{[ promptMessage ]}

Bookmark it

{[ promptMessage ]}

chapter 8 [article]

# chapter 8 [article] - Contents 1 Stochastic regressors 1.1...

This preview shows pages 1–4. Sign up to view the full content.

Contents 1 Stochastic regressors 1 1.1 New assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 OLS properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 2 Measurement errors 4 2.1 Sources and consequences . . . . . . . . . . . . . . . . . . . . . . 4 2.2 Friedman and consumption . . . . . . . . . . . . . . . . . . . . . 7 3 Instrumental variables 10 3.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 3.2 Practice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 1 Stochastic regressors Contents 1.1 New assumptions Contents Everything so far has been in the context of Model A : non-stochastic regressors unrealistic but useful for learning basics Now will move on to Model B : stochastic regressors will assume values of a regressor are random draws from its popula- tion probability distribution Implications for OLS? not necessarily serious, but potential for problems (measurement er- rors. . . ) solution: instrumental variables We will make the following assumptions in Model A : B.1 The model is linear in parameters and correctly specified ( A.1 ) B.2 The values of the regressors are drawn randomly from fixed populations note: we do not assume the regressors are independent of one another two (or more) regressors might have a joint probability distribution, and thus correlated sample values B.3 No exact linear relationship among the regressors ( A.2 )

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
B.4 The disturbance term has zero expectation: E ( u i ) = 0 ( A.3 ) B.5 The disturbance term is homoscedastic ( A.4 ) B.6 The values of the disturbance term have independent distributions ( A.5 ) B.7 The disturbance term is distributed independently of the regressors weaker assumptions possible here (p.242), but this one is sufficient very important: if this is violated, problems. . . B.8 The disturbance term has a normal distribution ( A.6 ) 1.2 OLS properties Contents Y i = β 1 + β 2 X i + u i (8.6) We’ve seen that the OLS estimator of β 2 in this simple regression can be written: b OLS 2 = n i =1 ( X i - ¯ X ) ( Y i - ¯ Y ) n i =1 ( X i - ¯ X ) 2 = β 2 + n X i =1 a i u i (8.7) where a i = X i - ¯ X n j =1 ( X j - ¯ X ) 2 When proving unbiasedness of OLS with nonstochastic X i , we used the fact that this implied the a i were also nonstochastic, so that E ( a i u i ) = a i E ( u i ) = a i × 0 Can’t do that now, since the X i and therefore the a i are stochastic this means that we can’t treat the a i as constants and take them out of the expectation E ( a i u i ) Instead, we invoke assumption B.7 of the u i being distributed indepen- dently of the X i Review chapter: if X i and u i are independent, then for any functions f ( . ) and g ( . ), we have: E { f ( X i ) g ( u i ) } = E { f ( X i ) } E { g ( u i ) } (8.10) Applying this result with f ( X i ) = a i and g ( u i ) = u i : 2
E ( a i u i ) = E ( a i ) E ( u i ) = E ( a i ) × 0 = 0 (8.11) (as long as B.7 holds and the errors u i are distributed independently of the X i ) This result allows us to prove unbiasedness of OLS: E ( b OLS 2 ) = β 2 + n X i =1 E ( a i u i ) = β 2 + n X i =1 E ( a i ) E ( u i ) (8.12) Since E ( u i ) = 0 by B.4 , we then have E ( b OLS 2 ) = β 2 Strictly speaking, need E ( a i ) to exist, so also need some variation in the X i as long as there is a constant term this is guaranteed by assumption B.3 Other finite-sample properties of OLS Conditional on the sample values of the regressors, the expressions for estimator variances are as before, e.g.:

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}