00	0.60	0.80	0.90	0.95
0.75	0.00	0.35	0.55	0.65	0.70
0.50	0.00	0.10	0.30	0.40	0.45
0.25	0.00	0.00	0.05	0.15	0.20
0.00	0.00	0.00	0.00	0.00	0.00

At v·s = 0.10 — say, v = 0.2 and s = 0.5: exemptions are rarely seen but heavily punished — the entire credibility space leaks. Even a flawless reputation (C=1.00) collapses to C_eff = 0. The punishment never fires, because the exemption is never seen. Severity protects nothing when visibility is zero. The institutional lesson of OMT is not that the ECB was willing to punish itself; it is that any override would have been seen, and seen instantly — conditionality, program audits, market scrutiny. Visibility is the vaccine; punishment is the booster that makes it work.

What the market pays

Bare-market p1 priced at the endogenously-credible capacity:

C	v·s=0.10	v·s=0.25	v·s=0.50	v·s=1.00	v·s=2.00
1.00	−44.1%	−59.3%	−54.2%	−51.3%	−49.4%
0.75	−44.1%	−63.1%	−60.6%	−58.4%	−57.6%
0.50	−44.1%	−53.2%	−63.4%	−62.6%	−62.0%
0.25	−44.1%	−44.1%	−48.5%	−58.0%	−62.2%
0.00	−44.1%	−44.1%	−44.1%	−44.1%	−44.1%

Two readings. First, the credibility-death world (v·s=0.10) is "calm" — p1 of −44.1% everywhere — but that calm is the C=0 world of Part 18: the market stopped believing, stopped levering, and prices the world with no backstop at all. Safety bought by killing the guarantee is not safety.

Second, look at the v·s=0.25 column — an environment where exemptions are rarely seen and punishment is weak. From nominal C=1.00 to C=0.25, p1 runs −59.3% → −63.1% → −53.2% → −44.1%: it deepens as credibility erodes, then lightens as C_eff falls through the credibility line into the no-backstop world. The cliff of Parts 18 and 23 appears again, and now its position is visibly institutional: the same market, the same temptation, different discipline — different tail.

And the steepest part of the cliff is even earlier than Part 23 showed. The v·s=2.00 row prices a 5% exemption rate: C_eff=0.95, p1 −49.4% — 20 of the 34.3 points of cliff damage already gone at 5% exemption. Part 23 said 64% of the damage sits in the first 10% of the loophole. It is worse: most of it sits in the first 5%. The credibility line is not approached gradually — the market falls toward it in the first visible doubt.

What the retail investor should do

1. Evaluate a backstop by its visibility, not its promises. The first question about any "we stand ready" is not how much it can buy — it is how fast an override would be seen. Independent audits, published reaction functions, conditional programs, free press, market monitoring: these are the actual vaccine. A promise with no visibility is a promise with v≈0, and v≈0 leaks the whole credibility space regardless of punishment.

2. Trust the line, not the person. North's lesson: credibility is an institutional property. The reliable backstop is not the one led by the most honest-sounding official; it is the one whose reneging is expensive. When the institutional line sits high (little transparency, weak constraints), price the market as if the promise may be empty — because below the line, it is.

3. Price the first doubt, not the full fall. The cliff's steepest step is its first: 5% exemption already eats 20 points of the worst 1%. Half-believing is the most dangerous state (Part 18), and a backstop that is credible but unobserved is precisely half-believing — the market prices the guarantee, and the guarantee is already leaking.

4. The spiral is the real risk. Credibility is a stock, not a stance. Every visible exemption draws down the stock, and below the line the drawdown accelerates. When assessing tail risk, do not ask "will they hold this time" — ask "how much credibility remains above the line".

The last garment

The retail garment (Parts 19–22), the policy backstop (Parts 23–24) — the series has now priced the whole stack: mechanism, behavior, commitment, exemption, and the institutions that keep promises. What is left is the question the first part raised and every part since has deferred: when the backstop is credible, the market leans on it — and the leaning itself changes the tail. The moral-hazard loop of Part 4 returns as the final act. That is the next part.

---

Code: crash_simulator_v10/policy_credibility.py (V10-P12), endogenous exemption on the Part 23 machinery, self-test and Monte Carlo included. Deterministic, 206 lines, no dependencies beyond the standard library.

Drafted with AI assistance; facts, figures, and errors are the author's own.

Currently available for freelance work — Python pipelines, quantitative risk tooling, and AI data automation. Reach me at [e-mail].

Photo by Giorgia Finazzi on Unsplash