A fixed-strategy bot is a dead bot walking. Markets drift. Volumes shift. Volatility regimes rotate. But if you change parameters every week, you’re curve-fitting in real time.
The solution is Bayesian updating. It’s a mathematical way of saying: I have a prior belief. I see new data. I update my belief but only in proportion to how much evidence I’ve seen.
Here’s how a model does it:
· Prior: The strategy’s long-term EV from the backtest period.
· Evidence: Each new trade’s outcome.
· Posterior: A weighted average, where the prior gets less weight as the sample grows.
After 5 new trades, the prior dominates. The model doesn’t flinch. After 50, the live data has real influence. After 100, if the live EV is significantly below the prior, a “Signal Freshness” yellow flag triggers, the model doesn’t stop trading, but reduces exposure until the evidence stabilises.
Bayesian thinking means never saying “the edge is dead” from a handful of losses, and never ignoring a real structural shift. The model walks that line every day.
I’ve put together a short PDF on Bayesian reasoning for traders no math degree required. Comment ‘BAYES’ and I’ll DM it.