A0151
Title: Bayesian method for causal inference in spatially-correlated multivariate time series
Authors: Subhashis Ghosal - North Carolina State University (United States) [presenting]
Abstract: Measuring the causal impact of an advertising campaign on sales is important for advertising companies. We propose a novel Bayesian method to infer causality which can also detect weak impacts. We compare two posterior distributions of a latent variable---one obtained by using the observed data from the test stores and the other one obtained by using the data from their counterfactual potential outcomes. The counterfactual potential outcomes are obtained from the data of synthetic controls given by a sparse linear combination of sale figures at many control stores over the causal period. We use a multivariate structural time series model to capture the spatial correlation between test stores. Stationarity is imposed on the local linear trend of the model to prevent the prediction intervals from being explosive. A two-stage algorithm is proposed to estimate the parameters of the model. In Stage 1, a modified EMVS algorithm is applied to select control stores. In Stage 2, an MCMC algorithm is used to obtain the samples of the rest parameters. We present extensive simulation results to show the effectiveness of the proposed method. The new method is applied to measure the causal effect of an advertising campaign for a consumer product sold at stores of a large national retail chain.
