Flat priors vs. ignorance priors in the analysis of the AR(1) model
通过蒙特卡洛模拟比较AR(1)模型中平坦先验与Phillips无知先验的表现，发现无知先验过度加权自回归参数大于1的值，扭曲似然函数中的样本信息，即使真实参数远小于1也会产生双峰后验分布。
Abstract The paper compares, by a Monte‐Carlo study based on an AR(1) model, the performance of the flat prior and the ignorance prior suggested by Phillips. It argues that the ignorance prior gives heavy weight to values of the autoregressive parameter p higher than 1, and hence distorts the sample evidence as summarized in the likelihood function. It yields bimodal posterior distributions, with the second mode at p higher than 1, even when the true value of p is substantially less than 1.