Thursday 9 May 2019, 3.00PM to 4:00pm
Speaker(s): Clifford Lam (LSE)
Host: Vanessa Smith
Abstract: In handling spillover effects using a spatial lag model, a researcher often faces the task of identifying a spatial weight matrix to use from a number of specifications. Wrongly specifying such a matrix can have detrimental effects on the estimation of other model parameters. We consider decomposing the spatial weight matrix as a linear combination of these specifications, plus a sparse adjustment matrix, which serves to compensate the inaccuracy in the best linear combination of spatial weight matrix specifications. We estimate all these from data, and provide full inferential theory on the non-zero estimated elements in the sparse adjustment matrix and other model parameters, including the coefficients in the estimated best linear combination. A further penalization step allow us to select which spatial weight matrix specifications are relatively important, and which are unimportant. In the special case of no specifications, the sparse adjustment matrix then serves as a sparse spatial weight matrix. A set of stock return data from NYSE is also analysed to demonstrate our results.
Location: A/D271 new Seminar Room
Admission: All welcome