Recent asset pricing methods estimate investor-level demand for financial assets. We show that leading demand systems omit the central mechanism of portfolio choice: substitution depends on payoff similarity under a latent measure reflecting the curvature of utility. Consequently, they do not enforce no arbitrage and may be severely biased—estimated elasticities can be near one even if true elasticities approach infinity. This reconciles low estimates with higher theoretical benchmarks. Controls eliminate bias only under global mean-variance optimization with exactly measured payoffs, or when payoffs never overlap. Estimated elasticities are not structural parameters and do not show that markets are inelastic.