Bankroll management: loss limits as the lesser irrationality
Bankroll management: loss limits as the lesser irrationality
One-line summary: Size risk to edge, volatility, and risk of ruin (Kelly). Humans overestimate edge and underestimate vol — and, worse, reload while on tilt. A precommitted loss limit is itself irrational for a rational actor, but it is less damaging than letting an aroused limbic system decide whether the losing streak is luck or skill.
What it is
Duke's poker definition of managing a bankroll: know your edge, know the volatility, know risk of ruin, and keep session risk inside what the total bankroll can survive ("essentially you're buying Kelly"). Two failure modes:
- Calibration of the inputs. "We generally overestimate our edge... We also underestimate the volume... we generally think we need a lot less money than we do, and then we're suddenly surprised that we're broke." The cold calculation can be brute-forced off-table.
- In-the-moment reload. Running out of the session stake means you have been losing, which means tilt: "my limbic system is now lit up" and the story becomes "I'm so unlucky... These people are really bad." That is the moment people put more money in.
Tilt (poker borrowing the pinball term): emotional arousal "shuts down" prefrontal reasoning "in the same way that a pinball machine just won't work anymore." Chip-by-chip P&L is both the training signal (every micro-decision has upside and downside) and the tilt engine (ticker-watching).
The loss-limit move is what she calls "stacking irrationalities": a purely rational actor would bet whenever they have edge. You are not that actor. Binding yourself to "this is how much I'm allowed to lose, period," and being accountable to a group (group-decision-hygiene) for exceptions, is less ruinous than the luck-vs-skill self-diagnosis while aroused — which otherwise spirals into the next day still "trying to get my money back."
Why it matters to psychology
This is dual-process / arousal-and-reasoning (PFC offline under emotion) plus a precommitment strategy (Ulysses contract) justified as the lesser of two biases. It is personal decision-making under uncertainty, not a market mechanism. The same pattern shows up as "I'll just this once" after a drawdown in any domain with a bankroll analogue (time, attention, reputation).
Evidence
- annie-duke in 2026-08-03-podcast-capital-allocators-best-of-decision-making-annie-duke-2018: "when you're emotionally lit up, it's like your brain is shaking and it shuts down. The prefrontal cortex shuts down your ability to reason in the same way that a pinball machine just won't work anymore."
- annie-duke in 2026-08-03-podcast-capital-allocators-best-of-decision-making-annie-duke-2018: "You recognize what your edge is, you know what the Vol is, so you understand what your risk of ruin is. And you make sure that however much money you have at risk in any given session is essentially you're buying. Kelly."
- annie-duke in 2026-08-03-podcast-capital-allocators-best-of-decision-making-annie-duke-2018: "We generally overestimate our edge. Let's be honest, we think we're better than we are. We also underestimate the volume."
- annie-duke in 2026-08-03-podcast-capital-allocators-best-of-decision-making-annie-duke-2018: "I know it's irrational to have a loss limit, but that irrationality is a lot less damaging to me in the long run than allowing myself to make these irrational decisions about whether it's a luck or skill issue that's causing me to lose right now, so that I press my position in places where I never should."
Implications
- Precommit the stop while unemotional; make exceptions reportable to someone who will not buy the hard-luck story.
- Do not treat "I still have edge, so I should reload" as available information while on tilt — that assessment is the thing the limit is protecting you from.
- Tilt is not only downswings; Duke notes upswings over-arouse too.
Open questions
- When is a loss limit just another resulting error (stopping because of a noisy P&L path) vs. a justified arousal hedge?
- How well does the Kelly analogy travel to non-monetary bankrolls (time, health) where "vol" is harder to estimate?