An Empirical Framework for Large-Scale Policy Analysis, with an Application to School Finance Reform in Michigan Maria Marta Ferreyra Tepper School of Business Carnegie Mellon University
Motivation How should we empirically analyze large-scale policies? This paper: Empirical framework for the analysis of large-scale policies Applies framework to study the Michigan school finance reform
K-12 Public School Finance Funding for K-12 schools comes from: Local districts (property taxes) State (income or sales tax) School finance reform since the early 70s Greater equity in funding across districts
Michigan School Finance Reform (Proposal A) Implemented in 1994 Local control centralized state control Tax Reform Reduced property taxes Increased sales and use taxes Revenue Reform State guarantees revenue = foundation allowance Revenue increases to low-revenue districts; revenue caps for high revenue districts
Proposal A (cont.) Goals: But other effects are also possible: Lower property tax burdens Lower revenue variation across districts But other effects are also possible: General equilibrium effects … and they can magnify the improvement
Empirical Framework for Large-Scale Policy Analysis Large-scale policies general equilibrium model Empirical content estimate model; fit in-sample data Serve to analyze actual policies Serve to analyze counterfactual policies model must fit out of sample data
Empirical Framework for Large-Scale Policy Analysis (cont.) Focus on Detroit metropolitan area Estimate a general equilibrium model for the Detroit metropolitan area using 1990 data Use parameter estimates to predict the effects of the actual reform in 2000 Compare predictions with 2000 data Simulate counterfactual school finance reforms
Contributions To literature on large-scale policy analysis Estimate and validate model Estimation rather than calibration Model validation: need a regime shift first validation of a Tiebout model Are counterfactuals plausible? In my counterfactuals, public good adjusts endogenously
Contributions (cont.) To the analysis of school finance reform General rather than partial equilibrium Structural estimation Current analysis consistent with reduced form studies for MI Study counterfactuals in school finance reform Analysis applicable to other metro areas (18 states currently litigating) and other types of policies
Model Builds upon Ferreyra (2007) and Nechyba (1999) Community Structure for Metro area: A continuum of houses and households D districts; one public school per district Each district has neighborhoods (housing qualities) A house gives entrance into the district’s school
Model (cont.) Households: One child per household; child goes to school Households vary in endowment and idiosyncratic preferences for locations
Model (cont.) Household preferences: c = consumption S = school quality K = neighborhood quality Households pay taxes: Local property tax State income tax
Model (cont.) Schools: School Funding: Vary in quality across districts Quality (achievement) is a function of: Spending per student x Peer quality q ( = avg. income) School Funding: Property taxes: Chosen by majority voting per district Paid by residential and non-residential owners Income taxes
Model (cont.) Households choose: Equilibrium: Location (and school) Property taxes Equilibrium: Who lives where district demographic compositions Rental Values School spending School peer qualities School qualities … such that nobody wishes to move or vote differently
Estimation Match 1990 data at the district level 83 school districts 4 parameters
Detroit Metropolitan Area in 1990
Detroit Metropolitan Area in 1990 (cont.)
Estimation (cont.) Challenge of estimating general equilibrium models: all equilibrium conditions must hold My approach: full solution estimation (Ferreyra 2007) Compute the equilibrium given 1990 exogenous variables Metro area income distribution Housing stock: quantity and quality across districts School finance regime Fit in-sample (1990) data well According to estimates, higher role for peer quality (household income) than for spending (0.87 out of 1)
Model Validation Compute equilibrium given 2000 exogenous data Model fits 2000 data reasonably well 2000 levels Changes between 90 and 00 Confidence in model for policy analysis and counterfactuals
Policy Analysis: Proposal A Recall: tax reform, and revenue reform Some revenue equalization Capitalization of tax and revenue reform Greater gains for low-income and urban districts Very little effect on district demographics Revenues do no affect school quality much Low housing quality in favored districts Achievement gap across districts shrinks little
Alternative Policies District Power Equalization: Property tax revenue per child = tax rate * property tax base per child To equalize effectively, set a very high guaranteed tax base fiscally very costly … yet tax rates differ across districts! Affects high-income districts … but it preserves local autonomy Not very effective to close the achievement gap
Alternative Policies (cont.) Uniform Foundation (full equalization) low foundation ($6,000) Very little effect High foundation ($15,000) Hurts high income districts and property values more No more effective than low foundation for achievement gap
Alternative Policies (cont.) Adequacy Costing-out studies Answer depends on achievement target for target = 30% of highest-achieving district: Income tax rate = 10% Partial-equilibrium estimate is 65% For target = 30%: City of Detroit gets $30,000 per student; other districts get funding b/w $20,000 and $100,000 per student Relocations are still limited Rental values fall in most places Most effective in achievement; also most costly
Alternative Policies (cont.) For target = 40%: Income tax rate = 45% No Child Left Behind targets a 100% proficiency rate by 2013/14 Prohibitively costly IF no other tools were used
Policy Analysis (cont.) Little room for revenue-based policies Extra money is not very effective under the current conditions Accountability Seemingly effective for low-income districts; low cost Raise peer quality Interdistrict open enrollment What peer quality measures Parental inputs and early childhood Non-cognitive skills
Concluding Remarks Empirical framework for large-scale policy analysis use the policy shift for out-of-sample prediction Here: estimate model with 1990 data; predict 2000. Ability to quantify g.e. effects Large-scale policies are very costly and may have unintended effects bring rigor to the policy analysis