Empirical application: young firm dynamics


Data
----
The Kauffman Firm Survey is confidential and is not in this package. Inside
the enclave, save three constructs into kfs_panel.mat:

    log_emp   log employment, 8 by N, rows 2004 to 2011
    rd        R&D employment over total employment, same size
    naics     two-digit NAICS code, 1 by N


Driver
------
run_kfs_application.m
    Screens the panel, estimates the model with each predictor, and writes
    the parameter, density forecast and point forecast tables, plus the
    figures. modrun 301. Each predictor draws 10,000 MCMC iterations and
    discards the first 5,000 as burn-in. plot_kfs_output.m writes the
    figures from the saved draws.

The specification is
    log emp_it = beta * log emp_i,t-1
                 + lambda_1i + lambda_2i * R&D_i,t-1 + u_it,
    u_it ~ N(0, sigma_i^2)
with T = 6 over 2004 to 2010 and 2011 held out. Firms whose R&D share exceeds
one, or that lack a long enough complete run to identify lambda_i, are
dropped. After this screen the paper has N = 503 firms and 9.32 percent
missing (i,t) cells. The initial log employment is standardized before it
enters the conditional priors.


Predictors
----------
The labels match Table 6. The sampler folders expose np_link.m, so the driver
selects a predictor by adding the corresponding folder to the path. These are
copies of the general_model samplers, with isfinite handling so that missing
(i,t) cells are skipped. NP-C/R is the benchmark for both the normalization
and the tests. The Oracle predictor is Monte Carlo only and does not apply
here.

heterosk_np_cr/          Heterosk-NP-C/R
homog.m                  Homog
homosk_np_c/             Homosk-NP-C
heterosk_flat.m          Heterosk-Flat
heterosk_param.m         Heterosk-Param
heterosk_np_disc/        Heterosk-NP-disc
heterosk_np_r/           Heterosk-NP-R
heterosk_np_c/           Heterosk-NP-C


Output
------
All files go to results/, prefixed results/modrun301_.

Tables
    _gibbs.txt _dfcst.txt _pfcst.txt Table 6 (beta, LPS) and the MSE table
    _descrip.txt                     Table 1
    _corr.txt                        Appendix correlations
Figures
    _datadens_logEmp.png _datadens_RD.png
    _pitmain_Homog.png _pitmain_NPCR.png
    _pit_*.png                       PIT for every predictor
    _pred_dist_firm1.png ... firm4.png
    _pred_dist_firm_Homog.png _pred_dist_firm_NPCR.png
    _corr41.png _corr42.png _corr12.png
                                     joint densities of lambda_i and y_i0
    _corr43.png _corr13.png _corr23.png
                                     joint densities with sigma_i^2
    _pred_dist_ind*.png              log average employment by NAICS
                                     (sectors with more than 10 firms);
                                     Construction and Retail Trade also
                                     as _main.png
.mat
    modrun301.mat                    workspace
    modrun301_est*.mat               draws
