1.32 Sharpe. 242 signals. A Bayesian fix to a framework everyone uses wrong.

Parametric Portfolio Policies (PPP) let you skip return modelling and go straight from signals to weights. Elegant — but there's a hidden cost: PPP treats estimated parameters as truth, ignoring estimation risk entirely.

This paper proves the consequences are not minor:

• PPP overstates expected utility — always,

• The overconfidence is worst exactly when signals are strongest,

• And the welfare cost grows with risk aversion — hurting conservative investors most.

The fix? Bayesian PPP (BPPP): place a prior on the policy and integrate over the posterior. No ad hoc penalties — just Jensen's inequality doing its job.

Results over 50 years with 242 signals and 6 factors:

• Sharpe: 1.32 (BPPP) vs. 1.05 (PPP) vs. 0.74 (market),

• At 50 bps costs: 0.99 vs. 0.60,

• Max drawdown: -24.5% vs. -37.2%,

• Bonus finding: factor-timing predictability is dense, not sparse — most signals contribute.