Ex-ante Moral Hazard of Unemployment Insurance...after increases in the potential duration of unemployment insurance by one month. Additionally, this paper analyzes the interaction

University of Brasilia

Economics and Politics Research Group–EPRG

A CNPq-Brazil Research Group http://www.econpolrg.com/

Research Center on Economics and Finance–CIEF Research Center on Market Regulation–CERME

Research Laboratory on Political Behavior, Institutions and Public Policy–LAPCIPP Master’s Program in Public Economics–MESP

Graduate Program in Economics–Pós-ECO

Ex-ante Moral Hazard of Unemployment Insurance

Artur Henrique da Silva Santos (STN) Maurício Soares Bugarin (UnB)

Paulo Roberto Amorim Loureiro (UnB)

Economics and Politics Working Paper 102/2020 September 10th, 2020

Economics and Politics Research Group Working Paper Series

http://www.econpolrg.com/

Ex-ante Moral Hazard of Unemployment Insurance

Artur Henrique da Silva Santos∗

Maurıcio Soares Bugarin†

Paulo Roberto Amorim Loureiro†

ABSTRACT

Unemployment insurance aims to maintain workers’ consumption while they are unemployed, butmay have undesirable effects on the beneficiaries’ labor supply. Theoretical and empirical studies haveprovided robust evidence of a reduction in the job search effort and an increase in the duration ofunemployment, i.e., ex-post moral hazard.By contrast, this paper analyzes the influence of unemployment insurance on the behavior of workerswhile employed, i.e., ex-ante moral hazard. We use regression kink design and differences-in-differencesto estimate the higher probability of termination after workers’ ensure eligibility for the benefit, andafter increases in the potential duration of unemployment insurance by one month. Additionally, thispaper analyzes the interaction of business cycles with ex-ante moral hazard. Contrary to what onewould expect, the probability of termination has a procyclical behavior. The insurance is more widelyused in periods of economic expansion than in recessions. According to the theoretical model, thisoccurs because, in expansion, it is easier to move out of unemployment after the insurance payout,whereas in recession, heading back to the job market is more challenging and, consequently, ex-antemoral hazard is less likely.

Keywords: unemployment insurance; ex-ante moral hazard; regression kink design; nonparametricestimation.JEL Classification: C41, D82, J22, J64

∗National Treasury Secretariat - [email protected]†University of Brasilia

1

1 INTRODUCTION

Most of the unemployment insurance literature has focused on analyzing ex-post moral hazard. In-deed, hundreds of theoretical and empirical studies have gathered strong evidence that unemploymentinsurance reduces job search effort and increases the duration of unemployment.

The present paper’s innovation is to conduct a unique analysis of the ex-ante moral hazard of unem-ployment insurance. It assesses whether workers bring about their own termination of employmentand would rather run the risk of remaining unemployed, having no earnings, and facing the uncer-tainty of whether they will get a new job. They make this choice only to be eligible for unemploymentinsurance and to have more leisure time while they are unemployed.

At first, this decision does not seem rational, or there would be a clearer preference risk. Neverthe-less, a sizable number of workers increase the likelihood of termination of their employment contractswhen becoming eligible for unemployment insurance and after an extension of the duration of unem-ployment insurance. That is enough to affect the rational decision of a substantial group of workersand to cause a discontinuous density of terminations during job tenure in a heavily populated countrysuch as Brazil.

Several studies have overlooked the evidence of cutoff manipulation when there is an increase inthe potential duration of unemployment insurance. This is also known as discontinuous density.It occurs with the increase in the number of individuals terminated from their jobs who apply forunemployment insurance immediately after the increase in its potential duration. This evidencewas observed by Meyer and Mok (2014) in New York; by Schmieder et al. (2016) in Germany; byLe Barbanchon et al. (2019) in France; and by Gerard and Gonzaga (2013) and Gerard et al. (2020)in Brazil. However, none of these studies claim that unemployment insurance is the cause of theincreased likelihood of termination after the cutoff. Only Gonzaga and Pinto (2014) suggested it.

We initially show strong evidence of ex-ante moral hazard by graphically representing discontinuousdensity and hazard rate peaks after the eligibility cutoffs and increases in the potential duration ofunemployment insurance by one month in Brazil.

After 2015, we find additional strong evidence that unemployment insurance leads to discontinuousdensity. An amendment removed unemployment insurance eligibility at six months of job tenure froma group of workers (former eligibility cutoff), and the discontinuous density has virtually disappeared.

In our quest for a plausible explanation for this body of evidence, we developed a theoretical modelto assess workers’ ex-ante moral hazard behavior and how it decreases with declining business cycles.According to the theoretical model, during an economic crisis, workers tend to work hard to keeptheir jobs as it could be challenging to find an equivalent job in case of termination. In a boomingeconomy, workers tend not to work so hard, contributing to their own termination so that they canreceive unemployment insurance. This behavior has a procyclical effect on unemployment insuranceexpenditures. Finally, this theoretical model analyzes why this effect is so strong in Brazil. Imagine ascenario where the replacement rate is greater than or equal to one for a considerable share of workers,and the level of informality is high. Under these circumstances, termination from formal employmentdoes not necessarily decrease income during unemployment. This effect makes a period of unemploy-ment even more attractive provided the new job is as good as the previous one and it is found quickly.

The analysis in the present paper is based on the Annual Report on Social Information (Rais) micro-data and the unemployment insurance administrative database. Both were obtained from 2011 to2016 from the then Ministry of Labor. Rais contains information only on employees who have beenterminated without due cause. To prevent loss of information, there are no sampling restrictions in

2

this study, i.e., we analyze data on all Brazilian employees who terminated a job. The analysis levelconsiders the employment relationship (employee at the firm, no duplicated records), accounting foraround 59 million terminations from 2011 to 2016.

To estimate the level of ex-ante moral hazard and infer the effect of discontinuous density on theprobability of termination, we developed a nonparametric Kaplan-Meier survival model. This modelestimates the probability of termination during job tenure. We estimate the slope changes in theprobability of termination around the cutoffs using the regression kink design (RKD). At six months’tenure, after eligibility for unemployment insurance, the probability of termination increased by 0.755percentage points (p.p.) monthly, when its average was 4%, representing a 19% monthly increase inthe probability of termination. After increasing the potential duration of unemployment insuranceby one month, there were more modest increases in the probability of termination at the cutoffs for12 and 24 months of job tenure. We estimate that the probability increased by 0.2 p.p. and 0.104p.p. at these cutoffs. Considering averages of 17.8% and 48.4%, the probabilities of terminationincreased by 1.1% and 0.21%, respectively.

Thereafter, we assessed the group of workers who lost their eligibility for unemployment insuranceafter the amendment to the law in 2015, even if they had six months of tenure. Until 2014, theseworkers increased their probability of termination by 1.28 p.p. after the cutoff for the six months’ jobtenure. As the average probability was 6.8%, there was a 19% monthly increase in the probability oftermination. With the loss of eligibility after March 2015, this probability increased by 0.19 p.p. Asthe average probability was 5%, the monthly increase in the slope was 3.8%. Therefore, after the lossof eligibility, the probability of termination decreased by 80%. Conversely, the group for which theunemployment insurance rule was not amended had a slight change in ex-ante moral hazard, wherethe monthly increase in the probability of termination went from 41% to 35%.

The ex-ante moral hazard at the cutoff for the 24 months’ job tenure decreased dramatically afterMarch 2015. However, the unemployment insurance rules did not change for this cutoff.

We believe that this reduction in ex-ante moral hazard could be due to the business cycles. Until2014, Brazil experienced a strong economic boom. From mid-2014 to 2016, Brazil faced an economiccrisis. This result possibly shows that moral hazard decreases during a crisis.

Finally, to estimate the effect of greater moral hazard free from business cycle movements, this studyproposes a differences-in-differences (DD) model for the estimated increase in probability obtainedwith RKD. The DD model uses the initial RKD estimates at the cutoff for the six months’ jobtenure at the local level, per year, allocating those workers who lost eligibility to the control groupand those who remained eligible to the treatment group. The DD model estimates that the ex-antemoral hazard of unemployment insurance increases the probability of termination of employmentby 2.6 p.p. As the probability of termination with a six months’ tenure was 4%, the moral hazardincreased the probability of termination by 65%.

Therefore, unemployment insurance generates two undesirable effects on labor supply: a larger num-ber of terminations and longer duration of unemployment, as described in the literature. Bothcontribute, in distinct ways, to increasing the unemployment rate.

This paper is not the first study on ex-ante moral hazard available in the literature. Topel (1983) wasthe first researcher to investigate it. He found evidence that the higher the replacement rate1 andgovernment subsidies to the unemployment insurance system, the higher the probability of employ-ment termination. The main studies on ex-ante moral hazard have been conducted in Canada2 and

1Ratio between the amount of unemployment insurance and the wage earned before termination.2Christofides and McKenna (1995), Green and Sargent (1998), and Baker and Rea Jr (1998) analyze increases in

3

Spain3 and they suggest an increase in the hazard rate of termination of employment4 after eligibilityfor unemployment insurance. There is ample evidence of that in Brazil, which is associated with ahigh level of informality.

Britto (2015) assessed the effect of an increase in unemployment insurance on the probability ofemployment termination5. By using an RKD for the minimum value of unemployment insurance inBrazil, he concluded that an increase in unemployment insurance reduces the probability of termina-tion. On the other hand, Carvalho et al. (2018) estimated differences-in-differences of an amendmentto the law and found out that loss of eligibility reduces the unemployment rate.

However, none of these studies estimate the ex-ante moral hazard of the probability of terminationgenerated by an employment survival model. This model is widely used to assess the ex-post moralhazard.The present paper innovates and makes a significant contribution to the scientific literatureby estimating ex-ante moral hazard using slope changes in the probability of termination, RKD, anddifferences-in-differences analysis.

As the paper explains density discontinuity, it is related to the literature on bunching (Saez, 2010),notches (Kleven and Waseem, 2013), and density discontinuity approach (Doyle Jr, 2007; Jales,2018). To explain discontinuities, this literature estimates a latent probability density function andproposes three hypotheses: (i) inexistence of spillover effects in the external area of discontinuity;(ii) density continuity, if untreated; and (iii) strong dependence on the theoretical density model.

Differently, in the present paper, we do not estimate a latent density function and we do not assumea theoretical density function model. We replace the theoretical model with the nonparametricsurvival model. Thus, we do not assume the hypotheses of inexistence of spillover effects and ofdependence on the theoretical density model. By estimating the effect of the policy using RKD, weadopt the hypothesis of density continuity with no treatment. Therefore, this paper contributes tothe literature by explaining discontinuous density with fewer hypotheses and no dependence on adhoc formulations. We test the hypothesis of continuity at different placebo cutoffs. We find evidencethat the hypothesis of continuity holds in the absence of treatment.

2 UNEMPLOYMENT INSURANCE IN BRAZIL

In Brazil, we have unemployment insurance and a severance indemnity fund, with mandatory con-tributions over time, known as Fundo de Garantia do Tempo de Servico (FGTS). Unemploymentinsurance is a cash benefit whose amount is calculated based on the wage earned in the previous job.

the hazard rate after eligibility in Canada.3Rebollo-Sanz (2012) assesses the effect of unemployment insurance on turnover in Spain. He evaluates the effects

of unemployment insurance on exit rates out of employment and out of unemployment and he estimates the peakhazard rate after eligibility for unemployment insurance.

4These studies develop survival models for employment analysis. In these models, the hazard rate denotes theintensity of terminations for a given period, considering that the employee has not been terminated from the job yet,and then we have,

h(t) = lim∆t→0

Pr (t+ ∆t > t|T > t)

∆t

Where, t is the job tenure, h(t) is the hazard rate, and T is the termination event.5However, this result is interpreted based on the assumption that unemployment insurance increases at the first

kink. But variations in the replacement rate are observed at this kink. If interpretation was based on the reductionof the replacement rate, the conclusion would be the opposite: lower replacement rates increase the probability oftermination. This was not the interpretation made by the author, but it is necessary to reflect on the actual treatmentgiven to unemployment insurance at the first kink, whether it can be compared to the other two kinks in the Brazilianunemployment insurance system and the effects of treatment at these kinks on the duration of unemployment.

4

The potential duration of unemployment insurance is estimated based on time worked in the past 36months prior to termination. FGTS consists of savings through monthly deposits made directly byfirms, and the worker is entitled to receive the total amount after the termination of employment.Both policies share the same purpose: to guarantee the worker’s earnings during the unemploymentperiod. The worker is only entitled to receive these benefits if he is terminated from the job, i.e.,terminated without cause. In this paper, we analyze only the effect of unemployment insurance ruleson the worker’s behavior while employed. However, FGTS compounds this effect.

Unemployment insurance is paid by the government and has drained public coffers since the early2000s owing to increasing spending. Spending on unemployment insurance has risen by 602% in 17years. As shown in Graph 1, public spending amounted to R$ 4.7 billion in 2001, but went up to R$33.0 billion in 2018, accounting for an increase of R$ 28.3 billion.

Graph 1: Formal unemployment insurance expenditures – 2001 to 2018.

Source: Siga-Brasil/Brazilian Senate. Metric: total amount paid.

As pointed out by (Meyer, 2002), one of the purposes of unemployment insurance is to serve as an“automatic stabilizer” in crisis periods. With cash transfers to the unemployed during recession pe-riods, the increase in unemployment is not expected to be fully applied to aggregate demand. Thus,unemployment insurance expenditures would be countercyclical. Nonetheless, Graph 2 shows thatwhen the unemployment rate in Brazil increases, the number of unemployment insurance recipientstends to decrease. Conversely, when unemployment decreases, the number of recipients tends toincrease sharply.

After several years of a generalized increase in mandatory public spending6, the Brazilian governmenthas experienced a fiscal crisis since 2014, with consecutive primary deficits, difficulty with spendingcuts, and revenue reduction as a result of the economic crisis. In late 2014, the Brazilian govern-ment enacted a provisional executive act7 to curb unemployment insurance expenditures until theCongress could decide on the new rule governing the unemployment insurance. Later, in June 2015,the Congress established the definitive rules8.

The provisional executive act was published in December 2014 and was expected to come into forceas of March 2015. As the new rules would disallow the eligibility of a given group of workers forunemployment insurance, it is plausible to assume an endogenous behavior that led workers to file

6Spending on public policies established by law, which the government is supposed to incur.7Provisional Executive Act 665/2014.8Act 13.134/2015.

5

Graph 2: Quarterly series and trends of the unemployment rate, GDP, and the number ofunemployment insurance beneficiaries.

Source: Data compiled by the authors based on SCN2000 and SCN2010/IBGE, PME/IBGE, PNADC/IBGE and

MTE data.

for unemployment insurance in advance, which occurred in January and February 2015. To eliminatethis bias, we excluded this 60-day period from our estimates and graphs on the behavior of workersbefore and after the amendment to the unemployment insurance rules.

The change in the rules increased the minimum job tenure requirements for those workers who werefiling for unemployment insurance for the first or second time. The period of analysis spanned 36months prior to termination. Table 1 summarizes the amendments made to unemployment insuranceeligibility rules.

Table 1: Summary of amendments to the unemployment insurance rules.

Recipient Until March 2015Between March 2015 andJune 2015

As of June, 2015

1st time· 4 months of insurance if jobtenure ≥ 18 and < 24 months

· 4 months of insurance if jobtenure ≥ 12 and < 24 months


· 5 months of insurance if jobtenure ≥ 24 months


2nd time· 4 months of insurance if jobtenure ≥ 12 and < 24 months




· 4 months of insurance if jobtenure ≥ 12 and < 24 months· 5 months of insurance if jobtenure ≥ 24 months

3rd time ormore times








Source: Data compiled by the authors.

6

Therefore, the rules did not change the rights for those filing for unemployment insurance for thethird time or more, and the cutoff for six months’ job tenure was kept for unemployment insuranceeligibility. However, those filing for the benefit for the first or second time needed longer job tenureand were no longer entitled to receive the insurance if they had six months’ job tenure.

3 EVIDENCE OF EX-ANTE MORAL HAZARD OF UN-

EMPLOYMENT INSURANCE

Graph 3 compares the density function and the hazard function for job tenure between 2011 and 2016,taking into account all workers terminated from their jobs. Both functions show a similar behavior.There is permanent termination discontinuity over the six-month period, temporary discontinuityover 12 months, and permanent discontinuity over 24 months. The intensity of termination discon-tinuity over six months is stronger than at the other cutoffs. These termination peaks coincide withthe unemployment insurance rules in force for most workers during that period. With a six months’job tenure, the worker becomes eligible for three months of insurance. With 12 and 24 months’ jobtenure, the worker adds one more month to the three months of insurance he is entitled to, receivingthe benefit for 4 and 5 months, respectively.

Graph 3: Density and hazard rate of terminations during job tenure.

Data: Data compiled by the authors from Rais database.

The graph suggests workers can contribute to their job termination when they become eligible for7

the unemployment insurance. In this section, we do not assess the magnitude of this increase andwhether it is statistically significant. However, Graph 3 suggests that unemployment insurance influ-ences the worker’s behavior while employed. In other words, ex-ante moral hazard is therefore likely.

Between 2011 and 2014, Brazil experienced economic growth. From 2014 to 2016, the country washit by recession. Concomitantly, from 2015 onwards, the government amended the unemploymentinsurance rules, increasing job tenure requirements for those filing for unemployment insurance forthe first or second time. Those workers were no longer eligible for unemployment insurance with sixmonths’ job tenure. Conversely, those filing for the insurance for the third time or more were stilleligible for unemployment insurance with six months’ job tenure.

Graph 4 compares termination density in those two groups for the periods 2011 to 2014 and March2015 to December 2016. The group of workers filing for the insurance for the first and second timeis affected by changes in the business cycle and unemployment insurance rules. The group of work-ers filing for the insurance for the third time or more is affected only by changes in the business cycle.

Graph 4: Termination density during job tenure per group of recipients for the periods 2011 to2014 and March, 2015 to December, 2016.

0.0

2.0

4.0

6re

lativ

e fre

quen

cy

0 3 6 9 12 15 18 21 24 27 30job tenure

3rd or more (before 31/dec/2014)0

.02

.04

.06

rela

tive

frequ

ency

0 3 6 9 12 15 18 21 24 27 30job tenure

3rd or more (after 01/mar/2015)

0.0

2.0

4.0

6

rela

tive

frequ

ency

0 3 6 9 12 15 18 21 24 27 30job tenure

1st or 2nd (before 31/dec/2014)

0.0

2.0

4.0

6

rela

tive

frequ

ency

0 3 6 9 12 15 18 21 24 27 30job tenure

1st or 2nd (after 01/mar/2015)

Source: Data compiled by the authors from Rais database.

Termination discontinuity with six months’ job tenure for those filing for the insurance for the firstand second time virtually disappeared after the changes in eligibility criteria in 2015. On the otherhand, discontinuity for those filing for the insurance for the third time or more times remained low.This evidence strongly suggests that unemployment insurance encourages workers to contribute to

8

their own termination from the job in order to be entitled to the benefit.

Density discontinuities of terminations at the 24-month cutoff decreased in both groups. This suggeststhat the economic crisis discouraged workers from contributing to their termination of employment.A possible explanation is that, in a recession, workers would have difficulty getting another job afterhaving received the unemployment insurance, and then they choose to work harder in their currentjobs to avert termination. The following section develops a theoretical model that assesses theseworkers’ behavior towards business cycles from an intertemporal standpoint.

4 THEORETICAL MODEL

4.1 Elements of the model

Suppose a worker can choose how hard he wants to work (at) in order to increase his productivity, andhow much effort he puts when unemployed (st) to search a job that pays better than his reservationwage. By giving his best in the job, the worker has the disutility cost ψe(at). When he puts a greatdeal of effort into looking for a job, he faces the disutility cost ψu(st). Both are strictly increasingand strictly convex functions, i.e.,

ψ′i (.) > 0, ψ′′i (.) > 0, i = e, u (1)

Suppose hypothetically that the decision to make a great deal of effort in employment reduces theprobability of being fired (πt), and the decision to effort in job search increases the probability ofmoving out of unemployment (pt), i.e.,

dπtdat

< 0;∂pt∂st

> 0 (2)

Also, assume, for simplicity, that efforts affect the probabilities of being fired and moving out ofunemployment9 linearly, i.e.,

d2πtda2

t

= 0;∂2pt∂s2

t

= 0 (3)

A worker may be employed or unemployed at t. In any case, the worker’s utility function takesinto consideration income and disutility in the current period, and next period expected utility, dis-counted at a factor β, 0 < β < 1. This study assumes a worker can move in and out of employmentseveral times. While widely used in the literature10, permanent job tenure, in which an employeekeeps his job until he retires, is not assumed herein.

When unemployed, the worker has a constant income (yu) that is exogenous to the model11. If theworker is terminated from the job after some time, he becomes eligible for unemployment insurance(b) for some time12, not being entitled to it in the subsequent period. When employed, the worker

9The hypothesis of linearity regarding the probability of moving out of unemployment is not necessary. However,it is assumed for the sake of simplicity.

10For other examples, see Shavell and Weiss (1979), Chetty (2008); and Schmieder et al. (2012).11This income may be derived from assets accumulated over time, from third-party donations, from government

cash transfer programs, or from an informal job. In this study, we assess the importance of informality based on thisincome.

12We simplified the model to represent the potential duration of unemployment insurance and the minimum jobtenure required from the worker for eligibility to be equal to 1 period.

9

receives a net of tax wage (w).

If the worker is employed in the initial period, the utility function is:

Vt = u(w) − ψe(at) + β[(1− πt(at))Vt+1 + πt(at)U

elt+1

](4)

If the worker is unemployed in the initial period and eligible for the unemployment insurance, theutility function is:

U elt = u(yu + b) − ψu(st) + β

[pt(st)Vt+1 + (1− pt(st))Une

t+1

](5)

If the worker is unemployed in the initial period but not eligible for the unemployment insurance,the utility function is:

Unet = u (yu)− ψu(st) + β

[pt(st)Vt+1 + (1− pt(st))Une

t+1

](6)

Subscript t denotes the current period. Superscript el or ne in the expected utilities denotes el-igibility or non-eligibility for the unemployment insurance. Utility Vt is the employed status att, whereas utility Ux

t is the unemployment status at t, x = el, ne. The utility function for theunemployed status changes after the unemployment insurance has been received one period, i.e.,U elt −Une

t = ∆u(b) = u (yu + b)− u(yu). Eligibility in the current period does not apply if the agentis employed.

Vt+1, U elt+1, and Une

t+1 denote the expected utility for the employed, unemployed and eligible, and un-employed but not eligible for the unemployment insurance, respectively, considering worker’s optimaldecisions from the subsequent period onwards. Note that Vt+1, U el

t+1 and Unet+1 are independent from

t, st, and at.

For the sake of future comparison, we also included the possibility of an employed worker not beingeligible at the end of the period. In this case, the utility is given by:

Vt = u(w) − ψe(at) + β[(1− πt(at))Vt+1 + πt(at)U

net+1

](7)

Finally, consider the hypothesis that the business cycle (Y ) expected by workers interferes in theirprobability of moving out of unemployment13. If the worker expects economic growth, then he ex-pects a lower probability of moving out of unemployment. If he expects an economic crisis, then theprobability of moving out of unemployment increases. In other words:

∂pt

∂Y> 0 (8)

To fully represent the probability of moving out of unemployment, denote it as a function of jobsearch effort, of the expected business cycle, and of the reservation wage (wR).

pt = p(st, Y , wR) (9)

13This hypothesis considers that economic growth boosts the demand for labor and tends to shift the distributionof wage supply for a given reservation wage outward to the right. Additionally, the crisis tends to shift the distributionto the left, as McCall (1970) shows for the effect of a qualification policy. In addition, this hypothesis is more naturalto accept in a scenario with wage stickiness. Hence, the reservation wage does not follow the change in the labordemand of firms and interferes with the probability of moving out of unemployment.

10

The literature defines wR as the minimum acceptable wage for an unemployed worker to accept a joboffer. This study relies on a broader concept for that variable. Consider wR the minimum acceptablewage for the worker, regardless of whether he is employed or unemployed. If he is unemployed, whathas been used in the literature applies, and the worker accepts the job offer if the wage is higherthan wR. If he is employed and the wage is lower than wR, the worker gets discouraged and can askto leave the job, look for another job, or be less diligent at work, leading to his termination. Thistheoretical model considers the latter alternative.

4.2 Optimal decision on worker’s effort

This study uses standard dynamic programming techniques, similarly to the study carried out byFerejohn (1986), assessing the optimal strategy of a policymaker. In this study, we consider that theworker will choose the effort at period t that maximizes the expected (discounted) utility from thatperiod on. We employ a model in which the worker’s initial status is “employed”.

At first, this study presents an equilibrium result when the worker chooses to keep his job. Furtherahead, we present a comparative statics result, showing that the level of job effort varies dependingon the parameters that characterize unemployment insurance eligibility, extension of potential dura-tion of unemployment insurance, and change in the business cycle.

To hold down the job, the wage should be higher than the reservation wage (wR). If the wage istoo low, the worker will think it is not worth keeping the job, and he will make little effort instead(at = 0). On the other hand, if level w is too high, the worker will think it pays off to hold down thejob and chooses an optimal at level.

The equilibrium outcome in which the worker prefers to keep the job only occurs if the utility, whileemployed, is higher than the utility when unemployed and eligible. Otherwise, the worker will notbe satisfied with his job and will choose at = 0. We can rearrange the terms of this condition andset up the following characterization of the best strategies for defining w and wR:

Vt > U elt ⇔ u (w) > u (wR) (10)

Where,

u (wR) = u (yu + b)− βπt∆u (b) + ψe (at)− ψu (st) + β(Unet+1 − Vt+1

)(1− pt − πt) (11)

which can be rewritten as

w > wR

Where,

wR = u−1u (yu + b)− βπt∆u (b) + ψe (at)− ψu (st) + β(Unet+1 − Vt+1

)(1− pt − πt) (12)

The worker will choose an action that maximizes his (discounted) utility from that moment on,choosing at to maximize the current value of flow utility. Obviously, if w < wR, then it is not possi-ble to be employed and the worker will always choose at = 0.

The worker’s ideal strategy consists of optimal choices at every t, considering optimal strategies inthe future, equivalent to a subgame perfect equilibrium.

11

Therefore, the optimal effort solution at t is obtained from the first-order condition of equation 11:

ψ′e (at) = βdπtdat

[Unet+1 + ∆u (b)− Vt+1

]= β

dπtdat

[U elt+1 − Vt+1

]= β

(−dπtdat

)[Vt+1 − U el

t+1

](13)

In an optimal effort situation, the marginal cost will be equal to the expected future marginal benefitfor increasing the chances of holding down the job. Thus, maximum job effort would result in amarginal disutility greater than zero, given that

ψ′e (at) =

+︷︸︸︷β +

−︷︸︸︷dπtdat−

−︷︸︸︷[U elt+1 − Vt+1

]≥ 0 (14)

The result of equation 13 should be stationary in the sense that at = a for every t and equation 13can be rewritten as follows,

ψ′e (a) = βdπ

da

[U el − V

](15)

Note that, if equation 4 is rewritten for period 0, including the optimal solution found in equation13, considering, therefore, that the probability of termination π0 = 0.

V0 = v(w) − ψe(a∗0) + β [V1] (16)

In this case, the strategies and returns are identical at 0 and 1. Moreover, the worker’s expectedutility does not depend on a0. Therefore, if a∗0 maximizes V0, a

∗1 should also maximize V1, and so on

and so forth for every t, such that at = a for every t.

Furthermore, we highlight an important fact in equation 15: a depends positively on the absolutevalue of difference

[U el − V

]. The higher the expected utility of remaining employed as opposed to

moving into unemployment, the greater the worker’s effort to keep his job.

4.3 Effect of Eligibility14 for the Unemployment Insurance

In equation 15, consider a worker who is eligible for the unemployment insurance:

ψ′e(a1)

= βdπ

da

[U el − V

](17)

When the worker is not eligible for the unemployment insurance, equation 15 can be rewritten as:

ψ′e(a2)

= βdπ

da

[Unl − V

](18)

Proposition 1 ψ′e (a2) > ψ′e (a1)

14The model assumes that after some time in employment, the worker is eligible for the unemployment insurance.Then, according to the model, it would make no sense to assess non-eligibility when the worker is in employment.However, in real life, the worker remains employed for some time before he becomes eligible. In this context, we findthis comparison to be relevant as it allows looking at the change in the worker’s behavior when he becomes eligible.Moreover, non-eligibility could be included in the model in the case of voluntary termination.

12

Proof Since U el = Unl + ∆u(b), we have U el > Unl.

Since dπda< 0, it follows that ψ′e (a1) < ψ′e (a2), Q.E.D.

Hence, as ψ(.) is strictly convex, then,

ψ′e(a2)> ψ′e

(a1)

=⇒ a2 > a1 (19)

So, when the worker is not eligible for the unemployment insurance, he works harder. This is sobecause, in the case of termination, he will not get the insurance, then he works harder to reducethe chances of being terminated from the job. In other words, unemployment insurance reduces theopportunity cost for losing the job, resulting in a reduction of the worker’s effort.

4.4 Effect of Extending the Potential Duration of the UnemploymentInsurance

For the sake of simplification15, we assume that expected utility when the worker is unemployed andeligible for two periods of potential duration of the unemployment insurance (U el

2SD) is higher thanthe expected utility with only one period (U el

1SD). Hence,

U el2SD > U el

1SD (20)

Thus, we can rewrite the equilibrium in equation 15, considering the scenario with two periods ofunemployment insurance payout:

ψ′e (a2SD) = βdπ

da

[U el

2SD − V]

(21)

And we can rewrite the equilibrium in equation 15, considering the scenario with one period ofunemployment insurance payout:

ψ′e (a1SD) = βdπ

da

[U el

2SD − V]

(22)

Proposition 2 ψ′e (a1SD) > ψ′e (a2SD)

Proof The proof is identical to the proof of Proposition 1.

Since ψ(.) is strictly convex, it follows that,

ψ′e (a1SD) > ψ′e (a2SD) =⇒ a1SD > a2SD (23)

Therefore, when the worker is eligible for the unemployment insurance for longer periods, he doesnot work as hard. This can be explained by the fact that he will get the unemployment insurancepayout for a longer time in the case of termination.

15We assume this simplification hypothesis because the model’s specification considers only one period for theunemployment insurance payout. To analyze this case without using this hypothesis, we should re-specify the wholemodel by considering three expected utilities of unemployment: (1) no eligibility: Unlt+1; (2) eligibility and one periodof unemployment insurance payout: Uelt+1, 1SD; and (3) eligibility and two periods of unemployment insurance payout:

Uelt+1, 2SD. In addition, we should specify when the worker becomes eligible for two periods of the unemploymentinsurance payout. This change would render the theoretical model unnecessarily complex and the result would be thesame, since Uelt+1, 2SD > Uelt+1, 1SD > Unlt+1 =⇒ Uel2SD > Uel1SD > Unl.

13

4.5 Effect of Business Cycles

In equation 15, recall that U elt depends on the probability pt and that pt = p(st, Y , wR). Therefore,

we may study the effect of an expected change in the business cycle. Indeed, deriving both sides byY 16 yields:

ψ′′e (a)da

dY= β

dπ

da

dU el

dY= β2dπ

da

dp

dY

[V − Unl

]=⇒

da

dY=

+︷︸︸︷β2

−︷︸︸︷dπ

da

+︷︸︸︷dp

dY

+︷︸︸︷[V − Unl

]ψ′′e (a)︸︷︷︸

+

≤ 0 (24)

i.e., if the worker expects economic growth, he tends to work less in the current period because thedemand for labor will be high, and if he is terminated from the job, he will be able to find anotherjob soon. If he expects an economic crisis, he works harder in the current job to reduce the chancesof termination and to keep his job during a period of low demand for labor.

4.6 Effect of Business Cycles: with High Level of Informality and HighReplacement Rate – Brazilian Scenario

Throughout the theoretical model, we consider the equilibrium solution occurs when the workerwould rather be employed than unemployed. We marginally assess the change in behavior using thecomparative statics results from the characterization of the equilibrium. However, it is also possiblethat he will prefer to be unemployed at t. In this case, in equation 13, consider U el

t+1 > Vt+1. So,

ψ′ (at) =

+︷︸︸︷β

−︷︸︸︷dπtdat

+︷︸︸︷[U elt+1 − Vt+1

]≤ 0 =⇒ at = 0 (25)

Hence, when the expected utility while unemployed is higher than while employed, the worker willprefer not to effort at work, thereby contributing to his termination from the job. This is a cornersolution. However, when does this scenario unfold? The following proposition sets up the conditionsfor when this tends to occur. The proof is presented in Appendix 1.

Proposition 3 In an economy with a high level of informality and with a generous unemploymentinsurance system (replacement rates close to 1) for a group of workers, U el

t+1 > Vt+1 is possible whenthere is no economic crisis.

In this scenario, when the economy is booming, the worker is not expected to work as hard becausehe is eligible for the unemployment insurance. In an expansionary phase, this occurs because he canmove out of unemployment shortly after he uses up the unemployment insurance if it is advantageousfor him to be in another formal job relationship again. With replacement rates close to 1 and witha high level of informality, he can have an income higher than or equal to the one he had whileemployed. Therefore, in this scenario, whenever eligible, the worker tends to contribute to his owntermination from the job in order to receive the unemployment insurance.

16Since d2πt

da2t= 0, then dπ

da is not a function of a.

14

However, during an economic crisis, the probability of getting another job is low. Therefore, havinga lazy behavior at work could lead to the worker’s termination with longer unemployment prospectthan the potential duration of unemployment insurance, i.e., he will possibly not receive the unem-ployment insurance and will not have a formal job for a long time. In this case, it is possible thatU elt+1 < Vt+1, i.e., the worker’s diligence will not be a corner solution equal to zero; it would possibly

be high, as shown in equation 13.

5 DATABASE AND METHODOLOGY

5.1 Database

This paper used the Annual Report on Social Information (Rais) microdata and the unemploymentinsurance administrative database, both obtained from 2011 to 2016 from the then Ministry of Labor.

Rais is a database of all employment relationships in Brazil. It identifies the worker, the employer,wage, job tenure, and other socioeconomic variables associated with the worker. At first, this studycollected employment information on the same worker, at the same firm, in the same time. There-fore, each observation from the database contains only one piece of information on the employmentrelationship of that worker, at that firm, and at the same time.

We also included the hiring and termination dates between 2011 and 2014, according to the GeneralRecords on Employed and Unemployed Individuals (CAGED) and to the unemployment insuranceadministrative database. Approximately 7.9% of the information could not be matched with thesedatabases. For these workers, we used the original information on job tenure from Rais, and we im-puted the information on the hiring date based on the mode of hiring dates for each month/year. Weused job tenure to estimate the termination date. In Appendix 2, we compare the graphs with andwithout imputation, and we show that this procedure did not bias the estimates, as the informationon job tenure was intact.

We collected information from Rais only on workers terminated from their jobs. We excluded data onvoluntary terminations, on statutory public servants, on military personnel, on individuals youngerthan 18 years, on individuals who did not earn a wage, and on workers with more than one em-ployment relationship on the termination date. Even with these exclusion criteria for non-eligibleworkers, the database contains 59 million terminations between 2011 and 2016. To avoid losinginformation and to be impartial, we decided to maintain all observations, analyzing the Brazilianpopulation rather than a sample. This sizable amount of information renders virtually all the esti-mates statistically significant. So, when analyzing the RKD tables, we verified whether the estimatewas close to zero, low or high, even though all of them were statistically significant.

Time analysis is the most important variable in a survival model. It represents the time before“failure.” In this paper, we developed a survival model for employment and used the job tenurevariable for time analysis. This variable assesses time up to unemployment. As we are interested inassessing the effects of unemployment insurance on the probability of termination, job tenure takesinto account the time necessary to become eligible for the unemployment insurance and the extensionof the potential duration of insurance payments.

According to the Brazilian unemployment insurance system, this information should consider theduration of employment relationships in the past 36 months, when the previous notice was issued,and a remaining balance of 15 working days (rounded up to 1 month). The Brazilian governmentperforms this roundup when the worker files for the unemployment insurance.

15

Using the Rais data on job tenure for the survival model does not represent the time analysis ofunemployment insurance. This Rais variable has been used in the literature17. There are two ob-stacles to the use of this variable as time analysis. First, the previous notice: which consists of theperiod provided by the employer allowing the worker to look for another job and get prepared forunemployment. This period varies according to job tenure. The firm decides whether it should awaitthe established previous notice period for termination or terminate the employment contract andpays the one-month wage upfront. In the case of termination before the end of the previous noticeperiod, this information is computed by the government as days worked, but the Rais database doesnot include these days. Second, the Rais database does not take into account the past 36 months.Finally, the roundup of 15 days is not contemplated by the Rais database.

On the other hand, job tenure information obtained from the unemployment insurance administra-tive database allows knowing about the length of time the government will take to pay the insurance.Therefore, we used the unemployment insurance administrative database and decided not to use Raisdata on job tenure. Appendix 2 compares these data.

Job tenure information provided by the unemployment insurance database is discrete, owing to theroundups at the time when the worker filed for the insurance. Thus, we chose a procedure to turnthe discrete variable into a continuous one. As this information is not available to those workers whodid not file for the insurance, we used a different procedure for recipients and non-recipients. Forrecipients, we used the job tenure indicated in the unemployment insurance administrative databaseand turned the discrete variable into a continuous variable by computing the time elapsed betweenthe hiring and termination dates. For non-recipients, we calculated the job tenure for the past 36months and performed the same transformation of the discrete variable into a continuous variableby computing the time elapsed between the hiring and termination dates. After this procedure,we added 15 days to the job tenure for all workers in order to represent the roundup made by thegovernment. Appendix 2 shows this procedure in detail.

5.2 Survival Model for Employment

To estimate the probability of termination based on job tenure, we used the nonparametric Kaplan-Meier survival model for employment. Hence, termination is the failure event (T ). The likelihood ofsurvival up to t represents the probability of not being terminated from the job during tenure, i.e.,

S(t) = Pr(T > t) (26)

The hazard function for being fired is the level of intensity of terminations, taking into account themonths of job tenure. Formally, it measures the probability of termination within a short timeframe,considering that it has not occurred yet, divided by this timeframe, i.e.,

h(t) = lim∆t→0

Pr (t+ ∆t > T > t|T > t)

∆t(27)

In our database, the failure event can take place only once. As the probability of survival measuresthe probability of keeping the job up to t; then, the complement of the probability of survival is theprobability of termination. In other words,

Pr(Termination) = 1− S(t) = 1− Pr(T > t) = Pr(T ≤ t) = F (T ) (28)

17Gerard and Gonzaga (2013), Gerard et al. (2020), Gonzaga and Pinto (2014), and Carvalho et al. (2018).

16

To estimate the probability of termination, we used the accumulated termination function calculatedby the Kaplan-Meier model. This is the straightforward explanation as to why RKD was used. As weidentified a priori density discontinuities, it is natural to expect the accumulated probability functionto have a slope change after the cutoff. As the accumulated function represents the probability oftermination, in practice, the slop change is a good estimate of ex-ante moral hazard.

5.3 Regression Kink Design (RKD)

As with the studies by Card et al. (2007) and Le Barbanchon (2016), this paper uses job tenure asforcing variable for assessment around the cutoffs. With RKD, we estimated slope changes betweenjob tenure (j) and the probability of termination after modification in unemployment insurance rules.For most workers assessed in this study, unemployment insurance increases the probability of treat-ment from the cutoff for 6 months’ tenure. After 12 and 24 months of job tenure, the potentialduration of unemployment insurance increases by 1 month. Thus, we explored the slope change ofthe probability of termination after cutoffs for 6, 12, and 24 months of job tenure.

This paper assesses one regression for each cutoff around the bandwidth. The regression estimationof each cutoff (c) seeks to estimate parameter γ, as described next.

P (Terminationi) = α0 + γTi (ji − c) + α1md (ji − c) + α2Tim

d (ji − c) + ui (29)

Where md(.) is a polynomial function of degree d; Ti is a dummy variable that indicates the rightside of the cutoff; Tim

d (wi − c) does not have degree 1.

The hypothesis of this method is the slope continuity, which assumes that individuals treated afterthe cutoff and those not treated before the cutoff would have the same slope in the absence of treat-ment. Therefore, any discontinuity in the slope would stem from treatment. Put formally:

γ = E

[dP (Termination)1

dj− dP (Termination)0

dj|T = 1, j ∈ h

]=

= limj→c+

E

(dP (Termination)

dj|T = 1, j ∈ h+

)− lim

j→c−E

(dP (Termination)

dj|T = 0, j ∈ h−

)(30)

Since P (Termination) = F (T ) = Pr(T ≤ t), then its derivative is the probability density function.In other words, the hypothesis of continuity of the derivative is construed as hypothesis of continuityof the density function in the absence of treatment. This hypothesis is identical to the one describedin the literature on bunching (Saez, 2010), notches (Kleven and Waseem, 2013), and density discon-tinuity approach (Doyle Jr, 2007; Jales, 2018). However, in the present study, we do not have toassume the hypotheses of no spillover effects and of the theoretical model of density.

We used RKD for all individuals and for the groups that filed for unemployment insurance for the1st and 2nd time and for the third or more times, sorting them into before 2014 and after March2015. Finally, as suggested by Card et al. (2015), we performed some placebo tests at no-kink pointsto check whether there are slope changes along other job tenure points. Appendix 3 presents moretests based on RKD, whose estimates vary according to the degrees of polynomials, bandwidth, andkernel, and graphs showing workers’ and firms’ observable characteristics.

17

5.4 Differences-in-Differences (DD)

As analyzed by the theoretical model, the ex-ante moral hazard of unemployment insurance interactswith business cycles. The effect estimated herein by RKD can be influenced by the effect of changeson business cycles.

This study applies DD to estimate the effect of moral hazard generated by the unemployment insur-ance rule independently from the effect brought about the change in business cycles. The dependentvariable is the estimated increase in the probability predicted by RKD at the cutoff for six months’job tenure. All RKD estimates and the respective survival model were estimated at the local levelannually, analyzing the effect of an increase in the probability of termination on the treatment andcontrol groups. The number of terminations in that municipality and year was used as a weightingfactor.

The treatment group consists of workers who filed for unemployment insurance for three or moretimes, as their right to receive the insurance was kept after the amendment to the law in 2015. Thecontrol group consists of workers who filed for unemployment insurance for the 1st or 2nd time.

Hence, DD calculates the differences before and after treatment (when t = 0 and t = 1, respectively)for each group, control and treatment. Later, the effect of treatment on the treated groups wasestimated by the differences in these two differences, as specified next:

DD = {E [Yit|Tit = 1, tit = 1]− E [Yit|Tit = 1, tit = 0]}− {E [Yit|Tit = 0, tit = 1]− E [Yit|Tit = 0, tit = 0]}

(31)

The effect of ex-ante moral hazard of unemployment insurance was estimated by measuring param-eter βDD in the following model:

Yit = α + ρTi + θtt + βDDTitt + εit (32)

Where Ti is the dummy for time-invariant treatment, tt is the time dummy for the period followingtreatment, and α, ρ, βDD, and θ are parameters estimated by the regression model. The parameterof interest is βDD.

With this econometric model, we can calculate the following conditional expectations:

E [Yit|Tit = 1, tit = 1, Xit ] = α + γXit + ρTit + θtit + βDDTittit + E [εit|Tit = 1, tit = 1, Xit] (33)

E [Yit|Tit = 1, tit = 0, Xit ] = α + γXit + ρTit + E [εit|Tit = 1, tit = 0, Xit] (34)

E [Yit|Tit = 0, tit = 1, Xit ] = α + γXit + θtit + E [εit|Tit = 0, tit = 1, Xit] (35)

E [Yit|Tit = 0, tit = 0, Xit ] = α + γXit + E [εit|Tit = 0, tit = 0, Xit] (36)

Hence,

moral hazard = βDD + E [εit|Tit = 1, tit = 1, Xit]− E [εit|Tit = 1, tit = 0, Xit]

− E [εit|Tit = 0, tit = 1, Xit] + E [εit|Tit = 0, tit = 0, Xit](37)

Under the hypothesis that E [εit|Tit, tit, Xit] = 0,

18

moral hazard = βDD (38)

This method allows assuming the existence of previous difference between the control and treatmentgroups. It permits the control group to be different from the treatment group provided that thesedifferent characteristics are constant over time. The only premise of DD, as pointed out by Foguel(2012, p. 75), is “that the time variation in the counterfactual average of the treated group should beequal to the variation observed in the control group average”, i.e.,

E[Y 0it |Tit = 1, tit = 1

]− E

[Y 0it |Tit = 1, tit = 0

]=

E[Y 0it |Tit = 0, tit = 1

]− E

[Y 0it |Tit = 0, tit = 0

] (39)

This hypothesis replaces the hypothesis of no density discontinuity in RKD. It allows individuals tobe different, provided this difference is time-invariant. To test this hypothesis, this study verifies theparallel trend assumption for both groups before and after treatment.

6 RKD RESULTS

Until 2014, all workers could be eligible with six months of job tenure, and they increased the po-tential duration of unemployment insurance by one month with 12 and 24 months of job tenure.Graph 5 shows this increase in the potential duration of unemployment insurance according to thecutoffs. As shown in Graph 3, there are density discontinuities and peak hazard rates after thesecutoffs, especially for the six months’ job tenure. Consequently, Graph 5 shows that the probabilityof termination increases the slope sharply after this cutoff. The slope change is not noticeable in thegraphic analysis after the cutoffs for 12 and 24 months of job tenure.

The graphic analysis reveals that eligibility for unemployment insurance seems to influence the in-crease in the probability of employment termination. As the termination decision is made by thefirm, this decision can be influenced by a change in the worker’s behavior, i.e., ex-ante moral hazardof unemployment insurance. However, the graphic analysis does not allow estimating the magnitudeof this increase, whether it is statistically different from zero and whether there are increases inprobability at the cutoffs for 12 and 24 months of tenure.

Table 2 displays these estimates and the inference analysis of RKD for all workers. After eligibilityfor unemployment insurance, the probability of termination increased by 0.755 p.p. per month, whenits average was 4%. In other words, a 19% monthly increase in the probability of termination. Therewere more modest increases in the probability of termination after the increases in the potentialduration of unemployment insurance by one month at the cutoffs for 12 and 24 months of job tenure.The probability must have increased by 0.2 p.p. and 0.104 p.p. at these cutoffs. Given that the aver-ages were 17.8% and 48.4%, the probability of termination increased by 1.1% and 0.21%, respectively.

19

Graph 5: Potential duration of unemployment insurance and the probability of termination ofemployment according to tenure, 2011-2016.

050

100

UI p

oten

tial d

urat

ion

0 3 6 9 12 15 18 21 24 27 30job tenure

UI potential duration

0.00

0.25

0.50

0.75

1.00

prob

abilit

y of

em

ploy

men

t ter

min

atio

n

0 3 6 9 12 15 18 21 24 27 30job tenure

P(employment termination)


Table 2: RKD estimates of the increase in probability of termination at the cutoffs for 6, 12, and 24months of job tenure for all workers from 2011 to 2016.

(1) (2) (3) (4) (5) (6)6 months 12 months 24 months

All workers - 2011-2016Increase in slope 0.00755*** 0.00816*** 0.002*** 0.0019*** 0.00104*** 0.00157***

(0) (0) (0) (0) (0) (0)P(Termination<cutoff) 0.04 0.04 0.178 0.178 0.484 0.484N 3,936,850 5,675,766 5,817,490 8,517,120 6,179,296 9,106,076bandwidth (each side) 2 3 2 3 2 3Kernel uniform triangular uniform triangular uniform triangular


Note: The values in brackets are standard errors. R2 statistics greater than 0.98. The probability of termination in

the cutoff was estimated by the constant. All polynomials of degree 2. Legend: * p<0.05; ** p<0.01; *** p<0.001.

Graph 6 shows the slope changes in the probability of termination by groups of applicants beforeand after the amendments to unemployment insurance rules. Those filing for the insurance for threeor more times were not affected by the change in the rule and still exhibited discontinuous density,as shown in Graph 4. They still have a perceptible slope change in the probability of terminationafter the cutoff for six months of tenure (Graph 6).

On the other hand, first- or second-time applicants lost their eligibility with six months of tenure

20

after March 2015. For instance, for first-time applicants, such eligibility could be obtained with 18months of tenure between March and June 2015, but only with 12 months of tenure after June 2015.Therefore, after the amendment to the law, it is not possible to notice a change in the slope of thecutoff for six months of tenure for these workers.

Graph 6: Probability of termination during job tenure per group of workers for the periods between2011 and 2014 and between March, 2015 and December, 2016.

0.00

0.25

0.50

0.75

1.00

prob

abilit

y of

em

ploy

men

t ter

min

atio

n

0 3 6 9 12 15 18 21 24 27 30job tenure

3rd or more (before 31/dec/2014)

0.00

0.25

0.50

0.75

1.00

prob

abilit

y of

em

ploy

men

t ter

min

atio

n0 3 6 9 12 15 18 21 24 27 30

job tenure

1st ou 2nd (before 31/dec/2014)

0.00

0.25

0.50

0.75

1.00

prob

abilit

y of

em

ploy

men

t ter

min

atio

n

0 3 6 9 12 15 18 21 24 27 30job tenure

3rd or more (after 01/mar/2015)

0.00

0.25

0.50

0.75

1.00

prob

abilit

y of

em

ploy

men

t ter

min

atio

n

0 3 6 9 12 15 18 21 24 27 30job tenure

1st ou 2nd (after 01/mar/2015)


Table 3 shows the estimates and inference analysis of RKD per group of workers before and afterthe amendment to the unemployment insurance law. Since Rais database records all terminationsof employment in Brazil and has 59 million terminations in this period, the estimates will probablybe statistically significant. Therefore, our interpretations are based on the estimate level. We willinterpret values that are very close to zero as having no effect, despite statistically different from zero.

The probability of termination increased significantly at the cutoff for 6 months of tenure for first-or second-time unemployment insurance applicants before the amendment to the law. Immediatelybefore this cutoff, the probability of termination was 6.8%. After 6 months of job tenure, the prob-ability of termination went up by 1.28 p.p. per month of tenure, i.e., a 19% monthly increase.

After the amendment to the law, this group lost eligibility for unemployment insurance with 6months of tenure. Table 3 reveals that the probability of termination increased by 0.19 p.p. permonth after this cutoff. As the probability of termination before the cutoff was 5%, this monthlyincrease in slope accounted for an increment of 3.8%. In other words, after the loss of eligibility, the

21


(1) (2) (3) (4) (5) (6)6 months 12 months 24 monthsFirst- and second-time applicants (before 12/31/2014)

Increase in slope 0.0128*** 0.01353*** 0.00221*** 0.00213*** 0.00104*** 0.00169***(0.00001) (0.00001) (0.00001) (0) (0) (0)

P(Termination≤ cutoff)) 0.068 0.068 0.255 0.255 0.535 0.535N 2,095,785 2,995,046 2,316,892 3,405,465 1,648,402 2,437,576

First- and second-time applicants (after 03/01/2015)Increase in slope 0.00195*** 0.00103*** 0.00232*** 0.00224*** -0.00049*** 0.00021***

(0.00001) (0) (0.00001) (0) (0) (0)P(Termination≤ cutoff) 0.05 0.05 0.175 0.175 0.47 0.47N 692,641 979,328 989,785 1,448,179 920,771 1,362,370

Applicant for the 3rd time or more (before 12/31/2014)Increase in slope 0.00708*** 0.00822*** 0.00181*** 0.00171*** 0.00249*** 0.00283***

(0.00001) (0) (0) (0) (0) (0)P(Termination ≤ cutoff) 0.017 0.017 0.123 0.123 0.446 0.446N 777,453 1,133,236 1,503,837 2,207,244 2,464,921 3,622,809

Applicant for the 3rd time or more (after 03/01/2015)Increase in slope 0.00385*** 0.0049*** 0.00184*** 0.00156*** 0.00068*** -0.00003***

(0.00001) (0) (0.00001) (0) (0.00001) (0)P(Termination ≤ cutoff) 0.011 0.011 0.13 0.13 0.123 0.123N 261,958 410,653 826,360 1,195,819 1,223,196 333,742bandwidth (each side) 2 3 2 3 2 3Kernel uniform triangular uniform triangular uniform triangular



the cutoff was estimated by the constant. All polynomials of degree 2. Legend: * p<0.05; ** p<0.01; *** p<0.001.

increase in the probability of termination decreased by 80%. Density discontinuity did not disap-pear altogether, but it decreased significantly after the loss of eligibility for unemployment insurance.

This reduction did not occur in the group of workers who filed for the insurance for the third time ormore times. The rules for payment of the insurance remained unchanged for this group after 2015.Before 2014, the increase in probability was 0.7 p.p. per month, when the probability of terminationwas 1.7%, i.e., a 41% increase. After March 2015, the probability went up 0.385 p.p. per month,when the probability of termination was 1.1%, i.e., a 35% increase. Hence, these relative increaseswere close to each other before and after the amendment to the law.

However, the increase in the probability of termination in percentage points decreased for thoseworkers who filed for unemployment insurance for the third time or more times. This reduction wasfollowed by a drop in probability before the cutoff for 6 months. In this paper, we associated thislower probability of termination with changes in the business cycles between the periods. Before2014, the Brazilian economy had been experiencing economic growth. From 2014 to 2016, Brazilwent through a period of recession. Counterintuitively, the probability of termination decreased withlonger tenure. We believe workers work harder during an economic crisis, and the effect of ex-antemoral hazard in percentage points decreases due to the interaction with business cycles.

Table 3 shows that the increase in the probability of termination at the cutoff for 12 months wasdifferent between the groups. For first- and second-time applicants for unemployment insurancebefore 2014, there was a 0.22 p.p. increase when the probability of termination was 25.5%, i.e., a0.86% increase. After March 2015, there was a 0.23 p.p. increase when the probability was 17.5%,i.e., a 1.3% increase. Therefore, the monthly increase in the probability of termination was closein percentage points before and after the amendment to the law. However, there was a significant

22

increase in percentage points. While the effect was much weaker than that observed at the cutofffor 6 months, the relative increase in the probability of termination indicates the effect of changes inunemployment insurance rules, which modified the eligibility for first-time applicants after June 2015to 12 months of tenure. Therefore, this evidences ex-ante moral hazard of unemployment insurance.

For those workers who applied for unemployment insurance for the third time or more at the cutofffor 12 months, the effect was close in percentage points and relative terms. Before 2014, there wasa 0.18 p.p. increase when the probability of termination was 12.3%, i.e., a 1.46% increase. AfterMarch 2015, there was a 0.18 p.p. increase when the probability was 13%, i.e., a 1.38% increase.

Regarding the cutoff for 24 months of tenure, the effect was negligible before 2014 for both groups.For first- and second-time applicants for unemployment insurance, the effect in relative terms hada 0.104% increase in the probability of termination. For those filing for the insurance for the thirdtime or more times, there was a 0.25% increase. After March 2015, while both groups continued toexhibit an increase by one month in the potential duration of unemployment insurance at this cutoff,the effect was virtually zero and slightly negative for some estimates. We also believe the changesin business cycles explain the reduction in the moral hazard of unemployment insurance at this cutoff.

Finally, Table 4 shows the estimates with placebo tests for the cutoffs. Despite statistically signif-icant estimates, they are much lower than those in Tables 2 and 3 because of the large database.Our interpretation is that these values are very close to zero, comparing these estimates with thosedisplayed in Tables 2 and 3, and observing that the sign changes according to the degree of the poly-nomial, bandwidth, and kernel. Thus, the estimates are statistically significant, but very small, theychange signs and are close to zero. So, we did not observe ex-ante moral hazard at the placebo cutoffs.


(1) (2) (3) (4) (5) (6)3 monthsincrease in slope 0.00178*** 0.00203*** 0.00286*** -0.00149*** -0.00045*** -0.00097***

(0) (0) (0) (0) (0) (0)9 monthsincrease in slope 0.00023*** -0.0004*** 0.00013*** -0.00019*** -0.00004*** -0.00054***

(0) (0) (0) (0) (0) (0)15 monthsincrease in slope 0.00052*** -0.00087*** 0.00067*** -0.00036*** -0.00053*** 0.00064***

(0) (0) (0) (0) (0) (0)20 monthsincrease in slope -0.00007*** -0.00135*** -0.00026*** -0.00087*** -0.0007*** -0.00035***

(0) (0) (0) (0) (0) (0)28 monthsincrease in slope 0.00026*** -0.00068*** 0.00034*** -0.00069*** -0.00053*** 0.00013***

(0) (0) (0) (0) (0) (0)30 monthsincrease in slope -0.00031*** -0.00118*** -0.00109*** -0.00028*** -0.00012*** 0.00042***

(0) (0) (0) (0) (0) (0)degree of polynomial 1 1 1 2 2 2bandwidth (each side) 1 1 2 2 2 3Kernel uniform triangular triangular uniform triangular triangular


Note: The values in brackets are standard errors. R2 statistics greater than 0.96. Legend: * p<0.05; ** p<0.01; ***

p<0.001.

Appendix 3 presents other tests from the RKD model, in which estimates vary in terms of degreesof polynomials, bandwidth, and kernel function, and graphs of worker’s and firm’s observable char-

23

acteristics.

7 DD RESULTS

Table 5 presents the differences-in-differences estimates for the RKD model, at the cutoff for sixmonths of job tenure. RKD estimates were calculated at the local level, yearly, and according to thetreatment and control groups. The DD method estimates the difference in slope between the controland treatment groups before and after the amendment to the unemployment insurance law in 2015.The number of terminations of employment recorded in municipalities in each group and year wasused as weighting factors. None of the RKD estimates at the local level considered terminationsbetween January and February 2015.

Model (1) calculates the average moral hazard before the amendment to the law, considering allterminations between 2011 and 2014. For after the amendment, this model considers the termina-tions between 2015 and 2016. Model (2) considers the years 2014 and 2015 to represent the averagemoral hazard before and after the amendment. This model uses a shorter time interval, with betterrepresentation of DD estimates, as it is less prone to macroeconomic changes with possible differenteffects between the treatment and control groups. Hence, our main estimate of ex-ante moral hazardof unemployment insurance increased by 2.6 p.p. the probability of termination when the workerwas eligible. Given that the probability of termination was 4% at the cutoff for six months (Table2), then the moral hazard increases the probability of termination by 65%.

Finally, model (3) tests the hypothesis of parallel trends before and after the amendment to the law.The coefficients of interactions between the year and treatment dummies estimate the differences intrends between the control and treatment groups18. Only the trend for 2012 was significantly lowerthan 1%. All the remaining differences in the trends before 2015 and in 2016 were nonsignificantat 5%. Only in 2015, year in which the law was amended, was there a remarkable and significantincrease in the probability of termination. This test provides strong evidence that the hypothesis ofparallel trends in the DD method is supported.

To estimate the increase of moral hazard before and after the amendment to the law using model(3), as in model (2), just subtract19 βI.T reated X I.2015 − βI.T reated X I.2014. The result is identical withthat of model (2): the probability of termination of employment for workers who maintained theireligibility for unemployment insurance at the cutoff for six months of tenure increased by 2.6 p.p.compared to those workers who lost eligibility with six months of tenure.

18

βI.Treated X I.2012 = E[Y |Treated, 2012]− E[Y |Treated, 2011]− E[Y |Control, 2012]− E[Y |Control, 2011]





19

βI.Treated X I.2015 − βI.Treated X I.2014 =

E[Y |Treated, 2015]− E[Y |Treated, 2014]− E[Y |Control, 2015]− E[Y |Control, 2014]

24

Table 5: DD in RKD model at the cutoff for 6 months of job tenure, estimated at the local level,annually, and according to the treatment and control groups.Variables (1) (2) (3)

DD between 2011and 2016

DD between 2014and 2015

DD with annualtrends

B DD 0.015*** 0.026***(0.0027) (0.0054)

I.Treated -0.004*** -0.003 -0.008***(0.0012) (0.0039) (0.0015)

I.After 2015 -0.026*** -0.027***(0.0014) (0.0025)

I.2012 -0.016***(0.0019)

I.2013 -0.009***(0.0017)

I.2014 -0.013***(0.0018)

I.2015 -0.04***(0.002)

I.2016 -0.022***(0.0021)

I.Treated X I.2012 0.011**(0.0037)

I. Treated X I.2013 0.0025(0.0035)

I. Treated X I.2014 0.006(0.0038)

I. Treated X I.2015 0.032***(0.0037)

I. Treated X I.2016 0.0055(0.0039)

Constant 0.017*** 0.0098*** 0.023***(0.0006) (0.0016) (0.0009)

N 4,841,588 1,118,351 4,841,588R2 0.0001 0.0001 0.0001


Note: The values in brackets are standard errors. Legend: * p<0.05; ** p<0.01; *** p<0.001.

8 CONCLUSION

This study gathered compelling evidence of ex-ante moral hazard of unemployment insurance inBrazil. It indicated a large increase in the probability of termination of employment after the eligi-bility of workers for the unemployment insurance. The first piece of evidence is density discontinuityafter the eligibility cutoffs and an increase in the potential duration of unemployment insurance. Thesecond piece of evidence arises after 2015, when the unemployment insurance rule changed the eligi-bility at six months of job tenure (eligibility cutoff) for one group of workers. Density discontinuity

25

after this cutoff virtually disappeared for this group, but it was kept for other workers. The thirdpiece of evidence concerns the DD method. It is observed in the parallel pattern of probabilities oftermination in the control and treatment groups up to 2014 and in 2016. However, in 2015, therewas a sharp reduction in the probability of termination of employment for workers with six monthsof tenure who lost their eligibility for the unemployment after the amendment to the law.

On the other hand, the study has some limitations regarding workers’ actions, which lead to an in-crease in the probability of termination. We modeled the moral hazard by the reduction in diligenceafter the worker became eligible or had a larger potential duration of unemployment insurance. How-ever, this moral hazard may occur as a peaceful settlement between employer and employee or evenas a fraudulent termination of employment, in which the employee keeps on working informally atthe firm (Van Doornik et al., 2018). The evidence of the type of moral hazard can be analyzed by in-vestigating the relationships between workers and firms, but this is not within the scope of our paper.

Finally, we highlight that ex-ante moral hazard needs to be further researched. It would be interestingto assess how this moral hazard occurs in different occupations, income brackets, economic activities,and firm sizes. Also, it would be possible to verify whether increases in the potential replacementrate interfere in ex-ante moral hazard. Therefore, it is a comprehensive topic that directly affectsthe labor market dynamics and should be analyzed in more detail by the scientific literature.

References

Baker, M. and S. A. Rea Jr (1998). Employment spells and unemployment insuranceeligibility requirements. Review of Economics and Statistics 80 (1), 80–94.

Britto, D. (2015). Unemployment insurance and the duration of employment: Evi-dence from a regression kink design.

Card, D., R. Chetty, and A. Weber (2007). Cash-on-hand and competing modelsof intertemporal behavior: New evidence from the labor market. The Quarterlyjournal of economics 122 (4), 1511–1560.

Card, D., A. Johnston, P. Leung, A. Mas, and Z. Pei (2015). The effect of unemploy-ment benefits on the duration of unemployment insurance receipt: New evidencefrom a regression kink design in missouri, 2003-2013. American Economic Re-view 105 (5), 126–30.

Carvalho, C. C., R. Corbi, and R. Narita (2018). Unintended consequences of unem-ployment insurance: Evidence from stricter eligibility criteria in brazil. EconomicsLetters 162, 157–161.

Chetty, R. (2008). Moral hazard versus liquidity and optimal unemployment insur-ance. Journal of political Economy 116 (2), 173–234.

Christofides, L. N. and C. J. McKenna (1995). Unemployment insurance and moralhazard in employment. Economics Letters 49 (2), 205–210.

Doyle Jr, J. J. (2007). Employment effects of a minimum wage: A density disconti-nuity design revisited.

26

Ferejohn, J. (1986). Incumbent performance and electoral control. Publicchoice 50 (1-3), 5–25.

Foguel, M. N. (2012). Diferencas em diferencas. Avaliacao economica de projetossociais 1, 69–83.

Gerard, F. and G. M. Gonzaga (2013). Informal labor and the cost of social pro-grams: Evidence from 15 years of unemployment insurance in brazil. Available atSSRN 2289880 .

Gerard, F., M. Rokkanen, and C. Rothe (2020). Bounds on treatment effects inregression discontinuity designs with a manipulated running variable. QuantitativeEconomics 11 (3), 839–870.

Gonzaga, G. and R. C. Pinto (2014). Rotatividade do trabalho e incentivos dalegislacao trabalhista. Technical report, Texto para discussao.

Green, D. A. and T. C. Sargent (1998). Unemployment insurance and job durations:seasonal and non-seasonal jobs. canadian Journal of Economics , 247–278.

Jales, H. (2018). Estimating the effects of the minimum wage in a developingcountry: A density discontinuity design approach. Journal of Applied Econo-metrics 33 (1), 29–51.

Kleven, H. J. and M. Waseem (2013). Using notches to uncover optimization frictionsand structural elasticities: Theory and evidence from pakistan. The QuarterlyJournal of Economics 128 (2), 669–723.

Le Barbanchon, T. (2016). The effect of the potential duration of unemploymentbenefits on unemployment exits to work and match quality in france. LabourEconomics 42, 16–29.

Le Barbanchon, T., R. Rathelot, and A. Roulet (2019). Unemployment insuranceand reservation wages: Evidence from administrative data. Journal of PublicEconomics 171, 1–17.

McCall, J. J. (1970). Economics of information and job search. The QuarterlyJournal of Economics , 113–126.

Meyer, B. D. (2002). Unemployment and workers’ compensation programmes: ra-tionale, design, labour supply and income support. Fiscal Studies 23 (1), 1–49.

Meyer, B. D. and W. K. Mok (2014). A short review of recent evidence on thedisincentive effects of unemployment insurance and new evidence from new yorkstate. National Tax Journal 67 (1), 219.

Rebollo-Sanz, Y. (2012). Unemployment insurance and job turnover in spain. LabourEconomics 19 (3), 403–426.

27

Saez, E. (2010). Do taxpayers bunch at kink points? American economic Journal:economic policy 2 (3), 180–212.

Schmieder, J. F., T. Von Wachter, and S. Bender (2012). The effects of extendedunemployment insurance over the business cycle: Evidence from regression dis-continuity estimates over 20 years. The Quarterly Journal of Economics 127 (2),701–752.

Schmieder, J. F., T. von Wachter, and S. Bender (2016). The effect of unem-ployment benefits and nonemployment durations on wages. American EconomicReview 106 (3), 739–77.

Shavell, S. and L. Weiss (1979). The optimal payment of unemployment insurancebenefits over time. Journal of political Economy 87 (6), 1347–1362.

Topel, R. H. (1983). On layoffs and unemployment insurance. The American Eco-nomic Review 73 (4), 541–559.

Van Doornik, B., D. Schoenherr, J. Skrastins, et al. (2018). Unemployment insur-ance, strategic unemployment, and firm-worker collusion. Technical report.

28

APPENDIX 1 – THEORETICAL MODEL COMPLEMENT

Proof of Proposition 3:

U elt+1 > Vt+1 =⇒

u(yu + b) − ψu(st+1) + β[pt+1Vt+2 + (1− pt+1)Une

t+2

]>

u(w) − ψe(at+1) + β[(1− πt+1)Vt+2 + πt+1U

elt+2

]Case 1: pt+1 = 1 and πt+1 = 1 (i.e., Y expansion and at+1 = 0)

U elt+1 − Vt+1 = u(yu + b)− ψu(st+1) + βVt+2 − u(w) + ψe(at+1)− βU el

t+2

= u(yu + b)− ψu(st+1)− [u(w)− ψe(at+1)] + βVt+2 − βU elt+2

= u(t) − u(t) + βVt+2 − βU elt+2

= u(t)− βu(t+ 1) − [u(t)− βu(t+ 1)] + β2(U elt+3 − Vt+2

)=

(1− β)

β[u(yu + b) − u(w) + ψe(at+1)− ψu(et+1)]

In a highly informal scenario, consider that the informal income (yu) is close to but lower than w.As the replacement rate is close to 1, we have b ≈ w. Since pt+1 = 1 and πt+1 = 1, consider that Yis high, at+1 = 0 and st+1 = s. So,

U elt+1 − Vt+1 =

(1− β)

β

+︷︸︸︷u(yu + b) − u(w)−ψu(s)

In this scenario, possibly, u(yu + b)− u(w) > ψu(s). Hence,

U elt+1 > Vt+1

Case 2: pt+1 = 1− πt+1 (intermediate scenario)

U elt+1 − Vt+1 = u(yu + b) − ψu(st+1) + β

[πt+1U

net+2 − U el

t+2

]− u(w) + ψe(at+1)

= u(yu + b) − u(w) − βπt+1u(b) + ψe(at+1)− ψu(st+1)

Since b ≈ w, yu ≈ w, 0 ≤ β ≤ 1 and 0 ≤ πt+1 ≤ 1. It is also possible that

u(yu + b) − u(w) − βπt+1u(b) + ψe(at+1)− ψu(st+1) > 0

In this scenario, then,

U elt+1 > Vt+1

Case 3: pt+1 = 0 and πt+1 = 0 (i.e., Y crisis and at+1 = amaxt+1 )

U elt+1 − Vt+1 = u(yu + b) − ψu(st+1) + βUne

t+2 − u(w) + ψe(at+1)− βVt+2

=

+︷︸︸︷u(t)− v(t) +

−︷︸︸︷β[Unet+2 − Vt+2

]In this scenario, unless the individual overvalues the short term, u(t) −v(t) > β

[Unet+2 − Vt+2

]. Hence,

U elt+1 < Vt+1

29

APPENDIX 2 – JOB TENURE: TIME ANALYSIS OF THE

SURVIVAL MODEL

Time analysis is the most important variable in a survival model. It denotes the time until “failure.”In this paper, we developed a survival model for employment. This variable indicates the time un-til unemployment. As we are interested in assessing the effects of unemployment insurance on theprobability of termination, job tenure considers the time necessary to be eligible for the insuranceand the increase in its potential duration.

According to the Brazilian unemployment insurance system, this information should take into ac-count the time worked over the past 36 months, previous notice period, and addition of 15 days tothe job tenure, which corresponds to a roundup to one month. The Brazilian government makes thisroundup at the time of application for the unemployment insurance.

The use of Rais data on job tenure for the survival model does not represent the time analysis ofunemployment insurance. The Rais variable has been used in the literature20. The first hindranceto using this variable for time analysis is related to the previous notice period, which consists of theperiod provided by the employer allowing the worker to look for another job and get prepared forunemployment. This period varies according to job tenure. The firm decides whether it should awaitthe established previous notice period for termination or terminate the contract of employment andpays the one-month wage upfront. In the case of termination before the end of the previous noticeperiod, this information is computed by the government as days worked, but the Rais database doesnot include these days. A second hindrance to the use of Rais data concerns the fact that it doescontemplate the past 36 months. Finally, the roundup of 15 days is not included in the Rais database.

On the other hand, the job tenure obtained from the unemployment insurance administrative databaseallows knowing the timeframe for the unemployment insurance payment. Therefore, this informa-tion seems to be more appropriate for the time analysis of the survival model. However, job tenureinformation from the unemployment insurance database is discrete due to the roundups performedwhen the worker applies for the insurance. Moreover, unemployment insurance non-recipients do nothave any records in the database.

To construct the job tenure variable necessary for the time analysis in the survival model, we used adifferent procedure for when the worker was an unemployment insurance recipient and for when hewas terminated from the job but was not a recipient.

For recipients, we used the number of days worked obtained from the unemployment insurance ad-ministrative database and transformed the discrete variable into a continuous one. To do that, weinitially calculated the days between hiring and termination dates. After that, we summed thesedays with the difference between the discrete time considered by the government and the roundupof days between these dates.

For non-recipients, we calculated the number of working days in the past 36 months and transformedthe discrete variable into a continuous one. To do that, we initially calculated the days betweenhiring and termination dates. After that, we summed these days with the difference between thenumber of days worked in the past 36 months and the roundup of days between these dates.

We then added 15 days to the data obtained in order to represent the roundup performed by the gov-ernment. Graph A2.1 compares these pieces of information. According to Gonzaga and Pinto (2014),discontinuity over 3 months occurs because of the probation period established by law. According

20Gerard and Gonzaga (2013), Gerard et al. (2020), Gonzaga and Pinto (2014), and Carvalho et al. (2018).

30

to these authors, negative discontinuity over 12 months occurs because terminations after this cutoffare overseen by unions and are liable to labor violation penalties. Nonetheless, with the use of timeanalysis data obtained in the present paper, these discontinuities change drastically. They indicatethe effects of unemployment insurance on the behavior of ex-ante moral hazard.

Graph A2.1: Density of terminations of employment during tenure for the variable used as timeanalysis in the survival model and for the tenure in the last job variable in Rais database.

0.0

2.0

4.0

6.0

8re

lativ

e fre

quen

cy

0 3 6 9 12 15 18 21 24 27 30job tenure

time analysis

0.0

2.0

4.0

6.0

8re

lativ

e fre

quen

cy

0 3 6 9 12 15 18 21 24 27 30job tenure

time at last job

Source: Data compiled by the authors based on Rais database.

Another procedure carried out before the analysis of terminations consisted in adding hiring andtermination dates from 2011 to 2014 in the Rais database. This information was obtained from theGeneral Records on Employed and Unemployed Individuals (CAGED) and from the unemploymentinsurance administrative database. Around 7.9% of the information could not be matched with thesedatabases. For these workers, we used the original information on job tenure from Rais, and we im-puted the information on the hiring date based on the mode of hiring dates for each month/year. Weestimated the termination date by adding the hiring date to the job tenure. Graph A2.2 comparesthe histograms of the databases with and without imputation. We believe this procedure did notbias the estimates, as the information on job tenure was intact.

31

Graph A2.2: Termination density during job tenure with and without imputation of hiring andtermination dates.

0.0

1.0

2.0

3.0

4re

lativ

e fre

quen

cy

0 3 6 9 12 15 18 21 24 27 30job tenure

including inputs

0.0

1.0

2.0

3.0

4re

lativ

e fre

quen

cy

0 3 6 9 12 15 18 21 24 27 30job tenure

without inputs


32

APPENDIX 3 – RKD COMPLEMENTARY TABLES AND

TESTS

This Appendix shows the estimates with variations in the degrees of polynomials, in bandwidth,in the kernel function, and graphs of workers’ and firms’ observable characteristics. The tables in-dicate that the estimates at the cutoffs are robust and consistent with variations of RKD parameters.

Table A3.1: RKD estimates of the increase in the probability of termination at the cutoff for 6 monthsof job tenure per group of workers for the periods between 2011 and 2014 and between March 2015and December 2016.RKD at 6 months (1) (2) (3) (4) (5) (6)P(Termination<6m) = 0.041 All workersincrease in slope 0.01075*** 0.01022*** 0.01151*** 0.00862*** 0.00887*** 0.00936***

(0) (0) (0) (0) (0) (0)N 1,626,212 1,544,411 3,245,555 3,322,787 3,245,555 4,787,830P(Termination<6m) = 0.068 1st- or 2nd-time applicants (before 12/31/2014)

increase in slope 0.01491*** 0.01451*** 0.01511*** 0.01282*** 0.01349*** 0.01354***(0.00001) (0.00001) (0) (0.00001) (0.00001) (0.00001)

N 1,019,939 968,849 2,046,995 2,094,513 2,046,995 2,993,156P(termination<6m) = 0.04 1st- or 2nd-time applicants (after 03/01/2015)

increase in slope 0.00376*** 0.00337*** 0.00457*** 0.0023*** 0.00223*** 0.00157***(0.00001) (0.00001) (0) (0.00001) (0.00001) (0)

N 159,459 151,579 306,492 313,482 306,492 448,281P(termination<6m) = 0.017 Applicants for 3rd time or more (before 12/31/2014)

increase in slope 0.00932*** 0.00857*** 0.01042*** 0.00709*** 0.00706*** 0.00823***(0) -0.00001 (0) -0.00001 -0.00001 (0)

N 382,659 363,298 758,039 777,092 758,039 1,132,643P(termination<6m) = 0.011 Applicants for 3rd time or more (after 03/01/2015)

increase in slope 0.00745*** 0.0069*** 0.00956*** 0.00434*** 0.00416*** 0.0052***(0.00001) (0.00001) (0.00001) (0.00001) (0.00001) (0.00001)

N 58,972 55,738 124,180 127,580 124,180 199,167Degree of polynomial 1 1 1 2 2 2bandwidth (each side) 1 1 2 2 2 3Kernel uniform triangular triangular uniform triangular triangular



the cutoff was estimated by the constant. Legend: * p<0.05; ** p<0.01; *** p<0.001.

33

Table A3.2: RKD estimates of the increase in the probability of termination of employment at thecutoff for 12 months of job tenure per group of workers for the periods between 2011 and 2014 andbetween March 2015 and December 2016.RKD at 12 months (1) (2) (3) (4) (5) (6)P(termination<12m) = 0.182 All workersincrease in slope 0.00207*** 0.00136*** 0.00124*** 0.00202*** 0.00245*** 0.00194***

(0) (0) (0) (0) (0) (0)N 2,444,032 2,330,412 4,635,205 4,748,633 4,635,205 6,961,564P(termination<12m) = 0.255 1st- or 2nd-time applicants (before 12/31/2014)

increase in slope 0.00106*** 0.00018*** -0.00072*** 0.00221*** 0.0027*** 0.00213***(0) (0) (0) (0.00001) (0.00001) (0)

N 1,191,721 1,138,002 2,259,927 2,314,096 2,259,927 3,401,534P(termination<12m) = 0.154 1st- or 2nd-time applicants (after 03/01/2015)



increase in slope 0.00274*** 0.00213*** 0.00252*** 0.00182*** 0.00223*** 0.00171***(0) (0) (0) (0) (0) (0)

N 773,820 736,318 1,464,975 1,502,231 1,464,975 2,204,956P(termination<12m) = 0.128 Applicants for 3rd time or more (after 03/01/2015)






34

Table A3.3: RKD estimates of the increase in the probability of termination of employment at thecutoff for 24 months of job tenure per group of workers for the periods between 2011 and 2014 andbetween March 2015 and December 2016.RKD at 24 months (1) (2) (3) (4) (5) (6)P(termination<24m) = 0.485 All workersincrease in slope 0.00185*** 0.00064*** 0.00156*** 0.00129*** 0.00146*** 0.00183***

(0) (0) (0) (0) (0) (0)N 2,552,821 2,428,812 4,964,376 5,088,894 4,964,376 7,499,319P(termination<24m) = 0.536 1st- or 2nd-time applicants (before 12/31/2014)


N 821,624 782,912 1,606,133 1,645,551 1,606,133 2,433,376P(termination<24m) = 0.449 1st- or 2nd-time applicants (after 03/01/2015)

increase in slope -0.00017*** -0.00121*** -0.00054*** -0.00034*** -0.00029*** 0.00039***(0) (0) (0) (0) (0.00001) (0)



N 1,239,717 1,179,727 2,400,696 2,460,902 2,400,696 3,616,869P(termination<24m) = 0.494 Applicants for 3rd time or more (after 03/01/2015)

increase in slope -0.00032*** -0.00095*** -0.0008*** -0.00033*** -0.00011*** 0.00019***(0) (0) (0) (0) (0.00001) (0)





We present the graphs of workers’ and firms’ observable variables. Except for schooling (in years), novariables had discontinuities around the cutoffs. Discontinuity in the schooling variable occurs afterthe cutoff for 6 months. It represents an ex-ante moral hazard behavior correlated with schooling.However, this behavior was not correlated with wages, as the graph of replacement rate does notexhibit discontinuity.

35

Graph A3.1: Analysis of discontinuities on workers’ characteristics.

.55

.6.6

5.7

Mal

e

0 3 6 9 12151821242730job tenure

Male

2830

3234

36Ag

e0 3 6 9 12151821242730

job tenure

Age

.45

.5.5

5.6

Whi

te e

thni

city

0 3 6 9 12151821242730job tenure

White ethnicity.7

.75

.8.8

5.9

Rep

lace

men

t Rat

e

0 3 6 9 12151821242730job tenure

Replacement Rate

9.8

1010

.210

.410

.610

.8Ye

ars

of s

choo

ling

0 3 6 9 12151821242730job tenure

Years of schooling

020

4060

8010

0Ye

ars

of e

xper

ienc

e0 3 6 9 12151821242730

job tenure

Experience


36

Graph A3.2: Analysis of discontinuities on firms’ characteristics.

.51

1.5

2Jo

b tu

rnov

er

0 3 6 9 12 15 18 21 24 27 30job tenure

Job turnover

.1.2

.3.4

.5Jo

b de

stru

ctio

n

0 3 6 9 12 15 18 21 24 27 30job tenure

Job destruction40

045

050

055

060

065

0N

umbe

r of e

mpl

oyee

s by

firm

0 3 6 9 12 15 18 21 24 27 30job tenure

Number of employees


37

The Economics and Politics (CNPq) Research Group started publishing its members’ working papers on June 12, 2013. Please check the list below and click at

http://econpolrg.com/working-papers/ to access all publications.

Number Date Publication

102/2020 09-10-2020 Ex-ante Moral Hazard of Unemployment Insurance, Artur Henrique da Silva Santos, Maurício Soares Bugarin and Paulo Roberto Amorim Loureiro

101/2020 01-30-2020 Motivação Partidária nas Transferências Voluntárias da União: O papel do Legislativo Federal, Fernanda L. Marciniuk, Maurício S. Bugarin and Débora C. Ferreira

100/2020 01-28-2020 The Paradox of Concessions in Developing Countries: A game-theoretic investigation of airport concessions in Brazil, Mauricio Bugarin and Frederico Ribeiro

99/2019 06-05-2019 Crime and political effects of a concealed carry ban in Brazil, Rodrigo Schneider

98/2019 05-08-2019 Polo Industrial de Manaus e Desempenho Socioeconômico: Uma Análise de Componentes Principais no Período de 2002 – 2015, Rogério da Cruz Gonçalves and Philipp Ehrl

97/2019 05-01-2019 Efeitos Econômicos da Zona Franca de Manaus, Rogério da Cruz Gonçalves and Philipp Ehrl

96/2019 04-25-2019 Closed Season e Seguro-Defeso:Análise de Incentivos e Avaliação de Impacto, Artur Henrique da Silva Santos, Maurício Soares Bugarin and Paulo Roberto Amorim Loureiro

95/2019 04-18-2019 Rules versus Standards in Developing Countries The Case for Clear and Precise Legal Norms on Eminent Domain Power, Hans-Bernd Schäfer

94/2019 02-20-2019 A influência da reeleição nas políticas fiscais subnacionais, Fernanda L. Marciniuk and Mauricio S. Bugarin

93/2018 12-10-2018 Electronic voting and Public Spending: The impact of enfranchisement on federal budget amendments in Brazil, Rodrigo Schneider, Diloá Athias and Maurício Bugarin

92/2018 12-04-2018 Aplicação da Lei de Benford na Detecção de Jogo de Planilha: O Caso do Estádio Nacional de Brasília, Carlos Alberto Cascão Júnior, Adriana Cuoco Portugal, Mauricio Soares Bugarin and Flávia Ceccato Cunha

91/2018 11-21-2018 Where’d You Get that Idea? Determinants of Creativity and Impact in Popular Music, Bernardo Mueller

90/2018 11-12-2018 A Previdência Complementar é, de fato, Complementar?, Marcos Nihari and Vander Lucas

89/2018 11-05-2018 Sobre Falácias de Moro, Nelson Gonçalves Gomes

88/2018 10-29-2018 Eficiência financeira das concessões de aeroportos no Brasil no período de 2015 a 2016, João Paulo de Castro Carisio Ribeiro, Marcelo Driemeyer Wilbert and Samuel de Resende Salgado

87/2018 10-19-2018 Leilões para conversão de dívida em investimento: Uma proposta da teoria de leilões para implantação dos Termos de Ajuste de Conduta (TAC) da ANATEL, Hélio Maurício Miranda da Fonseca and Maurício Soares Bugarin

86/2018 10-08-2018 Redistribution in Contemporary Japan: A Political Economy Analysis of Ide, Furuichi and Miyazaki (2016)’s “All for All” Fiscal Reform Proposal, Maurício Bugarin

85/2018 08-08-2018 Gestão Fiscal e Despesa de Pessoal: Análise dos Gastos Públicos Estaduais entre 2008 e 2016, Marcos Lima Bandeira, Paulo Augusto P. de Britto and André Luiz Marques Serrano

84/2018 07-11-2018 Atuação do poder legislativo estadual: análise dos incentivos dos deputados estaduais na atividade legislativa, Débora Costa Ferreira, Fernando B. Meneguin and Maurício Soares Bugarin

83/2018 07-04-2018 Eficiência, atratividade e dissuasão de acordos de leniência e combate à corrupção: o caso da Operação Lava-Jato, Rafael M. Gomes, Roberto H. de Aragão Neto, Maurício S. Bugarin and André L. Gama

82/2018 06-28-2018 Ciclos Políticos e Eleitorais na Execução do Programa Bolsa Família em Nível Municipal, Tiago Sousa Pereira, Moisés A. Resende Filho and Vander Mendes Lucas

81/2018 06-20-2018 Análise do IBS como Proposta de Reforma Tributária. Um Estudo Aplicado de Equilíbrio Geral Computável, Nathanael Pereira Costa and Vander Mendes Lucas

80/2017 09-27-2017 Impacto do Programa Ensino Médio Inovador em indicadores educacionais, Luís Felipe Batista de Oliveira and Rafael Terra


79/2017 09-14-2017 Uma Avaliação da Política de Privatização de Aeroportos Brasileira: Uma Abordagem por Controle Sintético, Caio Cordeiro de Resende

78/2017 08-30-2017 Tecnologia e Desempenho Escolar: Uma Avaliação de Impacto do Programa Um Computador por Aluno, Caio Cordeiro de Resende and Ana Carolina Zoghbi

77/2017 08-23-2017 Desregulamentação da franquia de bagagem despachada no Brasil: Uma Análise de Impacto Regulatório, Débora Alves Pereira Bastos, Rodrigo César Bessoni e Silva and Vander Mendes Lucas

76/2017 08-16-2017 Análise de Eficiência das IFES no Uso de Recursos Financeiros 2010-2015, Antonio Marcos Correia Melonio and Vander Mendes Lucas

75/2017 08-09-2017 Royalties do petróleo, reeleição e ciclos políticos: um estudo econométrico para as eleições, Eduardo Dornelas Munhoz e Moisés de Andrade Resende Filho

74/2017 08-02-2017 Responsabilidade fiscal, a atuação do Poder Judiciário e o comportamento estratégico dos governantes, Débora Costa Ferreira, Fernando B. Meneguin and Maurício Soares Bugarin

73/2017 07-26-2017 Sustentabilidade da Dívida dos Estados Brasileiros: 2005-2015, Hiromi Cristina Santos Doi and Vander Mendes Lucas

72/2017 07-19-2017 Hotelling’s product differentiation: an infinite-dimensional linear programming approach, Rodrigo Peñaloza

71/2017 06-16-2017 Escola japonesa ou escola brasileira? A inserção dos estudantes brasileiros na escola no Japão, Maurício Soares Bugarin

70/2017 04-17-2017 How the East was Lost: Coevolution of Institutions and Culture in the 16th Century Portuguese Empire, Bernardo Mueller

69/2017 04-10-2017 A reforma política sob o ponto de vista da análise econômica do direito, Pedro Fernando Nery and Fernando B. Meneguin

68/2016 07-14-2016 O dilema entre a eficiência de curto e de longo prazo no ordenamento jurídico e o impacto no crescimento econômico, Fernando B. Meneguin and Tomás T. S. Bugarin

67/2016 05-04-2016 A estrutura de capital de empresas brasileiras de capital aberto: uma análise de seus determinantes, João Pedro Bertani Catrib, Paulo Augusto P. de Britto and André Luiz Marques Serrano

66/2016 04-20-2016 Tests for history dependence in mixed-Poisson growth: Brazil, 1822-2000, and USA, 1869-1996, with an estimate of the world mixing distribution at start-up, Steve De Castro and Flávio Gonçalves

65/2016 04-13-2016 Piketty’s Prediction meets technical progress in Harrod-Domar’s Dynamics and Solow Swan’s Surrogate, Steve De Castro

64/2016 04-06-2016 Análise do impacto da alteração normativa na aposentadoria por invalidez no Brasil, Helvio Antonio Pereira Marinho, Moises de Andrade Resende Filho and Vander Mendes Lucas

63/2016 03-30-2016 Black movement: Estimating the effects of affirmative action in college admissions on education and labor market outcomes, Andrew Francis-Tan and Maria Tannuri-Pianto

62/2016 01-13-2016 Electronic voting and Social Spending: The impact of enfranchisement on municipal public spending in Brazil, Rodrigo Schneider, Diloá Athias and Maurício Bugarin

61/2015 12-02-2015 Alunos de inclusão prejudicam seus colegas? Uma avaliação com dados em painel de alunos da rede municipal de São Paulo, Bruna Guidetti, Ana Carolina Zoghbi and Rafael Terra

60/2015 12-02-2015 Impacto de programa Mais Educação em indicadores educacionais, Luís Felipe Batista de Oliveira and Rafael Terra

59/2015 10-21-2015 Eficiência de custos operacionais das companhias de distribuição de energia elétrica (CDEEs) no Brasil: Uma aplicação (DEA & TOBIT) em dois estágios, Daniel de Pina Fernandes and Moisés de Andrade Resende Filho

58/2015 10-14-2015 Determinantes do risco de crédito rural no Brasil: uma crítica às renegociações da dívida rural, Lucas Braga de Melo and Moisés de Andrade Resende Filho

57/2015 10-07-2015 Distribuição da riqueza no Brasil: Limitações a uma estimativa precisa a partir dos dados tabulados do IRPF disponíveis, Marcelo Medeiros

56/2015 10-01-2015 A composição da desigualdade no Brasil. Conciliando o Censo 2010 e os dados do Imposto de Renda, Marcelo Medeiros, Juliana de Castro Galvão and Luísa Nazareno

55/2015 09-24-2015 A estabilidade da desigualdade no Brasil entre 2006 e 2012: resultados adicionais, Marcelo Medeiros and Pedro H. G. F. Souza

http://econpolrg.files.wordpress.com/2017/08/eprg-wp-2017-76.pdf

http://scholar.google.com/citations?user=P6YXURsAAAAJ



http://scholar.google.com/citations?user=OO5-iGcAAAAJ


54/2015 09-24-2015 Reciclagem de plataformas de petróleo: ônus ou bônus?, Roberto N. P. di Cillo

53/2015 09-09-2015 A Progressividade do Imposto de Renda Pessoa Física no Brasil, Fábio Castro and Mauricio S. Bugarin

52/2015 07-03-2015 Measuring Parliaments: Construction of Indicators of Legislative Oversight, Bento Rodrigo Pereira Monteiro and Denílson Banderia Coêlho

51/2015 06-29-2015 A didactic note on the use of Benford’s Law in public works auditing, with an application to the construction of Brazilian Amazon Arena 2014 World Cup soccer stadium, Mauricio S. Bugarin and Flavia Ceccato Rodrigues da Cunha

50/2015 04-29-2015 Accountability and yardstick competition in the public provision of education, Rafael Terra and Enlinson Mattos

49/2015 04-15-2015 Understanding Robert Lucas (1967-1981), Alexandre F. S. Andrada

48/2015 04-08-2015 Common Labor Market, Attachment and Spillovers in a Large Federation, Emilson Caputo Delfino Silva and Vander Mendes Lucas

47/2015 03-27-2015 Tópicos da Reforma Política sob a Perspectiva da Análise Econômica do Direito, Pedro Fernando Nery and Fernando B. Meneguin

46/2014 12-17-2014 The Effects of Wage and Unemployment on Crime Incentives - An Empirical Analysis of Total, Property and Violent Crimes, Paulo Augusto P. de Britto and Tatiana Alessio de Britto

45/2014 12-10-2014 Políticas Públicas de Saúde Influenciam o Eleitor?, Hellen Chrytine Zanetti Matarazzo

44/2014 12-04-2014 Regulação Ótima e a Atuação do Judiciário: Uma Aplicação de Teoria dos Jogos, Maurício S. Bugarin and Fernando B. Meneguin

43/2014 11-12-2014 De Facto Property Rights Recognition, Labor Supply and Investment of the Poor in Brazil, Rafael Santos Dantas and Maria Tannuri-Pianto

42/2014 11-05-2014 On the Institutional Incentives Faced by Brazilian Civil Servants, Mauricio S. Bugarin and Fernando B. Meneguin

41/2014 10-13-2014 Uma Introdução à Teoria Econômica da Corrupção: Definição, Taxonomia e Ensaios Selecionados, Paulo Augusto P. de Britto

40/2014 10-06-2014 Um modelo de jogo cooperativo sobre efeitos da corrupção no gasto público, Rogério Pereira and Tatiane Almeida de Menezes

39/2014 10-02-2014 Uma análise dos efeitos da fusão ALL-Brasil Ferrovias no preço do frete ferroviário de soja no Brasil, Bruno Ribeiro Alvarenga and Paulo Augusto P. de Britto

38/2014 08-27-2014 Comportamentos estratégicos entre municípios no Brasil, Vitor Lima Carneiro & Vander Mendes Lucas

37/2014 08-20-2014 Modelos Microeconômicos de Análise da Litigância, Fabio Avila de Castro

36/2014 06-23-2014 Uma Investigação sobre a Focalização do Programa Bolsa Família e seus Determinantes Imediatos. André P. Souza, Plínio P. de Oliveira, Janete Duarte, Sérgio R. Gadelha & José de Anchieta Neves

35/2014 06-22-2014 Terminais de Contêineres no Brasil: Eficiência Intertemporal. Leopoldo Kirchner and Vander Lucas

34/2014 06-06-2014 Lei 12.846/13: atrai ou afugenta investimentos? Roberto Neves Pedrosa di Cillo

33/2013 11-27-2013 Vale a pena ser um bom gestor? Comportamento Eleitoral e Reeleição no Brasil, Pedro Cavalcante

32/2013 11-13-2013 A pressa é inimiga da participação (e do controle)? Uma análise comparativa da implementação de programas estratégicos do governo federal, Roberto Rocha C. Pires and Alexandre de Avila Gomide

31/2013 10-30-2013 Crises de segurança do alimento e a demanda por carnes no Brasil, Moisés de Andrade Resende Filho, Karina Junqueira de Souza and Luís Cristóvão Ferreira Lima

30/2013 10-16-2013 Ética & Incentivos: O que diz a Teoria Econômica sobre recompensar quem denuncia a corrupção? Maurício Bugarin

29/2013 10-02-2013 Intra-Village Expansion of Welfare Programs, M. Christian Lehmann

28/2013 09-25-2013 Interações verticais e horizontais entre governos e seus efeitos sobre as decisões de descentralização educacional no Brasil, Ana Carolina Zoghbi, Enlinson Mattos and Rafael Terra

27/2013 09-18-2013 Partidos, facções e a ocupação dos cargos de confiança no executivo federal (1999-2011), Felix Lopez, Mauricio Bugarin and Karina Bugarin


26/2013 09-11-2013 Metodologias de Análise da Concorrência no Setor Portuário, Pedro H. Albuquerque, Paulo P. de Britto, Paulo C. Coutinho, Adelaida Fonseca, Vander M. Lucas, Paulo R. Lustosa, Alexandre Y. Carvalho and André R. de Oliveira

25/2013 09-04-2013 Balancing the Power to Appoint officers, Salvador Barberà and Danilo Coelho

24/2013 08-28-2013 Modelos de Estrutura do Setor Portuário para Análise da Concorrência, Paulo C. Coutinho, Paulo P. de Britto, Vander M. Lucas, Paulo R. Lustosa, Pedro H. Albuquerque, Alexandre Y. Carvalho, Adelaida Fonseca and André Rossi de Oliveira

23/2013 08-21-2013 Hyperopic Strict Topologies, Jaime Orillo and Rudy José Rosas Bazán

22/2013 08-14-2013 Há Incompatibilidade entre Eficiência e Legalidade? Fernando B. Meneguin and Pedro Felipe de Oliveira Santos

21/2013 08-07-2013 A Note on Equivalent Comparisons of Information Channels, Luís Fernando Brands Barbosa and Gil Riella

20/2013 07-31-2013 Vertical Integration on Health Care Markets: Evidence from Brazil, Tainá Leandro and José Guilherme de Lara Resende

19/2013 07-24-2013 A Simple Method of Elicitation of Preferences under Risk, Patrícia Langasch Tecles and José Guilherme de Lara Resende

18/2013 07-17-2013 Algunas Nociones sobre el Sistema de Control Público en Argentina con Mención al Caso de los Hospitales Públicos de la Provincia de Mendoza, Luis Federico Giménez

17/2013 07-10-2013 Mensuração do Risco de Crédito em Carteiras de Financiamentos Comerciais e suas Implicações para o Spread Bancário, Paulo de Britto and Rogério Cerri

16/2013 07-03-2013 Previdências dos Trabalhadores dos Setores Público e Privado e Desigualdade no Brasil, Pedro H. G. F. de Souza and Marcelo Medeiros

15/2013 06-26-2013 Incentivos à Corrupção e à Inação no Serviço Público: Uma análise de desenho de mecanismos, Maurício Bugarin and Fernando Meneguin

14/2013 06-26-2013 The Decline in inequality in Brazil, 2003–2009: The Role of the State, Pedro H. G. F. de Souza and Marcelo Medeiros

13/2013 06-26-2013 Productivity Growth and Product Choice in Fisheries: the Case of the Alaskan pollock Fishery Revisited, Marcelo de O. Torres and Ronald G. Felthoven

12/2013 06-19-2003 The State and income inequality in Brazil, Marcelo Medeiros and Pedro H. G. F. de Souza

11/2013 06-19-2013 Uma alternativa para o cálculo do fator X no setor de distribuição de energia elétrica no Brasil, Paulo Cesar Coutinho and Ângelo Henrique Lopes da Silva

10/2013 06-12-2013 Mecanismos de difusão de Políticas Sociais no Brasil: uma análise do Programa Saúde da Família, Denilson Bandeira Coêlho, Pedro Cavalcante and Mathieu Turgeon

09/2013 06-12-2103 A Brief Analysis of Aggregate Measures as an Alternative to the Median at Central Bank of Brazil’s Survey of Professional Forecasts, Fabia A. Carvalho

08/2013 06-12-2013 On the Optimality of Exclusion in Multidimensional Screening, Paulo Barelli, Suren Basov, Mauricio Bugarin and Ian King

07/2013 06-12-2013 Desenvolvimentos institucionais recentes no setor de telecomunicações no Brasil, Rodrigo A. F. de Sousa, Nathalia A. de Souza and Luis C. Kubota

06/2013 06-12-2013 Preference for Flexibility and Dynamic Consistency, Gil Riella

05/2013 06-12-2013 Partisan Voluntary Transfers in a Fiscal Federation: New evidence from Brazil, Mauricio Bugarin and Ricardo Ubrig

04/2013 06-12-2013 How Judges Think in the Brazilian Supreme Court: Estimating Ideal Points and Identifying Dimensions, Pedro F. A. Nery Ferreira and Bernardo Mueller

03/2013 06-12-2013 Democracy, Accountability, and Poverty Alleviation in Mexico: Self-Restraining Reform and the Depoliticization of Social Spending, Yuriko Takahashi

02/2013 06-12-2013 Yardstick Competition in Education Spending: a Spatial Analysis based on Different Educational and Electoral Accountability Regimes, Rafael Terra

01/2013 06-12-2013 On the Representation of Incomplete Preferences under Uncertainty with Indecisiveness in Tastes, Gil Riella

Documents

Ex-ante Moral Hazard of Unemployment Insurance...after increases in the potential duration of unemployment insurance by one month. Additionally, this paper analyzes the interaction