Arpita Chatterjee, James Morley, and Aarti Singh, "Estimating
Household Consumption Insurance", Journal of Applied Econometrics,
Vol. 36, No. 5, 2021, pp. 628-635.

All files are zipped in cms-files.zip and are described below.

Because this is a replication paper of

Richard Blundell, Luigi Pistaferri, and Ian Preston, "Consumption
Inequality and Partial Insurance," American Economic Review 98(5),
December 2008, pp. 1887-1921,

it should be noted that the original data for household income and
consumption are available from

https://www.aeaweb.org/articles?id=10.1257/aer.98.5.1887

which currently links to

https://www.openicpsr.org/openicpsr/project/113270/version/V1/view

in order to download the data and replication files, including GAUSS
code. Full details of the original PSID data used by Blundell,
Pistaferri, and Preston (BPP) are provided in their paper.

The file cohA.txt included in the main directory of the CMS-files
folder contains data from BPP in the same format considered in their
GAUSS code and which can be used with our GAUSS code in the
Tables1and2 subfolder in the Full_replication_files_in_GAUSS folder
available in the main directory of the CMS-files folder in order to
replicate the GMM estimates in the first column of results in Table 6
of their paper and the GMM estimates in Tables 1 of our paper by using
our GAUSS program files dwmd_CMS_bba.gss and omd_CMS_bba.gss described
below and based on the original BPP GAUSS program file md_AER.prg
provided in the BPP replication files on the AER website (and also
included with our replication files for completeness).

There are 24710 observations corresponding to 1765 households from
PSID for the years 1979-1992. The data file includes indicators for
household id, year, and some dummy variables under the coh, ndrod,
yduy, yduc, and nmissd column headings used by BPP in estimation to
select observations and address missing observations. The two key
variables are under the duy and duc column headings and respectively
correspond to idiosyncratic income and consumption growth for a given
household in a given time period (and are set to 0 when data are
missing, with dummy variables under the respective yduy and yduc
headings equal to 0 when the the data are missing and 1 otherwise).

The file cohA_CMS_added1978.txt also included in the main directory of
the CMS-files folder contains the data in a format that can be used
with our GAUSS code in the Tables1and2 subfolder in the
Full_replication_files_in_GAUSS folder available in the main directory
of the CMS-files folder in order to replicate the QMLE estimates in
Table 1 of our paper by using our GAUSS program file qmle_CMS.gss.
There are 26475 observations corresponding to 1765 households from
PSID for 1978-1991. The file includes just two columns corresponding
to idiosyncratic income and consumption in log levels, with any
missing observations set to 999. Note: the changes in these log level
values, when observed in two consecutive periods, correspond exactly
to the corresponding observed values of duy and duc in the cohA.txt
data file used for the GMM estimation.

We have provided basic versions of code for QMLE using GAUSS (.gss),
Matlab (.m), or R (.R). The different programs should produce the same
estimates based on numerical optimization given the same starting
values. However, we note that the calculation of Huber-White standard
errors may differ from one computer or program to the next due to
precision issues with numerical derivatives and also that R is
considerably slower (about twice as long per iteration on the same
computer) than GAUSS or Matlab. The code for the corresponding
programs is given in the GAUSS, MATLAB, and R folders in the main
directory of the CMS-files folder, respectively. The data file
cohA_CMS_added1978.txt discussed above is also provided in these
folders, along with data files cohA_age1_CMS_added1978.txt,
cohA_age2_CMS_added1978.txt, cohA_high_CMS_added1978.txt, and
cohA_low_CMS_added1978.txt that contain only the corresponding
observations of data for the subgroups of young, old, college
education, and no college households, respectively. These files can be
used to replicate the QMLE estimates in Tables 1 and 2 of the paper
and Tables B1-B6 of the online appendix.

However, all of the results reported in the paper and online appendix
were generated using the full GAUSS code provided in the
Full_replication_files_in_GAUSS folder in the main directory of the
CMS-files folder. This code should provide the exact results reported
in Tables 1-3 of the paper and Tables B1-B8 of the online appendix.
Specifically, the Tables1and2 subfolder contains the dwmd_CMS_bba.gss
and omd_CMS_bba.gss program files that will reproduce the GMM
estimates in Tables 1 and B1 when using the data file cohA.txt and the
GMM estimates in Tables 2 and B2-B5 when using the data files
cohA_age1.txt, cohA_age2.txt, cohA_high.txt, and cohA_low.txt that
contain only the corresponding observations of data for the subgroups
of young, old, college education, and no college households,
respectively.

The Tables1and2 subfolder also contains the qmle_CMS.gss program file
that will reproduce the QMLE estimates in Tables 1 and B1 when using
the data file when using the data file cohA_CMS_added1978.txt and the
QMLE estimates in Tables 2 and B2-B5 when using the data files
cohA_age1_CMS_added1978.txt, cohA_age2_CMS_added1978.txt,
cohA_high_CMS_added1978.txt, and cohA_low_CMS_added1978.txt. The
qmle_CMS_FD.gss program file in the Tables1and2 subfolder will
reproduce QMLE estimates based only on data with consecutive
observations in levels reported in Table B6 of the online appendix
when using the data files cohA_CMS_added1978.txt,
cohA_age1_CMS_added1978.txt, cohA_age2_CMS_added1978.txt,
cohA_high_CMS_added1978.txt, and cohA_low_CMS_added1978.txt.

The Table3 subfolder contains further subfolders with program files
used to run the Monte Carlo experiments and the results reported in
Tables 3 and B7-B8. The Estimation to get shocks for MC subfolder
contains the program qmle_CMS_filtered_shocks.gss that reports
filtered or smoothed estimates of shocks from the Kalman
filter/smoother that are used for the empirical distribution in
generating artificial data for the Monte Carlo experiments. The
estimates of shocks are based on the same data considered in the
application, so the program file refers to the cohA_CMS_added1978.txt,
cohA_age1_CMS_added1978.txt, cohA_age2_CMS_added1978.txt,
cohA_high_CMS_added1978.txt, and cohA_low_CMS_added1978.txt data files
that are also included in the subfolder.

The Relication of Unadjusted MC Results folder contains
dgp_mle_gmm_bpp_shocks_large_sample.gss and
dgp_mle_gmm_bpp_shocks_small_sample.gss program files that generate
the properties of the estimators in the large and small sample
experiments, respectively, and make use of the estimated shocks in the
uc_bpp_tvv_shocks_eps.txt, uc_bpp_tvv_shocks_eta.txt,
uc_bpp_tvv_shocks_u.txt, and uc_bpp_tvv_shocks_v.txt files to simulate
data using empirical distributions of shocks. The
cohA_CMS_added1978.txt, cohA_age2_CMS_added1978.txt, cohA.txt, and
cohA_age2.txt data files are included only to ensure the simulated
data have the same number and location of missing observations as the
sample data. The Estimation of bias update folder contains the
qmle_CMS_shocks.gss program file that estimates empirical shocks for
bootstrap analysis based on the average large sample QMLE estimates
from the unadjusted Monte Carlo results. Then the
dgp_large_bias_qmle_dwmd_omd.gss and dgp_small_bias_qmle_dwmd_omd.gss
program files estimate the bootstrap bias correction given the large
and small samples. The various data files are included to estimate
empirical shocks and ensure simulated bootstrap samples of data have
the same number and location of missing observations as the sample
data.

If you use any of this code, please cite the paper.

James Morley

Email: james.morley [AT] sydney.edu.au

This version: 8 April 2020