Integrating Empirical Data in an Agent Based Model of Fiscal Federalism

Transcription

1 Integrating Empirical Data in an Agent Based Model of Fiscal Federalism Debra Hevenstone and Ben Jann University of Bern ECCS, Barcelona Sept. 16, 2013 ECCS, Barcelona Sept. 16, /

2 Question Topic: Fiscal Federalism ECCS, Barcelona Sept. 16, /

3 Question Question Topical Questions Are residential mobility (exit) and government responsiveness (voice) sufficient mechanisms to explain oft-observed features of fiscal federalism? Regressive Taxation Economic Segregation Constrained Spending How does equalization design impact fiscal federalism? Methodological Questions When designing an ABM how can we use empirical data to Limit arbitrary model assumptions? Validate model output? Generate realistic predictions? What is the balance between realism and parsimony? ECCS, Barcelona Sept. 16, /

4 Model Overview Model Overview INITIALIZATION 1. Landscape: 100* 100 grid (10,000 parcels) 2. Jurisdictions: 16 groups of 625 parcels 3. Households: 9000 households with: incomes preferences for public goods EXIT (MOVE) 1. Household: Pick a random vacant parcel Offer bid that marginally improving utility 2. Parcels of Land: Look through list of bids Accept highest bidder VOICE (VOTE) 1. Jurisdictions Propose high/low change in maximum tax rate Propose high/low change in tax progression People vote Local taxes are set 2. Federal government Propose high/low change in maximum tax rate Propose high/low change in tax progression People vote Federal taxes are set Equalization grants processed ECCS, Barcelona Sept. 16, /

5 Model Overview Model Overview poor households jurisdiction border rich households ECCS, Barcelona Sept. 16, /

6 Empirical Model Inputs Empirical Inputs (Institutional): Tax Curves tax = S(1 e yk ) y = income S = maximum tax k = tax progression k alt = ln( 3 1 ) k income at which one pays 2/3 of S ECCS, Barcelona Sept. 16, /

7 Empirical Model Inputs Empirical Inputs (Institutional): Equalization Formulae baseline horizontal NF horizontal vertical r j N r j + θ h N j ( x x j ) j N }{{} + r j + θ h N j ( x x j ) r j + R f N j (1/c j ) θv }{{} i N i(1/c i ) θv horizontal grant horizontal grant }{{} vertical grant N j population in j N total population R f total federal revenue r j revenue collected in j θ h horizontal redistribution parameter x j jurisdiction s per capita tax capacity x national per capita tax capacity θ v vertical redistribution parameter c j jurisdiction j s relative per capita revenue (c j = x j ) x j Based on a simplified version of real-life formulae Horizontal is based on difference in tax capacities Vertical is based on the ratio between tax capacities ECCS, Barcelona Sept. 16, /

8 Empirical Model Inputs Empirical Inputs (Voice Mechanism): Voting Governments propose random deviations in S & k Population votes directly (median voter model) Alternatively: Total utility maximization ( Rich Voice ) ECCS, Barcelona Sept. 16, /

9 Empirical Model Inputs Empirical Inputs (Exit Mechanism): Moving Pick an empty parcel Calculate price that would yield current utility Offer a bid of 1 franc less ( ( ) α ( ) ) 1 1 α 1 α p h 2 = y t 2 1 p 2 y t 1 h 1 1 Parcel offered to the highest bidder at the end of the round Only positive bids are taken With one jurisdiction, all bids are -1 and no one moves ECCS, Barcelona Sept. 16, /

10 Empirical Model Inputs Empirical Inputs (Agent Characteristics): Incomes Lognormal Swiss Income Distribution (2009) Simulated Distribution (lognormal µ =11,σ=1) Possibilities: Pareto, Weibull, generalized beta, gamma... ECCS, Barcelona Sept. 16, /

11 Empirical Model Inputs Empirical Input (Agent Characteristics): Preferences Randomly assigned from a normal distribution (µ =.2, σ =.05) Private goods preference: (1- public goods preference) local federal total 0 to 19% of income 0 to 21% of income 17 to 21% of income In Switzerland the total tax revenue = 28% GDP ECCS, Barcelona Sept. 16, /

12 Empirical Model Inputs Arbitrary Assumption: Agent Utility ( gj ) αl u l = + 1 (y i t i ) (y l (1 t l ) h l ) 1 α l n j n j }{{} i }{{} private goods public goods l household y l income of householdl t l tax of householdl h l housing cost of household l i index of households j jurisdiction g j grant to jurisdiction j n j number of households in jurisdiction j α l preference for public goods of household l 1 α l preference for private goods of household l We regularly assume Cobb Douglas utility w/ constant returns to scale...but it is still an assumption ECCS, Barcelona Sept. 16, /

13 Empirical Validation Empirical Validation: Tax Curves ECCS, Barcelona Sept. 16, /

14 Empirical Validation Empirical Validation: Tax Curves CH Data Simulation S Minimum Mean Maximum Correlation w/ p.c. income k alt Minimum 84,835 62,885 Mean 9, ,108 Maximum 299, ,879 Correlation w/ p.c. income Higher tax levels (no federal), similar range Similar progressivity level and range Similar correlation tax levels and wealth... but phase-ins more regressive Note: Simulation data is median voter model, equalization =.2, horizontal w/ no federal because in reality local and central revenue are not fungible; but this means local taxes are higher in the simulation to meet total public goods demand. ECCS, Barcelona Sept. 16, /

15 Empirical Validation Empirical Validation: Income Segregation Rich voice no equalization Median voter horizontal equalization (NF) =.1 Economic Segregation Median Voter Rich Voice Pure federalism Horizontal (w/ fed) Vertical Horizontal (no fed) Median income ratio in Swiss Cantons in 2001: 2.06 ECCS, Barcelona Sept. 16, /

16 Assessment Assessment How did empirical data help? Limited model assumptions Check that model results are reasonable How did empirical data hinder? Reality can be complex (tax curves, equalization formulae) This can make interpretation more difficult Is the interpretation limited to basis case(ch)? ECCS, Barcelona Sept. 16, /

17 Conclusion Conclusion Using empirical data probably improves ABM ECCS, Barcelona Sept. 16, /

18 Appendix: Topical Results Taste of Topical Results Rich Paradise Tyranny of the Majority Optimal Scenario (rich voice, no equalization) (median voter, vertical) (rich voice, hor w/ fed) ECCS, Barcelona Sept. 16, 2013 /