scoringRules (version 1.0.1)

run_casestudy: Run the case study in KLTG (2020), or a smaller version thereof

Description

Run the case study in KLTG (2020), or a smaller version thereof

Usage

run_casestudy(
  data_df,
  burnin_size = 5000,
  max_mcmc_sample_size = 5000,
  nr_of_chains = 3,
  first_vint = "1996Q2",
  last_vint = "2014Q3",
  forecast_horizon = 1,
  random_seed = 816
)

Value

Object of class "casestudy", containing the results of the analysis. This object can be passed to plot for plotting, see the documentation for plot.casestudy.

Arguments

data_df

data frame in the same format as the gdp data set in this package.

burnin_size

length of the burn-in period used for each forecast.

max_mcmc_sample_size

maximal number of MCMC draws to consider (integer, must equal either 1000, 5000, 10000, 20000 or 40000). Defaults to 5000.

nr_of_chains

number of parallel MCMC for each forecast date (integer, defaults to 3).

first_vint, last_vint

first and last data vintage (= time point at which forecasts are made). Default to "19962Q2" and "2014Q3", respectively.

forecast_horizon

forecast horizon to be analyzed (integer, defaults to 1).

random_seed

seed for random numbers used during the MCMC sampling process. Defaults to 816.

Author

Fabian Krueger

Details

The full results in Section 5 of KLTG (2020) are based on the following setup: burnin_size = 10000, max_mcmc_sample_size = 50000, nr_of_chains = 16, data_df = gdp, first_vint = "1996Q2", last_vint = "2014Q3", and forecast_horizon = 1. Since running this full configuration is very time consuming, the default setup offers the possibility to run a small-scale study which reproduces the qualitative outcomes of the analysis. Running the small-scale study implied by the defaults of run_study as well as the GDP data (data_df = gdp) takes about 40 minutes on an Intel i7 processor.

References

Krueger, F., Lerch, S., Thorarinsdottir, T.L. and T. Gneiting (2020): `Predictive inference based on Markov chain Monte Carlo output', International Statistical Review, forthcoming. tools:::Rd_expr_doi("10.1111/insr.12405")

See Also

plot.casestudy produces a summary plot of the results generated by run_casestudy run_casestudy uses ar_ms to fit a Bayesian Markov Switching model, recursively for several time periods.

Examples

if (FALSE) {
data(gdp)
cs <- run_casestudy(data_df = gdp, last_vint = "1999Q4")
plot(cs)
}