Bayesian Finance
Frequentists ask how often. Bayesians ask what to believe — and then update it the instant the data lands. This is the math that turns a vague hunch and a noisy backtest into a sharp, honest, evolving estimate, and the discipline that stops a flashy signal from fooling you into mistaking a 90%-accurate test for a 90%-real edge.
Probability as degree of belief, and the machinery for updating it with evidence — prior, likelihood and posterior; Bayes' rule and the base-rate fallacy; conjugate priors and sequential updating; precision-weighted return and volatility estimates; shrinkage and the Black–Litterman model; credible intervals versus confidence intervals; and MCMC for posteriors no formula can express.
You can’t rerun 2008. You get one noisy history, a head full of hunches, and you have to act anyway — which is exactly where frequentist “how often?” probability runs out of road and the Bayesian “given what I knew and what I just saw, what should I believe now?” takes over. This course is the math that turns a vague hunch and a noisy backtest into a sharp, honest, ever-updating estimate.
Here’s the apparatus you’ll build, from intuition to real practitioner tools:
- Prior, likelihood, posterior — the three objects every Bayesian argument is made of, and the rule that fuses what you believed before with what the data says.
- The base-rate fallacy — the most expensive mistake in quant trading: confusing P(data given hypothesis) with P(hypothesis given data). Reason in natural frequencies (“of 1000 signals…”) and watch a 90%-accurate test still spit out mostly false alarms.
- Conjugate priors & sequential updating — turn updating into one-line arithmetic: a Beta belief about a win rate sharpens by just adding wins and losses, where today’s posterior is tomorrow’s prior.
- Precision-weighted estimates — the Normal–Normal update, where a posterior mean is a precision-weighted blend of prior and data: noisy data barely nudges you, abundant data takes over.
- Shrinkage & Black–Litterman — the James–Stein result that pulling noisy estimates toward a common center beats trusting them, and the model that blends market equilibrium returns with investor views to tame the “error-maximizing” optimizer.
- Credible intervals vs. confidence intervals — finally say “there’s a 95% probability the true value is in this range” and be right, with the line between the two drawn cleanly.
- MCMC — sample a posterior with no closed form by wandering: propose, compare, accept-or-stay, repeat, and the histogram is the posterior.
This is the expert capstone of the quantitative ladder: by the end you’ll think in priors and posteriors by reflex, size your confidence to your evidence, and never again mistake a loud signal for a true one.
In this topic
- 1 What Bayesian Thinking Is Probability as degree of belief: how a prior, a likelihood, and Bayes' rule combine into a posterior — and why updating beliefs with evidence is the natural language of trading and risk. 9 min
- 2 Bayes' Rule & Base Rates Bayes' theorem worked in full: the formula, natural-frequency reasoning, the base-rate fallacy, and the prosecutor's fallacy that makes traders mistake a flashy signal for a real edge. 10 min
- 3 Conjugate Priors & Sequential Updating The Beta–Binomial conjugate prior made concrete: how a belief about a strategy's win rate updates trade-by-trade in closed form, why conjugacy is so convenient, and how a prior acts like pseudo-data. 10 min
- 4 Updating Return & Volatility Estimates The Normal–Normal conjugate update: how a prior on an asset's expected return combines with noisy sample data as a precision-weighted average, why more data and lower variance mean more weight, and how Bayesian estimates of volatility behave. 10 min
- 5 Shrinkage & Black–Litterman Why pulling noisy estimates toward a center beats trusting them: James–Stein shrinkage, shrinkage covariance estimators, and the Black–Litterman model that blends market-equilibrium returns with an investor's views as a Bayesian posterior. 10 min
- 6 Credible Intervals & MCMC Reading the whole posterior: what a Bayesian credible interval really says (and how it differs from a frequentist confidence interval), and how Markov-chain Monte Carlo draws samples to approximate posteriors no formula can express. 10 min
- 7 Bayesian Finance — Final Exam The graded final exam for Bayesian Finance: prior, likelihood and posterior; Bayes' rule and base rates; conjugate priors and sequential updating; precision-weighted return estimates; shrinkage and Black–Litterman; credible intervals and MCMC. 15 min
Mark course as finished
Done with every lesson? Lock it in — your progress is saved on this device.