The section download data from the paper.
download.file("https://raw.githubusercontent.com/ubcecon/ECON628_2018/master/paul/assignments/02/qsw-dataPrep.R",destfile="qsw-dataPrep.R")
download.file("https://raw.githubusercontent.com/ubcecon/ECON628_2018/master/paul/assignments/02/qsw-table2.R",destfile="qsw-table2.R")
download.file("https://raw.githubusercontent.com/ubcecon/ECON628_2018/master/paul/assignments/02/rlassoPanel.R",destfile = "rlassoPanel.R")
source("qsw-dataPrep.R")To estimate the parameters \(\beta_1\) and \(\beta_2\), we need the exogeneity assumption. \(E[\epsilon_{ijt}|Reform_{jt},Commercial_{i}] = 0\).
## Recreate Table 2 of QSW (2018)
if (!file.exists("qsw2018.Rdata")) {
source("qsw-dataPrep.R",echo=TRUE)
}
load("qsw2018.Rdata")
library(lfe)## Loading required package: Matrixlibrary(stargazer)##
## Please cite as:## Hlavac, Marek (2018). stargazer: Well-Formatted Regression and Summary Statistics Tables.## R package version 5.2.2. https://CRAN.R-project.org/package=stargazersource("rlassoPanel.R")
qsw$propre <- interaction(qsw$provincecheck, qsw$prefecturecheck)
col1_raw <- felm(pc1s ~ reform_cg2003
| newspaper_id + year | 0
| propre, data=subset(qsw, !is.na(realfdi) & realfdi>0),
na.action=na.exclude)
col2_raw <- felm(pc1s ~ reform_cg2003 + log(gdp) +
log(realfdi) + log(industrialshare) + log(population)
| newspaper_id + year | 0
| propre, data=subset(qsw, !is.na(realfdi) & realfdi>0),
na.action=na.exclude)
col3_raw <- felm(pc1s ~ reform_cg2003 + reform_cg2003:commercial
| newspaper_id + year | 0
| propre, data=subset(qsw, !is.na(realfdi) & realfdi>0),
na.action=na.exclude)
col4_raw <- felm(pc1s ~ reform_cg2003 + log(gdp) +
log(realfdi) + log(industrialshare) + log(population)
+ reform_cg2003:commercial | newspaper_id + year | 0
| propre, data=subset(qsw, !is.na(realfdi) & realfdi>0),
na.action=na.exclude)
qsw$outside <- qsw$tier > 2
col5_raw <- felm(pc1s ~ reform_cg2003 + reform_cg2003:commercial
+ reform_cg2003:outside + commercial:outside:reform_cg2003
| newspaper_id + year | 0
| propre, data=subset(qsw, !is.na(realfdi) & realfdi>0),
na.action=na.exclude)
col6_raw <- felm(pc1s ~ reform_cg2003 + reform_cg2003:commercial
+ reform_cg2003:outside + commercial:outside:reform_cg2003 + log(gdp) +
log(realfdi) + log(industrialshare) + log(population)
| newspaper_id + year | 0
| propre, data=subset(qsw, !is.na(realfdi) & realfdi>0),
na.action=na.exclude)
stargazer(col1_raw,col2_raw,col3_raw,col4_raw,type = "text" )##
## ============================================================================================
## Dependent variable:
## -------------------------------------------------------------------
## pc1s
## (1) (2) (3) (4)
## --------------------------------------------------------------------------------------------
## reform_cg2003 -0.002 -0.001 0.018** 0.019**
## (0.003) (0.003) (0.007) (0.007)
##
## log(gdp) 0.015 0.021
## (0.032) (0.032)
##
## log(realfdi) 0.007 0.007
## (0.010) (0.009)
##
## log(industrialshare) -0.012 -0.014
## (0.039) (0.039)
##
## log(population) 0.020 0.022
## (0.040) (0.043)
##
## reform_cg2003:commercial -0.024*** -0.024***
## (0.006) (0.007)
##
## --------------------------------------------------------------------------------------------
## Observations 872 872 872 872
## R2 0.832 0.833 0.841 0.842
## Adjusted R2 0.803 0.803 0.813 0.813
## Residual Std. Error 0.051 (df = 743) 0.051 (df = 739) 0.050 (df = 742) 0.050 (df = 738)
## ============================================================================================
## Note: *p<0.1; **p<0.05; ***p<0.01stargazer(col5_raw,col6_raw,type="text")##
## ======================================================================
## Dependent variable:
## ---------------------------------
## pc1s
## (1) (2)
## ----------------------------------------------------------------------
## reform_cg2003 0.034*** 0.036***
## (0.009) (0.009)
##
## log(gdp) 0.012
## (0.035)
##
## log(realfdi) 0.009
## (0.010)
##
## log(industrialshare) -0.027
## (0.040)
##
## log(population) 0.016
## (0.048)
##
## reform_cg2003:commercial -0.039*** -0.040***
## (0.008) (0.008)
##
## reform_cg2003:outside -0.020** -0.021**
## (0.010) (0.011)
##
## reform_cg2003:commercialTRUE:outside 0.019** 0.021**
## (0.009) (0.010)
##
## ----------------------------------------------------------------------
## Observations 872 872
## R2 0.842 0.844
## Adjusted R2 0.815 0.815
## Residual Std. Error 0.050 (df = 740) 0.050 (df = 736)
## ======================================================================
## Note: *p<0.1; **p<0.05; ***p<0.01## try a lasso specification
# construct x matrix
m <- felm(pc1s ~ reform_cg2003 + reform_cg2003:commercial +
poly(log(gdp), log(industrialshare), log(population),
log(realfdi), degree=5) +
year:(log(gdp) + log(industrialshare) + log(population) +
log(realfdi)) +
as.factor(year)
| newspaper_id | 0 | propre,
data=subset(qsw,realfdi>0), keepX=TRUE)## Warning in chol.default(mat, pivot = TRUE, tol = tol): the matrix is either
## rank-deficient or indefinited <- m$X[,grep("reform_",colnames(m$X))]
x <- m$X[,!grepl("reform_",colnames(m$X))]
y <- subset(qsw,realfdi>0 & !is.na(pc1s))$pc1s
id <- as.factor(subset(qsw,realfdi>0 & !is.na(pc1s))$newspaper_id)
# partial out x using lasso
eyx <- rlassoPanel(x, y=y, f=id)
ed1x <- rlassoPanel(x,y=d[,1],f=id)
ed2x <- rlassoPanel(x,y=d[,2],f=id)
# regress residuals of dependent variable on residuals of covariate(s)
# of interest
tmp <- data.frame(y=eyx$residuals, reform=ed1x$residuals,
reformC=ed2x$residuals, id=id)
col3 <- summary(felm(y ~ reform + reformC | id | 0 | id, data=tmp))