# 8 out of sample forecasting the linear regression

This preview shows pages 15–18. Sign up to view the full content.

8. Out-of-sample forecasting The linear regression model was introduced as a forecasting scheme. The question we now address is: How reliable is an out-of-sample forecast? Consider the linear regression model (2), and suppose we observe Then the forecast X n % 1 . of is where the OLS estimators are computed on the basis of Y n % 1 ˆ Y n % 1 ' ˆ α % ˆ β . X n % 1 , ˆ α and ˆ β the observations for j = 1,2,..., n . The actual but unknown value of is Y n % 1 = α + Y n % 1 β . X n % 1 % U n % 1 , so that the forecast error is: Y n % 1 & \$ Y n % 1 ' U n % 1 & ( \$ " & " ) & ( \$ \$ & \$ ). X n % 1 ' U n % 1 & j n j ' 1 1 n % ( X n % 1 & ¯ X )( X j & ¯ X ) ' n i ' 1 ( X i & ¯ X ) 2 . U j . (28) See the Appendix for the latter equality. It follows now from Lemma 3 that under Assumptions I through V, where Y n % 1 & ˆ Y n % 1 - N [0, σ 2 Y n % 1 & ˆ Y n % 1 ], F 2 Y n % 1 & \$ Y n % 1 ' F 2 n % 1 n % ( X n % 1 & ¯ X ) 2 ' n j ' 1 ( X j & ¯ X ) 2 . (29) See the Appendix. Denoting, \$ F 2 Y n % 1 & \$ Y n % 1 ' \$ F 2 n % 1 n % ( X n % 1 & ¯ X ) 2 ' n j ' 1 ( X j & ¯ X ) 2 , (30) it follows now similar to Proposition 6 that

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

View Full Document
16 Proposition 8 . Under assumptions I - V, ( Y n % 1 & ˆ Y n % 1 )/ ˆ σ Y n % 1 & ˆ Y n % 1 - t n & 2 . This result can be used to construct a 95% confidence interval, say, of Look up in Y n % 1 . the table of the t distribution the critical value of the two-sided t-test with n ! 2 degrees of t ( freedom. Then it follows from Proposition 7 that 0.95 ' P [ & t ( # ( Y n % 1 & \$ Y n % 1 )/ \$ F Y n % 1 & \$ Y n % 1 # t ( ] ' P [ & t ( \$ F Y n % 1 & \$ Y n % 1 # Y n % 1 & \$ Y n % 1 # t ( \$ F Y n % 1 & \$ Y n % 1 ] ' P [ \$ Y n % 1 & t ( \$ F Y n % 1 & \$ Y n % 1 # Y n % 1 # \$ Y n % 1 % t ( \$ F Y n % 1 & \$ Y n % 1 ] (31) Thus, the 95% confidence interval of is Y n % 1 [ ˆ Y n % 1 & t ( ˆ σ Y n % 1 & ˆ Y n % 1 , ˆ Y n % 1 % t ( ˆ σ Y n % 1 & ˆ Y n % 1 ]. Observe from (30) that increases with and so does the width of the ˆ σ Y n % 1 & ˆ Y n % 1 ( X n % 1 & ¯ X ) 2 , confidence interval. Thus, the father is away from the more unreliable the forecast X n % 1 ¯ X , ˆ Y n % 1 of becomes. Also observe from (30) that and that gets close to Y n % 1 ˆ σ Y n % 1 & ˆ Y n % 1 \$ ˆ σ , ˆ σ Y n % 1 & ˆ Y n % 1 ˆ σ if n is large because lim n 64 ' n j ' 1 ( X j & ¯ X ) 2 ' 4 . 9. Relaxing the non-random regressor assumption As said before, the assumption that the regressors X j are non-random is too strong an assumption in economics. Therefore, we now assume that the X j ‘s are random variables. This requires the following modifications of the Assumptions I-V: Assumption I * : The pairs are independent and identically distributed . ( X j , Y j ), j ' 1,2,3, .... , n , Assumption II * : The conditional expectations are equal to zero : E [ U j | X j ] E [ U j | X j ] / 0. Assumption III * : The conditional expectations do not depend on the X j 's and are E [ U 2 j | X j ] finite, constant and equal : ( This is called the homoscedasticity E [ U 2 j | X j ] / σ 2 < 4 . assumption .)
17 Assumption IV * : Conditional on X j , U j is N (0, σ 2 ) distributed . The Assumptions I * and II * imply that for j = 1,..., n , E [ U j | X 1 , X 2 ,..., X n ] / 0, (32) and similarly the Assumptions I * and III * imply that for j = 1,..., n , E [ U 2 j | X 1 , X 2 ,..., X n ] / F 2 . (33) Because (loosely speaking) conditioning on is effectively the same as X 1 , X 2 ,..., X n treating them as given constants, most of the previous propositions carry over: Proposition 9. Under Assumptions I * -IV * , Propositions 1 and 4 through 7 carry over, and the results in Propositions 2 and 3 now hold conditional on X 1 , X 2 ,..., X n .

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 ]}

### What students are saying

• As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

Kiran Temple University Fox School of Business ‘17, Course Hero Intern

• I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

Dana University of Pennsylvania ‘17, Course Hero Intern

• The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

Jill Tulane University ‘16, Course Hero Intern