I’m trying to replicate the result of HLW 2017. The authors use a sequential maximum likelihood approach, whereas I’m trying to estimate it using Bayesian estimation. I’m able to obtain fairly close estimations for function coefficients, but state variable estimations are off.
Notably, there’s a big jump at the first few time periods and the rest of the estimation falls under HLW’s. HLW_init.pdf (19.6 KB)
Is there a way to place constraints on state variables? Or perhaps the problem lies somewhere else. I’ve also tried with or without using the same initial value that HLW had used with dyanre 4.7’s new feature, but both results in big jumps at first few values of state variable.
P.S. the state variable of interest r* is not modeled directly in the program, but modeled as 4*g+z. In this case, z has unusually large value at the first few time periods.