Game Theory · GT-32
A threat only constrains a rational opponent if carrying it out would actually be in your own interest when the moment comes.
A threat is credible if the threatening party would genuinely still want to carry it out, given their own payoffs, at the exact moment they'd need to. A threat is non-credible if, once the bluff is called, the threatening party would rationally prefer not to follow through — meaning a sophisticated opponent should correctly ignore it, regardless of how forcefully it's delivered.
The formal distinction is central to Thomas Schelling's 1960 analysis and is operationalized precisely through Selten's subgame-perfect equilibrium concept, which mathematically identifies exactly which threats survive rational scrutiny.
The Mechanism
Two threats, one that survives scrutiny and one that doesn't
The test is entirely mechanical, not rhetorical: would the threatener genuinely prefer to carry out the threat, given their own payoffs, at the exact moment the bluff is called? If yes, it's credible and should deter. If no, a rational opponent should call the bluff and proceed as if the threat doesn't exist.
01 · VOLUME AND CONFIDENCE ARE NOT CREDIBILITY
A common and costly confusion in real negotiations
How forcefully or confidently a threat is delivered has no bearing on whether it's actually credible — credibility is entirely a function of the threatener's own incentives at the moment of truth, not their tone. Sophisticated counterparts learn to evaluate threats on this basis and ignore the delivery entirely.
02 · COMMITMENT DEVICES ARE THE STANDARD FIX FOR NON-CREDIBLE THREATS
Converting 'wouldn't want to' into 'can't avoid'
A threat that wouldn't rationally be carried out can sometimes be converted into a credible one by removing the threatener's own future flexibility to back down (see Commitment Devices) — publicly locking in consequences, delegating enforcement to a third party who has no discretion, or structuring irreversible contractual triggers all convert an otherwise idle threat into a genuinely credible one.
03 · REPUTATION EXTENDS CREDIBILITY ACROSS REPEATED INTERACTIONS
A single incident's cost can be outweighed by future reputational value
A threat that would be individually costly to carry out in one specific instance can still be rationally credible if the threatener interacts repeatedly with many counterparts and values their reputation for following through — carrying out a costly threat once can be worth it purely to make all future threats credible against everyone else who's watching.
Where It Fails / Inversion
Where it fails / inversion
Real counterparts don't always correctly evaluate credibility — psychological factors (anger, loss aversion, a desire to appear consistent) can lead people to carry out threats even when it's individually costly and irrational to do so in the moment, meaning some 'non-credible by the math' threats are followed through anyway. Relying purely on the formal credibility calculation to call every bluff carries real risk when the other party isn't behaving as a pure rational actor.
How To Use It
Worked example · calling a supplier's threat to cut off supply
A key supplier threatens to cut off your company's supply entirely unless you accept a price increase. Before capitulating, evaluate credibility directly: would the supplier genuinely prefer to lose your business (and any associated reputation among other customers who might hear about it) rather than negotiate, or is this a bluff aimed at avoiding a harder negotiation? If losing your account would meaningfully hurt the supplier's own revenue and reputation more than a negotiated compromise would, the threat is likely non-credible, and calling it (offering a firm counter, or credibly signaling you have alternative supply lined up) is the rational response.
How to use it
Before responding to any threat in a negotiation, run the credibility test explicitly: would the other party genuinely prefer to carry this out, given their own true incentives, if you called their bluff right now? If the honest answer is no, don't let confident delivery or emotional pressure substitute for that analysis — call it. If the answer is yes, take it seriously regardless of how it's delivered.
See Also