: i oti nese TI hay ue pa ru waa e Sey & if te. Pike e @ eet ae coe @ fare the quest Commer forannn reat _ Aabmoes cadodhr chy 5 We 2 Gad hold Ka i rahe he oan dy-the. de tua. hea 4, ite, ig ie bedpertrem hy tre ge Be | aay. Aeocloumlg F Inestreys bred Boned Ses at wae angunn2& shook tet lobes? anand \ sez 19 anton Aabghre> ine diker, See P eae rascal) angen SOV - QH oF u53F2 janizetes @ wx oneursad ting, Argkeroes cade. S peas nly Aaloa| ; —Aee aaranicl oy* 1 Ar wt AN the explanatory ae in eh, SAGO “Ys cal 5 ta \Udrelhe. ee Soa: Mra prt Petet Wa © k Sogpesh Cy ae og ile: oe Ly eX, | ex — 3 Gg iS Se Whew ef Mra iS pupte = fprd % be aaah oF Gnd LrcHelokat. An ily bodys — . Ue TMB eau beer | Fetlowing abe Flaked Xi ia sag ac Jak eo Shor oben, acunt gS ae nage ts ele B thet a2 more map\ Vine Oreck & Pa. Scie ahs Gccerdans Ao the Ualvsd ot | Ke ots Spotted beret; | dr too At ee saw JoAain: + nod & ee oe Ayrokl, Surs3 & Aa a tne, Daseer he see ges, Res; -% i QgrsrP” ae ess team Hh mi (Oh eA ae | eek & (maeshjae Ty . ie 2 ab— z io ee, ; eee en Ta | or: a eer cf paranLers| fr be ibrnabed mnctdnng wdeart Ste, Comeudr “toa AO Ac RES2/Af ete. | oO Noes feacloyers g attonad oc Eas Z | o Sf, 4a 1S arr wad A be narra thig. venotly ay) dunbuted C (pred we | awd © tie bm & ram 2acadi— a tata Be alle 5 thon. Le Car be photon ade wh, oh an ® eS | awe ng. FAs teh tt, emanate Anne eee af eael if hec- 24), ans? ot f- a ee Rss. (iss gee Show: oh oud 78 thar WA sagen ey RSS) 5 Dae \Avgptiness 5 ad te Ts pepe, seem, Pree. eee mel J. | 2 RSS> tae oe As be. SrvoLey, hom RSS), Wwe Ae TH Bes Orn hee Be mo Poteet hy Grou. stra, “the tert stotefe RSS2 fe s Tha C) Central Obeanok oud are. oyatred Ae stospen, Be | aceen UL oke the i Feta a lpokttu22t, and. tre Morge YWagrante. Ger | Naneth, Ean Tho oleh ty % 1 jth. Groh Les om rF ee yee 5 tegt Fe AO ee : - Acco d on Lie GA eS Chat nn. | eehnrentay the peat op the tert Pee). | ’ “Ge 10 145 387 é 9110695) 43 046 6S the anus ar ra erA- The we hh Aupod ers, | he Buolh voMauig gle? One eo (2 Sn fen sp athomna tos whnah wobriery | | 5‘ ee i eee eae ped pater tte bh Pe yf Nee "4 tine oe Aug, es Whom ue iB fee o [- pdt Cpe etch) | Henk nut Apo Zz rot | eres cas fh tee Gre Bi aI rag uotra be | ee Mite Canta exper panne, i ef | Quontot ips bey Gath jd and ses Siupaest Hort Se rs aie oT \ Zz g j= abad | Ane. aap a rs QD | 16 oy pena Oo alecut a | Bet eee Fi fos C= fe qo Coe 1 ae So & | ; aden fog” —-to income for 2 cross section of 30 families. Suppose we postulate that consumption expenditure is linearly related to income but that heteroscedasticity is present in the data. We further postulate that the nature of heteroscedasticity is as given in Eq. (1.5.10). The necessary reordering of the data for the application of the test is also presented in Table 11.3. Dropping the middle 4 observations, the OLS regressions based on the first 13.and the last 13 observations and their associated residual sums of squares are as shown next (standard errors in the parentheses). Ragression based on the first 13 observations: f= 3.4094 + 0.6968X, ee spothetlcal Data on Consumption Expenditure ¥{5} and Income 2X{§) o> Iuetsate the Golefeld- Quiudt Test é | To illustrate the Goldfelel_Quandt test, we present in Table 11.3 data on consumption expenditure in relation Data Ranked by Values y x y x 33 80 55 80 65. 100 70 85 70 85 7s 50 80 no 65. 100 73 120 74 105 a4 115 80 10 98 230 8a 5 95 140 8 120 50 125 90 125 rs 90 98 130 74 105, 95 140 no 160 108 145 13 180 3 150 125 165 no 160 108 145 125 165 | Middle 4 15 180 us 180. | observations 140 25 130 185 120 200 135 190 145 240 120 200 130, 135 140 205 152 220 144 210 144 210 152 220 Ws 245 140 225 180 260 137 230 as 190 14s 240 140 208 17s 245; 178 265, 139 250 191 270 180 260 137 230 178 265 189 250 191 270 Regression hased on the last 13 observations ¥, =~ 28.0272 + O.7941K, (30.6421) (0.1319) From these results we abtain RsS/df 1836.8/71 RSS\/df 37717711 4-407 07681 RSS; = 19368 df= 11 The critical F value for 11 numerator and F (=a) value cr sds the critical value, we m. er, if the level of signi ticity. (Why?) Note that the jominator df at the 5 pervert level iy 2.82 Since theestim: conclude that thee a hetoras ince is fixed at T percent, we may ri value of the observed tis 0.014 dasticity in the error vari tthe assumption of hoThie Following regressions were fited using the Shanghai sc cost data introduced in Section 6.4 (standard errers in poreaeses) Q 5 \/ Casr =24,000+ 3390" (27,000) (50) COST =si,000 —4.0000CC + 152N 1 2B4NOCC R= 068 (31,000) ¢41,000) 160) (5) where COST is the annual cost of running & school, is the auther of students, OCC is s dummy variable defined to be 0 for regular schools and { for occupational schools, and NOCK is 8 slope dummy variable defined as the product of N and OCC. There are 74 schools in the sample, With the data sorted by 1, the regressions ate fitted again for the 26 savaliestand 26 largest schools, the residual sum of squares being ac shawn in the table Zesmalien 26 lapet Pest segreesion. Taxio® —— s4axi0" Second regression T1013 Rx a? & @ Perform a Goldfeld. Quandt test for heteroscedusticity for the To models and. with reference to Figure 6.5, explain why the problem of heteggscedasticity is less severe in the second mode nse) rezet.one ES ds RS, rte e sinallest schools and RSS; for the 26 largest schools. In the frst regression, RSSYRSS, ~ (54.4 10!°UC7.8 10") = 697. There are 24 degrees of freedom in each subsumple (26 observations, 2 parsietersestimated), The critical value of F424) is approximately 3.7 a the 01 percent level, sd so we reject he ml hypothesis of bomascedasticty af that level. In he second regreision, RSSURSS, ~ (13.8% 10°)(6 7 10!9}-=2 06, There are 22 degrees of freedom in cach subsample (26 obser { parameters estimated). The critical value of F(2222) is 2.05 a the $ percent level aud s0 we (ust) de not hypethesis of homoscedasicty ect the aul a significance level Why is the problen ofheteroscedasticity les im Exercise AS? for the on veal that regular schuuls ref onseduetite Hie tae 5588 of observ 4asiequssey fanioets | 42/04s6&- Nawrh Aradanws a Procthemat us , Bemeoww & PF tnonntheteroscedastic, ir pective of whether the disturhance term is traly heteroscedastic ar not. The first regression takes ao account of this and the Goldielé-Quandt testtherefore indicates significant heteroscedasticity. In the second regression the problem of sapparent heteroscedasticity does not arise because the intercept and slope du variables allow sepaiaie imphet regressiom fines for the twa types of school Looking closety «the diagram, the observations for the occupational schools exhibit a classic partern of true heteruscedasticay, and this would be canfirmed by 4 Guldfeld Quand: test confined to the subsample of Additional Exercise 482) ose schools (see However the observations for the regular Schools appear to be homoscedastic and this accounts for the fact that we did not (quite) reject the null hypothesis of homoscedasticity for the sumbined sample. A (b) “The file educ dia in the beteroscedastic data sets folder on che ‘website contains intemational cross-section data on expenditure on education, EDUC, gn and population, #07, tora somple of 38 countries in 1997. EDUC and GDP are measured in US $ million and POP is measured in thousands. Download the data set, plot « scatter diagram of EDUC on GOP, and comment on whether the data set appears to be subject to heteroncedasticity. Som the data set by GDP and perform a Goldfeld-Quandt test for heteroscedusticity. manning regressions using the subsamples of 14 countries with the sanallest and greatest GDP, Y () seer mene tn ct hn ‘grnss domestic product, GDP, for the 38 countries in the sample The observations exhibit heteroscedasticity. Sorting them by GDP and regressing EDUC on GDP for the subsamples of (4 countries ss domestic p 4&4 eqns 3¢>/ e764) 43) 04 SSE WAL Ecen ows ge‘with smallest and greatest GP, the residual sum of squares for the first and second subsamples, denoted RSS, and respectively, are 1,660,000 2nd 63,113,000 respectively. Henes, y= RSS _ 93185000 2 : 3302 RSS, 16640 { ce “The eritial value oF FUU2I2) a the OL percent level is 7.007 some reject the null hypothesis of homoscedasticity (c) Repeat ere +s, using the Goldield-CQuane testo investigate whether aeaing by population o by GDP, or whether rani the tsrcebss Sop itni: cn ok lice tn heteroscedasticity. Compare the results ofregressions using the entire sample and the alternative specifications Dividing through by population, PGR, the model becomes EDUC POP 1p GOP, Pop ** POP” Por svith expenditure en edtuestion per capi lenoted EDUCPOP, hypothesized ta be a Function of gross domestic product per capita, GDPFOF, and the reciprocal of population, POPREC, with no intercept. Sorting the sample by GOPPOP and mnning the regression for the subsamples of 14 countries with smallest and largest GDPPOP, RSS; © $6.541 and RSS; = 1,915,515. Now Laissis #0212) e 54541 = 25.08 Thus thee model is still subject to heteroscedasticity af the 0.1 percent level. This is evidesit in the Figure 2.000 if | po 2 1000 | ‘ |Tus again we would not reject the null hypothesis of homoscedasticity 3 0 " 1 og GDF Expenditure on edueation and GDP. logarithmic he third and tourth models both appear to be fr hicteroscedasticity. How-de we choode between them? We wil sicily. Howse we chomse between Gent ew > Gqafaine the repression resuks, shown For te wo models withthe fall sample: - 8 0233 = 0.1349 ~ 0.1108 SOUCGOP | | -230.0823 9878309 O4B1S93 .0026696 by 381 Se In equation form, the first regression rs EDLC - 01-244 GDP ~ GoP (0.004). 98.5) Multiplying through by GDM, it way be rewritten ~ 35.1 omsane Ivimplics that expenditure on education accounts for 48 percent of gross domestic product at the «margin. The constant docs not have any'sensible interpretation, We will compare this with the oxpur ‘rom an OLS egression tka hererascedasticity yakes no artempt to eliminate esrier0 s45819.2 coat. neenese isn. 4683 The slope coefficient, 0.48, 1s identical to two decimal piaces ‘This is not entircly a surprise, since heterascedasticity does not sive ise 1 bias and so there should be no systematic difference between the estimate from an-OLS regression aad that trom a specification that eliminates heteroscedasticity OF course, itis ¢ surprise that the eslimates are s0 close. Generally there would be some random difference, and of course the OLS estimate would tend to be less accurate, 4 this ease, the main difference is in the estimated standard error. That for the OLS regression is actually smaller thu that for the regression of EDUCGDP on GDPREC, Tutt is misteading, tis incorcectly calculated and me know thot, since OLS is inefficient. the true standard error for the OLS cstimate is actually larger. ‘The Jogarithmuc regression in cuuation form is lngEDUC = $03 © LIT log GDFIncquation form, the first regression 1S 0.048 2351 SDP xo.o0d) (885) Mulkiptying through by GDP tt may be rewritten | / EDUC = =234.1 +0.048GDP that expenditure @n education accounts for 4.8 percent of rors domestic product atthe margin. The constant docs not have any sensible inerpretation. We will compare this withthe oalput ‘fom an OLS regression thar makes no attempt ro eliminate beteroscedasticity: = reg E00C Gur sumber of obs = 38 osTLe+0$ Gasai9-2 coef. Sta. Err. sense 0021208 oa 3482 3. “eqns | "160 46%5 311-8899 Sis 792,621 474.684 “The slope coefficient, 0.48, is identical to twa decimal paves “This ie not entirely a surprise, since heteroscedusticity does Hat sive rise to bus and so there should be no systematic ifference etween the estimate from an OLS regression and that from = specification tha eliminates heteroscedesticly. Of course, itis a surprise thatthe estimates are so close. Generally there wovld be ‘some random difference, and of course the OLS cstimate would tendo be Fewsmecurate, In this case, the main difference is in the xtimated standard error. That for the OLS regression ip acutally ‘cnaller than that for the regression of EDUCGDP on GDPREC tot iti nisleading: This incorrectly calculated and we know That vince OLS is inefficient, the true standart error for the OLS estimate is actually ‘The logarithmic regression in equation form t fogEDUC = 503+ LIT CDP qos2) 07) smplvine that the elasticity af expendiaure on education with ef annonIngermrabne and: Poets Aards. a> o better, mtesAsng. poms fR- Aurtter errr: fo enannpe ay ane = eS oe. Sg tv aso. pee Pek c-2-| Blot No Gre Novs deehi- 35° qe1e4us sé? AMN06 Ag YB10K66E-S&S To illustrate the Goldfeld-Quandt test, we present in Tabt to income for a cross section of 30 families. Suppos#We postulate that consumption expenditure is linearly Telated to income but that heteroscedasticity is present in the data. We further postulate that the nature of zdata on consumption expenditure in relation heteroscedasticity is as given i test is also presented in Table Dropping the middle 4 observations, the OLS regressions based on the first 13 and the last 13 observations and their associated residual sums of squares are as shown next (standard errors in the parentheses). Regression based on the first 13 observations: : ‘The necessary reordering of the data for the application of the = 34094 + 0.6968%, (8.7049) (0.0744) BBB7 RSS) = 377.17 dha ‘Hypothetical Das on Coasumpdon Expendicure ¥ }) and Income X{$) 10 Huxtrare the Goldfeld-Qunadt Test Data Ranked by : X Values y x y x 35 80 35 80 65 100 70 85 70 85 75 90 80 vo 5 100 73 120 74 105, 84 ns 80 0 98 130 B4 115 \ 95 140 79 120 30 125 90 125 75 90 98 130 4 105 95 140 0 160 108 145 na 150 n3 180 128 165 no 160 108 145 125 165 | Middle 4 us 180 ms 180 | observations 140 Bs 130 185 120 200 135 190 143 240 120 200 130 185 140 205 152 220 144 210 144 210 152 220 Ws 245 140 225 180 260 137 230 135 130 145 240 140 205. 175 285 178 265 189 250 191 270 180 260 137 BO 178 265 189 250 191 270 Regression based on the last 13 observations ~ 28.0272 + 0.7941X, (B0.6421) (0.1319) 07681 RS From these results we abtain 5368 0 di = 17 RSSz/df 1536.8/77 RSSifdl 37797711 4.07 The critical F value for 11 numerator and 11 denominator dl ar the 5 pe A) value exceeds tte critical value, we may conclude that ther Mowever, if the level of significance is fixed at | percent, we may n ticity. (Why?) Note that the p value of the observed Ais 0.014 ni level is 2.82 Since thexy s heleroscedasticity 1 thre errr ject the assumption of fact seas