Reference no: EM132473246
Assignment -
Instructions - In what follows, the 'full' model considers x1 and x2, while the truncated model refers to a specification that only uses x1.Z1, Z2, Z3 and Z4 will be candidates for instrumental variables.
This is the Var-Cov matrix that I used for simulating the data:
Correlations
|
Const
|
x1
|
x2
|
Z1
|
Z2
|
Z3
|
Z4
|
eps
|
Const
|
1
|
0
|
0
|
0
|
0
|
0
|
0
|
0
|
x1
|
0
|
1
|
0.12
|
0.5
|
0
|
0.2
|
0.01
|
0.6
|
x2
|
0
|
0.12
|
1
|
0.4
|
0
|
0
|
0.001
|
0.02
|
Z1
|
0
|
0.5
|
0.4
|
1
|
0.6
|
0
|
0
|
0.6
|
Z2
|
0
|
0
|
0
|
0
|
1
|
0.4
|
0
|
0
|
Z3
|
0
|
0.2
|
0
|
0
|
0.4
|
1
|
0
|
0
|
Z4
|
0
|
0.01
|
0.001
|
0
|
0
|
0
|
1
|
0
|
eps
|
0
|
0.6
|
0.02
|
0.6
|
0
|
0
|
0
|
1
|
Here are the "true coefficients" I used to generate y4 and y5:
y4 = 0.48 + 1.18*x[1] + 0.72*x[2] + eps
y5 = 0.48 + 2.2*x[1] + 1.74*x[2] + 0.4*eps
Question - IV for starters
Compare the estimates of the full OLS model and the IV Regression. -Why are the estimated coefficients different?
Looking at the Var-Cov Matrix, what bias would you expect for x1 and what bias for x2?, calculate! (Hint: the X - matrix for calculating the bias consists of x1 and x2., i.e. you can disregard the z's and the constant for this calculation. E(X,u) is given by the "eps" - column)
Which of the X Variables is endogenous with y (e.g.asimultaneityproblem)?
You have 4 candidates that you can use as an instrumental variable for X1, but
- One is itself endogenous.
- One is a weak instrument
- One is irrelevant and
- Only one is valid.
Which is which?
Now try instrumenting for x2!
Before you go ahead, consider for each z if you can or cannot use it as instrument. Explain, why not, or, if you can, explain what the assumptions would be and explain the exclusion restriction.
Refer to the "IV-assumptions" in the Angristbook or on the slides for your argument.
Independent of your answer in the previous points: consider using Z4 as an instrument for X2:
- Do 2SLS, describe which steps you have to take?
- Write a code that separately runs the first stage and the second stage regression.
- Which of the two assumptions can you test? Is it satisfied?
- Rehash the MM-IV Estimator that we saw in class.
- What variables will you include in the Z matrix?
- Write a code that directly implements the estimator in matrix notation.
- Now interpret the coefficient estimates that you got In the direct IV and In the 2SLS.
- Share any thoughts, questions or confusion on this exercise. (Hint: it's likely that the final estimation result is not satisfactory - maybe you can see why, but don't worry if you cannot.)