Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 4

Problem 3 : a. From these data estimate the following regression model: ln Ct = 1 + 2 ln It + 3 ln Lt + 4 ln Ht + 5 ln At + ut Interpret the results. Solution : .

gen c1 = ln(c) . gen i1 = ln(i) . gen l1 =ln(l) . gen h1 = ln(h) . gen a1 = ln(a) . reg c1 i1 l1 h1 a1

Interpretation : The model can be written as Chat = -1.50 + .47 ln(I) + .28 ln (L) -.01 ln(H) + .44 ln(A) A high F statistics implies that the model as a whole is statistically significant. The t statistics of I , L and A are high enough to suggest that they all are statistically significant on a stand alone basis too . The R2 and the adjusted R2 are also in the high 90s suggesting a good model. b. Obtain the residuals and standardized residuals from the preceding regression and plot them. What can you surmise about the presence of autocorrelation in these residuals? Solution : . predict r, resid . predict r1, rstandard . plot r r1

The graph suggests that there is a good chance for the presence of autocorrelation. c. Estimate the DurbinWatson d statistic and comment on the nature of autocorrelation present in the data. Solution : . tsset year time variable: year, 51 to 80 delta: 1 unit

. dwstat Durbin-Watson d-statistic( 5, 30) = .9549404 D statistic from Table ( 95 % confidence ) = 1.138 Since d statistic < d critical ( table value ) , the model has a positive first order autocorrelation. d. Carry out the runs test and see if your answer differs from that just given in c. Solution:

Since P value is about 1 % , we can confirm that there is a positive autocorrelation. This is in line with what we found using the durbin-watson d test.

e. How would you find out if an AR(p) process better describes autocorrelation than an AR(1) process? twoway (connected r year, sort)

We can see that mostly a positive residual is followed by another positive residual and likewise a negative residual is followed by another negative residual indicating that it is a AR(1) process.

We also check a partial autocorrelation plot .

. pac r

A breusch godfrey test can also be conducted to test AR(p)

Here we can conclude the model could be AR(2) as the p value decreases when we move from AR(1) to AR(2) and there after increases. Hence this model could be said as AR(P)

You might also like