Inference in Some Reduced Rank Structures and Piecewise Linear Forms
This seminar will give a brief overview of four lines of work; two using continuous piecewise forms and two using reduced rank structures. In the first paper, we investigate non-linearities in the impact of management practices on firm performance using Gaussian process and a continuous piece-wise linear approach with probabilistically smoothed endogenous breaks. In all cases we find significant evidence of a U-shaped relationship, with the biggest returns to management occurring when management practices are at their highest levels. In the second paper, we develop methods to estimate the continuous piecewise linear model (CPLM) and to conduct inference on the model.
In the third paper, we take advantage of a common empirical feature of large system state space models to reduce the dimension of the model and permit more efficient inference. Unlike existing approaches, the model has a specification that allows for formal inference.
The last section does not discuss a paper but rather outlines a new line of work using g-priors and I am looking for an interesting application. Appropriate applications would be to situations with either fat data (k > n) or highly collinear data.