Econ 326 (002) Winter Session, Term II, January 2021 M. Vaney Problem Set 1 - Solutions 1. Show the following equalities hold: (x = (a) Pn i=1 (xi 1 n x) = 0 n X (xi Pn i=1 n X x) = i=1 xi i=1 n X = i=1 n X = Pn i=1 xi (yi y) = Pn i=1 (xi x)(yi n X n X xi xi Pn i=1 Pn n X i=1 n X 1 n yi ) x i=1 i=1 (b) xi ; y = i=1 xi n xi = 0 i=1 y) (xi x)(yi y) i=1 = n X [xi (yi y) x(yi y)] i=1 = = n X i=1 n X xi (yi xi (yi y) y) i=1 but Pn i=1 (yi y) = Pn i=1 yi n X x(yi i=1 n X x (yi y) y) i=1 Pn i=1 = y = ny n X xi (yi ny = 0 y) i=1 2. Consider the following random sample of size n = 5 drawn from a population described by the population regression function y = 0 + 1 x + u : xi 2 5 8 10 15 1 yi 8 14 16 18 24 (a) Compute x and y. x = 8 y = 16 (b) Complete a table that shows (xi each observation. x), (yi xi 2 5 8 10 15 y) yi (xi 8 6 14 3 16 0 18 2 24 7 x) (yi 8 2 0 2 8 y); (xi (xi 36 9 0 4 49 x)2 x)2 and (xi (xi 48 6 0 4 56 x)(yi x)(yi y) for y) (c) Compute estimators ^ 1 and ^ 0 . ^ 1 = = ^ 0 P5 x)(yi i=1 (xi P5 x)2 i=1 (xi y) 114 = 1:16 98 = y ^ 1x = 16 1:16(8) = 6:69 (d) Complete a table that shows y^i ; u^i and xi u^i for each observation. xi 2 5 8 10 15 yi y^i u^i xi u^i 8 9:02 1:02 2:01 14 12:51 1:49 7:44 16 16 0 0 18 18:33 0:33 3:27 24 24:14 :14 2:14 3. For the regression in question #2: xi 2 5 8 10 15 yi y^i u^i SST SSR x2i 8 9:02 1:02 64 1:0404 4 14 12:51 1:49 4 2:2201 25 16 16 0 0 0 64 18 18:33 0:33 4 0:108 9 100 24 24:14 :14 64 0:019 6 225 (a) Determine the R2 for the regression. SST = 136; SSE = 132:612; SSR = 3:388 R2 = 2 132:612 = :975 136 (b) Compute the Standard Error of the regression (estimate of the standard deviation of the disturbances). Pn 2 u^ SSR 3:389 2 ^ u = i=1 i = = = 1:13 n 2 n 2 5 2 q p ^ 2u = 1:1207 = 1:06 (c) Compute the variance of the ^ 1 OLS estimator. var( ^ 1 ) = P5 2 u i=1 (xi ^ 2u P5 i=1 (xi = x)2 x)2 1:13 = :0115 98 = (d) Compute the variance of the ^ 0 OLS estimator. var( ^ 0 ) = Pn 2 x2 Pn i=1 i 2 n i=1 (xi x) (1:13) (418) (5) (98) = :963 var( ^ 0 ) = 4. For the regression in question #2 and using results from question #3: (a) Construct the 90% con…dence interval for ^ 0 . For the 90% con…dence interval, tcrit :05;3 = 2:353: The ols estimate of 0 = :10: We will use the critical value tcrit;n 2 2 = is used as the center point and we add and subtract q crit t ;n 2 var( ^ 0 ) 2 ^0 6:69 6:69 q var( ^ 0 ) p (2:353)( :963) tcrit;n 2 2 p (2:353)( :963) 4:38 0 0 p 6:69 + (2:353)( :963) 9:00 (b) Test the hypothesis that the slope coe¢ cient is equal to zero at the = :05 level of signi…cance. Be sure to specify both null and alternative hypotheses. We are conducting a two-tailed test with hypotheses: 3 H0 : 0 =0 H1 : 0 6= 0 Using a t-test: null ^ 1:16 0 1 tcalc = q1 = p = 10:817 :0115 ^ var( 1 ) The test statistic follows a T -distribution with n 2 = 3 degrees of freedom. At the = :05 level of signi…cance the critical value is tcrit;n 2 2 = tcrit :025;3 = 3:182 The null hypothesis can be rejected at the = :05 level of signi…cance. The slope coe¢ cient is signi…cantly di¤erent from zero. 5. Consider the savings function sav = 0 + 1 inc + u p u = inc e where e is a random variable with Efeg = 0 and var(e) = independent of inc. 2 e. Assume that e is (a) Show that Efujincg = 0: Is the conditional mean assumption satis…ed? Efujincg p = Ef inc ejincg conditioning on inc then p inc will be known and constant so p = incEf ejincg and with e independent of inc; Ef ejincg = Efeg = 0 Efujincg = 0 (b) Show that var(ujinc) = 2 e inc: Is the assumption of homoskedasticity satis…ed? p var(ujinc) = var( inc ejinc) and again, with the conditioning on inc = p 2 inc var(ejinc) = (inc) var(ejinc) 2 e inc this is not constant, variance will increase with income so homoskedasticity is not satis…ed. 4 6. ^ 0 is the o.l.s. estimator of Prove the following: (a) Ef ^ 0 g = 0 from the population regression function y = 0 ^ =y 0 0+ 1 x+u. ^ x 1 from the population regression function (PRF) yi = + 0 1 xi + ui taking the average over a sample of size n y= 0 + 1x + Pn i=1 ui n and substituting this into the expression for ^ 0 Pn i=1 ui ^ 1x ^0 = 0 + 1x + n Pn i=1 ui ^ = 0 ( 1 1 )x + n and taking expectations conditional on x1 ; : : : ; xn 1X )g + Efui g 1 n i=1 n Ef ^ 0 g = and with Ef( ^ 1 have 1 )g 0 xEf( ^ 1 = 0 ( ^ 1 is unbiased estimator of Ef ^ 0 g = 1) and Efui g = 0 we 0 unbiased 7. Consider a random sample of size n = 6. f(xi ; yi ) : i = 1; 2; 3; 4; 5; 6g (a) What are the 2 restrictions placed on the residual terms (^ ui ) in constructing o.l.s. estimators for ^ 0 and ^ 1 ? 6 X u^i = 0 i=1 and 6 X xi u^i = 0 i=1 (b) What number, , of the u^i are independent? There are 2 restrictions so there will be = n 5 2 = 4 independent residuals. (c) Let xi = i for i = 1; : : : ; 6 (i.e. x1 = 1; x2 = 2; : : : ; x6 = 6). Let the …rst residuals be independent. Use the restrictions from (a) to write expressions for the dependent residuals (^ u +1 ; : : : ; u^6 ) as a function of the independent residuals (^ u1 ; : : : ; u^ ). From the …rst restriction we can write an expression for u^6 as a function of u^1 ; : : : ; u^5 6 X u^i = 0 i=1 u^6 = u^1 u^2 5 X u^i = u^3 u^4 u^5 i=1 and the second restriction along with the values of the x0i s gives 6 X xi u^i = 0 i=1 using the known values of xi (1; 2; : : : ; 6) u^1 + 2^ u2 + 3^ u3 + 4^ u4 + 5^ u5 + 6^ u6 = 0 substituting the expression for u^6 into this u^1 + 2^ u2 + 3^ u3 + 4^ u4 + 5^ u5 + 6( u^1 u^2 u^3 u^4 u^5 ) = 0 u^1 + 2^ u2 + 3^ u3 + 4^ u4 + 5^ u5 6^ u1 6^ u2 6^ u3 6^ u4 6^ u5 = 0 5^ u1 4^ u2 3^ u3 2^ u4 u^5 = 0 which gives u^5 = 5^ u1 4^ u2 3^ u3 2^ u4 u^6 = u^1 u^2 u^3 u^4 ( 5^ u1 u^6 = 4^ u1 + 3^ u2 + 2^ u3 + u^4 4^ u2 and 3^ u3 2^ u4 ) The two dependent residuals as functions of the other 4 : u^5 = 5^ u1 4^ u2 3^ u3 2^ u4 u^6 = 4^ u1 + 3^ u2 + 2^ u3 + u^4 The same problem is solved again, this time keeping the 4 known residuals together in a summation. Separating out the two unknown residuals from the summation we can write the restrictions as ! 4 X u^i + u^5 + u^6 = 0 i=1 6 and the second restriction 6 X xi u^i = 0 i=1 with xi = i for all i 6 X i^ ui = 0 i=1 4 X i^ ui + 5^ u5 + 6^ u6 = 0 i=1 this gives 2 equations with 2 unknowns u^5 and u^6 4 X u^i + u^5 + u^6 = 0 i=1 4 X i^ ui i=1 ! + 5^ u5 + 6^ u6 = 0 Taking 5 times the …rst equation and subtracting the two equations (eliminating u^5 and solving for u^6 ) ! ! 4 4 X X 0 = 5^ ui + 5^ u5 + 5^ u6 i^ ui + 5^ u5 + 6^ u6 u^6 = u^6 = i=1 4 X 5^ ui i=1 4 X (5 i=1 4 X i^ ui i=1 i)^ ui i=1 u^6 = 4^ u1 + 3^ u2 + 2^ u3 + u^4 Taking 6 times the …rst equation and subtracting the two equations (eliminating u^6 and solving for u^5 ) ! ! 4 4 X X 0 = 6^ ui + 6^ u5 + 6^ u6 i^ ui + 5^ u5 + 6^ u6 0 = u^5 = i=1 4 X (6 i=1 4 X i=1 i)^ ui + u^5 (i 6)^ ui 5^ u1 4^ u2 i=1 u^5 = 3^ u3 7 2^ u4 8. Given y= then adding and subtracting 0 0 + 1x +u from the RHS y = 0 + 1x + u + 0 y = ( 0 + 0 ) + 1 x + (u Let the new error be e = u The new intercept is ( 0 + The slope term is still 1 . 0; 0 ). 0 0) then E feg = E fu 0g = 0 0 = 0. 9. Consider the simple regression model y = 0 + 1 x + u and o.l.s. estimators ^ 0 and ^ 1 (unbiased). Let 1 be the estimator of 1 obtained when the intercept is assumed to be zero: yi = 1 xi + ei . n o (a) Find E 1 . Is 1 always a biased estimator of 1 ? For a regression through the origin the residuals will be ei = yi min n X 1 xi . Using o.l.s. e2i i=1 n X min (yi 1 @ @ : 1 1 = 2 1 xi ) 2 i=1 n X (yi 1 xi )xi =0 Pni=1 xi y i Pi=1 n 2 i=1 xi Before taking expectations, we can substitute yi = 0 + 1 xi + ui from the PRF: Pn i=1 xi (P0 + 1 xi + ui ) n 1 = 2 i=1 xi Pn 2 Pn Pn x + x + i=1 xi ui i 1 0 i=1 Pn i=12 i 1 = i=1 xi Pn Pn xi xi u i i=1 + 1 + Pi=1 n 1 = 0 Pn 2 2 i=1 xi i=1 xi Taking expectations conditional on the xi : Pn n o xi E 1 = 0 Pni=1 2 + i=1 xi The estimator will be biased unless 0 = 0 or x = 0: (b) Compare the variances of the two estimators for 8 1 1 ( ^ 1 and 1 ): We know that var( ^ 1 ) = Pn 2 i=1 (xi To …nd var( 1) we can use 1 Pn xi = 0 Pni=1 2 + i=1 xi With population parameters var( 1) 0 = = = = Pn i=1 xi ui 1 + Pn 2 i=1 xi constant and conditional on the x0 s Pn xi u i var Pi=1 n 2 i=1 xi ! n X 1 xi u i P 2 var ( ni=1 x2i ) i=1 n X 1 x2i var(ui ) Pn 2 2 ( i=1 xi ) i=1 n X 1 x2i 2 Pn 2 2 ( i=1 xi ) i=1 Pn 2 2 xi Pn i=12 2 ( i=1 xi ) and = x)2 1 2 = Pn i=1 x2i ^ 1 ) can be written as Pn (xi Comparing the variances: the denominator of var( i=1 P x)2 = ni=1 x2i nx2 and provided x 6= 0; this will be less than the denominator P of var( 1 ) = ni=1 x2i . var( 1 ) var( ^ 1 ). 10. The simple regression model y = 0 + 1x + u gives o.l.s. estimators ^ 0 and ^ 1 . (a) Consider a regression of c1 yi on c2 xi . Find the estimators ~ 0 and ~ 1 and show that they can be expressed as function of ^ 0 and ^ 1 . Given c1 and c2 are constants then the average of c1 yi = c1 y and the average of c2 x1 will be c2 x. We can write: Pn x c2 x)(c1 yi c1 y) ~ = i=1 (c P2n i 1 c2 x)2 i=1 (c2 xi 9 factoring out ~ 1 = = = = = Pn c c (x x)(yi i=1 P2n 1 2i x)2 c (x Pni=1 2 i c2 c1 i=1 (xi x)(yi P c22 ni=1 (xi x)2 P c2 c1 ni=1 (xi x)(yi P c22 ni=1 (xi x)2 P c1 ni=1 (xi x)(yi Pn c2 x)2 i=1 (xi c1 ^ c2 1 y) y) y) y) For ~ 0 ~ 0 = c1 y = = = = ~ 1 c2 x c1 ^ c1 y c2 x c2 1 c1 y c1 ^ 1 x c1 (y c1 ^ 1 x) c1 ^ 0 (b) What does this imply about the e¤ect of a linear scaling of variables, such as a change of units, on estimates. Slope coe¢ cient will be scaled in direct proportion to any scaling of the dependent variable (c1 ), and in an inverse fashion to scaling of the explanatory variable ( c12 ). Intercept term will be scaled in direct proportion to the dependent variable scaling (c1 ). (c) Consider a regression of c1 + yi on c2 + xi . Find the estimators that they can be expressed as function of ^ 0 and ^ 1 . 0 and 1 and show Similar to (a) above. Note that the average of c1 + yi will be c1 + y (likewise for c2 + xi ). The equation for 1 will have ((c2 + xi ) (c2 + x)) = (xi x) and ((c1 + yi ) (c1 + y)) = (yi y). Both c1 and c2 will drop out of the equation for ^ 1 so we …nd that 1 = 1 . For 0 0 with 1 = (c1 + y) 1 (c2 + x) = y 1 x + c1 1 c2 = ^1 0 = y ^ 1 x + c1 ^ 1 c2 = ^ 0 + c1 c2 ^ 1 10