Monte Carlo simulation: baseline model with random effects and
cross-sectional homoskedasticity


Driver
------
run_baseline_mc.m
    Generates the data, estimates the model with each predictor, and writes
    the point and density forecast tables. One run loops over the three
    f0(lambda_i) experiments: degenerate, skewed, and bimodal.
    Set is_par = 1 for the full experiment (n_sim = 1000, parallel pool) or
    is_par = 0 for a single repetition with the "f0 vs posterior" figure.
    Each repetition draws 40,000 MCMC iterations per predictor and discards
    the first 20,000 as burn-in.


Predictors
----------
oracle.m        knows the common parameter rho and the distribution f(lambda_i)
homog.m         pooled OLS, no individual heterogeneity
flat.m          flat prior on individual heterogeneity
param_normal.m  f(lambda_i) modeled as a normal distribution
np_disc.m       f(lambda_i) modeled as a discrete distribution via a Dirichlet
                process prior
np_r.m          f(lambda_i) modeled as a continuous distribution via a
                Dirichlet process mixture prior


Output
------
All files go to results/, prefixed results/modrun101_.

Tables
    ispar*_out.txt                   MSE and LPS; one file, three f0
                                     blocks
Figures (is_par = 0 only)
    pi_shade_lambda{2,3}_est{4,6}.png
                                     f0 vs posterior for Param and NP-R
.mat
    lambda*_ispar*.mat               workspace after each f0
    lambda*_est*.mat                 draws; written only when is_par = 0
