STATA Assignment Solution on Monte Carlo Experiment

* initialize
clear
set more off
set mem 2g

* set working directory
cd /* enter the path here */

*--------------------------
* Q1. Patent application 
*--------------------------

* import data
use "chem_patents_maindataset.dta"

* 1(a) 
summarize grntyr
* Comment: The data covers the sample period from 1939 to 1975

* 1(b) 
tabulate grntyr, subpop(treat)
* Comment: The firms started to receive licensing in year 1919

* 2
summarize count_usa
* Comment: maximum = 68, minimum = 0, mean = 0.349

* 3(a) 
drop if grntyr != 1919
. histogram count_cl
(bin=38, start=0, width=.39473684)
* Shape of distribution is nearly normal

* 3(b)
drop if grntyr != 1919
. histogram count_cl,freq
(bin=38, start=0, width=.39473684)

* there are 38 bins of non zero classes

* 4 create dummy variables from grntyr
tab grntyr, gen(td)                                         

* 5 finish the command below
reg count_usa treat, robust /* equation 1 */
reg count_usa count_cl, robust /* equation 2 */

* 6 finish the command below.
reg count_usa treat td1-td64, robust  /* equation 3 */
reg count_usa count_cl td1-td64, robust /* equation 4 */
* Reason why we don't include all the dummies (from td1 to td65) is that it results in strong correlation and for linear regression there
*should not be any correlation between variables. So in case of linear regression we alway consider (n-1) dummy variables for n 

* 7(a) We assume that there exists no correlation between count_citt (instrument variable) and error, so we use IVregress to get unbiased beta parameter
* Also, we assume that errors are heteroscadastic and independant among observations
* 7(b) finish the command below. It should look like "ivregress 2sls ... (... = ...), robust"
 ivregress 2sls count_usa treat (count_cl = count_cl_itt),robust
* IV estimate will be helpful to get un biased estimation of beta when there exists correlation of variable with error term

* 7(c) Run two-stage least squares manually. 
reg count_cl count_cl_itt,robust   /* equation 5 */
predict double count_cl_hat      /* get the fitted value and store it in count_cl_hat */
reg count_usa count_cl,robust                            /* equation 6 */
* We have now  got beta of 0.16 where as earlier we have 0.94 that is higher effect is shown before than now.

*---------------------------------
* Q2. Monte Carlo simulations
* N = 1000
* beta0 = 1, beta1 = 3, gamma = 2
*---------------------------------

* initialize
clear 
set more off 

* set seed (this ensures that the experiment can be replicated)
set seed 1234

* create temporary file to store experiment results
capture postclose tempid
postfile tempid beta_hat using ols_estimate, replace 
/* declares the variable names and the filename of a (new) Stata dataset where results are to be stored. */

* run the Monte Carlo simulation
forvalues i = 1(1)10000 {       /* perform the experiement 10000 times (number of simulations) */
    drop _all                  /* drop all the variables in memory */
    qui set obs 1000           /* set the number of observations (N) */ 

    gen e = rnormal()          /* generate e from a standard normal distribution */
    gen u = rnormal()          /* generate u from a standard normal distribution */
    gen x=e+(2*u)                       /* fill the gap here (generate x according to x = e + 2u) */
    gen y=1+(3*x)+u                       /* fill the gap here (generate y according to y = 1 + 3x + u) */

    quietly reg y x            /* quietly regress y on x, suppressing all output */
    post tempid (_b[x])        /* post beta_hat to mydata */
    }
postclose tempid

* clear data and import simulation data
clear
use ols_estimate               /* ols_estimate contains the result from the monte carlo experiment*/ 

* summarize beta_hat so its mean is stored in r(mean)
quietly sum beta_hat

* check the mean of beta_hat across replications
display r(mean)
* beta_hat follows normal distribution

* plot
twoway hist beta_hat, title("sampling distribution of beta_hat")       

*---------------------------------
* Q2. Monte Carlo simulations
* N = 1000
* beta0 = 1, beta1 = 3, gamma = 0
*---------------------------------

* initialize
clear 
set more off 

* set seed (this ensures that the experiment can be replicated)
set seed 1234

* create temporary file to store experiment results
capture postclose tempid
postfile tempid beta_hat using ols_estimate, replace 
/* declares the variable names and the filename of a (new) Stata dataset where results are to be stored. */

* run the Monte Carlo simulation
forvalues i = 1(1)10000 {       /* perform the experiement 10000 times (number of simulations) */
    drop _all                  /* drop all the variables in memory */
    qui set obs 1000           /* set the number of observations (N) */ 

    gen e = rnormal()          /* generate e from a standard normal distribution */
    gen u = rnormal()          /* generate u from a standard normal distribution */
    gen x=e                       /* fill the gap here (generate x according to x = e) */
    gen y= 1+(3*x)+u                       /* fill the gap here (generate y according to y = 1 + 3x + u) */

    quietly reg y x            /* quietly regress y on x, suppressing all output */
    post tempid (_b[x])        /* post beta_hat to mydata */
    }
postclose tempid

* clear data and import simulation data
clear
use ols_estimate               /* ols_estimate contains the result from the monte carlo experiment*/ 

* summarize beta_hat so its mean is stored in r(mean)
quietly sum beta_hat

* check the mean of beta_hat across replications
display r(mean)
* beta hat has normal distribution this shows that its not biased

* plot
twoway hist beta_hat, title("sampling distribution of beta_hat")                    

*---------------------------------
* Q2. Monte Carlo simulations
* N = 10
* beta0 = 1, beta1 = 3, gamma = 0
*---------------------------------

* initialize
clear 
set more off 

* set seed (this ensures that the experiment can be replicated)
set seed 1234

* create temporary file to store experiment results
capture postclose tempid
postfile tempid beta_hat using ols_estimate, replace 
/* declares the variable names and the filename of a (new) Stata dataset where results are to be stored. */

* run the Monte Carlo simulation
forvalues i = 1(1)10000 {       /* perform the experiement 10000 times (number of simulations) */
    drop _all                  /* drop all the variables in memory */
    qui set obs 10             /* set the number of observations (N) */ 

    gen e = rnormal()          /* generate e from a standard normal distribution */
    gen u = rnormal()          /* generate u from a standard normal distribution */
    gen x = e                       /* fill the gap here (generate x according to x = e) */
    gen y = 1 +(3*x) + u                       /* fill the gap here (generate y according to y = 1 + 3x + u) */

    quietly reg y x            /* quietly regress y on x, suppressing all output */
    post tempid (_b[x])        /* post beta_hat to mydata */
    }
postclose tempid

* clear data and import simulation data
clear
use ols_estimate               /* ols_estimate contains the result from the monte carlo experiment*/ 

* summarize beta_hat so its mean is stored in r(mean)
quietly sum beta_hat

* check the mean of beta_hat across replications
display r(mean)  /* 2.99 is mean value from stata output*/

* plot
twoway hist beta_hat, title("sampling distribution of beta_hat")    
* sampling distribution follows normal distribution.