CONCORDIA UNIVERSITY Department of Economics ECON 437 SECTION A/ ECON 537 SECTION A ECONOMICS OF PUBLIC EXPENDITURE Professor Georgi Boichev Summer 2020 – ASSIGNMENT 1 Due: Thursday, July 9th at 16:00 in my Moodle mailbox TOTAL POINTS: 75 TEAM SCORE: ___________________________________________________ SUBMISSION INSTRUCTIONS: Submit the responses of your team, of either 2 or 3 members, in a single PDF file. Late submissions, submissions in files other than a PDF, individual submissions and/or submissions of teams of more than 3 members will be disqualified and receive a score of 0 pts. The submission of identical responses across teams is a form of academic dishonesty. Question 1 (34 pts.) There are two individuals, A and B, with quasi-linear preferences over a private good, π₯π , and over a public good, πΊ, represented by the following utility function, π(π₯π , πΊ) = π₯π + ln πΊ. Each of them has income, ππ , which can be either spent on the private good, π₯π , or contributed to the public good, πΊ. 1 unit of individual contributions, ππ , from either individual is used to produce 1 unit of the public good, πΊ. The social planner has a Rawlsian social welfare function given by π(ππ΄ , ππ΅ ) = min{5ππ΄ , 5ππ΅ }. a) (3 pts.) Derive the Lagrangean corresponding to the optimization problem faced by individual B. The objective function of the Individual B’s optimization problem is max π(πΊ, π₯π΅ ) = π₯π + ln πΊ subject ππ΅ ,π₯π΅ to her budget constraint π₯π΅ + ππ΅ ≤ ππ΅ and the production function of the public good πΊ = (ππ΄ + ππ΅ ). Substituting the production function directly into the objective function and keeping the budget constraint as a constraint yields the Lagrangean: max πΏ = π₯π΅ + ln(ππ΄ + ππ΅ ) − π[π₯π΅ + ππ΅ − ππ΅ ] ππ΅ ,π₯π΅ ,π b) (3 pts.) Calculate the first-order conditions. ππΏ 1 = −π =0 πππ΅ ππ΄ + ππ΅ ππΏ = 1−π = 0 ππ₯π΅ ππΏ = ππ΅ − π₯π΅ − ππ΅ = 0 ππ 1 c) (2 pts.) Derive the best response function of individual B, ππ΅ . From ππΏ ππ₯π΅ ππΏ 1 +π π΄ π΅ = 1 − π = 0, π∗ = 1. Substituting π∗ = 1 into ππ = π π΅ − π = 0 yields ππ΄ + ππ΅ = 1. Individual B’s best response function is ππ΅ = 1 − ππ΄ . d) (4 pts.) Calculate the demand functions for π₯π∗ and πΊ ∗ . To solve for πΊ ∗, first we need to determine the individual contributions, ππ΄ and ππ΅ using the best response functions ππ΅ = 1 − ππ΄ and ππ΄ = 1 − ππ΅ , where we exploit the symmetry in the optimization problem they are facing. Solving this system of two equations with two unknowns yields a continuum of Nash equilibria (ππ΄∗ , ππ΅∗ ) such that ππ΄ + ππ΅ = 1 and ππ ∈ [0, 1]. Only one of them is a symmetric Nash equilibrium, ππ΄∗ = ππ΅∗ = 0.5. The rest of the answer key uses the numerical values corresponding to the symmetric Nash equilibrium. Substituting the individually optimal contributions into the production function yields πΊ ∗ = (ππ΄∗ + ππ΅∗ ) = (0.5 + 0.5) = 1 Finally, substituting ππ΅∗ = 0.5 into the budget constraint yields π₯π΅∗ = ππ΅ − 0.5. e) (2 pts.) State the conceptual difference between the best response function, ππ΅ , and the function, ππ΅∗ . For an arbitrary value of the individual contribution of person A, the best response function, ππ΅ , specifies a best response contribution by person B in the sense that it generates the highest utility for person B among all affordable values of ππ΅ . In contrast, the function, ππ΅∗ , is the Nash equilibrium contribution to the public good chosen by person B that depends only on numerical and/or parameter values. f) (2 pts.) Briefly explain whether the bundle (ππ΄∗ − 0.5, ππ΅∗ ) constitutes a Nash equilibrium. The bundle (ππ΄∗ − 0.5, ππ΅∗ ) does not constitute a Nash equilibrium since person A has a unilateral incentive to deviate from ππ΄ = ππ΄∗ − 0.5 and choose ππ΄ = ππ΄∗ . g) (2 pts.) Assuming that ππ΄ = ππ΅ , calculate for what income level πΊ ∗ = 1. For ππ < 0.5, there is a corner solution and all income is being spent on the public good, i.e. π₯π∗ = 0 and ππ∗ = ππ . By exploiting the symmetry in preferences and income levels, the production function for the public good becomes πΊ ∗ = 2ππ . πΊ ∗ = 1 is supplied whenever ππ ≥ 0.5 for both individuals. 2 h) (2 pts.) State whether the individual contributions, ππ , would differ across two high-income individuals who earn different income levels, ππ > 2. Justify your answer using your findings from parts d) and e). For ππ < 0.5, there is a corner solution and all income is being spent on the public good, i.e. π₯π∗ = 0 and ππ∗ = ππ . In contrast, for ππ > 0.5, ππ∗ = 0.5, i.e. individual contributions are independent of income. Since both individuals earn ππ > 0.5, the individual contributions are the same even when different income levels. i) (4 pts.) If ππ΄ = 2 and ππ΅ = 1, calculate the optimal bundle for each individual and their utility level. Since ππ > 0.5, πΊ ∗ = 2. Using the demand functions of the private good for an interior solution, π₯π΅∗ = ππ΅ − 0.5 = 1 − 0.5 = 0.5 and π₯π΄∗ = ππ΄ − 0.5 = 2 − 0.5 = 1.5. The utility levels are π(π₯π΄ , πΊ) = π₯π΄ + ln πΊ = 1.5 + ln 2 = 1.5 + 0.7 = 2.2 and π(π₯π΅ , πΊ) = π₯π΅ + ln πΊ = 1.5 + ln 2 = 0.5 + 0.7 = 1.2 j) (2 pts.) Write down the social planner’s problem. The social planner maximizes the social welfare function max π(πΊ, π₯π΄ , π₯π΅ ) = min{5ππ΄ , 5ππ΅ } πΊ,π₯π΄ ,π₯π΅ subject to the resource constraint: π₯π΄ + π₯π΅ + πΊ ≤ ππ΄ + ππ΅ and the reservation utility of one individual, e.g. person B. π(π₯π΅ , πΊ) = π₯π΅ + ln πΊ ≥ Μ π’Μ Μ π΅Μ , where Μ π’Μ Μ π΅Μ = 1.2. k) (2 pts.) Show that the social welfare function implies that π₯π΄ = π₯π΅ . The Rawlsian social welfare function is maximized whenever its two argument are equalized, i.e. 5ππ΄ = 5ππ΅ , which implies that ππ΄ = ππ΅ and, in turn, π₯π΄ + ln πΊ = π₯π΅ + ln πΊ. Cancelling out ln πΊ from both sides yields π₯π΄ = π₯π΅ . l) (2 pts.) Use the resource constraint and your result from part i) to find an expression of individual A’s participation constraint where πΊ is the only endogenous variable contained in it. Substituting π₯π΄ = π₯π΅ into the resource constraint yields, 2π₯π΄ + πΊ = 3, or equivalently, π₯π΄ = 1.5 − 0.5πΊ. Substituting π₯π΄ = 1.5 − 0.5πΊ into A’s utility function yields π π΄ (πΊ) = 1.5 − 0.5πΊ + ln πΊ. m) (2 pts.) Briefly explain why the Rawlsian welfare function captures only equity considerations. The Rawlsian social welfare function captures only equity considerations in the sense that the utility levels of the two individuals are equalized. For this numerical problem, this implies that the consumption allocation of the private good will be equalized irrespective of individual income levels. 3 Due to the non-excludability property of a public good, both individuals also consume the same amount of the public good. n) (2 pts.) Use your result from part j) to calculate the socially efficient quantity of the public good. The FOC πππΆ ππΊ = −0.5 + 1 πΊ = 0 implies that πΊ ∗ = 2. o) (2 pts.) Verify that the Samuelson condition yields the same πΊ ∗ as in part k). π΄ π΅ ππ ππΊ,π₯ + ππ ππΊ,π₯ = ππ ππΊ,π₯ π΄ π΅ πππΊπ΄ πππΊπ΅ + = ππ ππΊ,π₯ πππ₯π΄π΄ πππ₯π΅π΅ 1 1 1 Substitute πππΊ = πΊ , πππ₯π΄ = 1 and πππ₯π΅ = 1 into the Samuelson condition to calculate πΊ + πΊ = 1, which yields πΊ ∗ = 2. p) (2 pts.) Briefly explain why the decentralized provision of the public good is (in)efficient for this economy. The decentralized problem of the amount of the public good is underprovided relative to the socially efficient level. This is due to the free-riding problem captures by the discrepancy that each person consumes πΊ but contributes only the fraction ππ as an input for its production. Question 2 (41 pts.) You are tasked to determine which one of two policy options yields better results in reducing corruption at the village level. Option 1: top-down monitoring by an upper-level government; Option 2: grassroots participation of citizens in village council accountability meeting. Corruption is measured as a reduction in the percentage of missing materials in infrastructure projects, e.g. road construction. As part of an experiment, some villages are randomly assigned to be audited by the federal government (group A), while the remainder are not (group N). A public economist uses data on monitoring type π·π and the amount of missing materials, ππ , to estimate the population regression function (PRF) ππ = π½1 + π½2 π·π + π’π by using the following sample: π π· (observed) π (observed) π· (hypothetical) π (hypothetical) 1 1 5 0 3 2 0 6 1 2 4 3 1 7 0 1 4 0 8 1 4 The benchmark group of villages are subject to top-down monitoring. a) (2 pts.) Write down the optimization problem of the method of least squares. Μ1 , π½ Μ2 ) = Σi π’Μπ 2 min π(π½ Μ1 ,π½ Μ2 π½ b) (3 pts.) Briefly describe the nature of the optimization problem by specifying the dependent, explanatory and exogenous variables from a statistical standpoint. To have the regression best fit the data, minimize the sum of squared of squared residuals, Σi π’Μπ 2 , (dependent variable) by choosing the optimal quantities of the sample estimators of the intercept and Μ1∗, π½ Μ2∗ (explanatory variables), as a function of the numerical values of the the slope coefficients, i.e. π½ exogenous variables available in the data, ππ and ππ . c) (4 pts.) Show that Μ1 ,π½ Μ2 ) ππ(π½ Μ1 ππ½ Μ1 − π½ Μ2 π·π ) = 0 by using the chain rule and the = −2Σπ (ππ − π½ summation rules where appropriate. Do NOT skip steps. Μ1 , π½ Μ2 ) = min Σi (ππ − π½ Μ1 − π½ Μ2 π·π )2 π(π½ Μ1 π½ Μ2 π½ Μ1 − π½ Μ2 π·1 )2 + (π2 − π½ Μ1 − π½ Μ2 π·2 )2 + (π3 − π½ Μ1 − π½ Μ2 π·3 )2 + (π4 − π½ Μ1 − π½ Μ2 π·4 )2 = (π1 − π½ The chain rule: Μ1 ,π½ Μ2 ) ππ(π½ Μ ππ½1 = Μ1 ,π½ Μ2 ) ππ’ ππ(π½ Μ1 Μ1 Μ1 ππ’ ππ½ + Μ1 ,π½ Μ2 ) ππ’ ππ(π½ Μ2 Μ1 Μ2 ππ’ ππ½ + Μ1 ,π½ Μ2 ) ππ’ ππ(π½ Μ3 Μ1 Μ3 ππ’ ππ½ + Μ1 ,π½ Μ2 ) ππ’ ππ(π½ Μ4 Μ1 , Μ4 ππ’ ππ½ where π’Μπ = ππ − Μ1 − π½ Μ2 π·π . π½ Μ1 , π½ Μ2 ) ππ(π½ Μ1 − π½ Μ2 π·1 )(−1) + 2(π2 − π½ Μ1 − π½ Μ2 π·2 )(−1) + 2(π3 − π½ Μ1 − π½ Μ2 π·3 )(−1) = 0 = 2(π1 − π½ Μ1 ππ½ Collecting terms yields: Μ1 ,π½ Μ2 ) ππ(π½ Μ ππ½1 Μ1 − π½ Μ2 π·π ) = 0. = −2Σπ (ππ − π½ Distributing the summation operator yields: Μ1 ,π½ Μ2 ) ππ(π½ Μ ππ½1 Μ1 − π½ Μ2 Σπ π·π ) = 0, where = −2(Σπ ππ − ππ½ Μ1 is an additive constant and π½ Μ2 is a multiplicative constant. π½ n n n n n d) (2 pts.) Calculate Σπ=1 ππ , Σπ=1 ππ2, Σπ=1 π·π , Σπ=1 π·π2 , Σπ=1 π·π ππ . π 1 2 3 4 Σπ ππ 5 6 7 8 26 π·π 1 0 1 0 2 ππ2 25 36 49 64 174 π·π2 1 0 1 0 2 5 π·π ππ 5 0 7 0 12 e) (2 pts.) Use your results from part c) as well as Μ1 ,π½ Μ2 ) ππ(π½ Μ ππ½2 Μ1 − π½ Μ2 π·π )π·π = 0 to = −2Σπ (ππ − π½ Μ1 and π½ Μ2 . calculate the estimates of π½ FOCs become: Μ1 ,π½ Μ2 ) ππ(π½ Μ1 ππ½ Μ1 − 26 + 2π½ Μ2 = 0 (1) and = 4π½ Μ1 ,π½ Μ2 ) ππ(π½ Μ2 ππ½ Μ2 − 12 + 2π½ Μ1 = 0 (2). = 2π½ Solve the FOCs as a system of two equations with two unknowns. Μ1 − 14 = 0. Solve this resulting equation for π½ Μ1 : π½ Μ1 = 7. Subtract equation (2) from (1) to obtain: 2π½ Μ1 = 7 into equation 2: 2π½ Μ2 − 26 + 4(7) = 0. Solve this resulting equation for π½ Μ2 : π½ Μ2 = Substitute π½ −1. Μ1 is always equal to the conditional To check your work: For a regression with a single binary regressor, π½ Μ1 + π½ Μ2 is always equal to the conditional mean of the dependent variable for the π·π = 0 group and π½ mean of the dependent variable for the π·π = 1 group. f) (2 pts.) Calculate the true parameter, π½1 . π½1 = πΈ(π0π |π·π = 0) = π01 + π03 3 + 1 = =2 2 2 g) (4 pts.) Calculate the causal effect measured by the true parameter π½2 . Use appropriate mathematical notation before substituting any numerical values. Do NOT skip steps. π½2 = πΈ(π1π |π·π = 1) − πΈ(π0π |π·π = 1) = π11 + π13 π01 + π03 5 + 7 3 + 1 − = − =6−2=4 2 2 2 2 h) (4 pts.) State and briefly interpret each of the two terms of the causal effect by specifying the relevant group(s), the state(s) of nature and the outcome of interest. Be pedantic about the clarity and precision of your response. πΈ(π1π |π·π = 1) – expected percentage reduction of missing materials for the villages who (chose to) be enrolled in the grassroots participation mechanism observed in the actual state of nature. πΈ(π0π |π·π = 1) – expected percentage reduction of missing materials for the villages who (chose to) be enrolled in the grassroots participation mechanism observed in the counterfactual top-down mechanism. 6 i) (2 pts.) Calculate the selection bias. Do NOT skip steps. πΈ(π0π |π·π = 1) − πΈ(π0π |π·π = 0) = j) π01 + π03 π02 + π04 3 + 1 6 + 8 − = − = 2 − 7 = −5 2 2 2 2 (2 pts.) Briefly explain why Olken (2007) is concerned that the geographic terrain might lead to making an erroneous conclusion about its effectiveness in reducing corruption. The villages in audit group tend to be located disproportionately in the mountains, where roads are more likely to be washed away during heavy rainfall. The observed effect captures both the influence of monitoring type and all other influences including geographic terrain om missing materials. If the impact of the geographic is larger than that of monitoring type, the sign of the observed effect that we estimate would be the opposite of the true impact (the causal effect). k) (2 pts.) Briefly explain why Olken (2007) conducts an experiment to find the causal effect instead of relying on your method in part g). Relying on non-experimental data means that we cannot distinguish between the causal effect and the selection bias since hypothetical data does not exist. The experimental data uses randomization, e.g. tossing a coin, to assign villages to a monitoring mechanism, i.e. stripping them of the ability to choose the monitoring mechanism. Randomization ensures that in a reasonably large sample all factors other than the monitoring mechanism will have no impact on the amount of missing materials, e.g. approximately equal share of villages located in the mountains for each assigned group. l) (2 pts.) For the data point (π2 , π·2 ), calculate the deterministic component πΈ(π2 |π·2 ), the error term π’2 , the predicted value πΜ2 and the residual π’ Μ. 2 πΈ(π2 |π·2 ) = π½1 + π½2 π·2 = 2 + 2(0) = 2 π’2 = π2 − πΈ(π2 |π·2) = 6 − 2 = 4 Μ1 + π½ Μ2 π·2 = 7 + 1(0) = 7 πΜ2 = π½ π’ Μ2 = π2 − πΜ2 = 6 − 7 = −1 m) (3 pts.) In a single graph: - Draw the sample regression function, the population regression function and all data points, i.e. (π·π , ππ ). - For the data point (π·2 , π2 ), illustrate the decomposition of π2 into its deterministic component πΈ(π2 |π·2 ) and its error term π’2 ; 7 - For the data point (π·2 , π2 ), illustrate the decomposition of π2 into its predicted value πΜ2 and its residual π’ Μ. 2 The graph appears on the last page of this file. Μ1 ) = n) (5 pts.) Suppose that the sample variances of the intercept and slope coefficients are π£ππ(π½ Μ2 ) = 4 respectively. Test the hypothesis at the 90 % level of significance that the 2 and π£ππ(π½ centralized provision of public goods is at least twice as effective in reducing the amount of missing materials relative to the corresponding decentralized provision. In your answer, - (2 pt.) Specify the appropriate hypotheses and statistical test. - (2 pt.) Demonstrate your calculations, including the appropriate expansion of the statistical test’s numerator and denominator. Do not skip steps. - (1 pt.) Provide a conclusion. To specify the hypothesis, use the fact that π½1 measures the average of π corresponding to the topdown mechanism, while π½1 + π½2 measures the average of π corresponding to the grassroots participation mechanism. What we would like to find is that 2π½1 > π½1 + π½2 , which is equivalent to : π½1 > π½2 . π»0 : π½1 ≤ π½2 π»π΄ : π½1 > π½2 π‘= Μ2 − π½ Μ1 − π½2 − π½1 Μ2 − π½ Μ1 π½ π½ = = Μ2 − π½ Μ1 − π½2 − π½1 ) Μ2 − π½ Μ1 ) π π(π½ π π(π½ Μ2 π½ Μ Μ √π£ππ(π½1 ) + π£ππ(π½2 ) π−πΎ = −1 − 0 √ 6 4−2 = −1 = −0.58 1.73 There is insufficient evidence to reject π»0 since π‘ = −0.58 < 2.92 = π‘πΌ=0.1,π−πΎ=2 . o) (2 pts.) Briefly explain why there is a discrepancy between your answers in parts f) and g) and those in part n). In part n), we use only sample information that turns out to be from a biased sample. In contrast, we use the true population parameters in parts f) and g). 8