Mester (130 3) °F Econ, On such a samp bN do We 1¢ households wh Now, suppose ou enn ised estimator of ee iS over 25 years old. If we use OLS eters les only . same oft Saving equation? Eggi COUPES witout children. Can we estimate al ofthe 10t hav Unbiased est have of the families that imator of g,7 save more than Tk 500, ley lopment). From that data set we want to hy nancial performance (here we define it as level of profit, which is a continuous Suited econometric specification, why the problem of sample selection might }OW can we correct such bias? 3. (a) Win oyu i belp of a practical example, explain Propensity Score Matching (PSM) method and the 'PS of conducting it. Briefly explain different types of matching methods in case of PSM. [5+2] (b )Define and explain ATE and ATET. 4. (a) If case ofa binary dependent variable, why a probit model is considered superior to a linear probability model? Explain ©) Suppose we have data for estimating a probit model of the probability that a child goes to a private School as a function of child’s ability, measured by the score of a certain test. A simple probit model can therefore be defined as: P(private=1|score)=®(By+Biscore). The results of this model is as follows: __| Coefficient — - t-value - 1 [Bo “1.82 12.84 Bi 0.05 11.76 Assuming ‘score’ as a continuous variable with a mean value of 31, what will be the marginal effect of ability on the probability of the child going to private school? 5. Why panel data is considered superior to cross section data? Illustrate and compare different estimation methods of modelling panel data. 6. Consider a simple model to estimate the effect of laptop computer (Lptp) ownership on grade point average (GPA) at a university (Lptp is a binary variable indicating laptop ownership). Suppose, we estimate the following model: GPA=BotPiLptptu (a) Why might laptop ownership be correlated with the error term (u)? (6) Explain why Lptp is likely to be related to parents” annual income. Does this mean parental income is a good IV for Lptp? (©) Suppose that, two years ago, the university gave grants to buy laptop to roughly one-half of the incoming students, and the students who received grants were randomly chosen, How do you think this information can be utilized to construct an IV for Lptp?of : 1. 3. wef University of Dhaka Department of Economics: 12” Batch, 7 Semester, Final Examination 2021 Time: 3 hours; Total Marks: 50 es a of married women as well as the wage offer 9s i Wbrium condition imposed, 4 ly function ] PPRictivity variables, With the equ the following manner: aget Bi; kidsh6*Bianwifeine tur x pert B23 expersa+ U2 fa) Let us consider labour function of hours nd oth the structural eq) age)+ Biot ried hours? B20 Bareduet of the wot kidsht6" is the number of children less Nand -educ’ and “exper” are years les, except ‘hours’ and fo estimate the parametres “ition and the rank condition. hours=ai log log(wage) an in years nwage incor ectively. All. vat Tow will you pro fhe order con r ogenous. x years old, * mand prior experien "are assumed to be ations? Explain your answer using ¢ “log(w of these equ [6] (b) What is simultancity bias? Hlustrate. [4] a dummy variable for whether ‘school grade point ave dy be the number of 20 students, the followin: -1.17+0.24 GPA+0.00058 SAT+0.073 study) A fixed at 3.0 and SAT fixed at 1200, what is ability for someone Who spent 10 hours per ? (7 a university graduates in time. thd score of SAT examination urs spent per week by the student 2 logit model is obtained: a student at (a) Let graduate Let GPA and SAT be high respectively. Also assume that stu studying. Suppose that, using data on 4 A, SAT, study) = AC Holding GP raduation prol hho spent 5 hours per week? 1G P* (graduat where A(,) is the logit function: the estimated difference in the gt week for studying and someone w! tances, first difference under what circumst =xplain. [3] arried women, 428 of whom are participating ng table shows the probit estimates of labour experience and its square, age, of children greater than 5 years (b) In case of panel data estimation method. estimation is preferred to fixed ef! estimation (a) Consider that we have a data set of 753 m: in the main stream labour market. The followi n (inlf) on family income, education, force participatior dummy of having children less than 5 years old and dummy old: Dependent Variabl inlf Independent Variables Probit (MLE) Famine [-012 (.005) Educ Te (.025) Exper 123 L (019) Exper2 ~.0019 _| (0006) Age -.053 .008) Kidsless5 -.868 (119) |te reat — 036 (on) 370 ot i (509) a — Now consider a women with famine= 20.13, edue= 12. kidsgreatS Constant exper = 10.6 and age = 42.5 and - For that woman, what is the estimated change in the probability of working, from zero to one child less than 5 years age? [7] while goir (b) With a practical example, illustrate Ordered Probit Model. [3] (a) Briefly explain the Random Effect method of estimation. [3] tion’ in econometric estimation? in which some of the respondents n program’ and from that data set ipation in that program contributes positively towards better health outcome of the participants. Discuss whether the problem of sample selection might arise in this context and how can we correct it? [7] (b) What do you understand by the problem of ‘sample sel Let us assume that, we have data of a household surve: participated in a ‘health and hygiene awareness campai we want to find out whether pa (a) Consider that we have conducted a household level survey and wanted to estimate the effect of financial inclusion (FI) on women’s empowerment (EMP) where the latter is proxied by an index. Under what circumstances, Fl can be endogenous? If we have information on distance of the household from the nearest bank (DST) then how can this information be utilized to resolve the endogeity problem? Illustrate. Note: In this example, by FI we consider access to financial products and services available in banks, mobile financial services and micro finance institutions] [7] (b) Briefly explain the ways you can test endogencity in a regression model. B) Consider the following dynamic panel model: Vie = OYira + Ui + Mee (For simplicity, we ignored any regressors other than lagged dependent variable) Here, u; ~ I/D(0, 02) is individual specific feature vig ~ HID (0,02) is idiosyncratic error term And u; and v;eare independent of each other. why OLS estimator will be inconsistent and biased. (a) Expl: (b) Why fixed effect estimator will be inconsistent and biased’ Explain. 2 * (©) Sa how GMM (Arellano and Bond, 1991) will produce consistent and unbiased 2 estimator. iests u way for more efficient 2 (@) Aretlano-Bover (1995) and Blundell-Rond (1998) sui prator, How? estimator than Aretlano-Bond (1991) est {e) Assume the errors are heteraskedastic, We performed two-step difference GMM. Consider the following STATA output Interpret results of Arellano-Bond test for ARC 1) and ARQ). as) ii, Interpret results of Hansen test for overidentiti (1s) ion restrictions, De(w wth KALE ALD ys ysth yst2 yrl976 yr2977 yr1978 yr1979 yr3980 yr298t ye1987 yr1989 yri9ee) Cowi-type (missinge®, separate instruments for Wevrienty each period unless collapses) ARG) Ain firat differences: & AR(2) in thet differences: = Arellane-fong test tor Sargan test of overia. restesetions: cmi2(25) 67.58. Prob > en? (Nor robust, But not weakened by many instruments.) h2025) 31.38 prob > emi? = oar? (Robust, but weakened by many instruments.) 7. (a) Discuss experimental and non-experimental studies. (2.5) (b) Differentiate between Average Treatment Effect (ATE) and Average Treatment E the Treated (ATT) (c) Briefly explain the steps of Propensity Score Matching (PSM). (4) (d) The study aims to find the effect on real earnin thered working experience and received counseling. Using *psmatch2’ to ie the average ect on Q) (re78) of participation to a program: where they conduct nearest neighbor matching with replacement and ‘Iscore2" to estima ed (ATT) we got the following result. Interpret the following (1.5) treatment effect on the trea result: = psmatch2 4, outcone(re78) pscore(tscore2) Difference SE. Tostat Treated Controls Variable Sa V778 4s4.61433 =13.17 2209.11372 0.77 6349,1435 21553.9209 -1520. re7B Unmatched 1435 4558.2016 1690. 9419 art | 63 S.E. does not take into account that the propensity score is estimated, 8. (a) Briefly explain the concept of consistency and asymptotic normality. (5) (b) Consider the OLS estimator # = ({x/x)71Ex/ yi. Prove that B is consistent estimator (5) FEE aE II EE IS IE TEE TE ISOS O OD IOS OEISORIORISR RISB RR ee wo ws x| | University of Dhaka, Department of Economics BSS (Hons.), 11th Batch, 7” Semester, Final Examination, 2020 ECON 405 (Advanced Econometrics 1) [Total Marks: 2; Sect a ars 2S Secon A Answer any ONE out of the following 3 broad questions. 1* 10= 10. Section S SE out of the following 5 short questions. 3*5=15: Total time: 90 Minutes] SECTION A 1. (a) Consi (2) Consider a binary choice mode! of unemployment (a dumms variable which is | if the person is unemployed and § y otherwise) with male (a dummy which is | if the person is a male). age (10 years) and degree (a dummy variable which is | if the person holds a university degree) as ¥ s je me on 5 33 respec! unemployment=-1.574+0.21 Smale-0.0362age-0.947degree (standard error) (0.172) (0.103) (0.004) (0.193) N=7867: Pseudo R-square=0.0315 At sample mean, for this model, what is the effect of gender on unemployment probability? (7) * ®) HF you want J estimate a model of individual level of savings from a survey data where the dependent varia le savings is continuous over strictly positive values but is zero for a nontrivial ion of the sample. then what type of model you should use? Briefly explain B) 2. What do you understand by the problem of ‘sample selection” in econometric estimation? Let us assume that, we have data of a firm level survey in which some of the firms are reported to have Spent in R&I) (research and development). From that data set we want to find out whether spending in R&D contributes positively towards firm’s financial performance (here we define it as jevel of profit, which is a continuous variable). Discuss, with required econometric speciticalion. why the problem of sample selection might arise in this context and how can we correct such bias? 3. What is simultaneity bias? Explain. In case of simultaneous equation, discuss the Order and Rank conditions of identification. SECTION B 4. With the help of a practical example, explain Propensity Score Matching (PSM) method and the relevant steps of conducting it 5, Consider a student's choice of optional subjects in 7” Semester BSS Economics at Dhaka University. He/she has three alternative courses (course A, B, and C) and has to choose only one course. If the choices are not ranked, then what type of econometric model should you use to understand the factors affecting his/her choice? Explain briefly 6. Illustrate the three stage least square (3SLS) estimation method. 7. Suppose, we want to estimate the effect of class attendance on student's performance 1 Evcond0S course and we estimate the following model Performance=fo attendanes ISCGPA ¢ fhyscore 1 Here, attendance is attendance in the class, performan ti student performance, HSCGPA is his/her GPA in HSC eam and score is score in university admission test, Do you think attendance is. an endogenous wariable? Explain, If we have information on another variable distance “hich is the distance trom stustents’ residence 16 the campus then what conditions must be satisfied to consider distance as a valid instrument for attendance? is standardized outcome on & final exam or8 Incase of panel data. compare Fixed Effect and Random Effect estimation methods.2 yarnariment of Economics, University of Dhaka Courses Adv iestet 9" Batch) Final Examination 2018 * Advanced Econometrics 1; Course No: ECON 405 Time: 3 hours; Total Marks: 50 nswer any Fe the f 4" 10=44 Suppose we have dat: ata has any effect on voaubadk Household Survey and we want to understand whether access to credit ‘ood security situation and we have estimated the following model tion, FSecurity=Py+B,Credi Bots 'Credit+B:Age+p,A gesq+f,HHSize+f,HHIncome+fi.HeadEductu Here, FSecurit il denoiing whether hoch pet caPita food consumption expenditure: Credit is « dummy variable <8) and its square: Hi[Sne oi feceives credit or not; Age and Agesq are age of the household head HeadEduc is years of cedar an”, Of the household; HHIncome is per capita household income; (a) Do you think Cred ucation of household head and u is error component of the model. (b) If we have ingens an endogenous variable? Briefly Explain nearest banks then nation on another variable disiance-which is distance of households from Credit? E fat Conditions must be satisfied to consider distance as a valid instrument for plain eu 2. Let i hours and et labour supply function of married women as well as the wage offer as a function of ‘equation: ‘er productivity variables. With the equilibrium condition imposed, the structural a 's can be expressed in the following manner: hours=a, log(wage)+B,o+ B, i 0 neduct B,,age+ B,, kidsh6+P,snwifeinc+u, log(wage)=a: hours+B.o+ By.educ+ B;,exper* Bs expersq+u; Here, the variable ‘age’ is age of the woman in years, *kidsIt6” is the number of children less than six years old, ‘nwifeinc’ is woman’s nonwage income and ‘educ’ and ‘exper’ are years of education and prior experience, respectively. All variables, except ‘hours’ and “log(wage)" are assumed to be exogenous. How will you proceed to estimate the parametres of these equations? Explain your answer using the order condition and the rank condition. (10) 3. Let us assume the following coefficient estimates of the Probit Maximum Likelihood Estimation method. Variable Probit MLEs Intercept 184 (058) X; 1.35 (056) X ~ 114 (.069) Xs 095 (.068) Xa 324 (081) i % ~677 (109) Xe +339 (.073) If we want to keep the variables X,-X; at their median value (median values being 1, 0.5, 0.5, 0 and 0 respectively) then what will be the impact on the dependent variable when X, takes a value of 0 and | respectively? (10) [Please turn over] od™N NN 4. Let us assume that we have a data 460 of 616 workers and of which 444 workers reported to have et ved positive pension benefits (here benefits are in dollar value). If we ‘estimate a Tobit model Of pension benefitthen we get the following result: Explanatory Variable Tobit Coefficients Experience (in years) 5.20 6.01) Ree (in years) -4.64 (3.71) = Tenure (in years) 36.02 1. 456) Education (in years) 93.21 00.89) Depends (number of dependent) 35.28 1.92) Married (dummy if married=1) 53.69 _ 1.73) — ‘White (dummy if white=1) 144.09 (102.08) Male (dummy if male=1) 308.15 _ (69.89) Constant 71252.43 (219.07) N 616 Log Likelihood Value -3672.96 ° 677.74 married=0, white=1 and If we assume, experience=tenure=10, age-3>, ‘education=16, depends i ith the given characteristics? male=I then what is the expected pension benefit for a white male wi (Hint: for Tobit model, E(y)=(AP/9) xB + 0988/0)] (10) conducting Propensity Score Matching. (5.5) (45) 5, (a) Explain the steps of. (b) Discuss different matching methods in evaluating a program. 6. Why do you think panel data is considered superior to cross-sectional data? Explain how Fixed od uP heterogeneity bias. In case of panel data, under Effect estimation can take care of the pro! what circumstances, first difference cetumation is preferred to fixed effect estimation? (3+4+3) (5x2=10) 7. Write short notes on any two of the following: (a) Differences in Differences Estimator (b) Panel GMM (c) Pseudo R-square r WO of the folloy xxample, illustrate Ordered Probl GLS estimator?Why might a system GLS be 1. With the help of appropriate ¢? 2. In case of a system of equations, how can you get preferred to a system OLS? 3, [let us assume that, we have data of a firm level survey in which some of the firms We reported to have spent in R&D (research and development). From that data set we want to find out whether spending in R&D contributes positively towards firm’s financial performance (here we define it as evel of profit, which is a continuous Yariable), Discuss, with required econometric specification, why the problem of sample selection might arise in this ‘context and how can we correct such bias? 4. How can you test endogeneity in an ‘econometric model? E plain. enecosennnnannccssaaassnnneeseeeeee® PRRRAREAARER EEA IID