| Abstract: | Akaike's BAYSEA approach to seasonal decomposition is designed to capture the respective merits of several pre-existing adjustment techniques. BAYSEA is computationally efficient, requires only weak assumptions about the data-generating process, and is based on solid inferential (namely, Bayesian) foundations. We present a model similar to that used in BAYSEA, but based on a double exponential rather than a Gaussian error model. The resulting procedure has the advantages of Akaike's method, but in addition is resistant to outliers. The optimal decomposition is obtained rapidly using a sparse linear programming code. Confidence bands and predictive intervals can be obtained using Gibbs sampling. |
