PROBLEM 23: INSTRUMENTAL VARIABLES
OBJECT OF THE PROBLEM
To run a model using instrumental variables.
INTRODUCTION
Here is an example of the use of instrumental
variables. He uses information from one industry to create an
instrument to be used in analyzing a second industry. In this
case he takes the ratio of money wages to product price for the
knitting-mill product industry and uses it to analyze the
furniture manufacturing industry. We are satisfied that wages
in the knitting industry are unrelated to measurement errors in
the furniture industry, but are related to wages in the furniture
industry.
PROBLEM PROGRAM
The following program can be used for the problem:
//PROB23 JOB (AF,A699),yourname,REGION=4096K,TIME=1
/*OPENBIN
//S1 EXEC SAS,WORK=100
//SYSIN DD *
DATA LORENZ; INPUT FW FP LVRATIO KP KW;
X=FW/FP; X=LOG(X);
Y=(1/LVRATIO)*(1/FP);
Y=LOG(Y);
ZSTAR=KW/KP;
ZSTAR=LOG(ZSTAR);
CARDS;
3515 1.0000 .1706 1.0000 2698
2668 1.5470 .2385 1.4891 2260
1834 .9482 .3678 1.0346 1548
1713 .8195 .3857 .7358 1487
1288 .8941 .5040 .7713 1169
1342 .9437 .5228 .8990 1021
1078 .6646 .6291 .6030 802
738 1.8260 .7200 2.2570 845
448 1.4590 .9415 .9720 364
471 1.7580 .9017 1.2458 546
464 2.2300 1.0863 1.3901 523
PROC MEANS;VAR X ZSTAR;
OUTPUT OUT=WORKERS MEAN=MX MZSTAR;
PROC PRINT DATA=LORENZ;
DATA CRAYPO; SET WORKERS;
DO I=1 TO 11; OUTPUT; END;
DATA CALC; MERGE LORENZ CRAYPO;
Z=ZSTAR*MX/MZSTAR;
YSTAR=Z*Y; XSTAR=X*Z;
PROC PRINT DATA=CALC;
PROC REG DATA=CALC; MODEL Y=X;
PROC REG DATA=CALC; MODEL YSTAR=Z XSTAR/NOINT;
//
DISCUSSION OF THE PROGRAM
Between the first DATA statement and the CARDS statement we create the variables X, Y, AND Z* used by Kmenta. Using PROC MEANS we then calculate the mean of X and Z* and output them to a data set called WORKERS. We need the means in order to construct our instrumental variable later in the program. Beginning with data step CRAYPO we calculate our instrumental variable Z. First we use a DO, OUTPUT, and END statement to create a column of mean observations for X and ZSTAR. This is needed to transform the data. We transform X and Y by multiplying them by Z. Z is derived using ZSTAR and the means of X and ZSTAR.
We then run two models in REG. The first model represents
the original equation. The second is the instrumental variables
model. Note that Z takes the place of the intercept in the second
model.
INSTRUCTIONS FOR WRITE-UP
Give an intuitive and econometrics notation explanation of each
step of the program, and an intuitive explanation of what the
problem is trying to accomplish. Write down the estimated models
(original model and instrumental variable model). Analyze them
in terms of t and F-statistic, R-square, etc. Give an intuitive
explanation of each model. Label all output. read page 356 in
kmenta and compare your results to his. Label the data by nation
on the two PROC PRINT output pages. Under what circumstances is
Kmenta's technique valid? In general, what properties does the
OLS estimators have and what properties does the OLS instrumental
variable estimators have?