A talk by Bradley Setzler, a Ph.D. student at the University of Chicago and author of the Julia/Economics blog. The long title is:
Efficiently Estimating Human Development in Julia by Maximizing the Likelihood of a Nonlinear Recursive Panel Filter with a Time-Varying Parameter Space
I demonstrate in Julia the maximum likelihood estimation of the panel data Kalman Filter with time-varying parameter space, estimated initial state, and interpretable units of the state. In particular, I present:
(1) A review of the theory of estimating and motivation for using the Kalman Filter;
(2) Construct the parameter space to automatically adjust in response to the data structure;
(3) Simulate the data generating process presumed by the use of the Kalman Filter;
(4) Compute the sample likelihood across time and across individuals, including the initial state likelihood using background data;
(5) Obtain the maximum likelihood estimated Kalman Filter through numerical optimization;
(6) Run each individual through the estimated Kalman Filter to obtain optimal predictions of the final state (i.e., the standard use of the Kalman Filter); and,
(7) Use the final state predictions to obtain interpretable units of the state through “anchoring” on relevant outcomes, then use this to correct the units of the model parameters.
...Woohoo! Should be a good time.