Minimum Wage, Informality and Economic

Development

Jin Ho Kim

April 20, 2019

Abstract: There is a lack of consensus in the existing literature on the extent to which the

minimum wage in�uences the overall labor market in developing countries, where a large

share of the population engages in informal sector economic activity. This paper shows that

an increase in the minimum wage can induce (i) an increase in the formal sector economic

activity, (ii) an increase in the wage (iii) reduced pro�t margin �rm gains over hiring each

worker and (iv) an increase in the number of non-compliant formal sector �rms which are

paying less than the minimum wage or providing no health insurance. This occurs when �rms

in the formal sector engage into monopsonistic competition over workers and when there is

substantial income heterogeneity among informal sector laborers. Using historical minimum

wage change in Indonesia during 2000~2014, the paper �rst tests this hypothesis empirically,

and then builds a search theoretical model that incorporates this empirical evidence. The

equilibrium search model features (i) monopsonistic competition among �rms for workers,

(ii) non-compliance behavior of �rms on the minimum wage law, and (iii) heterogeneous

�rm and worker productivity and heterogeneity in worker's reservation wage. Our structural

model provides a rich mechanism to analyze the e�ect of the legal minima on labor market

in developing countries.

Key Words: Minimum Wage, Informality, Non-compliance, Monopsony

JEL Classi�cation: J38; J42; O17

1

I. Introduction

How does the minimum wage a�ect the shape of the formal sector economy in developing

countries? This question interests many policymakers as the minimum wage policy has been

the single most widely implemented labor protection law in developing countries. Though the

question has been extensively studied among development and labor economists, literature

has not yet reached a consensus since the e�ect of the law depends on the institutional

structure of the countries and often developing countries do not have enough quality data

to control for these institutional set up. Also, a large portion of informal sector economic

activiy1, imperfection in formal sector labor market, and non-compliant behavior of �rms to

existing labor protection regulations, which are prevalent in developing countries add even

more complication for analysis. Existing literature on minimum wage usually focused on its

e�ect on limited aspects of labor market outcome, but often failed to show its overall e�ect

on the labor market. Both theoretical and empirical study is much needed to further provide

a comprehensive understanding of the law and to correctly evaluate its impact on overall

labor market.

Indonesia is an ideal case study to address this research question in that (i) there has

been a large variation in minimum wage levels across provinces and years and (ii) there

is exceptional quality panel data, Indonesian Family Life Survey (IFLS henceforth) and

Manufacturing Survey (IS henceforth), to study the underlying features of the labor market.

In this paper, we use the historical minimum wage hike in Indonesia between 2000 and 2014

to empirically and theoretically investigate the e�ect of the law on the overall labor market.

To this end, we start our analysis by detailing and explaining the characteristics of the

informal and formal sector labor market, and the non-compliance issue in Indonesia. Our

analysis shows that the informal business sector in Indonesia is too small to ever formally

register and operate in the formal sector, consistent with the viewpoint of survival view of La

Porta and Shleifer (2014) and Rothenberg et al. (2016). 2However, we still observe a large1Informal economic activity refers to a business that is not legally registered with the government. These are primarily,

small, household-run businesses that often lag in productivity behind formal �rms, which are legally registered.2The literature on informality is large and diverse, with many controversies over the mechanism driving informal employment

2

income heterogeneity among informal sector laborers, in which a sizable portion of them earn

more than formal sector wage earners. This feature shows that if there is a formalization

mechanism, it should be through an occupational transition from informal sector job to

formal sector wage-earning jobs. Our data also shows evidence of labor market imperfection

in the formal sector labor market such as imperfect information or a positive gap between the

value of labor productivity and wage. As for the non-compliance issue of the labor market

regulation, our data shows (i) a third of full-time workers do not receive minimum wage (ii)

only 20-25 percent of full-time workers receive mandatory health bene�ts.

This underlying labor market structure and non-compliance issue motivates us to test

several hypotheses. In the second step of the investigation, we conduct a regression analysis

to �nd the e�ect of the historical minimum wage hike in Indonesia on the overall labor

market. Using individual-level panel data, we adopt a di�erence-in-di�erence approach in

which we allow for individual and time �xed e�ects. Our �ndings indicate that the minimum

wage hike caused (i) some informal sector workers to move to formal sector jobs; (ii) the wage

of the formal sector workers and the average wage in manufacturing plants to increase�both

initially sub-minimum wage-earning workers (sub-minimum wage paying plants) and over-

minimum wage earning earners (over-minimum wage paying plants); (iii) the pro�t gain

that manufacturing �rms earn from hiring each individual to decrease and (iv) some �rms

to circumvent paying higher employment costs by hiring part-time workers, or by hiring

full-time workers without providing mandatory health insurance.

Our empirical �ndings portray various features of the Indonesian labor market that are

largely shared by other developing countries. This motivates us to conduct the third step

of analysis. We construct a coherent structural model that embodies our empirical �ndings.

There have been important theoretical works that contain these labor market features in

a uni�ed way but the currently existing models cannot explain all our empirical �ndings.

Our structural model provides plausible mechanisms under which a minimum wage hike

in developing countries. Two of these mechanisms are the survival view and the competitive view. The survival view arguesthat informal sector workers were �ltered out of the formal sector labor market due to their low productivity. The competitiveview argues that informal sector workers rationally opted into the informal sector labor market in order to earn a higher income.

3

can (i) increase wages of both the initially sub-minimum wage paid group and the initially

over-minimum wage paid group, (ii) increase the formal sector employment (iii) decrease

pro�t margin from hiring each worker and (iii) increase non-compliance behavior of �rms for

existing labor protection law.

Our work contributes to two strands of the literature. First, our study on the e�ect of

the minimum wage complements a long-standing debate on how a minimum wage a�ects

labor market outcomes. There exists studies analyzing the impact of the minimum wage on

employment in developing countries, both in the formal and informal sectors (for example,

Nunez (2004) for Columbia; Gindling & Terrell (2007) for Costa Rica; Alaniz et al. (2011) for

the case of Nicaragua; Fajnzylber (2001), Lemos (2004a, 2004b, 2009) for Brazil; Dinkelman

& Ranchhod for South Africa(2012)). This literature generally agrees that minimum wage

policies increase wages, and decrease or have no e�ect on formal sector employment. For

the case of Indonesia, previous empirical evidence does not give a consistent answer about

employment and generally agreed results on wages (Rama (2001), Islam & Nazara (2000),

Suryahadi et al. (2003), Del Carpio et al. (2012), Harrison & Scorse (2010), Alatas &

Cameron (2008), Comola & de Mello (2011), Magruder (2013)). Our work contributes

to the existing empirical literature in that we further provide evidence for the e�ect of

minimum wage on employment, wage for di�erent group of workers, non-compliance to its

own regulation and other labor protection regulation, and also on pro�t margin of �rm from

hiring workers, measured by Pigou's E. To date, empirical literature that studies the e�ect

of minimum wage on other related labor protection law is rare. Also, to the best of our

knowledge, we do not have any empirical study that investigates the e�ect of minimum wage

on pro�t margin of �rms.

Second, our theoretical framework extends the existing search model with monopsonistic

competition (Burdett and Mortensen (1998)) and provides a rich mechanism to study for

the labor market in developing countries.3. Importantly, our structural model assumes La

3 To date, Burdett and Mortensen model has been extended by numerous authors to analyze labor markets in developed

economies where informal economy is not assumed. Look van den Berg and Ridden 1998; Postel-Vinay and Robin, 2002; Jolivet

4

Porta and Shleifer's viewpoint (2014) on informality in that informal �rms are too incapable

to legally register and operate formally. In this viewpoint, formality may occur through

occupational choice from informal sector entrepreneur jobs to formal sector wage jobs. This

assumption on informality di�erentiates our modeling exercise from the existing model that

also extends Burdett and Mortensen. For instance, Meghir et al. (2015) extends the original

Burdett-Mortensen (1998) model by introducing �rm's endogeneous choice between formal

sector and informal sector economic activity. In their modelling exercise, they do not assume

a formality mechanism through occupational shift between informal sector entrepreneur and

formal sector wage earner.

Our model also incorporates non-compliance behavior of the �rms against labor protec-

tion regulation that has been emphasized in the models that follow industrial organization

literature. For instance, Ullyea (2018) models not only for �rms's choice over formal or in-

formal sector economic activity, but also introduces a non-compliance behavior of formally

registered �rms on existing labor market regulation. He uses the model to study for the two

kinds of non-compliance behavior of �rms against existing regulations, and its rami�cations

on aggregate output and TFP. Basu et al. (2009) also introduces imperfect compliance of

the �rms on the existing labor law, informal sector economic activities and heterogeneous

worker preference to provide a coherent mechanism for the impact of legal minima on labor

market in developing countries. However, the literature does not contain a mechanism that

explains the multiplication e�ect of minimum wage regulation, which is widely documented

in policy papers (For instance, see Cunningham (2007)). We incorporate non-compliance

behavior of the �rms into the existing search literature to allow for the multiplication e�ect

of the binding minimum wage regulation. Combining with heterogeneity in �rm and worker

productivity and also a heterogeneity in the workers' outside options, our model provides a

rich mechanism to study for labor supply and demand, �rm's non-compliance behavior and

wage dispersion.

et al., 2006; Moscarini and Postel-Vinay, 2013; Lise et al., 2016; Lise and Robin, 2017; Moser and Engbom, 2018.

5

Outline. The paper proceeds as follows. Section 2 introduces data and provides de-

scriptive statistics on the informal and formal sector labor markets, and minimum wage

in Indonesia�its history, relative stance on income distribution and non-conformity to the

regulation. Section 3 provides an estimation strategy and regression results. In Section 4,

we construct an equilibrium search model that is consistent with existing labor market and

replicates the e�ect on minimum wage that we found in section 3. Section 5 concludes.

2. Labor Market Facts

2.1. Data and de�nitions

This section describes the two data sets used for the analysis of the e�ect of the minimum

wage on employment and non-compliance to labor regulation from 2000 to 2014. The �rst

data set consists of three separate surveys conducted by the Indonesian Family Life Survey

(IFLS) in 2000, 2007 and 2014 (�Wave 3,� �Wave 4,� and �Wave 5�). The IFLS covers

83 percent of the total population and contains over 30,000 individuals living in 13 out of

the 27 provinces, primarily on the west side of the country. IFLS contains rich individual-

level information which allows us to construct individual-level panel data, and also use

various individual level controls for in regression analysis. The sample we use for the analysis

comprises of the legally de�ned working-age population aged between 15 and 65 years during

the period from 2000 to 2014. All these samples are restricted to individuals for whom all

variables were available for two consecutive years so as to control for individual speci�c

e�ects. The sample is further restricted to individuals whose earnings and household assets

are between the 1st and 99th percentile of real wages and the real value of household assets.

This leaves 56,134 valid observations.

Indonesia's National Statistics Agency (Badan Pusat Satistik, BPS) identi�es enterprises

as legal entities if they are listed by the Ministry of Manpower, and registered �rms are

considered formal enterprises. Unfortunately IFLS data does not include identi�ers for the

6

legal classi�cation of work, and we only follow previous literature to establish private or

government workers as formal sector workers. However, IFLS data provides an excellent

source of information on the informal sector employment. ADB Reports (2010) document

95 percent of self-employed workers as informal sector workers and 98 percent of casual

workers and unpaid family workers as informally hired, which is directly identi�ed in the

IFLS survey. IFLS also contains additional information on whether wage earners receive

health insurance bene�t. As Manpower regulation states that every legally registered �rm

should provide health insurance for full-time workers, we can use IFLS data to study for non-

compliance behavior of �rms on minimum wage and health insurance regulation accordingly.

We complement the IFLS data with the Indonesian manufacturing survey (IS) to study for

the formal sector economic activity. Data spans from 2000 until 2009, and it contains detailed

information on a wide set of plant-level characteristics which fully characterize formal sector

economic activity in the manufacturing sector. The data set contains information on plant-

level variables such as wages and other bene�ts paid to workers, �rm age, the number of

workers hired in the manufacturing job and the non-manufacturing job, total capital stock,

investment, total materials and fuels purchased, and total revenues. Especially, we use IS

data to estimate plant-level productivity, which then will be used to study the labor market

imperfection. Appendix A provides a detailed description of our variable construction.

2.2. Workers in the formal and informal sector economy

The literature on the informal sector in developing countries does not reach a consensus

on its characteristics. The survival view regards the informal sector as segmented from the

formal labor market due to its inability to compete. The parasite viewpoint and De Soto's

viewpoint regard the informal sector as a rational decision to avoid taxation or the result

of being blocked from moving into the formal sector due to a high cost of entry. IFLS data

seems to suggest that the informal sector in Indonesia aligns with the survival viewpoint on

average, though there is still heterogeneity in the informal sector.

7

Table 1 provides information on individuals who work in the formal and informal sector

accordingly. The majority of workers (60 percent) are involved in informal sector employ-

ment, which shows that the labor market in Indonesia is thin. Informal workers usually work

in companies of less than �ve people. Their education level lags behind when compared to

employees in the formal sector. This table suggests that i.) informal sector workers are less

productive compared to formal sector workers (segment viewpoint for labor market; Rothen-

berg et al. (2014)) ii.) there are signi�cant overlaps between the two sectors even within the

narrowly de�ned industries, which suggests that there is still a margin to regard the informal

sector economy as competitive.

8

Figure 1 illustrates the distribution of formal versus informal workers by income decile.

Informal workers constitute 78 percent of the lowest income decile. However, we still �nd a

signi�cant overlap between the income of informal and formal sector workers. Speci�cally, 30

percent of the individuals in the top income decile work in the informal sector. This feature

suggests that even though overall characteristics of informal sector workers in Indonesia align

with the survival viewpoint (La Porta & Shleifer (2014)), there is still enough heterogeneity

among the informal sector economy that we can infer a sizable portion of informal sector

workers may choose to stay in the informal sector economy.

2.3. Labor market imperfect in the formal sector

While there is general consensus on the existence of labor market imperfections and non-

negligible market power employers can have over workers, empirical investigation on labor

market imperfections in the context of developing countries is thin. Literature attributes the

source of labor market imperfection to preference heterogeneity over jobs, mobility costs,

and imperfect information (Bhaskar et al (2002)). Two recent rounds of IFLS data (IFLS4

and 5) contain useful survey information that indicates imperfect information as the main

source of market imperfection. The survey recorded how respondents found their current

9

job.

Even with only two rounds of the data, we observe the signi�cance of imperfect informa-

tion. IFLS data shows that 42-45 percent of total respondents found their job through friends

or relatives, and 45-48 percent of formal sector workers found their positions through friends

or relatives. Considering that we can potentially regard other answers such as 'contacted

company' and 'contacted by company' as a sort of personal connection mechanism, the table

shows the limited information job seekers have in their search for a job. As we only have

two rounds of data to study for the imperfection in the labor market, we also use IS data to

complement our understanding of labor market imperfection. We follow existing literature

and use standard structural estimation methods to estimate the production function and

calculate the marginal product of labor and then directly compare to paid wage (Dobbe-

laere and Jacques (2013), Dobbelaere et al. (2015), Petrin and Sivadasan (2013) Olley and

Pake (1996), Blundell and Bond (1998), Levinsohn and Petrin (2003), Wooldridge (2009),

Ackerberg et al. (2015), Brummund (2013)).

10

11

Tables 3 and 4 show the main descriptive statistics for manufacturing �rms using the

Indonesia manufacturing survey. To illustrate labor market imperfection, we use a standard

measure, Pigou's E, the normalized gap between the value of worker's marginal product of

labor and wage (E = R′(L)−W (L)W (L)

where R′(L) is marginal product of labor andW (L) is wage.

Look at Appendix B). With no labor market imperfection, pro�t-maximizing employers

should hire workers until the marginal product of labor is equal to the wage. Therefore,

a higher value of Pigou's E indicates more severe labor market distortions. We construct

Pigou's E by using the value-added per worker and output per worker information from the

survey. As the survey does not have information for value-added or output by production

laborers or non-production laborers separately, we use a standard econometric method to

estimate production function to calculate Pigou's E for production workers and for non-

production workers respectively. We estimate the production function at the 2-digit industry

level, calculating marginal productivity of labor for production and non-production workers

respectively, which we then directly compare to average wage each �rm pays to its worker.

12

We �nd a sizable gap between the value of labor product and wage across all industries,

especially among non-production workers. The gap may re�ect monopsonistic behavior of

�rms due to search friction, or worker's heterogeneous preference. The gap may be caused

by high unexpected labor cost due to rigid labor market regulation, discouraging �rms from

hiring more workers. Regardless of the role of labor market regulation, this sizable gap

between the marginal product of labor and wage re�ect an imperfect formal sector labor

market. We test how minimum wage a�ects the gap in our empirical analysis.

2.4 Minimum Wage and Non-Compliance

The role of the minimum wage on formal sector employment in developing countries has been

a heated debate among policymakers. The literature does not reach a consensus because the

e�ect of the law depends on the social and economic context. The case for Indonesia is

not an exception. This section introduces minimum wage policy in Indonesia: its history,

attributes, relative stance on income distribution, and its e�ect on compliance to labor

market regulation.

A minimum wage has been required by law since 1970 in Indonesia, though was rarely

implemented until Western customers put pressure on the Indonesian government in the

1990s (See Harrison and Scores for a more detailed discussion). During the �rst half of

the 1990s, alongside rapid economic expansion, real minimum wage grew quickly but this

growth slowed in the second half of the 1990s. Especially due to depreciation in the currency

during the Asian �nancial crisis in 1997, real minimum wages declined by 30 percent in 1998.

Beyond its impact on the decline of the real minimum wage, the Asian Crisis also provided the

political and economic impetus that led to the demise of Suharto, the dictator of Indonesia

from 1967 to 1998, and the subsequent political transformation that led to the enactment

of the decentralization laws of 1999. These laws allowed each provincial government to

make autonomous policies in consideration to the local economy, including the determination

of minimum wage rates. Since then, the level of the minimum wage has been set and

13

annually updated by discussion among provincial tripartite wage councils�representative of

the ministry of manpower, local employers, and unions.

The process of setting minimum wages is mostly based on negotiations and is weakly

linked to the technical assessments of the cost of living increases. Though the technical basis

to calculate the cost of a decent living for workers (Kebutuhan Hidup Minimal, KHL) exists

as an input for determining minimum wages, the in�uence of the KHL on minimum wage

is relatively small in practice. The negotiation based procedure brought large variations in

the minimum wage across provinces and years (Fig 1). Especially, during 2013, relative to

the years between 2006 and 2012 when the minimum wage grew by 7.6 percent per year on

average, unions were more successful in their negotiations to raise local minimum wages and

there has been a 43.7 percent increase in Jakarta and 49.7 percent in East Kalimantan.

14

It is well-known fact that the level of minimum wage is quite close to the median wage

in developing countries, and Indonesia data also demonstrates this feature. Table 5 records

the ratio of the minimum wage to the median of full-time wage, the part-time wage and the

pro�t by province and year. The table indicates that the range of the ratio for full-time

workers spans from 80 percent to 85 percent across the years. It also shows the income gap

between full-time wage earners and the rest of the working group has been widening across

years, which may be attributed to the increased minimum wage that only applies to formal

sector employment.

15

Figure 3A shows the kernel density for the wage distribution and the self-employed pro�t

distribution, where the distributions are normalized by the minimum wage. It is striking to

�nd that the wage density curve is relatively stable across the years, even though there has

been a rapid increase in average real minimum wage across years (�gure 1). The stability of

the normalized kernel density shows that the wage distribution has been moving alongside

minimum wage across years. The minimum wage in Indonesia does not function as a safety

net to protect vulnerable workers as documented by the World Bank Report (2010) and ILO

(2014). Rather, the Indonesian minimum wage appears to be a wage-setting mechanism for

negotiation.

Figure 3B shows the distribution of the mean wage for manufacturing �rms, and together

with �gure 2A, the graphs illustrate the signi�cance of non-compliance on minimum wage.

Nearly 35 percent of the full-time workers in the IFLS data do not receive the minimum

wage. Further, 40 percent of the average wage payment in each manufacturing plant is less

than the minimum wage. The non-compliance ratio is signi�cant because the Manpower

Law requires all employers to pay the minimum wage to full-time employees. This suggests

weak enforcement of the minimum wage regulations. 4

4The Manpower Law stipulates that if an employer violates the minimum wage regulation they face imprisonment for between1 and 4 years. In addition, they are required to pay monetary compensation between Rp 10,000,000 and 400,000,000. However,the e�ectiveness of the Manpower Law is limited due to its high monitoring costs.

16

Another notable feature in the Figures 3A and 3B is the spike in the wage distribution

near the minimum wage, which also is a typical characteristic of the wage distribution in

developing countries that implement a minimum wage law. (Basu et al. 2009).5 This

spike near the minimum wage suggests that even though the enforcement to the law is not

complete, there is still a portion of �rms that pay around or exactly the minimum wage to

employees.

5The IFLS data only has information on total compensation. It includes wages and other bene�ts such as bonuses, medicalcare, and health insurance. We cannot disentangle these other bene�ts from the pure wage. However, we can infer from thegraph that when we exclude the other bene�ts the minimum wage will get closer to the median wage.

17

Table 6 shows the non-compliance rate of companies by (i) �rm size and (ii) education

level of employees. We use the poverty measurement index applied to minimum wage non-

compliance, and report the ratio for non-compliance incidence and for the gap between

minimum wage and sub-minimum wage. Consistent with the literature, as �rm size grows,

the non-compliance incidence and gap get smaller. This can be due to potentially higher

productivity of large �rms, or a higher likelihood of being monitored by the government.

Likewise, people with high education tend to receive more than the minimum wage. Despite

the labor market imperfections, it is evident that highly educated workers have increase

bargaining power compared to poorly-educated workers.

There are potential e�ects of a minimum wage law on compliance with other labor pro-

tection regulations such as health insurance. This has been emphasized in recent work as the

minimum wage is one of many interrelated labor protection laws a�ecting the total cost of

o�ering formal employment (Basu et al. (2015)). For example, employers can respond to an

increased minimum wage by reducing other monetary or in-kind components of compensation

such as pension bene�ts and health bene�ts. In the case of Indonesia, the Manpower Law

stipulates that employers should cover health insurance for full-time workers. 6 Employers

may circumvent these laws by hiring full-time workers without providing health insurance

or by hiring part-time workers in order to avoid paying both the minimum wage and health

insurance.

6Full-time workers are de�ned as any employee working more than 40 hours a week.

18

Figure 4 illustrates non-compliance to labor regulation is quite serious. Only 10-20 percent

of formal sector workers receive health insurance. This �gure raises the empirical question

that the minimum wage may have decreased health insurance bene�ts and increased part-

time workers.

3. The consequence of minimum wage

The descriptive statistics in the previous section portrays important features of the la-

bor market and the minimum wage in Indonesia that are also commonly found in other

developing countries. The theoretical literature predicts that the employment e�ect of the

minimum wage law critically hinges on characteristics of the informal versus formal sector

labor market, the relative stance of the minimum wage on wage distribution, and on the

enforcement rate of the law. For instance, if the minimum wage is considerably higher than

the competitive market wage, and the enforcement rate is strong, labor demand and formal

sector employment declines. However, a properly set minimum wage, which is still less than

the competitive market wage, can create an increase in labor supply without reducing labor

demand, which can increase formal sector employment (Card and Kruger (1996), Burdett

and Mortensen (1998)). Also, it is predicted that if enforcement of the minimum wage law

is imperfect, then monopsonistic �rms who have considerable bargaining power can still op-

timize their job-o�ering behavior by circumventing existing labor market regulation. These

19

�rms may o�er a sub-minimum wage, intentionally hire part-time employees to legally avoid

the law, or optimize employment cost by not providing mandated bene�ts such as health

insurance. If enforcement is su�ciently low such that only large �rms are easily monitored

and comply to the minimum wage, then middle-sized �rms may o�er sub-minimum wage

jobs which are still pro�table to the �rms and su�ciently attractive to draw labor demand.

As for the wage e�ect, though most literature �nds a positive e�ect of the minimum wage on

the wage distribution, there is not much research that studies the e�ect of the minimum wage

on the sub-minimum wage in formal sector employment, and the theoretical predictions and

empirical �ndings are somewhat di�erent. Considering that a signi�cant portion of formally

hired workers receives wages below the legal minimum, we are also interested in the e�ect

on sub-minimum wage workers. This section is devoted to testing these hypotheses. We �rst

introduce our estimation strategy and then discuss the regression result.

3.1. Estimation Strategy

3.1.1. Two-way �xed e�ect

To estimate the relationship between minimum wage and labor market outcome variables,

we use a standard two-way �xed e�ect speci�cation, using individual or plant-level panel

data.

(1) yijt = α + βMWjt + γXijt + ηMVjt + λi + δt + uijt

Here i indicates an individual or a plant, j is the province of the respondent and t

represents time. MWjt represents the log of minimum wage and varies by time and province.

MVjt represents province-speci�c macro variables. Xijt represents individual controls. With

the IFLS data, Xijt contains the education level and its square, age and its square, a dummy

for urban residence, and household assets. With the IS data, Xijt represents �rm-speci�c

controls such as the log of a �rm's age, the percentage of the �rm owned by foreigners,

the percentage of output exported, the percentage of the �rm owned by the local and the

central government, capital used by �rms, intermediate goods, and the output of the �rm.

20

δt represents time �xed e�ects which control for year-speci�c macroeconomic shocks, and λi

represents the individual or the �rm-level �xed e�ects depending on the sample used.

The individual level panel data lessens the concern for endogeneity bias, which is more

severe when cross-province or cross-district data is used to identify the e�ects of the minimum

wage on the outcome of interest. As mentioned above, in Indonesia, minimum wages are

carefully targeted by the local government in consideration of the overall provincial economy

and as such, it is known that the provincial minimum wage tends to be set higher when the

provincial unemployment rate is low and GDP per capita is high. Thus, estimation based on

province or district level aggregated data is more likely to su�er from upward bias or reverse

causality. Using disaggregated individual (plant) panel data resolves the reverse causation

issue that aggregate-level data su�ers�individual (plant) �xed e�ect speci�cation assumes

that a province's minimum wage does not respond to changes in outcomes of an individual

(plant), but individual (plant) outcome responds to changes in the provincial minimum wage.

This assumption is likely to hold as speci�c individuals (plants) cannot in�uence the level of

minimum wage.

To reduce omitted variable bias, we further control for provincial speci�c macro variables,

MVjt, log income per capita and the unemployment rate. It is a standard practice to assess

the robustness of panel data estimates of province policy with the inclusion of province-

speci�c time trends, which account for heterogeneous time e�ects across the minimum wage

jurisdiction (Allegretto et al. (2017)). However, given that we have only three rounds

of data where the period between surveys is 7 years, we cannot limit the identi�cation

information from the deviation around province-speci�c linear time trends. Instead, we

address the heterogeneous provincial macroeconomy that can a�ect the level of minimum

wage, by controlling provincial log income per capita and unemployment rate.7

The coe�cient of interest is a di�erence-in-di�erences estimator of the e�ect of changes

in the minimum wage on outcome yijt. The province-year speci�c variation of the minimum

7Meer and West (2016) point out that if there is a dynamic e�ect of the minimum wage on the trend itself, then includingjurisdiction time trends will attenuate estimates of the treatment e�ect.

21

wage is the main source of variation. The identifying assumption is that after controlling

for individual observed and unobserved traits and the provincial-speci�c macroeconomy, the

outcome of interest would have followed a similar trend across provinces, if not for the

di�erential changes in the minimum wage regime. The coe�cient β measures the elasticity

of the outcome of interest with respect to real minimum wage.

3.1.2. Non-compliance to the minimum wage

There are several issues to deal with when examining the employer's incentive to comply with

the minimum wage. First, the �rm's willingness to observe the minimum wage regulation

depends on the level of the minimum wage and the intensity of government surveillance.

There is not always data available to measure the intensity of government monitoring (Adres

2018). Second, it is challenging to clearly identify the control and treatment groups. For

instance, we want to test how a �rm's size a�ects the observance of the law. Individuals

may change their job in order to work a larger �rm if it is more likely to comply with the

minimum wage. If this is the case, then the control and treatment groups are contaminated

by these individuals. Third, �rm-level data has a misreporting issue. Firms may not report

their wage payment truthfully if they violate the minimum wage law.

It is possible to address the misreporting and group identi�cation issues by using individual-

level panel data (IFLS). Speci�cally, using individual-level data alleviates the systematic

misreporting issue and panel data allows the treatment and control groups to be identi�ed.

Here we are interested in whether individuals who work in di�erent size �rms have di�erent

probabilities of receiving less than the minimum wage. Several empirical works point out

the relevance of �rm size on minimum wage compliance; larger �rms are subject to more

government monitoring and penalties.8The IFLS data also illustrates this point.

8Harrison (2010) shows the impact of exogenous enforcement through the anti-sweatshop movement on wage growth inIndonesia. The result shows that targeted foreign-owned �rms under the high intensity of surveillance increased their wagepayment compare to small �rms.

22

Panel A of �gure 3 illustrates the time trend of relative ratio of the minimum wage over

the median wage. Panels B and C show compliance with the minimum wage regulation by

�rm size. As mentioned above, successful labor union negotiations caused a surge of the

minimum wage in 2013. The Jakarta Report described this unusual surge as an unexpected

shock to most �rms and there is a steep increase in the minimum wage-median wage ratio

in 2014. Regardless of the shock, the non-compliance ratio for laborers in the small �rms

(1-4 employees) steadily increased across the years, which contrasts to laborers in the large

�rms (>200 employees) where the non-compliance ratio decreased across years. The mini-

mum wage shock seemed to increase the non-compliance ratio for middle-sized �rms where

government surveillance was relatively weak compared to large �rms. This hypothesis is

formally tested with the following regression speci�cation.

(2) BMWist = α + βDist + γXist + λi + δt + uist

Here s is the �rm-size category, and BMWist is a binary indicator that identi�es if worker

i in the province �rm-size category s at time t is paid below the minimum wage. Dist is

23

the interaction of the treatment group indicator and year 2014 indicator. We regard the

2013 event as an exogenous policy shock to �rms. The treatment and control groups were

created using a subsample of full-time formal sector workers who remained in a similar-

sized �rm for more than two consecutive rounds. The control group consists of full-time

workers who remained in �rms with more than 200 employees. The treatment group consists

of full-time workers who remained in �rms with 5-199 employees. This regression tests

how �rms whose expected �ne payment are less compare to the control group respond to

an unanticipated minimum wage hike. The method assumes that in the absence of the

unexpected minimum wage change in 2013, the compliance ratio in medium-sized �rms

would have behaved similarly to the large-sized �rms.

The coe�cient on the interaction term, β, captures the average di�erence in non-compliance

to the minimum wage law across the treatment and control groups before and during 2014.

An expanded version of this equation is also estimated where the treatment identi�er in-

teracts with dummy variables for each year. This tests the parallel trend assumption of

di�erence-in-di�erencem, and thus support a di�erence-in-di�erence strategy to test for par-

tial compliance with the minimum wage law. In the next session, we report estimates of

the minimum wage impact on non-compliance, employment, wages, hours, and Pigou's E

respectively.

3.2. Labor market outcomes

In this sub-section, we report our empirical results. We report the e�ect the minimum wage

has on employment, wage, and �rm's optimizing behavior in response to increased labor

cost. Also, we show how the monopsonistic margin, measured by Pigou's E, is a�ected by

the minimum wage.

24

Table 7 presents the regression results of various categories of employment on real mini-

mum wages using two-way �xed e�ects. Binary indicators for each category of employment

are constructed. The regression results measure the probability of being in each employment

category compared to being non-employed or being in another category of employment. The

regression results show that a 10 percent increase in the minimum wage is related to a 1

percent increase in formal sector employment. This can be further broken down to show

that a 10 percent increase in the minimum wage is related to a 0.5 percent and 0.4 percent

increase in full-time formal sector and part-time formal sector employment respectively. In-

formal sector workers show the opposite trend. The results show that a 10 percent increase

in the minimum wage is related to a 1 percentage point decrease in informal sector jobs. The

result suggests that informal sector workers �nd their way into the formal sector in response

to a minimum wage increase.

25

Table 8 shows the e�ect the minimum wage has on plant-level employment. We perform

three sets of analysis by regressing on the total employment, employment for production

workers, and employment for non-production workers. The results do not show an e�ect for

minimum wage on the overall manufacturing sector employment, but they show a strong and

positive e�ect on non-production workers. According to the IS survey, the aggregate number

of workers hired for production is �ve times more than non-production. Considering this, the

results show that the positive employment e�ect on non-production workers has been evened

out by the negative employment e�ect on production workers. There is a positive relationship

between the minimum wage and the total employment when controlling for macro variables,

though it is not statistically signi�cant. The results indicate that there is no disemployment

e�ect on the manufacturing sector due to the minimum wage hike. Considering that the

manufacturing sector employment only consists of 14 percent of the total employment (IFLS),

the results from IFLS and IS do not contradict each other. Overall, the results suggest an

overall positive e�ect of the minimum wage on formal sector employment. Considering that

there has been a 112 percentage point increase in real minimum wage on average during

2000~2014, our results suggest a rather large impact of minimum wage on formal sector

employment during this time period.

These �ndings are consistent with Magruder (2013) and Hohberg and Lay (2015), who

also used IFLS data. The results di�er slightly from Harrison and Score (2010) and are

not consistent with Del Carpio et al. (2015), Comola and Mello (2011), or Ken (2015).

The works of Harrison and Score (2010) and Del Carpio (2010) use the same data and their

regression speci�cation of OLS with �rm-level �xed e�ects is identical. Using the IS data, Del

26

Carpio et al. (2010) discovered a small but statistically signi�cant negative impact on both

production and non-production employment. The results of this paper and the work of Del

Carpio et al. di�er because di�erent time periods were used in the analysis. Del Carpio et

al.'s analysis contains the years when the Indonesian economy experienced a �nancial crisis,

the demise of Suharto, and the decentralization of the bureaucratic regime. The inclusion

of this tumultuous time period could confound Del Carpio et al.'s results. Labor protection

regulations amplify the negative employment e�ect during an economic downturn. During

a crisis, when �rms contract their labor demand, the minimum wage can be higher than the

market wage, which generates the disemployment e�ect. In contrast to the Del Carpio et al.

analysis, we restrict our sample periods to when the economy stays on a steady growth phase.

This allows us to identify the impact of the minimum wage on steady-state employment. The

Indonesian economy, from 2000 to 2014, did not experience a major downturn but shows

a steady increase in gross domestic product per capita. The work of Comola and Mello

(2011) also includes a period of �nancial crisis in their analysis. Their estimation also uses

district-level aggregated data from a di�erent data source. Aggregated data is known for its

loss of information (Bertrand et al. (2004)), and is also subject to endogeneity issues when

it is applied to the minimum wage analysis as mentioned above.Following Card and Krueger

(1995)), Ken (2015) constructed fraction of formal sector workers who are a�ected by the

minimum wage; by calculating ratio of workers whose current wage is over minimum wage,

but below to the next year's minimum wage, he constructed province-speci�c real minimum

wage e�ect. However, the minimum wage has been a wage-setting mechanism in Indonesia,

and the level of minimum wage a�ects the whole wage distribution (WB (2010)), not just

workers whose wages are close to minimum wage. Ken's measure only focuses on individuals

on the minimum wage threshold, this by construction will ignore the wider implications that

the minimum wage has on the whole wage distribution.

27

Tables 9 and 10 report the e�ect of the minimum wage on the average wage, which is

largely consistent with the existing literature. Using the IFLS sample, we �nd a point esti-

mate of 1.2-1.7 percent wage increase for formal sector workers in response to a 10 percent

increase in minimum wage. Using the IS data, we observe a 2 percent wage increase when

there is a 10 percent increase in the minimum wage. In addition to the overall e�ect, it

28

is interesting to note the heterogeneous e�ect the minimum wage has on workers that are

initially paid below the minimum wage and workers who are initially paid more than the

minimum wage. We are interested in this empirical result for these two groups of workers

since the existing empirical results and theoretical predictions di�er.9 Our regression results

agree with previous empirical works. Both groups increase their wage in response to the

minimum wage, but the group initially paid below the minimum wage receives a larger per-

centage increase in pay. With the IFLS data, the coe�cient of the log minimum wage is

0.214 for the sub-minimum wage workers and 0.129 for the over-minimum wage workers. IS

data also shows the coe�cient of 0.406 for a sub-minimum wage paying plant in the initial

period, which is almost twice as large as the coe�cient for an over-minimum wage paying

plant. This larger percentage change in wage is likely to represent the initial wage gap be-

tween the two groups. This empirical result illustrates the possible role of the minimum wage

as a wage-setting mechanism in several developing countries. This wage setting mechanism

applies to both the formal sector workers and the sub-minimum wage earners. When we

conduct a regression analysis of the income of self-employed workers, we do not �nd out

statistically signi�cant e�ect. These empirical results on income infer that an important

reason of sorting into formal sector job is in wage increase in response with minimum wage.

These �ndings are consistent with the previous literature that discusses the e�ect of the

minimum wage on a monopsonistic labor market when the minimum wage regulation is as-

sumed to have perfect compliance. An important empirical �nding of the current work is that

we �nd a similar e�ect of the law under the presence of partial regulation compliance and

when a large portion of workers receive a sub-minimum wage. The overall wage distribution,

including both sub-minimum wage and above-minimum wage, responds positively to a min-

imum wage increase. Also, a positive employment e�ect is found. These �ndings necessitate

the extension of the pre-existing model that predicts the e�ect of the minimum wage on a

monopsonistic labor market when the regulation is imperfectly monitored, a large portion of

9Basu et al. (2009) predicts that if there is a punishment for violating the minimum wage law and the �ne increases inproportion to the amount of gap between minimum wage and sub-minimum wage, initially non-complying �rms will furtherreduce wage in response to minimum wage. However, empirical work often �nds that wage increases for the sub-minimum wagegroup in response to the hike of legal minimum wage (Hohberg and Lay (2015)).

29

workers receive a sub-minimum wage, and the majority of workers engage in informal sector

economic activity. Before we report the �rm's optimizing behavior, we analyze the relation

between the minimum wage and the monopsonistic margin as measured by Pigou's E.

Table 11 shows the negative relationship between the minimum wage and the monopson-

istic margin, as measured by Pigou's E with the di�erent estimation method. As explained in

Section 2, the minimum wage and other labor protection regulations might work as barriers

to �rms since they would not hire more workers due to the future uncertainty of labor costs.

However, the gap between the marginal revenue of labor and wage could be caused by the

intentional choice of monopsonistic employers. Therefore, minimum wage regulations could

work as a correcting tool. We study this hypothesis by studying the relationship between

minimum wage and Pigou's E measure. Our preferred Pigou's E measure is made using the

value-added per worker; using the various production function estimation method to calcu-

late labor productivity and Pigou's E measure should satisfy several identifying assumptions

and current estimation methods still do not reach to the consensus (see Blundell and Bond

(2002), Ackerberg et al. (2015), Wooldridge (2009)). Our estimation shows that a 10 per-

cent increase in the minimum wage is related to a 1.6-2.8 percent decrease in the Pigou's

E measure. This result is robust to the di�erent Pigou's E measures used. This suggests

that increasing the minimum wage reduces the monopsonistic margin, even with partially

monitoring.

30

3.3. Non-compliance and �rm behavior

In this section, we study how �rms responded to the unexpected minimum wage spike in

2013. We expect that �rms found ways to reduce employment costs since the regulation

was not perfectly enforced and since the �rms had bargaining power over employees. In

Table 7, it was shown that the minimum wage had a positive e�ect on part-time hiring and

hiring without paying for health insurnace. Part-time hiring allows �rms to circumvent labor

protection laws and avoid paying the minimum wage. Now we report the �rm's compliance

with the minimum wage regulation.

Table 12 reports estimation results for equation (2), which tests non-compliance practice

of medium-sized �rms (treatment group) compare to large-sized �rms (control group) in

response to unexpected minimum wage surge in 2013. The identi�cation strategy assumes

that in the absence of changes in the Indonesian minimum wage policy, compliance practice

of medium-sized �rms would have behaved similarly to the large-sized �rms. We also test

an expanded version of the regression where the treatment identi�er interacts with dummy

variables for each year. Results in the table show that in response to the minimum wage

hike in 2013, medium-sized �rms, compared against large-sized �rms, did not comply with

the minimum wage increase. The non-compliance ratio increases by 0.6-0.7 percent per

every 10 percent increase in the minimum wage. This result is robust to the inclusion of

dummy interaction terms. This �nding indicates that medium-sized �rms tend to increase

31

non-compliant behavior in the presence of the unexpected minimum wage shock.

In sum, our empirical analysis �nds a few important results. With a minimum wage hike,

1. The number of informal sector workers decreased, and full-time workers and part-

time workers increased. With plant-level data, we have evidence that employment for non-

manufacturing workers increased. Overall, the evidence shows that the minimum wage in-

duced laborers to switch into the formal sector economy.

2. Workers' wage increased. Workers with an initial sub-minimum wage increased their

wage more (in percentage) when compared to workers who initially made more than the

minimum wage. This result is consistent with studies of Indonesia and other countries.

The minimum wage increased the whole wage distribution. However, it does not show a

statistically positive e�ect on the pro�t of the self-employed. So, this result shows that the

minimum wage increase does make formal sector jobs more attractive than informal sector

jobs.

3. The increase in the minimum wage reduced the monopsonistic margin. This was robust

to the various ways of measuring Pigou's E.

4. Non-compliance increased. Medium-sized �rms tended not to pay the minimum wage

in response to the steep minimum wage hike in 2013.

5. Firms employ various ways to reduce employment cost such as (i) hiring part-time

worker (ii) not-complying to the minimum wage and (iii) not paying health insurance.

4. Equilibrium Model

We now develop a stationary equilibrium model that shows the key empirical results we

introduced in the previous section. To capture the monopsonistic behavior of the �rms,

our model extends Burdett and Mortensen (1998) by allowing for di�erences in �rm and

32

worker ability and also allows for heterogeneity in the workers' outside option. The Burdett-

Mortensen model provides a theoretical foundation for a positive employment e�ect of the

minimum wage law even when there is no dominant monopsonist in the labor market. We

combine this model with Basu et al. (2010) by introducing a sub-minimum wage payment

mechanism in order to capture the empirical observation of the wage distribution spike

near the minimum wage. The model contributes to the existing literature by introducing

a non-compliant wage o�er into the framework of Burdett-Mortensen. This generates the

empirically relevant wage distribution and the positive employment e�ect in response to a

minimum wage hike.

4.1. Environment

We study a stationary economy cast in continuous time. The measure of workers in the

labor market z is indicated by mz, whereas the measure of employers is normalized to 1.

For the following discussion and problem of the �rm, we de�ne expected earned wage, ω̃ =

ω + κmax{0, ωmin − ω} where ω is the �rm's o�ered wage. When �rms pay more than

minimum wage, the earned wage becomes ω (ω̃ = ω). When �rms pay less than the minimum

wage, the expected wage payment is the combination of the o�ered wage and the minimum

wage weighted by the punishment ratio, κ (ω̃ = (1−κ)ω+κωmin; where κ is the punishment

ratio). We see that expected punishment increases with the punishment ratio, the gap

between the minimum wage and the o�ered wage, and the employment level; κ(ωmin−ω)n(ω̃).

Note that the minimum wage increase may not a�ect the equilibrium earned wage, ω̃∗ =

ω∗+κzmax{0, ωmin−ω∗} in some labor market z if the minimum wage and its enforcement

rate are signi�cantly low. As one can see, worker's earned wage ω̃∗ = ω∗ + κzmax{0, ωmin−

ω∗} can be targeted by the �rm's optimal wage o�er ω∗; �rms can respond to the minimum

wage hike by adjusting ω∗ to target the same, ω̃∗, and thus the equilibrium earned wage may

not be a�ected by minimum wage hike at all. There will be an e�ect when the increased

minimum wage and the penalty are high enough that �rms are forced to pay κzωmin, even

33

though they can pay wages less than κzωmin, to attract workers the minimum wage has a

real e�ect. (ω̃∗ < κzωmin). We use the earned wage concept ω̃ which is distinguished from

the o�ered wage as it is the earned wage that determines a worker's occupational choice

between the formal and informal sectors of employment.

4.2. Workers

The problem for workers is a straightforward adaptation of Burdett and Mortensen (1998).

We assume that workers joining the labor market are composed of (i) current employees in

the formal sector (ii) workers in the informal sector. 10 Among employees in the formal

sector, there are workers who receive a sub-minimum wage.

Workers di�er in their permanent ability level and their opportunity cost of employment.

Worker's ability, z, is distributed as T (·) over support [z, z], andHz(x) denotes the proportion

of workers in the labor market, z, whose opportunity cost of employment, or earnings in the

informal sector, is no greater than x. We assume that a worker's opportunity cost is positively

related with their ability, so if z1 < z2, then Hz1(x) �FOSD Hz2(x). The labor market is

segmented in the sense that di�erent worker types search in separate markets while �rms can

decide which labor market to join and what wage to o�er in each market. Search is a random

process as workers do not direct their search towards speci�c �rms, and it occurs from both

informal and formal sector workers in each labor market segment. Workers maximize their

lifetime income discounted at a rate ρ.

Individuals receive job o�ers according to a Poisson process with arrival rate λsz where

s = i, e. Let λiz denote the arrival rate for the informal sector laborer, and λez be the arrival

rate for those currently working in the formal sector. We assume that the instantaneous job

arrival rate for hired workers in the formal sector are the same�either they are hired with the

legal wage or not. This means that within the speci�c z labor markets, those hired formally

and those hired �o� the book� are not segmented and they compete directly against each10When respondents answer their primary activity was searching for a job, we de�ne them as unemployed. In our sample,

unemployed workers are less than 1percent, so unlike Burdett and Mortensen, we do not assume unemployed workers.

34

other. The assumption is clearly a limitation and we employ it for reasons of tractability.

Considering that our model allows for arrival rates that vary with worker's ability, z, we do

not see the assumption as too restrictive.

Firms strategically post wage o�er ω with consideration of worker's earned wage, ω̃, as

the worker's decision rule is to compare their reservation wage and their earned wage, ω̃.We

de�ne the distribution of the �rm's expected wage payment o�er, Fz(ω̃). 11 12Formal sector

jobs will be terminated exogenously with δz ratio, or endogenously by laborers moving ahead

to the better paying formal sector jobs. Let Sz be the value function of an agent who works

in the informal sector, and Wz(ω̃) be the value function of an agent who works in the formal

sector with ω̃. The worker with ability z receives bz in case he chooses to work in the informal

sector. Then the following Bellman equations can be formulated.

(2) ρSz = bz + λiz´ ωωmax{Wz(x)− Sz, 0}dFz(x)

(3) ρWz(ω̃) = ω̃ + λez´ ωω̃(Wz(x)−Wz(ω̃))dFz(x) + δz[Sz −Wz(ω̃)]

From these equations the reservation wage can be derived as follows:

(4) Rz = bz + (λiz − λez)´ w̄Rz

1−Fz(x)ρ+δz+λez(1−Fz(x))

dx

As Wz(ω̃) is increasing in ω̃ whereas Sz is independent of it, there is a unique reservation

wage, Rz, such that Wz(ω̃) ≷ Sz as ω̃ ≷ Rz. The decision rule of agents is to become a

wage-earner in the formal sector if ω̃ > Rz; and remain self-employed if ω̃ < Rz.

Now, we de�ne the steady-state measure of the informal sector and the labor supply. To

simplify our argument, we assume λiz = λez = λz, and reservation wage of a type bz worker as

Rz = bz. Let Iz(b) denote the steady-state measure of informal sector workers with ability

11The wage package for those hired informally and formally can di�er. For example, it is often the case that formal sectorworkers receive bene�ts other than their �nancial remuneration such as insurance subsidies. We address this di�erence inbene�ts by de�ning wage as the entire monetary compensation for the worker. The wage is after tax (if it is levied) but beforesocial security deduction. Social security is considered part of their compensation as it entitles them to a pension and healthbene�ts.

12Note that allowing for wage gap transfer in the case of being monitored ensures that the optimal sub-minimum wageincreases along with the minimum wage as our data illustrates comovement of the minimum wage and the wage distribution.

35

z and outside option b. Given the reservation wage, the �ows of workers into and out of the

informal sector will be equal. At the steady state, λz(1 − Fz(b))Iz(b) = δz(1 − Iz(b)), or

Iz(b) =δz

δz+λz [1−Fz(b)]. As the density of type b is Hz(b), the steady-state measure of informal

workers willing to accept a wage o�er less than or equal to x conditional on the wage o�er

distribution Fz is

(5) Iz(x|Fz) =´ xb

(δzmz

δz+λz [1−Fz(b)]

)dHz(b)

Let the steady-state number of workers employed with a wage no greater than ω̃ be given

by Gz(ω̃)(mz − Iz), where Iz = Iz(b|Fz) is the total ratio of informal sector workers, and

Gz(ω̃) is the realized earned wage distribution of formal sector workers. At the steady-state,

the �ow of workers leaving employers o�ering a wage no greater than ω̃ equals the �ow of

workers returning to such employers,

(6) (δz + λz(1− Fz(ω̃))Gz(ω̃)(mz − Iz) = λz´ ω̃b(Fz(ω̃)− Fz(x))dIz(x|Fz)

From (5), we have [1 + kz(1 − Fz(x))]dIz(x|Fz) = mzdHz(x), we can now express (6) as

follows:

Gz(ω̃)(mz − Iz) =kz´ ωb (Fz(ω̃)−Fz(x))dIz(x|F )

(1+kz(1−Fz(ω̃))= kzmz

(1+kz(1−Fz(ω̃))2

´ ω̃b(Fz(ω̃)− Fz(x))dHz(x)

As Fz(ω̃)− Fz(x) represent the probability that an o�er received by type x is acceptable

and less than or equal to ω̃, and the density of type x is Hz(x), mz

´ ω̃b(Fz(ω̃)−Fz(x))dHz(x)

represents the expected number of workers that the �rm can hire with less than, ω̃. From

this expression, we derive mz

´ ω̃b(Fz(ω̃)− Fz(x))dHz(x) = mzHz(ω̃)Fz(ω̃).

The steady-state number of workers earning a wage in the interval [ω̃−ε, ω̃] is represented

by dGz(ω̃)(mz−Iz), while dFz(ω̃) is the measure of �rms o�ering an expected wage payment,

36

ω̃, in the same interval. Thus, the measure of workers per �rm o�ering a wage, ω̃, at the

steady state can be expressed as

(7) nz(ω̃|Fz) = (mz−Iz)dGz(ω̃)dFz(ω̃)

= kzmzHz(ω̃)(1+kz [1−Fz(ω̃)])2

This equation shows that ∂nz(ω̃|Fz)∂ω̃

> 0.

4.3. Firms

Heterogenous �rms whose productivity is p join multiple labor markets, z ∈ Z, with di�erent

wage posting strategies, considering the level of minimum wage and the enforcement rate.

Firms commit to paying a wage ωz for the remainder of the match. They operate a linear

production technology combining nz using workers of di�erent types to produce �ow output.

y(p, {nz}z∈Z) = p´z∈Z znzdz

The model assumes perfect segmentation of the labor market and perfect substitution of

di�erent ability for total production. Therefore, entrepreneurs can maximize their aggregate

pro�t by maximizing pro�t in each labor market separately.

(8) πz(ω̃) = maxω̃≥kωmin {(pz − ω̃)nz(ω̃|Fz, Hz)}

where nz(ω̃) is the labor supply at wage ω̃, given Fz and Hz. Firms are monitored

and are subject to pay a �ne if they pay less than the minimum wage and get caught by

authorities. Imperfect monitoring of the minimum wage law will create pro�table opportuni-

ties for �rms to ignore the regulations and hire uno�cially. Note that the total punishment,

κz(ωmin−ω)nz(ω̃), is proportional to the number of workers hired and the wage gap between

the minimum wage and the wage o�ered. κzωmin is the lowest noncompliant wage �rms are

37

expected to pay. In di�erent segments of the labor market, the measure of workers available

with expected wage ω̃ in labor market z is derived in equation (7) at the steady-state equi-

librium. In other words, employers decide wage to maximize (8), considering the expected

wage payment distribution, Fz(ω̃), and the proportion of workers whose opportunity cost of

employment is no greater than ω̃, Hz(ω̃).

4.4. Equilibrium

The stationary search equilibrium is a set of reservation policies functions {Rz(b)}z∈Z ; re-

ceived wage distributions {Fz(ω̃)}z∈Z ; �rm sizes {nz(ω̃)}z∈Z ; self-employment rates {Iz}z∈Z

such that given {Hz(x)}z∈Z , ωmin and κ,

1. Worker optimality: An Individual with ability z will learn ω, κ,Fz(ω̃), b, and the set of

reservation policies that solves her occupational choice.

2. Entrepreneur optimality: Taken Fz(ω̃) as given and knowing Hz(ω̃), kz, and mz, the

wage policy in each market solves the entrepreneurs' problem.

3. Labor market consistency: The self employment rates are consistent with Iz(x|Fz) =´ xb

(δ

δz+λz [1−Fz(b)]

)dHz(b).

4. Aggregation: The wage distribution in each segment of the labor market will be

determined.

4.5. Equilibrium characterization

The critical characteristics of the equilibrium wage and the employment in our model closely

follow those of Burdett and Mortensen. We feature some of the characteristics below.

Proposition 1: In the given labor market z, workers in the more productive �rms earn

higher wages than workers in the less productive �rms.

Proof : Let ω̃1 and ω̃2 be the equilibrium wage of the �rms whose productivity is p1 and p2

accordingly. Assume that p2 > p1. Then,

38

(p2z − ω̃2) kzmzHz(ω̃2)(1+kz [1−Fz(ω̃2)])2

≥ ((p2z − ω̃1) kzmzHz(ω̃1)(1+kz [1−Fz(ω̃1)])2

> (p1z − ω̃1) kzmzHz(ω̃1)(1+kz [1−Fz(ω̃1)])2

≥(p1z − ω̃2) kzmzHz(ω̃2)

(1+kz [1−Fz(ω̃2)])2

⇔ (p2 − p1)zkzmzHz(ω̃2)

(1+kz [1−Fz(ω̃2)])2> (p2 − p1)z

kzmzHz(ω̃1)(1+kz [1−Fz(ω̃1)])2

⇔ ω̃2 > ω̃1

As Mortensen (1990) proved, this property is also satis�ed for the case of continuous

productivity of employers, and there is a unique equilibrium wage associated with each

productivity type. This implies that the market distribution of wage o�ers is a transformation

of the underlying distribution of employer productivity. Now we derive the equilibrium

wage associated with the employer's productivity from the producers' pro�t maximization

problem. From

π(p, z, ω̃|H,F ) = (pz − ω̃)n(ω̃) = (pz − ω̃) kzmzHz(ω̃)[1+kz [1−FZ(ω̃)]]2

An equilibrium solution to the search and wage posting game satis�es the following �rst

order condition:

(9) (pz − ω̃)[n′(ω̃|H,F )n(ω̃|H,F )

]= (pz − ω̃)

[H′z(ω̃)[1+kz [1−Fz(ω̃)]]+2kzF ′

z(ω̃)Hz(ω̃)[1+kz [1−Fz(ω̃)]]Hz(ω̃)

]= 1

Let Jz(p) denote the proportion of employers with productivity no greater than p. Assume

Jz is continuous and di�erentiable with support [p, p]. A wage o�er distribution Fz is an

equilibrium solution to the wage posting game if and only if it satis�es (9) for all p ∈ (bzz, p]

where bz denote in�mum of the support for Hz(b). As all wage o�ers must be at least as

great as the lowest worker reservation wage, bz, only employers with productivity pz ≥ bz

can make a pro�t and participate in the labor market z. Hence without loss of generality

assume that pz ≥ bz.

Given that there is a one-on-one matching between �rm's productivity and the equilibrium

wage distribution, the proportion of workers whose opportunity cost of employment is no

39

greater than ω̃, Hz(ω̃), can be also expressed in terms of the �rm's productivity. From

Fz(ω̃(p)) = Jz(p), we can derive ω̃(p) = F−1z (Jz(p)). We substitute this into Hz(ω̃(p)), so

that Hz(ω̃(p)) == Hz(F−1z (Jz(p))) = (Hz ◦ F−1

z ◦ Jz)(p) = Qz(p). Thus, Qz(p) denotes the

proportion of workers that a �rm with productivity p can attract. Thus from Fz(ω̃(p)) =

Jz(p) and Hz(ω̃(p)) = Qz(p), we can derive

F ′z(ω̃(p))ω̃′(p) = J ′z(p) and H

′z(ω̃(p))ω̃

′(p) = Q′z(p)

Substituting these into equation (9), we get

(pz − ω̃)[H′(ω̃)[1+kz [1−Fz(ω̃)]]+2kzF ′

z(ω̃)Hz(ω̃)[1+kz [1−Fz(ω̃)]]Hz(ω̃)

]= (pz − ω̃(p))

[[1+kz [1−Jz(p)]]Q′

z(p)+2kzJ ′(p)Qz(p)[1+kz [1−Jz(p)]]Qz(p)ω̃′(p)

]= 1

This equation can be rearranged as

(pz − ω̃(p))[Q′z(p)

Qz(p)+ 2κzJ ′

z(p)1+kz(1−Jz(p))

]= ω̃′(p)

Now let us de�ne Dz(p) = −2log(1+kz(1−Jz(p))) and Sz(p) = log(Qz(p)). Thus D′z(p) =

2kzJ ′z(p)

1+kz(1−Jz(p))and S ′z(p) =

Q′z(p)

Qz(p). We can re-write the above equation as

(pz − ω̃z(p)) [S ′z(p) +D′z(p)] = ω̃′z(p)

We can de�ne Kz(p) = Sz(p) + Dz(p) so that K ′z(p) = S ′z(p) + D′z(p). Rewriting the

equation, we get

(pz − ω̃z(p))K ′z(p) = ω̃′z(p)

Multiplying the above equation with the integrating factor, µz(p) = eKz(p), on both sides

and rearranging, we get

40

[ω̃z(p)µz(p)]′ = pzµ′z(p)

Integrating both sides, we get

ω̃z(p)µz(p) = z´xµ′z(x)dx+ A

⇐⇒ ω̃z(p)eKz(p) = z

´ pbzz

xK ′z(x)eKz(x)dx+ A

From(xeKz(x)

)′= eKz(x)+xK ′z(x)e

Kz(x), we deduce´ pbzz

xK ′z(x)eKz(x)dx =

´ pbzz

[xeKz(x)

]′dx−

´ pbzz

eKz(x)dx, and thus we can rewrite the above equation as

ω̃z(p) = pz + e−Kz(p)[A− beKz(

bzz

)]− e−Kz(p)z

´ pbzz

eKz(x)dx

As ω̃z(bzz) = bz, A = beKz(

bzz

). We can re-write the above equation as

ω̃z(p) = z[p− e−Kz(p)

´ pbzz

eKz(x)dx]

ω̃z(p) = z[p− e−(Sz(p)+Dz(p))

´ pbzz

e(Sz(x)+Dz(x))dx]

(10) ω̃z(p) = z[p−´ pbzz

[1+kz(1−Jz(p))]2Qz(x)[1+kz(1−J(x))]2Qz(p)

dx]

Thus, ω̃z(p) represents the equilibrium earned wage that corresponds with the produc-

tivity level and monotonically increases with p. Now let us denote the lowest equilibrium

wage in wage distribution Fz as ω̃∗z. For the case ωmin ≤ ω̃∗z ≤ pz, all laborers are hired with

the legal wage. If ω∗z ≤ ωmin ≤ pz, some employer may hire workers with a sub-minimum

wage, whereas others hire with the legal wage. It is possible that all workers in that labor

market be hired with an illegal wage. From equation (10), the following propositions are

easily derived.

Proposition 2: Minimum wage hike increases wages in the labor market z where

ω̃∗z ≤ κzωmin

41

Proof :

∂ω̃z(p)∂ωm

=

[z[

1+kz(1−Jz(p))1+kz(1−J(κωm

z))

]2] [

Qz(κωmz )Qz(p)

]> 0

Thus, for the labor market with a binding minimum wage (ω̃∗z ≤ κzωmin), the equilibrium

wage earning distribution Fz is stochastically increasing in κωmin.

Proposition 3: Minimum wage hike increases employment of the �rms whose productivity

is greater than κzωminz

in the labor market z(p > κzωmin

z

), while it pushes out �rms from

the market whose productivity less than κzωminz

,(κzωmin

z> p). 13

Proof : From equation (7) and Proposition 2, we can deduce the following:

∂nz(ω̃)∂ωmin

= kzmzhz(ω̃)(1+kz(1−Fz(ω̃)))+2kzfz(ω̃)Hz(ω̃)(1+kz(1−Fz(ω̃)))3

∂ω̃∂ωmin

From proposition 2, we know that the minimum wage increase a�ects the whole wage

distribution in a �rst-order stochastically dominant way, thus ∂nz(ω̃)∂ωmin

> 0 for �rms whose

productivity is greater than κzωminz

(p > κzωmin

z

). This portion of increased employment is

due to the decrease in the ine�cient informal sector whose wage is less than the reser-

vation wage even though the worker's contribution to the employer's revenue exceeds the

worker's opportunity cost of employment. Note that total employment e�ect of minimum

wage is ambiguous as there is a disemployment e�ect due to pushed-out �rms. That is, from

nz(ω̃|Fz, Hz) =kzmzHz(ω̃)

(1+kz [1−Fz(ω̃)])2, we can compare the aggregated amount of employment due

to minimum wage increase. Comparing (11) with (12),

(11)´ z0

´ ω̃max1

bznzdFz(ω̃)dT (z) =

´ z0

´ ω̃max1

bz

kzmzHz(ω̃)(1+kz [1−Fz(ω̃)])2

dFz(ω̃)dT (z)

(12)´ z0

´ ω̃max2

κωminnzdFz(ω̃)dT (z) =

´ z0

´ ω̃max2

κωmin

kzmzHz(ω̃)(1+kz [1−Fz(ω̃)])2

dFz(ω̃)dT (z)

where ω̃max2 > ω̃max1 due to multiplication of minimum wage, minimum wage increase has

a positive (negative) employment e�ect if´ z

0

´ ω̃max2κωmin

nzdFz(ω̃)dT (z) ≷´ z

0

´ ω̃max1bz

nzdFz(ω̃)dT (z).

13One can look at this from the �rm's viewpoint: a �rm with productivity p will post wages in all labor market z that satisfyz > κzωmin

p. Thus, with a minimum wage increase, some �rms may not �nd workers in a certain labor market z.

42

We can also express informal sector workers in z labor market before and after minimum

wage as

(13) mz −´ ω̃max1bz

kzmzHz(ω̃)(1+kz [1−Fz(ω̃)])2

dFz(ω̃)

(14) mz −´ ω̃max2κωmin

kzmzHz(ω̃)(1+kz [1−Fz(ω̃)])2

dFz(ω̃)

and then compare accordingly.

Proposition 4: Minimum wage hike increases the non-compliance ratio among the formal

sector.

Proof : From Proposition 2, we also know that the minimum wage hike does not increase

the wage distribution by the same magnitude of the minimum wage increase.

∂ω̃z(p)∂ωm

=

[z[

1+kz(1−Jz(p))1+kz(1−Jz(κωm

z))

]2] [

Qz(κωmz )Qz(p)

]< 1

Combining with proposition 3, we can deduce that the minimum wage increase generates

a higher non-compliance ratio to the minimum wage law in the formal sector.

Proposition 5: For labor market z where the minimum wage is binding, increase in

minimum wage boosts remunerations of lower paid workers more than those of higher paid

worker.

Proof : As equation (10) establishes that ω̃z(p) monotonically increases in p, we only need

to show that an increase in wage due to a minimum wage hike decreases in p.

∂ω̃z(p)∂ωmin

∂p= −z 2κJ ′

z(p)Qz(p)[1+kz(1−Jz(p))]+[1+kz(1−Jz(p))]2

[Qz(p)]2Qz(κωmz )

[1+kz(1−Jz(κωmz

))]2 < 0

This establishes the empirical fact that a minimum wage increase a�ects initially submin-

imum wage earners more than those earning the legal wage.

43

5. Conclusion

In this paper, we analyzed the role of the minimum wage on the labor market in developing

countries where a majority of workers are involved in informal sector economic activities and

where the formal sector labor market is characterized with imperfection. The starting point

of the investigation was to study the key features of the formal and informal labor markets in

Indonesia. Similar to other developing countries, Indonesia has a substantial proportion of its

labor force involved in informal sector economic activities. The formal sector labor market

in Indonesia is imperfect because of informational frictions and the sizable positive gap

between the marginal revenue of labor and the wage payment. These features are generally

shared with other developing country labor markets. However, Indonesia may di�er from

other developing countries because of its notable heterogeneity in earned income. We use

unique historical Indonesian minimum wage data from 2000 to 2014 to conduct a regression

analysis. The results show that an increase in the minimum wage has a positive impact on

employment and on the wage. Interestingly, the wage of both the initially sub-minimum wage

paid workers and the over-minimum wage paid workers increased. Our regression results also

show a negative relationship between the minimum wage and Pigou's E, which suggests that

the minimum wage acts as a correcting mechanism in an incomplete labor market where

the employers undercut wages and protections for workers. We also �nd that employers

increased non-compliant behavior with the existing labor protection regulations in order to

compensate for the increased wage payment.

To help understand our ample empirical �ndings in a coherent manner, we construct a

structural search model in the spirit of Burdett and Mortensen (1998). The key feature

of the model is incorporating the employers' non-compliant behavior into the framework

of Burdett and Mortensen. Introducing heterogeneous �rms, worker productivity, and the

worker's outside option allows for a rich mechanism that can explain both labor supply

and demand in the formal sector. A binding minimum wage generates spillover e�ects on

the whole wage distribution, generated by monopsonistic competition. This causes i.) a

44

wage increase for both the sub-minimum and over-minimum wage workers, ii.) workers

from both the informal sector and workers in the small �rms to sort into larger sized �rms,

iii.) a reduced gap between the marginal revenue of labor and wage, and iv.) a higher

non-compliance ratio with the existing labor market regulation. Our structural model is

grounded in the documented facts of the existing labor market and also generates the labor

market responses we found in our empirical study.

Our �ndings point to interesting future work. First, our model can easily be expanded

to study the occupational choice between a formal sector entrepreneurial job and a formal

sector wage-earning job, which we have abstracted from. A large and unexpected increase

in the minimum wage gives more incentives to individuals to earn a formal sector wage-

earning job, and less incentive to become an entrepreneur in the formal sector. Including

the occupational choice of individuals into the existing model can shed additional light on

the study of labor protection regulation. Second, our study leads us to further investigate

how �rms respond to the existing labor market regulations. Our current analysis abstracts

from other labor protection regulations such as unions or severance costs, and also from the

uncertainties brought by labor market regulations. Including these factors into the existing

model can enrich our understanding of the e�ect of labor protection regulations on the �rm's

decision making and the overall economy. For instance, our current analysis on the minimum

wage suggests that while a binding minimum wage pushes less productive �rms out of the

labor market, large �rms that absorb laborers tend to hire part-time or illegally. These two

con�icting e�ects do not provide a de�nitive answer on the e�ect of the regulation on the

overall productivity. In future research, we can extend this line of analysis to other labor

market institutions, the response of �rms, and the overall economic productivity.

45

References

Ackerberg, Daniel, C. Lanier Benkard, Steven Berry, and Ariel Pakes. 2007. �Econometric

Tools for Analyzing Market Outcomes.� In Handbook of Econometrics, Vol. 6A, edited by

James Heckman and Edward Leamer, 4171�76. Amsterdam: Elsevier

Alatas, Vivi, and Lisa A. Cameron. 2009. �The Impact of Minimum Wages on

Employment in a Low-Income Country: A Quasi-Natural Experiment in Indonesia,�

Industrial and Labor Relations Review, Vol. 61, No. 2. 201�223

Basu, Arnab K., Chau, Nancy H. and Ravi Kanbur. 2010. �Turning a Blind Eye: Costly

Enforcement, Credible Committment, and Minimum Wage Laws,� Economic Journal, Vol.

120, Is. 543.

Brummund, Peter. 2012. �Variation in Monopsonistic Behavior Across Establishments:

Evidence from the Indonesian Labor Market.� Working Paper. Ithaca, NY: Cornell

University

Burdett, Kenneth and Dale T. Mortensen. 1998. �Wage Di�erentials, Employer Size, and

Unemployment,� International Economic Review, Vol. 39, No.2

Card, David and Alan B. Krueger. 1994 �Minimum Wages and Employment: A Case

Study of the Fast-Food Industry in New Jersey and Pennsylvania,� American Economic

Review, Vol. 84, No. 4: 772-93.

Cahuc, Pierre, Fabien Postel-Vinay, and Jean-Marc Robin. 2006. �Wage Bargaining with

On-the-Job Search: Theory and Evidence.� Econometrica 74 (2): 323�64.

Cunningham, Wendy. 2007 �Minimum Wages and Social Policy-Lessons from Developing

Countries.� World Bank Policy Report, 2007

Engbom, Niklas and Moser, Christian. 2017 �Earnings Inequality and the Minimum Wage:

Evidence from Brazil.� CESifo Working Paper Series No. 6393

De Soto, Hernando. 1989. The Other Path: The Economic Answer to Terrorism. New

York: Harper & Row

Gabriel Ulyssea. 2018. Firms, Informality, and Development: Theory and Evidence from

Brazil. American Economic Review 108:8, 2015-2047.

46

Harrison, Ann, and Jason Scorse. 2010. �Multinationals and Anti-sweatshop Activism,�

American Economic Review, 100(1): 247-73.

Hallward-Driemeier, M., Rijkers, B., 2013. Do crises catalyze creative destruction?

Firm-level evidence from Indonesia. Review of Econonomics and Statistics 95 (5),

1788�1810.

La Porta, Rafael, and Andrei Shleifer. 2014. �Informality and Development.� Journal of

Economic Perspectives 28 (3): 109�26.

Leal Ordóñez, Julio César. 2014. �Tax Collection, the Informal Sector, and Productivity.�

Review of Economic Dynamics 17 (2): 262�86

Levinsohn, James and Amil Petrin. 2003. �Estimating Production Functions Using Inputs

to Control for Unobservables,� Review of Economic Studies, Vol. 70.

Manning, Alan. 2003. Monopsony in Motion: Imperfect Competition in Labor Markets.

Princeton University Press; Princeton, NJ.

Meghir, Costas, Renata Narita, and Jean-Marc Robin. 2015. �Wages and Informality in

Developing Countries.� American Economic Review 105 (4): 1509�46.

Neumark, David and William Wascher, �Employment E�ects of Minimum and

Subminimum Wages: Reply to Card, Katz, and Krueger,� Industrial and Labor Relations

Review, 1994, 47, 497� 512.

van den Berg, Gerard J. and Geert Ridder, �An Empirical Equilibrium Search Model of the

Labor Market,� Econometrica, 1998, 66 (5), 1183�1221.

Olley, G.S. and Ariel Pakes. 1996. �The Dynamics of Productivity in the

Telecommunications Equipment Industry,� Econometrica, Vol. 64, No. 6.

Petrin, A. and Sivadasan, J. �Estimating Lost Output from Allocative Ine�ciency, with an

Application to Chile and Firing Costs.� Review of Economics and Statistics, Vol. 95

(2013), pp. 286�301.

Pigou, Arthur C. 1924. The economics of welfare. 2nd Edition. London: Macmillan and

Co.

Ulyssea, Gabriel. 2010. �Regulation of Entry, Labor Market Institutions and the Informal

Sector.� Journal of Development Economics 91 (1): 87�99.

47

Appendix

A. Construction of Household Asset Variable (IFLS) and Capital Variable (IS)

Minimum Wage: We use CPI to de�ate nominal minimum wages. The BPS provides

constructed CPI for di�erent cities across the country. Matching the CPIs of the capital

city with each province, we have created a CPI measure for provinces across years. We

choose 2000 as the base year.

Self-reported Income (IFLS): The self-reported income variables are annualized to be

consistent with minimum wages prescribed by law for annual wage income. IFLS data only

contains information on the total wage (monetary remuneration and other bene�ts), and

we cannot dissect monetary remuneration from other bene�ts. All these values are adjusted

by a province-level CPI published by the Indonesian Central Bureau of Statistics (BPS)

Household Asset (IFLS) is the addition of the whole di�erent kinds of asset values.

These are house occupied by this household, other house/building, non-agricultural land,

livestock/�shpond, vehicles (cars, boats, bicycles, motorbikes), household appliances (radio,

tape recorder, TV, fridge, sewing or washing machine, computer), saving/certi�cate,

receivables, Jewelry, Household furniture and utensils). There is some sample whose asset

value is missing in all categories for wife and husbands. Considering that the questionnaire

contains comprehensive items including the value of utensils, it is reasonable to assume

that those samples are misreported. We do not include those samples for our regression

analysis.

IFLS is consisted of several di�erent books and respondents sometimes choose to answer in

book 2 or book3. Unfortunately, categories of asset listed in book3 of IFLS5 is not

consistent with book2 of IFLS5 and the rest of the IFLS series. That is, it does not contain

information on several asset values which are available in the previous rounds. These are

poultry, livestock/�shpond, hard stem plant not used for farm or non-farm business,

vehicles, household appliances, furniture, and utensils. To deal with the missing

information, we impute the missing value by applying standard Oaxaca method. Since we

have information for a sample who answered in book2, we aggregate for the list of items

that are in the book3, and the list of the items not listed in book3. Using these two values,

we proceed with standard Oaxaca method, and impute values for the missing items for the

information in book3 and construct household asset value that is comparable with samples

48

who answered in book2. The value of the household assets are de�ated by province level

CPI.

Education Level (IFLS) is divided into 4 categories. 0. No education, 1. Elementary, 2.

Middle School, 3. High School, 4. University or above.

Capital (IS) is measured as the estimated value of machinery and equipment at

December 31 of the year in question. When the capital value is not reported, we use the

reported value of the capital in the previous year for constructing missing capital value.

We assume that Kit = 0.9Ki,t−1 + I i,t−1 where I is an investment for machinery and

equipment. Kit and Iit are the real value where we used price de�ator based on Wholesale

Price Indices for new machinery and equipment.

Output, Material, and Fuel (IS) are measured as the total reported value of output

produced, raw materials and fuels used by the plant during the calendar year respectively.

These were de�ated to 2000 rupiah using sector-speci�c de�ators based on Wholesale Price

Indices provided to us by Peter Brummund.

Average Total Production Workers is the average number of workers, paid and

unpaid, used per working day.

Other Total Production Workers is the average number of all others, paid and unpaid,

used per working day.

Average Total Wage was constructed as the sum of cash wages/salary and in-kind

bene�ts per production worker and per non-production workers de�ated to 2000 rupiah

using provincial consumer price index obtained from BPS.

Average Wage was constructed as the cash wages/salary and per production worker and

per non-production workers de�ated to 2000 rupiah using provincial consumer price index

obtained from BPS.

Average Other Bene�t was constructed as the total in-kind bene�t per production

worker and per non-production workers de�ated to 2000 rupiah using provincial consumer

price index obtained from BPS.

Firm Age was constructed using the di�erence between the survey year and the year the

�rm reported the �start of commercial production. (Driemeier and Rijkers (2013)

49

B. Pigou's E

B.1. Concept

We follow Brummund (2013) to study for labor market imperfection and �rm's monopsonistic

behavior. Pigou's E is a widely used measurement to capture labor market imperfection.

Under the assumption of the perfect market, �rms are expected to hire laborers under the

marginal value of the last hired one equals to wage �rm pays to the worker. If there exists

a gap, especially when marginal revenue of hiring one more labor is greater than wage

payment, then this can be an indicator for labor market imperfection. Pigou's E captures

this distortion by normalizing the gap with wage payment.

E = R′(L)−W (L)W (L)

where R(L) is the revenue function, R′(L) marginal revenue product of labor and W (L)

the inverse labor supply curve. Under the perfectly competitive labor market, Pigou's E will

equal to 0, and the index will increase as the normalized gap widens. The distortion can be

caused by monopsonistic labor market caused by lack of competition in the product market,

insu�cient information, workers' preference heterogeneity (See Robinson (1933) Card and

Krueger (1993), Burdet and Mortensen (1998), Moser and Engbom(2018) among many). On

the contrary, the gap between marginal revenue of labor and wage can be induced by the

rigid labor market regulation such that though �rms want to hire until the marginal revenue

of labor equals to wage, they are discouraged to do so due to high employment cost.

If we assume that Pigou's E shows positive sign due to monopsonistic behavior of the

�rms, we have an intuitive interpretation of Pigou's E in relation to the elasticity of the

labor supply curve,ε = WL′(W )L(W )

. This measure has been estimated in several previous works

to show evidence of a monopsonistic labor market. We can deduce the relationship based on

the �rm's optimization behavior where the marginal revenue of labor is equal to the marginal

cost of payment. R′(L) = W (L) +W ′(L)L.

E = R′(L)−W (L)W (L)

= W ′(L)LW (L)

= ε−1

50

One can see that when the elasticity of the labor supply curve approaches in�nity (per-

fectly elastic) that Pigou's E approaches zero, and �rm do not have monopsonistic market

power.

B.2. Calculating Pigou's E

We use the existing production function estimation method to calculate marginal revenue

product of labor, and then directly obtain Pigou's E. There are vibrant research on the

production function estimation using plant-level panel data. Following Petrin and Sivadasan

(2013), we apply Wooldridge-LP method for our main estimation method as the method not

only address the simultaneity problem from Levinsohn-Petrin (2003) and Olley and Pakes

(1996), it also addresses the multicollinearity issue pointed by Ackerberg, Caves and Fraser

(2015); as we are interested in investigating endogenous decision of labor input in relation

with minimum wage regulation, we cannot use OP or LP method where the method does

not allow for �rms to make endogenous labor choice. Also, though the method assumes

unobserved productivity to be scalar, still it has a more robust assumption on unobserved

productivity as it assumes unobserved productivity to follow Markov process, not the linear

AR(1) process presumed by System GMM approach (Blundell and Bond). Another advan-

tage of using Wooldridge-LP method is when we have variables such as capital that does

not vary much; by utilizing moment conditions from di�erenced equations, System GMM

approach occasionally generates an unreasonably low coe�cient on capital (sometimes even

negative) when the use of capital does not vary much. In contrast, as Wooldridge-LP GMM

method do not form moment conditions from di�erenced data, we can get a more reason-

able coe�cient on capital. Also, as the Wooldridge-Lp set up only uses one-step estimation

procedure to earn coe�cients, it is easy to attain fully robust standard errors and it is more

e�cient than the two-stage estimation procedures (OP, LP, ACF).

Now we describe the Wooldridge-LP approach by positing Cobb-Douglas production func-

tion:

yit = βslmit + βul

oit + βkkit + βmmit + βeeit + εit

51

where yit is the log of real output, lmit is the log of number of manufacture employee, l

oit is

the log of number of other employee, mit is the log of real value of intermediate materials,

eit is the log of of fuels used and error term εit is assumed equal to:

εit = ωit + ηit

with ωit the transmitted component of the �rm-speci�c productivity shock that is un-

observed by econometricians and causes endogeneity, and ηit representing the �rm-speci�c

i.i.d. productivity shock or measurement errors.

Following Petrin and Sivadasan (2013), we assume labor as state variable due to rigid

labor protection regulation including minimum wage, and following LP, we assume mit as of

the proxy variable:

ωit = g(xit,mit), where xit = {lit, lit, kit}

Assumption 1. Strict monotonicity condition, mit = f(ωit, xit), can be inverted such that

ωit = f−1(xit,mit) = g(xit,mit). ωit is a function of the state variables and the proxy

variable (material)

Assumption 2. Unobserved productivity follows the Markov process, ωit = ωi,t−1 + ait

where ait is i.i.d. innovation.

Assumption 3. Current Productivity shock, ait = ωit − E(ωit|ωi,t−1), is uncorrelated with

the current state variables.

Assumption 4. Lagged state and proxy variables are uncorrelated with current productivity

shock

Under these assumptions, we can deduce E[ωit|xit−1, ei,t−1,mi,t−1, ...., ei1, x1,m1] = E[ωit|ωi,t−1] =

h(g(ki,t−1, lui,t−1, l

si,t−1, mi,t−1)), which then can be used to re-write the above equation as

yit = βslmit + βul

oit + βkkit + βmmit + βeeit + h(g(ki,t−1, l

ui,t−1, l

si,t−1, mi,t−1)) + ait + ηit

Following Petrin and Sivadasan (2013), we approximate h(g(ki,t−1, lui,t−1, l

si,t−1, mi,t−1))

with second order polynomial. As for instruments, we use �rst and second lag of fuel and

second order lags of manufacturing labor and other labor. We estimate the production

function separately by two-digit industry.

52

The coe�cient estimates are summarized in the Table B1. These coe�cient estimates

appear reasonable, with materials and production workers having the highest coe�cients,

followed by other workers, capital and then fuels. Coe�cients on capital, which can be

unreasonably low in some �xed estimates are positive, though we have several negative

coe�cients on fuels. With these industry-speci�c estimates for the parameters of the Cobb-

Douglas production function, we generate �rm-year speci�c measures for the average revenue

product of each �rm, which then be used to calculate Pigou's E. Table 4 is the summary

result of our calculation.

C. Robustness Check: Migration

We also report the estimation results without sample who migrated to di�erent province from

the original province they began their sample. Migrating population could skew statistics

53

if informally employed individuals crossed the provinces to search for higher-paying formal

sector jobs or if unemployed workers migrate out of higher minimum wage provinces to

search for jobs. We conduct robustness analysis excluding samples who migrated from the

initial place where observation began. This analysis will allow us to look at how much our

estimation could be contaminated by migration. The migrating population is approximately

1.6 percent for our 3 rounds of the IFLS sample. The positive e�ect on the formal sector

employment slightly decreases when we exclude the migrated population. This is evidence

that some people migrated into the provinces with a higher minimum wage to earn a formal

sector job. We also see the same pattern with the wage regression. This can be more evidence

that a higher minimum wage improves the labor market environment, as it attracts a migrant

population. Also, even without the migration sample, our results are largely consistent with

our main results.

54

55