How to obtain estimation results from context?

Dynare Julia
1

Many thanks for the Julia version of Dynare, it is a pleasure to be using Dynare from Julia!

I am interested in running some Monte Carlo experiments, and I would like to access the estimated parameters in Julia, after doing maximum posterior or MCMC.

Starting with maximum posterior, I can see the estimated parameters in the context.results structure, but I can’t find a way to pull out the estimated parameter vector, to use in Julia. Similarly, for MCMC, I don’t see a way to access the posterior quantiles or the raw chain.

Thanks!

2

Actually, studying the source code for the estimation function, I see it now, in part:

results = context.results.model_results[1]
julia> results.estimation.posterior_mode
7-element Vector{Float64}:
 0.9923027290016925
 1.88954160889908
 0.9075452454187274
 0.008870398634074989
 0.6907623612618923
 0.0200331318674717
 0.33131850116134115

I don’t find the posterior quantiles or the raw chains, however. Perhaps they are not saved?

3

Sorry, currently there is no accessor functions to access the estimation results.

When the posterior mode is computed, the following fields are available in context.results.model_results[1].estimation

  • posterior_mode
  • posterior_mode_std
  • posterior_mode_covariance
    By default, the optimization is done with transformed_parameters set to true and the parameters whose prior takes its value on a subset of R are transformed so as take their value on R.
    When transformed_parameters is true, in addition, the following results are stored:
  • transformed_posterior_mode
  • transformed_posterior_mode_std
  • transformed_posterior_mode_covariance

Currently we only store the MCMC chains in <modname>/output/mcm_chain_N.jls where N starts at one and increase every time that you run estimation or rwmc_compute!() for the same model (actually with the same context).
You can recover the chain with

using Serialization
C = deserialize("<modname>/output/mcm_chain_1.jls")

for example.

Please open another issue to request accessor functions for estimation results.