EV, Bayes, Kelly: three tools that cannot rescue a bad estimate

Put a formula next to a trade and it suddenly looks more scientific. The weak point usually sits inside the formula.

For a simplified binary share, expected value depends on your estimated probability and the price paid. If the estimate is wrong, precise arithmetic produces a precise answer to the wrong question.

Update beliefs with actual evidence

Bayesian thinking asks how new evidence should change a prior belief. It does not mean treating every fresh headline as decisive.

Record what you believed before the update, what new information arrived, and how reliable that information is. Repeating the same report through several outlets is not several independent observations.

Sizing magnifies estimation errors

Kelly-style sizing is highly sensitive to the probability input. A small overestimate of an edge can recommend far more exposure than the evidence supports.

Real markets also involve costs, correlated positions, changing quotes, and limited depth. A simple single-trade formula does not automatically capture those constraints.

Keep a calibration record

Group past estimates into probability bands and compare them with outcomes. Events you called 70% should succeed around that rate over a suitable sample, with considerable short-run noise.

Use formulas to make assumptions explicit and to stress-test them. Do not use them as decoration around a gut feeling.

The most valuable number in the calculation may be the uncertainty around your own estimate.

Further reading: Prices and order books · Resolution rules · Current fees.

Explore the markets: Open Polymarket (referral link).