L1 Estimation: On the Optimality of Linear Estimators

Consider the problem of estimating a random variable X from noisy observations Y = X + Z, where Z is standard normal, under the L-1 fidelity criterion. It is well known that the optimal Bayesian estimator in this setting is the conditional median. This work shows that the only prior distribution on X that induces linearity in the conditional median is Gaussian. Along the way, several other results are presented. In particular, it is demonstrated that if the conditional distribution P-X|Y = y is symmetric for all y, then X must follow a Gaussian distribution. Additionally, we consider other Lp losses and observe the following phenomenon: for p is an element of [1, 2], Gaussian is the only prior distribution that induces a linear optimal Bayesian estimator, and for p is an element of (2, infinity), infinitely many prior distributions on X can induce linearity. Finally, extensions are provided to encompass noise models leading to conditional distributions from certain exponential families.