clear all
set more off
set maxvar 10000
set mem 500m

*Set path
***


*cd ---- specify path to \data_replication\psid\

local procdata  "./proc_data/"
local rawdata  "./raw_data/"

cd proc_data

*Create consumption panel
*!!!!!!!!!!!!!!!!!!!!! consumption files are generated by the do-file 'consumption.do'
use cons1999.dta, clear
forval x=2001(2)2011{
append using cons`x'.dta
}

sort hid year
xtset hid year

save health_costs_for_income.dta, replace



*import income data
use for_reg.dta, clear
*merge with health costs
joinby hid year using health_costs_for_income.dta, unmatched(both)
tab _merge
drop if _merge!=3

*Prepare for regressions
*take out health costs
*first deflate costs
*Define the CPI annual average for each relevant year. 
*These are defined for the relevant year of the survey (e.g cpi1999 is for 1998)
local cpi1999 = 163.01
local cpi2001 = 172.19
local cpi2003 = 179.87
local cpi2005 = 188.91
local cpi2007 = 201.56 
local cpi2009 = 215.25
local cpi2011 = 218.09
gen heal_cons_deflated=heal_cons
forval x=1999(2)2011{
replace heal_cons_deflated=heal_cons_deflated*(`cpi1999' / `cpi`x'') if year==`x'
}
*take out equivalized! costs
*replace inc=inc-heal_cons_deflated/equiv

drop if inc<0


drop if year>2007
drop if age<25 & year==1999
drop if age>85 & year==2007
bys hid: gen nyear=[_N]
keep if nyear==5

*Generate the age polynomial
gen a2=(age^2)/10

*Generate year dummies
tab year, gen(time_dummy)

*Convert income into logs and keep only the relevant variables
replace inc=log(inc)
keep inc time_dummy* age a2 hid year

*Create retirement dummy
gen retired=1 if age>=65
replace retired=0 if retired!=1

*Find fit before retirement
reg inc age a2 time_dummy* retired



************************************************
************************************************
*Report the coefficients of the age polynomial
estimates table, keep(age a2 retired) b
************************************************
************************************************

predict inc_hat 


*Compute the deviation after permanent differences have been removed
gen x_tilda=inc   //use just log income



*Clean all observations that are above or below a threshold !!!!!!!!!!!!
*Make years consecutive
replace year=1 if year==1999
replace year=2 if year==2001
replace year=3 if year==2003
replace year=4 if year==2005
replace year=5 if year==2007

xtset hid year

*Compute mean by household (across time)
bysort hid: egen mean_hid_inc=mean(inc)
*Generate deviation from mean
gen dif_mean=inc-mean_hid_inc
*Find the top and bottom 1 percent
sum dif_mean, d
gen p1=r(p1)
gen p99=r(p99)
*Remove observations that are in the tails
replace x_tilda=. if dif_mean<=p1
replace x_tilda=. if dif_mean>=p99

*################################################
*FIND AUTOCOVARIANCES
*################################################






preserve

*Generate lag structure to compute the autocovariances
gen lag1_x_tilda=L.x_tilda
gen lag2_x_tilda=L2.x_tilda
gen lag3_x_tilda=L3.x_tilda
gen lag4_x_tilda=L4.x_tilda

*Compute autocovariances
*1)Variance
correlate x_tilda x_tilda, covariance
gen v0=r(Var_1)
*2)Lags
forval x=1(1)4{
correlate x_tilda lag`x'_x_tilda, covariance
g float v`x'=r(cov_12)
}
keep v*
duplicates drop

************************************************
************************************************
*Export results for autocovariances
save autocov_final_new_NO_health_ALL_v3.dta, replace
************************************************
************************************************

restore

gen lag1_x_tilda=L.x_tilda
keep x_tilda lag1_x_tilda

gen dif_x_tilda=x_tilda-lag1_x_tilda

egen SD = sd(dif_x_tilda)

egen KURT = kurt(dif_x_tilda)

egen SKEW = skew(dif_x_tilda)

keep SD KURT SKEW


duplicates drop 

merge using autocov_final_new_NO_health_ALL_v3.dta

drop _merge

*keep only the relevant moments and order them according for table 2
keep SD v0 v1 v2
replace SD = round(SD, 0.01)
replace v0 = round(v0, 0.01)
replace v1 = round(v1, 0.01)
replace v2 = round(v2, 0.01)

order v0 v1 v2 SD

cd ..

cd output

outsheet using tab2_income.csv, c replace