ECO00047M Econometrics 1 Spring 2020-21
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
ECO00047M - Spring
MSc Examinations 2020-21
Econometrics 1
Section A
1. Indicate which of the following five statements are true (T) and which are false (F). For example, if statement (n) is false, your answer should read, ”(n) F”. For all regression models the regressors are considered fixed and non-constant.
(a) Let A be a fixed n × k matrix where k < n and there is at least one square sub-matrix of A of order r < k whose determinant is not equal to zero. The matrix AA is positive semi-definite. (1 mark)
(b) Consider the linear regression model yi = α+βxi +εi, i = 1, 2,...,n, where E (εi) = 0 and E (ε) = σ2 for each i, and E (εiεj) = 0 for all i j. The model is estimated using a large sample size and the null hypothesis H0 : α = 0 is tested. It is found that the null hypothesis is not rejected. The appropriate conclusion is to drop the intercept and run the regression again. (1 mark)
(c) Consider the linear regression model yi = βxi + εi, i = 1, 2,...,n, where E (εi) = 0 and E (ε) = σ2 for each i, and E (εiεj) = 0 for all i j. Let ei, i = 1, 2,...,n, be the residuals of the estimated regression model. It holds that ei = 0. (1 mark)
(d) Consider the linear regression model yi = βxi + εi, i = 1, 2,...,n, where E (εi) = 0 and E (ε) = σ2 for each i, and E (εiεj) = 0 for all i j. The R2 for the model is given by R2 = 1 − (1 mark) (e) Consider the linear regression model yi = α+βxi +εi, i = 1, 2,...,n, where E (εi) = 0 and E (ε) = σ2 for each i with σ2 unknown, and E (εiεj) = 0 for all i j. Using this model for n = 40 we have predicted that a household with $2,000 weekly income would spend $287.61 on food. If the standard error of our forecast is 90.63 then a 95% prediction interval suggests that a household with $2,000 weekly income will spend approximately between $104.14 and $471.08 on food. (1 mark)
2. Indicate which of the following five statements are true (T) and which are false (F). For example, if statement (n) is false, your answer should read, ”(n) F”. For all regression models the regressors are considered fixed and non-constant.
(a) Consider the regression model y = Xβ + ε, where E(ε) = 0, E(εε) = σ2In, and the error terms ε are uncorrelated. Applying the orthogonal projection matrix X(XX)− 1X to the predictive value of the estimated regression does not affect its value. (1 mark)
(b) Consider the regression model y = Xβ+ε, where E(ε) = 0, E(εε) = σ2In, and the error terms ε are uncorrelated and normally distributed. Let b denote the OLS estimator of β. For the statistic associated with testing H0 : βj = βj0 we have that
bj − βj0 |
∼ N(0, 1), |
2cj (XX)− 1cj |
where βj and bj are the jth element of β and b respectively, appropriate selection vector.
and cj= [0, 0,..., 1, 0,...] is the (1 mark)
(c) Consider the regression model yi = β0 +β1x1i+β2x2i+εi where E (εi) = 0 and E (ε) = σ2 for each i, and E (εiεj) = 0 for all i j. Let 1 be the estimator obtained when x2i is omitted from the equation. The bias in 1 is negative if β2 < 0 and x1i and x2i are negatively correlated. (1 mark)
(d) The following assumptions are required for parameter estimation of the model yi = α + βxi + εi, i = 1, 2,...,n, by ordinary least squares: E(εi) = 0; Var(εi) = σ2; E(εiεj) = 0 for i j;εi ∼ N(0,σ2). (1 mark)
(e) Consider the regression models: Model 1: yi = β1 + β2x2i + εi , and Model 2: yi = α1 + α2x2i + α3x3i + εi, i = 1,...,n, where E (εi) = 0, E (ε) = σ2 for each i, and E (εiεj) = 0
for all i j. Model 2 must have an R2 at least as high as that of Model 1. (1 mark)
3. Consider the simple linear regression model
yi = α + βxi + εi, i = 1, 2,...,n,
where E (εi) = 0 and E (ε) = σ2 for each i, and E (εiεj) = 0 for all i j .
(a) Instead of the usual least squares estimator you would like to consider the estimator = . Investigate whether this estimator is unbiased and derive its variance. (3 marks) (b) Show that the estimator is less efficient than the least squares estimator. (3 marks)
4. Consider the equation
salesi = β1 + β2pricei + β3adverti + β4advert + εi, i = 1, 2,...,n,
where the disturbances are normally distributed with E (εi) = 0 and E (ε) = σ2 for each i, and E (εiεj) = 0 for all i j. Describe how you would construct an F- test of the hypothesis H0 : β3 + 3.8β4 = 1 against the alternative H1 : β3 + 3.8β4 1, using the restricted and unrestricted model formulation without the use of any matrix notation (both the restricted and unrestricted regressions need to be clearly specified) and indicate the reference distribution of your test statistic. (6 marks)
5. Consider a nonlinear growth model, yt = β1 + β2 exp( −β3t) + εt , t = 1,...,n, where βk , k = 1,..., 3, are unknown parameters of interest, and εt is a random disturbance with zero mean and constant variance. Suppose one wants to use the nonlinear least squares method to estimate the parameters, βk , k = 1,..., 3. Write down the objective function of the nonlinear least squares method and derive the first order conditions. (6 marks)
6. Consider a dynamic regression model, yt = β1 + β2xt + β3yt − 1 + εt , t = 1,...,n, where xt is an explanatory variable whose values are fixed, yt − 1 is the value of yt lagged by 1 time period, and εt is a random disturbance with zero mean and constant variance. Com- ment on the properties of the OLS estimator of β = (β1,β2,β3) when there is autocor- relation in εt. Suggest a potential remedial measure for autocorrelation in such a model. (Word limit: 100 words) (6 marks)
7. Consider the linear model yi = β1 + β2xi + εi , i = 1,...,n, where xi is an explanatory variable whose values are fixed, εi is a random disturbance satisfying E(εi) = 0 and E(ε) = σ2v, where σ2 is a parameter whose value is unknown and vi is a variable with known non- constant values. Describe the properties of the OLS estimator of β = (β1,β2) for this model and explain how you can obtain an improved estimator of β via the weighted least squares method. (Word limit: 100 words) (6 marks)
Section B
ln(wagei) = β1 + δ1femalei + β2educi + δ2femalei × educi + ui (8.1)
where i = 1, 2,...,n that includes a dummy variable for gender such that femalei = 1 if the
person is a female, 0 if the person is a male. The variable educ is the years of education. Model (8.1) was estimated providing the following results:
ln(wagei) = 1.739 − 0.3319 femalei + 0.0539 educi − 0.0027 femalei × educi + i
RSS = 393 R2 = 0.243 n = 2000
Describe fully how you would test whether the equation for wage determination is the same
for men and women at the 1% significance level. (10 marks) (b) Consider next the alternative model of wage determination
ln(wagei) = β1 + δ1femalei + θ1mediumi + θ2largei
+ϕ1femalei × mediumi + ϕ2femalei × largei + β2educi + ui (8.2)
that includes two dummy variables medium and large, defined as mediumi = 1 if the firm has 50 to 199 workers, 0 otherwise, largei = 1 if the firm has more than 199 workers, 0 otherwise. The reference category corresponds to small firms with up to 49 workers. The dummy variable for gender and the variable educ are defined the same as in (a).
Model (8.2) was estimated providing the following results:
ln(wagei)
RSS
= 1.624 − 0.262 femalei + 0.361mediumi + 0.179largei
−0.159femalei × mediumi −0.043 femalei × largei + 0.0497educi + i (0.050) (0.051) (0.0024)
= 359 R2 = 0.308 n = 2000
Describe fully how you would test whether small firms discriminate against women in terms of pay more or less than larger firms. (10 marks)
9. A researcher considers the model, yi = β1 + β2x + β4x3i + εi, for household expenditure data taken from a US survey, where y denotes the fraction of total consumption expenditure spent on food, x2 denotes total consumption expenditure (measured in $10,000) in a house- hold, and x3 denotes household size. Panel 9.1 below shows some estimation results she obtains for this model.
(a) Assuming that the classical assumptions about disturbances (i.e., zero mean, constant variance, and no correlation) are satisfied, use relevant results in Panel 9.1 to test the null hypothesis H0 : β3 = 1/2 using a 5% significance level. Also construct a 95% confidence interval for β3. (6 marks)
(b) Describe how the researcher may carry out an LM test of the hypothesis H0 : β3 = 1/2 using the nR2 form. (Word limit: 150 words) (7 marks)
(c) Suppose that, after conducting the above tests, the researcher decides to use the model, yi = β1 + β2x + β4x3i + εi, for the household expenditure data. But she suspects that the disturbance variance is non-constant and it depends on the household size x3. De- scribe how she may carry out a Breusch-Pagan test of heteroskedasticity in this case. (Word limit: 150 words) (7 marks)
Panel 9.1. Dependent variable: Y Model: Y = β1 + β2X3 +β4X3 Method: Nonlinear Least Squares; Included Observations: 54 |
||||
|
Estimated Coefficient |
Standard Error |
t-Statisitc |
Probability |
1 |
0.4467 |
0.0836 |
5.3456 |
0.0000 |
2 |
-0.2508 |
0.0812 |
-3.0887 |
0.0033 |
3 |
0.4279 |
0.2012 |
2.1267 |
0.0384 |
4 |
0.0150 |
0.0014 |
10.6111 |
0.0000 |
R-squared: 0.8517 Sum of Squared Residuals: 0.0224 Log likelihood: 133.6349 |
10. A researcher considered the following model for determination of aggregate consumer expenditure in the UK between the first quarter of 1970 and the fourth quarter of 1995:
log(Ct) = β1 + β2 log(It) + β3 log(Wt) + β4rt + εt , (10.1)
where log denotes the natural logarithm, Ct is the real household expenditure (in millions of pounds), It is the real household disposal income (in millions of pounds), Wt is the real net financial wealth of the personal sector (in billions of pounds), rt is bank base interest rate (in %).
(a) Panels 10.1 and 10.2 show some results obtained for equation (10.1). Do the results provide evidence of autocorrelation for equation (10.1)? What is the likely sign for the first- order autocorrelation coefficient if autocorrelation is present? (6 marks)
2022-01-08