計(jì)量經(jīng)濟(jì)學(xué)(英文版).課件

1、Chapter 4 Statistical Properties of the OLS EstimatorsXiAn Institute of Post & TelecommunicationDept of Economic & ManagementProf. Long第1頁(yè)，共69頁(yè)。yt = household weekly food expendituresSimple Linear Regression Modelyt = b1 + b2 x t + e tx t = household weekly incomeFor a given level of x t, the expect

2、edlevel of food expenditures will be: E(yt|x t) = b1 + b2 x t4.2第2頁(yè)，共69頁(yè)。1. yt = b1 + b2x t + e t2. E(e t) = 0 E(yt) = b1 + b2x t 3. var(e t) = s 2 = var(yt)4. cov(e i,e j) = cov(yi,yj) = 05. x t c for every observation6. e tN(0,s 2) ytN(b1+ b2x t, s 2) Assumptions of the SimpleLinear Regression Mod

3、el4.3第3頁(yè)，共69頁(yè)。The population parameters b1 and b2are unknown population constants.The formulas that produce thesample estimates b1 and b2 arecalled the estimators of b1 and b2.When b0 and b1 are used to representthe formulas rather than specific values,they are called estimators of b1 and b2which ar

4、e random variables becausethey are different from sample to sample.4.4第4頁(yè)，共69頁(yè)。If the least squares estimators b0 and b1are random variables, then what are theirmeans, variances, covariances andprobability distributions?Compare the properties of alternative estimators to the properties of the least

5、squares estimators. Estimators are Random Variables ( estimates are not )4.5第5頁(yè)，共69頁(yè)。 The Expected Values of b1 and b2 The least squares formulas (estimators)in the simple regression case:b2 =nSxiyi - Sxi SyinSxi2 -(Sxi)2 2b1 = y - b2xwhere y = Syi / n and x = Sx i / n (4.1a)(4.1b)4.6第6頁(yè)，共69頁(yè)。Substi

6、tute in yi = b1 + b2xi + e ito get:b2 = b2 +nSxiei - Sxi SeinSxi -(Sxi)22The mean of b2 is:Eb2 = b2 +nSxiEei - Sxi SEeinSxi -(Sxi)22Since Eei = 0, then Eb2 = b2 .4.7第7頁(yè)，共69頁(yè)。The result Eb2 = b2 means thatthe distribution of b2 is centered at b2.Since the distribution of b2 is centered at b2 ,we say

7、thatb2 is an unbiased estimator of b2. An Unbiased Estimator 4.8第8頁(yè)，共69頁(yè)。The unbiasedness result on the previous slide assumes that weare using the correct model.If the model is of the wrong formor is missing important variables,then Eei = 0, then Eb2 = b2 . Wrong Model Specification 4.9第9頁(yè)，共69頁(yè)。 Un

8、biased Estimator of the Intercept In a similar manner, the estimator b1of the intercept or constant term can beshown to be an unbiased estimator of b1 when the model is correctly specified.Eb1 = b14.10第10頁(yè)，共69頁(yè)。b2 =nSxiyi - Sxi SyinSxi -(Sxi)22(4.3b)(4.3a)Equivalent expressions for b2:Expand and mul

9、tiply top and bottom by n:b2 =S(xi - x )(yi - y )S(xi - x )24.11第11頁(yè)，共69頁(yè)。 Variance of b2 Given that both yi and ei have variance s 2,the variance of the estimator b2 is:b2 is a function of the yi values butvar(b2) does not involve yi directly.S(x i - x)s 22var(b2) =4.12第12頁(yè)，共69頁(yè)。 Variance of b1 nS(

10、x i - x)2var(b1) = s 2Sx i2the variance of the estimator b1 is:b1 = y - b2xGiven4.13第13頁(yè)，共69頁(yè)。 Covariance of b1 and b2 S(x i - x)2cov(b1,b2) = s2- x If x = 0, slope can change without affectingthe variance.4.14第14頁(yè)，共69頁(yè)。 What factors determine variance and covariance ?1. s 2: uncertainty about yi va

11、lues uncertainty about b1, b2 and their relationship.2. The more spread out the xi values are then the more confidence we have in b1, b2, etc.3. The larger the sample size, n, the smaller the variances and covariances.4. The variance b1 is large when the (squared) xi values are far from zero (in eit

12、her direction).5. Changing the slope, b2, has no effect on the intercept, b1, when the sample mean is zero. But if sample mean is positive, the covariance between b1 and b2 will be negative, and vice versa.4.15第15頁(yè)，共69頁(yè)。Gauss-Markov Theorem Under the first five assumptions of the simple, linear regr

13、ession model, the ordinary least squares estimators b1 and b2 have the smallest variance of all linear and unbiased estimators of b1 and b2. This means that b1and b2 are the Best Linear Unbiased Estimators (BLUE) of b1 and b2.4.16第16頁(yè)，共69頁(yè)。implications of Gauss-Markov1. b1 and b2 are “best” within t

14、he class of linear and unbiased estimators.2. “Best” means smallest variance within the class of linear/unbiased.3. All of the first five assumptions must hold to satisfy Gauss-Markov.4. Gauss-Markov does not require assumption six: normality.5. G-Markov is not based on the least squares principle b

15、ut on b1 and b2.4.17第17頁(yè)，共69頁(yè)。G-Markov implications (continued)6. If we are not satisfied with restricting our estimation to the class of linear and unbiased estimators, we should ignore the Gauss-Markov Theorem and use some nonlinear and/or biased estimator instead. (Note: a biased or nonlinear est

16、imator could have smaller variance than those satisfying Gauss-Markov.)7. Gauss-Markov applies to the b1 and b2 estimators and not to particular sample values (estimates) of b1 and b2.4.18第18頁(yè)，共69頁(yè)。Probability Distribution of Least Squares Estimators b2 N b2 ,S(xi - x)s 22b1 N b1 , nS(x i - x)2s 2 S

17、xi24.19第19頁(yè)，共69頁(yè)。 yi and e i normally distributed The least squares estimator of b2 can beexpressed as a linear combination of yis:b2 = S wi yi b1 = y - b2x S(x i - x)2where wi = (x i - x)This means that b1and b2 are normal sincelinear combinations of normals are normal.4.20第20頁(yè)，共69頁(yè)。 normally distr

18、ibuted under The Central Limit TheoremIf the first five Gauss-Markov assumptionshold, and sample size, n, is sufficiently large,then the least squares estimators, b1 and b2,have a distribution that approximates thenormal distribution with greater accuracythe larger the value of sample size, n.4.21第2

19、1頁(yè)，共69頁(yè)。 Consistency We would like our estimators, b1 and b2, to collapse onto the true population values, b1 and b2, as sample size, n, goes to infinity.One way to achieve this consistency property is for the variances of b1 and b2 to go to zero as n goes to infinity.Since the formulas for the vari

20、ances of the least squares estimators b1 and b2 show that their variances do, in fact, go to zero, then b1 and b2, are consistent estimators of b1 and b2.4.22第22頁(yè)，共69頁(yè)。 Estimating the variance of the error term, s 2et = yt - b1 - b2 x tSett =1T2n- 2s 2 = s 2 is an unbiased estimator of s 2 4.23第23頁(yè)，

21、共69頁(yè)。The Least Squares Predictor, yo Given a value of the explanatory variable, Xo, we would like to predicta value of the dependent variable, yo.The least squares predictor is:yo = b1 + b2 x o (4.4) 4.24第24頁(yè)，共69頁(yè)。Inference in the Simple Regression ModelChapter 55.1第25頁(yè)，共69頁(yè)。1. yt = b1 + b2x t + e t

22、2. E(e t) = 0 E(yt) = b1 + b2x t 3. var(e t) = s 2 = var(yt)4. cov(e i,e j) = cov(yi,yj) = 05. x t c for every observation6. e tN(0,s 2) ytN(b1+ b2x t,s 2) Assumptions of the Simple Linear Regression Model5.2第26頁(yè)，共69頁(yè)。Probability Distribution of Least Squares Estimators b1 N b1 ,n S(x t - x)2s2 Sx t

23、2b2 N b2 ,S(x t - x)s225.3第27頁(yè)，共69頁(yè)。s 2=n - 2et2SUnbiased estimator of the error variance:s 2s 2(n - 2)n - 2cTransform to a chi-square distribution: Error Variance Estimation 5.4第28頁(yè)，共69頁(yè)。We make a correct decision if:The null hypothesis is false and we decide to reject it.The null hypothesis is tru

24、e and we decide not to reject it.Our decision is incorrect if:The null hypothesis is true and we decide to reject it. This is a type I error.The null hypothesis is false and we decide not to reject it. This is a type II error.5.5第29頁(yè)，共69頁(yè)。b2 N b2 ,S(x t - x)s22Create a standardized normal random var

25、iable, Z, by subtracting the mean of b2 and dividing by its standard deviation:b2 - b2 var(b2)Z = N(0,1)5.6第30頁(yè)，共69頁(yè)。Simple Linear Regressionyt = b1 + b2x t + e t where E e t = 0yt N(b1+ b2x t , s 2) since Eyt = b1 + b2x t e t = yt - b1 - b2x t Therefore, e t N(0,s 2) .5.7第31頁(yè)，共69頁(yè)。Create a Chi-Squa

26、ree t N(0,s 2) but want N(0,1) .(e t /s) N(0,1) Standard Normal .(e t /s)2 c2(1) Chi-Square .5.8第32頁(yè)，共69頁(yè)。Sum of Chi-SquaresSt =1(e t /s)2 = (e1 /s)2 + (e 2 /s)2 +. . .+ (e n /s)2 c2(1) + c2(1) +. . .+c2(1) = c2(N) Therefore, St =1(e t /s)2 c2(N)5.9第33頁(yè)，共69頁(yè)。 Since the errors e t = yt - b1 - b2x t a

27、re not observable, we estimate them with the sample residuals e t = yt - b1 - b2x t.Unlike the errors, the sample residuals arenot independent since they use up two degrees of freedom by using b1 and b2 to estimate b1 and b2. We get only n-2 degrees of freedom instead of n.Chi-Square degrees of free

28、dom5.10第34頁(yè)，共69頁(yè)。Student-t Distributiont = t(m)ZV / mwhere Z N(0,1)and V c(m)25.11第35頁(yè)，共69頁(yè)。t = t(n-2)ZV / (n- 2)where Z = (b2 - b2) var(b2)and var(b2) = s 2S( xi - x )25.12第36頁(yè)，共69頁(yè)。t = ZV / (n-2)(b2 - b2) var(b2)t = (n-2) s 2s 2(n- 2)V = (n-2) s 2s 25.13第37頁(yè)，共69頁(yè)。var(b2) = s 2S( xi - x )2(b2 - b2)

29、 s 2S( xi - x )2t = =(n-2) s 2s 2( n-2)(b2 - b2) s 2S( xi - x )2 notice thecancellations5.14第38頁(yè)，共69頁(yè)。(b2 - b2) s 2S( xi - x )2t = =(b2 - b2) var(b2) t = (b2 - b2) se(b2)5.15第39頁(yè)，共69頁(yè)。Students t - statistic t = t (n-2)(b2 - b2) se(b2) t has a Student-t Distribution with n-2 degrees of freedom. 5.16第

30、40頁(yè)，共69頁(yè)。Figure 5.1 Student-t Distribution(1-a)t0f(t)-tctca/2a/2red area = rejection region for 2-sided test5.17第41頁(yè)，共69頁(yè)。probability statementsP(-tc t tc) = 1 - aP( t tc ) = a/2P(-tc tc) = 1 - a(b2 - b2) se(b2)5.18第42頁(yè)，共69頁(yè)。Confidence IntervalsTwo-sided (1-a)x100% C.I. for b1:b1 - ta/2se(b1), b1 +

31、ta/2se(b1)b2 - ta/2se(b2), b2 + ta/2se(b2)Two-sided (1-a)x100% C.I. for b2:5.19第43頁(yè)，共69頁(yè)。Student-t vs. Normal Distribution1. Both are symmetric bell-shaped distributions.2. Student-t distribution has fatter tails than the normal.3. Student-t converges to the normal for infinite sample.4. Student-t c

32、onditional on degrees of freedom (df).5. Normal is a good approximation of Student-t for the first few decimal places when df 30 or so.5.20第44頁(yè)，共69頁(yè)。Hypothesis Tests1. A null hypothesis, H0.2. An alternative hypothesis, H1.3. A test statistic.4. A rejection region.5.21第45頁(yè)，共69頁(yè)。Rejection Rules1. Two

33、-Sided Test: If the value of the test statistic falls in the critical region in either tail of the t-distribution, then we reject the null hypothesis in favor of the alternative.2. Left-Tail Test:If the value of the test statistic falls in the critical region which lies in the left tail of the t-dis

34、tribution, then we reject the null hypothesis in favor of the alternative.2. Right-Tail Test:If the value of the test statistic falls in the critical region which lies in the right tail of the t-distribution, then we reject the null hypothesis in favor of the alternative.5.22第46頁(yè)，共69頁(yè)。Format for Hyp

35、othesis Testing1. Determine null and alternative hypotheses.2. Specify the test statistic and its distribution as if the null hypothesis were true.3. Select a and determine the rejection region.4. Calculate the sample value of test statistic.5. State your conclusion.5.23第47頁(yè)，共69頁(yè)。practical vs. stati

36、stical significance in economicsPractically but not statistically significant:When sample size is very small, a large average gap between the salaries of men and women might not be statistically significant. Statistically but not practically significant:When sample size is very large, a small correl

37、ation (say, r = 0.00000001) between the winning numbers in the PowerBall Lottery and the Dow-Jones Stock Market Index might be statistically significant.5.24第48頁(yè)，共69頁(yè)。Type I and Type II errorsType I error:We make the mistake of rejecting the null hypothesis when it is true.a = P (rejecting H0 when i

38、t is true).Type II error:We make the mistake of failing to reject the null hypothesis when it is false.b = P (failing to reject H0 when it is false).5.25第49頁(yè)，共69頁(yè)。Prediction IntervalsA (1-a)x100% prediction interval for yo is:yo tc se( f )se( f ) = var( f )f = yo - yoS(x t - x)2var( f ) = s 2 1 + +1

39、n(x o - x)25.26第50頁(yè)，共69頁(yè)。Two-sided Hypothesis5.275.6 HYPOTHESIS TESTING: 1 THE CONFIDENCE-INTERVAL APPROACH第51頁(yè)，共69頁(yè)。5.28One-Sided or One-Tail TestH0:2 0.3 and H1:2 0.3Sometimes we have a strong a priori or theoretical expectation (or expectations based on some previous empirical work) that the alte

40、rnative hypothesis is one-sided or unidirectional rather than two-sided, as just discussed.Thus, for our consumptionincome example, one could postulate that H0:2 0.3 and H1:2 0.3Perhaps economic theory or prior empirical work suggests that the marginal propensity to consume is greater than 0.3. Alth

41、ough the procedure to test this hypothesis can be easily derived from earlier works, the actual mechanics are better explained in terms of the test-of-significance approach discussed next.第52頁(yè)，共69頁(yè)。5.29 Before concluding our discussion of hypothesis testing, note that the testing procedure just outl

42、ined is known as a two-sided, or two-tail, test-of-significance procedure in that we consider the two extreme tails of the relevant probability distribution, the rejection regions, and reject the nullhypothesis if it lies in either tail. But this happens because our H1 was a two-sided composite hypo

43、thesis; 2= 0.3 means 2 is either greater than or less than 0.3. But suppose prior experience suggests to us that the MPC is expected to be greater than 0.3. In this case we have: H0:2 0.3 and H1:2 0.3. Although H1 is still a composite hypothesis, It is now one-sided.第53頁(yè)，共69頁(yè)。 To test this hypothesi

44、s, we use the one-tail test (the right tail), as shown in Figure 5.5. The test procedure is the same as before except that the upper confidence limit or critical value now corresponds to t = t.05, that is, the 5 percent level.As Figure 5.5 shows, we need not consider the lower tail of the t distribu

45、tion in this case. Whether one uses a two- or one-tail test of significance will depend upon how the alternative hypothesis is formulated, which, in turn, may depend upon some a priori considerations or prior empirical experience. 5.30第54頁(yè)，共69頁(yè)。5.31第55頁(yè)，共69頁(yè)。HYPOTHESIS TESTING: 2 THE TEST-OF-SIGNIFI

46、CANCE APPROACHTesting the Significance of Regression Coefficients: The t Test An alternative but complementary approach to the confidence-intervalmethod of testing statistical hypotheses is the test-of-significance approach developed along independent lines by R. A. Fisher and jointly by Neyman and

47、Pearson. Broadly speaking, a test of significance is a procedure by which sample results are used to verify the truth or falsity of a null hypothesis. The key idea behind tests of significance is that of a test statistic (estimator) and the sampling distribution of such a statistic under the null hy

48、pothesis. The decision to accept or reject H0 is made on the basis of the value of the test statistic obtained from the data at hand. As an illustration, recall that under the normality assumption the variable follows the t distribution with n-2 df. If the value of true 2 is specifiedunder the null

49、hypothesis, the t value can readily be computedfrom the available sample, and therefore it can serve as a test statistic. 5.32第56頁(yè)，共69頁(yè)。5.33第57頁(yè)，共69頁(yè)。5.34第58頁(yè)，共69頁(yè)。5.35第59頁(yè)，共69頁(yè)。5.36第60頁(yè)，共69頁(yè)。APPLICATION OF REGRESSION ANALYSIS:THE PROBLEM OF PREDICTIONOn the basis of the sample data of Table 3.2 we obtained the following sample regression:Y = 24.4545+0.5091Xi 5.37第61頁(yè)，共69頁(yè)。5.38第62頁(yè)，共69頁(yè)。第63頁(yè)，共69頁(yè)。5.39第64頁(yè)，共69頁(yè)。5.40第65頁(yè)，共69頁(yè)。SUMMARY