What Is The Worst This Position Can Cost Me?

Most position-sizing math starts from an expected return and a volatility estimate. A useful complementary question, cheap to ask and easy to skip, is simply: if everything about this trade that could go wrong does go wrong at once, how much do I actually lose? Not the 95th-percentile loss from a model, but the genuinely worst plausible outcome — the stop doesn't fill, the hedge doesn't hold, liquidity disappears exactly when you need to get out. Sizing a position so that number is survivable is a different discipline from sizing it so the expected outcome looks good.

Why expected loss isn't the same question

A position can have an attractive expected value and still be sized in a way that ruins you, because expected value averages over outcomes you might never actually face in sequence — a strategy that wins small most of the time and loses catastrophically rarely can have a great average return and still blow up the one time the rare loss hits. Worst-case thinking asks a narrower, more concrete question: strip out the probability weighting entirely and just look at the single worst number. If that number is one you can absorb without changing your business, the position is sized fine regardless of how attractive the average outcome looks. If it isn't, no amount of favorable expected value justifies the size.

A trader selling out-of-the-money options for premium income had a model showing a strongly positive expected return per trade, with losses occurring only in the tail. Asked "what's the worst case," the honest answer was that a large enough gap move could produce a loss several multiples of the total premium collected over the entire program to date — a number the expected-value framing had never surfaced because it was weighted by a small probability. Resizing the position down until that worst case became merely painful instead of business-ending left the expected return per trade almost unchanged, because the probability of the bad outcome was small, but it changed what happened to the trader on the one day it occurred.

The question is deliberately blunt and doesn't require a probability distribution to answer — that's the point. It's a sanity check that sits alongside, not instead of, formal risk models.