Haroon Mumtaz and Laura Sunder-Plassmann, "Non-linear Effects of Government Spending Shocks in the US. Evidence from State-level Data", Journal of Applied Econometrics, Vol. 36, No. 1, 2021, pp. 86-97. The panel dataset for the paper is provided in a "comma-delimited" text file called ms-data.csv. This CSV file is in DOS format. It is zipped in the file ms-data.zip. Unix/Linux users should use "unzip -a". The file ms-code.zip contains Matlab and STATA computer programs to produce the estimates reported in the paper. All program files are ASCII files in DOS format. Unix/Linux users should use "unzip -a". Description of Data (MS-DATA.csv) The data are provided in stacked panel format with the time-series for each cross-sectional unit (i.e. US state) stacked on top of one another. The first column of the data file has an index to identify US states. This index takes values from 1 to 50 and represents the 50 states used in the analysis. The time-series runs from 1950Q1 to 2016Q4 for all states apart from Alaska, Michigan and Hawaii where the start dates are 1960Q1, 1956Q1 and 1958Q1, respectively. The second column contains (annual) state-level government spending where missing values represent unobserved quarterly data that is to be estimated. The third column contains quarterly state real income. The fourth column contains quarterly state level employment. Columns 5 to 10 of the data set contain quarterly aggregate US variables that are used as exogenous regressors in the empirical model. Column 5 contains Federal government spending. Column 6 contains Federal government revenue. Column 7 contains real GDP per capita. Column 8 contains the GDP deflator. Column 9 contains the 3 month Treasury bill rate. Column 10 contains the spread between the yield on BAA corporate bonds and a 10 year government bond rate. The final column of the data set contains an initial estimate of quarterly state-level government spending. This initial estimate is used by the estimation code as a starting value for the estimation of the unobserved data in column 1 of this dataset. Source: The data on state level government spending are obtained from the Census Bureau annual survey of state and local government finances. All remaining variables are taken from the FRED database available at https://fred.stlouisfed.org/ Description of estimation code in MS-CODE.zip 1) Matlab Programs for data-construction (Sub-folder DATA) (a) Arrange_benchmark_hall_transform.m. This file loads raw data and arranges it in stacked panel format. It writes the data in csv format (ms_data.csv) and also saves it as a Matlab dataset. (b) run_mfvar.m. This code runs a mixed-frequency Bayesian VAR model to provide an initial estimate for state-level government spending. This initial estimate is saved in the final column of ms_data.csv. (c) arrange_data_for_MFVAR.m. This file creates the dataset that is used by the mixed frequency Bayesian VAR in (b) above. This program needs to be run before running run_mfvar.m. 2) Matlab Programs for estimating the panel threshold VAR (sub-folder Panel_TVAR) (a) Estimate_new_starting_values.m. This file estimates starting values for the main MCMC algorithm and should be run first. (b) Estimate_new.m. This file runs the main MCMC algorithm to estimate the panel threshold VAR model. (c) getirf1SD_regime1_H.m estimates the average impulse response in regime 1 (d) getirf1SD_regime2_H.m estimates the average impulse response in regime 2. (e) Compute_multipliers_tr_h.m. This program uses the impulse responses estimated to calculate the fiscal multipliers in each regime. (f) getirf1SD_regime1_H_byState.m computes the impulse response in regime 1 for each of the 50 states. (g) getirf1SD_regime2_H_byState.m computes the impulse response in regime 2 for each of the 50 states (h) Load_state_irfs.m extracts the estimated state-level impulse responses for analysis and plotting. (i) Load_state_multipliers.m calculates the fiscal multiplier for each state. (3) Matlab Programs for estimating the fixed coefficient panel VAR (Sub-folder fixed_coefficient) (a) estimate_new.m This file runs the MCMC algorithm to estimate a panel VAR with fixed coefficients (4) STATA program for cross-sectional regressions (sub-folder XSECTION) (a) xsection.do carries out the regression analysis reported in Table 1 in the paper. The Matlab and STATA files may require file paths to be changed. Contact: h.mumtaz [AT] qmul.ac.uk