[PPT] - The accuracy of the bunching method under optimization frictions: PowerPoint Presentation

SLIDE 1

The accuracy of the bunching method under

ptimization frictions:

Students’ labor market frictions

Tuomas Kosonen (Labour Institute for Economic Research) Tuomas Matikka (VATT Institute for Economic Research) IFS Workshop, March 1, 2016

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 1 / 29

SLIDE 2

Motivation

Bunching method is used to discover earnings elasticities from behavioral responses to kinks and notches (Saez AEJ:EP 2010, Bastani and Selin JPubE 2014, Kleven 2015)

Relate the size of the excess mass in the income distribution to the size

f tax incentives

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 2 / 29

SLIDE 3

Motivation

The importance of optimization frictions in attenuating responses (Chetty et al. QJE 2011, Chetty Ecta 2012, Kleven and Waseem QJE 2013)

Mitigated observed bunching responses

Frictions are defined as factors that affect behavior but outside of the standard labor-leisure framework Types of optimization frictions:

Labor market and search frictions: imprecise control of (annual) income (Chetty et al. QJE 2011) Awareness of rules and optimization ability: lack of these attenuate responses (Chetty et al. AER 2009, Chetty et al. AER 2013, Currie and Grogger 2001 and Kleven and Kopczuk AEJ:EP 2008)

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 3 / 29

SLIDE 4

Introduction

Key question: How different optimization frictions affect

ptimization behavior and the bunching estimator?

Study subsidy in Finland

Applies to all higher education students (university, polytechnic) Eligibility depends on earnings: exceeding an income limit results in losing disposable income → the notch Notches create dominated regions where no one should willingly locate under any standard preferences

A reform shifted out income limits

Global vs. local effects of the limit Potential to disentangle different optimization frictions

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 4 / 29

SLIDE 5

Introduction

We find both large excess bunching and a mass of students in the dominated region

Students are aware of the notch, but frictions seem to matter too

The reform shifts out the whole income distribution → a global response

The income limit appears to affect behavior further away from the notch (in addition to the local response) Bunching at the old notch disappears → students in general aware of the income limits

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 5 / 29

SLIDE 6

Introduction

Divided sample results

Optimization failures and ability explain some of the nonstandard responses Suggestively, these types of frictions only attenuate local responses

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 6 / 29

SLIDE 7

Institutions

Marginal income tax rate schedule

.2 .3 .4 .5 .6 Marginal tax rate 20000 40000 60000 80000 100000 Taxable income Note: Marginal tax rates include the average flat municipal tax rate and average social security contributions

Year 2007

Marginal income tax rate schedule

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 7 / 29

SLIDE 8

Institutions

Study subsidy

Study subsidy approx. 500e per month when studying at a university

r polytechnic

Default: 9 subsidy months per year

Max. 55 months per degree (in 2007)

Eligibility depends on

annual gross income: Default limit 11,800e

income limit depends inversely on the number of study subsidy months applied

credit points (5pts/month)

In Finland, students typically work part-time during their studies

Effective third-party reporting of wage income

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 8 / 29

SLIDE 9

Institutions

Study subsidy

The notch: Exceeding the income limit results in reclaiming the subsidy of one month (with interest)

Carefully monitored by the Social Insurance Institution Additional subsidy month is reclaimed per every 1,300e over the limit

The reform: Income limits increased by 30% in 2008

Default 9 subsidy months: 9,200e→11,800e

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 9 / 29

SLIDE 10

Institutions

Study subsidy notch

11100 11200 11300 11400 11500 11600 11700

Disposable income, euros

400
300
200
100

100 200 300 400

Gross income relative to notch point

9 months of study subsidy, year 2007

Disposable income around the study subsidy notch

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 10 / 29

SLIDE 11

Methods

Behavioral responses to kinks and notches

Under standard labor-leisure preferences, some individuals should respond to local discontinuities in their budget sets if average elasticity is significant If these responses occur, we should see individuals clustering around kinks and notches However, frictions might eliminate or mitigate observed responses

In theory, frictions can be thought of as adjustment costs or fixed costs Example: small structural elasticity and large frictions → no observed responses Different (set of) frictions might lead to different patterns of responding

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 11 / 29

SLIDE 12

Methods

Bunching at a kink point

(1- )z z

k

k+dz

Indiff. curve

for type L

Indiff. curves

for type H Slope 1-τ1 Slope 1-τ2

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 12 / 29

SLIDE 13

Methods

Bunching at a notch point

z j j+ ∆z (1- )z

Slope 1-τ

j+∆zD

∆τ

Indiff. curve

for type L

Indiff. curves

for type H

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 13 / 29

SLIDE 14

Methods

Empirical methods

Bunching estimation

Calculate elasticity by comparing behavioral responses from the size of the excess mass (bunching) to the incentives (Saez AEJ:EP 2010, Kleven and Waseem QJE 2013, Chetty et al. QJE 2011) Even if ETI is not correctly estimated, the bunching response is informative about behavioral responses

Optimization frictions

bunching response smaller than in a frictionless world → Kleven and Waseem (QJE 2013) correct for this by utilizing the mass in the dominated region

True response to a notch could be more global

Frictions not only attenuating bunching → We present new thoughts

n extending the bunching method

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 14 / 29

SLIDE 15

Methods

Data

Register-based data on the universe of taxpayers in Finland in 1999-2011 (at the moment student data up to 2009)

Based on linked employer-employee database (FLEED) provided by Statistics Finland Data include detailed income tax and income transfer variables from

fficial registers (Tax Administration, Social Insurance Institution)

Allows for accurate analysis of bunching around different kinks and notches

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 15 / 29

SLIDE 16

Results

Study subsidy notch

2000 4000 6000 8000 10000 12000 Frequency

50
40
30
20
10

10 20 30 40 50 Distance from the notch Observed Counterfactual

Excess mass: 2.046 (.227), Share in the dominated region: .906 (.032) Upper limit: 26 (4.613)

Study subsidy notch, all students 2000 4000 6000 Frequency

50
40
30
20
10

10 20 30 40 50 Distance from the notch Observed Counterfactual

Excess mass: 1.903 (.269), Share in the dominated region: .881 (.039) Upper limit: 23 (3.24)

Study subsidy notch, students with default subsidy (9 months)

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 16 / 29

SLIDE 17

Results

MTR bunching for current and former students

5000 10000 15000 20000 25000 30000 Frequency

50
40
30
20
10

10 20 30 40 50 Distance from the kink Observed Counterfactual

Excess mass: -.09 (.107), Elasticity: -.006(.007)

First MTR kink, all students 2500 3000 3500 4000 4500 5000 Frequency

50
40
30
20
10

10 20 30 40 50 Distance from the kink Observed Counterfactual

Excess mass: -.106 (.084), Elasticity: -.001(.001)

Last MTR kink, university/polytechnic graduates 1000 2000 3000 4000 Frequency

50
40
30
20
10

10 20 30 40 50 Distance from the kink Observed Counterfactual

Excess mass: -.656 (.183), Elasticity: -.044(.012)

First MTR kink, students who previously bunch at the notch

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 17 / 29

SLIDE 18

Results

Road map

We showed that notches create behavioral responses and that

ptimization frictions tend to attenuate these responses

Next: Global vs. local responses

Utilize the reform to study the extent and shape of behavioral responses Alternative counterfactuals

Divided sample results

Divide students according to study credits, correlated with optimization ability Compare new and old students and rules, look into awareness about the rules more carefully Position relative to the notch prior to the reform (bunchers, dominated region)

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 18 / 29

SLIDE 19

Global vs. local responses

Students with 9 subsidy months in different years

200 400 600 Frequency 5000 10000 15000 20000 Income 2004−2005 2006−2007 2008−2009

Students with 9 subsidy months and 9 subsidy months in base year

Income distribution in different years

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 19 / 29

SLIDE 20

Global vs. local responses

Students with 9 subsidy months in 2007 - 2008

100 200 300 400 Frequency 2000 4000 6000 8000 10000 12000 14000 16000 18000 Income 2007 2008

Students with 9 subsidy months in year t and t−1

Income distribution in different years

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 20 / 29

SLIDE 21

Global vs. local responses

Alternative shape of distribution

1500 3000 4500 6000 Students 1000 2500 4000 5500 7000 Non−students 1500 4500 7500 10500 13500 16500 19500 Income Students Non−students

Students: 9 subsidy months Non−students: part−time workers, aged 19−24

Income distributions for students and non−students

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 21 / 29

SLIDE 22

Global vs. local responses

Alternative shape of distribution: Dif-in-dif-style

.00005 .0001 .00015 Density 1500 4500 7500 10500 13500 16500 19500 Labor income Students 04−05 Students 08−09 Non−students 04−05 Non−students 08−09

Students with 9 subsidy months Non−students: part−time workers aged 19−24

Income distributions for students and non−students

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 22 / 29

SLIDE 23

Divided sample results

Room for optimization errors?

We found that some of the optimization frictions originate from the labor market and could be more global in nature Students also aware of the study subsidy system in general Any room for learning / inattention / optimization error type of frictions that attenuate behavior?

These issues can be studied utilizing the local approach

We next utilize study subsidy credits to investigate if some students do not fully understand all rules (although aware of them)

The relevant income concept (annual gross income) How to accurately calculate and predict annual income

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 23 / 29

SLIDE 24

Divided sample results

Divided sample results: study credits

100 200 300 400 500 Frequency −50 −40 −30 −20 −10 10 20 30 40 50 Distance from the notch Study subsidy notch, study credits 0−25th percentile 500 1000 1500 2000 2500 Frequency −50 −40 −30 −20 −10 10 20 30 40 50 Distance from the notch Study subsidy notch, study credits 75−100th percentile

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 24 / 29

SLIDE 25

Divided sample results

Utilizing the reform: new and old students and rules

500 1000 1500 2000 Frequency −50 −40 −30 −20 −10 10 20 30 40 50 Distance from the notch Study subsidy notch, new students before the reform 500 1000 1500 Frequency −50 −40 −30 −20 −10 10 20 30 40 50 Distance from the notch Study subsidy notch, new students after the reform

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 25 / 29

SLIDE 26

Divided sample results

Utilizing the reform: new and old students and rules

500 1000 1500 2000 Frequency −50 −40 −30 −20 −10 10 20 30 40 50 Distance from the notch Study subsidy notch, old students before the reform 200 400 600 800 1000 1200 Frequency −50 −40 −30 −20 −10 10 20 30 40 50 Distance from the notch Study subsidy notch, old students after the reform

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 26 / 29

SLIDE 27

Divided sample results

Bunching/dominated region before the reform

100 200 300 400 500 Frequency −50 −40 −30 −20 −10 10 20 30 40 50 Distance from the notch Study subsidy notch, students who bunched before the reform 50 100 150 200 250 Frequency −50 −40 −30 −20 −10 10 20 30 40 50 Distance from the notch Study subsidy notch, students in the dominated region before

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 27 / 29

SLIDE 28

Summary

Conclusions

Study subsidy evidence shows that incentives matter (bunching) and at the same time optimization frictions attenuate behavior (many students in dominated region) The reform evidence shows that the notch does not only induce local bunching, but also a more global shift in the income distribution

The notch appears to induce behavioral responses well below the income limit Presumably, ETI is not calculated correctly by just utilizing local bunching estimates This, and the disperse bunching, also indicate that taxpayers have difficulties in controlling their income precisely

Optimization ability also prevalent in explaining behavioral responses

Correlation between study credits and optimization behavior Use data on secondary school test scores in the future

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 28 / 29

SLIDE 29

Summary

Thank you for your attention!

Tuomas Matikka (VATT) Students and optimization frictions IFS Workshop 2016 29 / 29