Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level...

60
Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG 1 April 10, 2010 Abstract: This paper uses Indonesian manufacturing census data from 1991-2001 to examine the impact of the East Asian crisis on the efficiency of resource allocation. The hypothesis that crises may have a silver lining by improving resource allocation is rejected. There was large excess churning; gross job creation and destruction far exceeded the amount required to achieve net employment reallocation and the correlation between productivity and employment growth weakened. Moreover, firm productivity became a less important determinant of survival, contradicting the idea that crises accelerate creative destruction by ―cleansing‖ out less productive firms. Decomposing aggregate productivity growth reveals within firm adjustment dominated, rather than reallocation between firms. Credit market imperfections could account for the attenuated relationship between productivity and survival. Indeed, firms more exposed to credit market volatility exited more and shed labor, but amongst them, relatively productive firms were better capable of weathering the shock; the attenuation of the link between productivity and exit was greatest for firms that lacked access to credit to start with. There does not, however, appear to be long term scarring in the process of creative destruction. The link between productivity and survival is restored post-crisis, although the link with employment growth remains less strong. Key words: Financial crisis, creative destruction, firm survival, productivity decompositions, capital market imperfections JEL Codes: G01, O47, J21, D21 1 We would like to thank Andrew Waxman for exceptional assistance in the preparation of the dataset and paper. We would also like to thank Luis Serven, Ana Fernandes, Fitria Fitrani, Leo Iacovone, Kai Kaiser, Alex Korns, David Newhouse, Rifa Rufiadi and William Wallace for their assistance in obtaining the data and in discussing the policy environment in Indonesia. We also thank the participants of the Economist Forum and a DECRG Macroeconomics and Growth seminar, particularly Pierella Paci and John Giles, for their useful comments. Jagadeesh Sivadasan kindly shared the code used to implement the Ackerberg Caves Frazer procedure. The views expressed here are those of the authors and do not necessarily represent the views of the World Bank, its Executive Board or member countries. Corresponding author: Mary Hallward-Driemeier, [email protected].

Transcript of Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level...

Page 1: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

Do Crises Catalyze Creative Destruction?

Firm-level Evidence from Indonesia

Mary Hallward-Driemeier, DECRG

Bob Rijkers, DECRG1

April 10, 2010

Abstract: This paper uses Indonesian manufacturing census data from 1991-2001 to examine

the impact of the East Asian crisis on the efficiency of resource allocation. The hypothesis that

crises may have a silver lining by improving resource allocation is rejected. There was large

excess churning; gross job creation and destruction far exceeded the amount required to achieve

net employment reallocation and the correlation between productivity and employment growth

weakened. Moreover, firm productivity became a less important determinant of survival,

contradicting the idea that crises accelerate creative destruction by ―cleansing‖ out less

productive firms. Decomposing aggregate productivity growth reveals within firm adjustment

dominated, rather than reallocation between firms. Credit market imperfections could account

for the attenuated relationship between productivity and survival. Indeed, firms more exposed

to credit market volatility exited more and shed labor, but amongst them, relatively productive

firms were better capable of weathering the shock; the attenuation of the link between

productivity and exit was greatest for firms that lacked access to credit to start with. There does

not, however, appear to be long term scarring in the process of creative destruction. The link

between productivity and survival is restored post-crisis, although the link with employment

growth remains less strong.

Key words: Financial crisis, creative destruction, firm survival, productivity decompositions,

capital market imperfections

JEL Codes: G01, O47, J21, D21

1 We would like to thank Andrew Waxman for exceptional assistance in the preparation of the dataset and paper. We would also like to thank Luis Serven, Ana Fernandes, Fitria Fitrani, Leo Iacovone, Kai Kaiser, Alex Korns, David Newhouse, Rifa Rufiadi and William Wallace for their assistance in obtaining the data and in discussing the policy environment in Indonesia. We also thank the participants of the Economist Forum and a DECRG Macroeconomics and Growth seminar, particularly Pierella Paci and John Giles, for their useful comments. Jagadeesh Sivadasan kindly shared the code used to implement the Ackerberg Caves Frazer procedure. The views expressed here are those of the authors and do not necessarily represent the views of the World Bank, its Executive Board or member countries. Corresponding author: Mary Hallward-Driemeier, [email protected].

Page 2: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

2

1. Introduction

Firm dynamics and resource allocation are increasingly recognized to be crucial determinants of

countries‘ comparative economic success and long-run productivity growth. Cross-country studies

based on aggregate data furthermore suggest that plant-dynamics are also key determinants of the

depth and duration of crises (Bergoeing et al. ,2004, Collier and Goderis, 2009) Yet, while crises are

periods of intensified adjustment, firm-level evidence on the impact of crises on resource allocation

is limited. Moreover, their impact is theoretically uncertain. The idea that crises may accelerate the

Schumpetarian (1939) process of creative destruction by ―cleansing‖ out unproductive arrangements

and freeing up resources for more productive uses features prominently in macro-models (see e.g.

Hall, 1995, Caballero and Hammour, 1994, 1999 and Gomes et al., 1997). On the other hand, some

recent papers suggest that, rather than being cleansing, crises ―scar‖ the economy and undermine

long-run productivity growth by exacerbating market imperfections and destroying productive firms

(see Barlevy (2002)).2 Which view is correct matters for policy, since at issue is whether or not there

is a tradeoff between minimizing the short-term impact of crises and maximizing long-run growth

prospects; if crises are cleansing policies to dampen short-term impacts may obstruct long-run

recovery and be not only costly, but also counterproductive. By contrast, if crises are scarring,

policies to minimize short-term impacts are consistent with maximizing long-run growth prospects

(see Paci et al., 2009).

Given these competing paradigms, empirical analyses become all the more important in

determining whether there is evidence of market imperfections and whether their effects are

exacerbated during crises such that the cleansing hypothesis is overturned. By examining how the

East Asian crisis affected manufacturing firm dynamics in Indonesia this paper attempts to

empirically discriminate between the cleansing and scarring hypotheses and to redress the lacuna in

the empirical evidence on the impact of crises on firm dynamics. The empirical basis for our analysis

is the Indonesian manufacturing census from 1991-2001, a uniquely detailed firm-level panel dataset

that spans the periods before, during and after the East Asian crisis. We provide both aggregate and

firm-level evidence to test for the effects of the crisis on the efficiency of resource allocation and the

reallocation process itself. We examine job flows and assess how the relative contributions of entry,

exit, and reallocation between and within firms to aggregate productivity growth change during and

2 In addition, Ouyang (2009) has argued that crises may undermine long-run productivity growth by destroying potentially productive firms during their infancy.

Page 3: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

3

after the crisis. These decompositions are complemented by firm-level analysis of exit and

employment growth patterns, to assess whether and how firm dynamics differed during the crisis

compared to pre- and post-crisis periods. The firm-level analysis enables us to disentangle the role of

productivity from that of other firm-characteristics, such as size, ownership, export-orientation and

sector. In addition, we explore whether firm dynamics were affected by exacerbating credit market

imperfections, by looking at how the impact of the crisis varied across sectors varying in terms of

their dependence on external finance, and within sectors depending on whether or not firms used

loans to finance investments. Furthermore, we examine whether crisis-induced changes in the

reallocative process persist during the recovery.

Previous work on allocative efficiency has largely avoided periods of crisis, and work on

crises has rarely examined firm responses. By using firm-level data to examine the impact of a major

crisis on resource allocation this paper thus combines and contributes to several strands of literature

that have evolved fairly separately up until this point. The available evidence on the impact of crises

on resource allocation typically relies on aggregate data. The few previous attempts to examine the

cleansing hypothesis using firm-level data – discussed in detail in section 2 - have either failed to

examine changes in the reallocative process itself or suffer methodological shortcomings. Second,

existing plant-level studies of reallocation dynamics have demonstrated that business cycles are

important determinants of both the pattern and pace of reallocation (see surveys by Bartelsman and

Doms, 2000, Caves, 1998, and Syverson, 2010) but typically deliberately exclude crisis-periods. It is

therefore not clear to what extent the conclusions based on them generalize to crisis times. Thirdly,

micro-studies of the impact of crises on labor market outcomes mostly rely on household and labor

market data (see e.g. McKenzie 2003, 2004; Fallon and Lucas, 2002; Manning, 2000, Beegle et al.

1999). While such data are useful for documenting who becomes unemployed and how much their

earnings decrease, these types of data are ill-suited for examining the impact of crises on labor

demand, which requires understanding how productively labor can be employed – and whether

shifts are coming from changes in net entry or incumbents adjusting. Moreover, by enabling us to

document and analyze heterogeneity in firm vulnerability, establishment-level data can help us assess

which workers were likely to be most impacted and how.

Examining the impact of the East Asian crisis on the Indonesian manufacturing sector

provides a very interesting case study of how financial crises reverberate through the real economy.

Strong manufacturing performance had been an important driver of growth until the onset of the

crisis and manufacturing was one of the hardest hit sectors. Examining the mechanisms by which

Page 4: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

4

the financial crisis affected manufacturing firms helps highlight transmission mechanisms likely to be

relevant in the current global crisis, which is also characterized by a sharp contraction of demand,

reduced access to credit, and uncertainty. However, the current crisis is not characterized by

depreciations of a magnitude similar to those observed during the East Asian crisis and, is global,

rather than regional in nature. Focusing on Indonesia is particularly revealing as the size and

archipelagic geography of Indonesia provides us with cross-sectional and temporal variation which

we can exploit in our econometric strategy. In addition, Indonesia has a uniquely rich longitudinal

census of manufacturing firms that allows us to examine these issues.

Our main findings can be summarized as follows: the data do not support the cleansing

hypothesis, which predicts that the link between productivity and survival should become stronger

during the crisis, and that reallocation should occur between rather than within firms (i.e. that the

crisis would hurt relatively unproductive firms, disproportionately). On the contrary, they reveal that

the dominant mode of adjustment was within firms, rather than across firms. While the crisis led to

a spike in exit and induced excessive employment reallocation, productivity was only weakly

correlated with survival and employment growth during the crisis. Rather than the crisis raising the

productivity threshold for survival, it was more indiscriminate in terms of the productivity of firms

driven out of business. Firms more exposed to changing credit market conditions were more likely

to exit and shed labor, but amongst them, relatively productive firms were better capable of

weathering the shock. Thus, the attenuation of the association between productivity and exit was

greatest for firms that lacked access to credit to start with. Overall, the crisis was clearly destructive

in the short-run. During the recovery, aggregate allocative efficiency improved somewhat, as did

entrants‘ productivity. The protective power of productivity against exit was restored, but the

correlation between productivity and employment growth remained weaker than it had been pre-

crisis.

The remainder of this paper is organized as follows. The next section reviews the related

literature and elaborates on our approach. Section three describes the data and the context.

Decompositions of aggregate productivity growth and job flows are presented in section four, while

section five presents firm survival and employment growth models. A final section concludes.

2. Relation to Existing Literature and Approach

Page 5: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

5

Before explaining how we aim to empirically discriminate between the cleansing and scarring

hypotheses, we briefly review the theoretical models that underpin the idea that crises improve the

efficient allocation of resources as well as empirical evidence for this hypothesis. In addition, we

discuss recent theoretical and empirical challenges to the cleansing hypothesis and review the

existing firm-level evidence on the impact of crises on resource allocation.

2.1 Related Literature

According to Schumpeter (1939), business cycles are driven by a process of creative destruction by

which innovative, high-productivity firms drive relatively unproductive firms out of business. Some

prominent macro-models predict that recessions may speed up this process by ―cleansing‖ out

unproductive firms and freeing up resources for more productive uses (see e.g. Hall, 1995, Caballero

and Hammour, 1994, 1999; and Gomes et al., 1997). The precise mechanism can vary, but the idea

is that the additional competitive pressure caused by crises facilitates efficiency enhancing

reallocation. For example, firms will fire the least productive employees first and may be able to

attract more highly skilled workers to their firms as the number of applicants rises; banks may

allocate credit more efficiently as a result of increased scrutiny and competition; labor unions may be

more willing to accept employment losses or wage cuts. To be clear, these models do not predict

that crises will enhance aggregate welfare or suggest they are inherently desirable, but rather suggest

that, by improving the efficiency of resource allocation, crises may have a silver lining.

Empirical work using firm-level data supports the general claim of the importance of

creative destruction in improving productivity. Indeed, the increased availability of firm-level micro-

data has led to a burgeoning body of evidence on the microeconomic determinants of growth. There

is now rich empirical support for the idea that differences in the allocation of resources across

establishments that differ in productivity may be an important factor not only in accounting for

cross-country differences in output per worker, but also in differences in long-run growth (see e.g.

Restuccia and Rogerson, 2008; Hsieh and Klenow, 2009).

Although they typically exclude crisis periods, these firm-level studies raise questions about

how the process of creative destruction varies over time, with more ambiguous results regarding

whether the cleansing effect holds, even in periods of mild downturns. Studies of reallocation and

productivity growth that use panel data on manufacturing firms (surveyed in Bartelsman and Doms,

2000 and Caves, 1998, and Syverson, 2010) indicate that the business cycle is an important

Page 6: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

6

determinant of both the pattern and the pace of aggregate productivity growth. However,

decompositions of aggregate productivity growth only weakly support the hypothesis that allocative

efficiency increases during normal cyclical downturns (see, inter alia, Griliches and Regev, 1995) and

Baily et al. (1992), who demonstrate that within-firm adjustment was the major driver of

productivity growth in Israeli and U.S. manufacturing industries).3 Since such studies typically

exclude extreme economic events, it is uncertain to what extent lessons based on them apply during

times of crisis. Moreover, it is not clear to what extent conclusions based on developed country

datasets generalize to developing country contexts (see e.g. Bartelsman, Haltiwanger and Scarpetta

(2004) and Loayza et al. (2005) for empirical evidence of differences in the reallocation process

across developed and developing countries).

This microeconomic evidence that challenges the view that downturns can be cleansing is

supported by theoretical models that demonstrate that, in the presence of market imperfections,

downturns may hamper, rather than facilitate, adjustments. Barlevy (2002) for example, has pointed

out that crises may well obstruct the process of creative destruction by exacerbating labor and credit

market imperfections, whereby the expected efficiency-enhancing adjustment that is supposed to be

facilitated does not materialize. He argues that credit market imperfections are more likely to be

binding for relatively efficient producers, since highly efficient production arrangements are more

vulnerable to financing constraints as a result of their typically higher fixed costs. Crisis-induced

exacerbation of credit constraints would therefore hurt efficient firms disproportionately,

contradicting the creative destruction hypothesis. In addition, crises may increase labor market

frictions by augmenting search costs, which lead to lower average worker-firm match quality.

Moreover, these models have shown that the efficiency of surviving jobs deteriorates, leading to a

sullying effect of recessions. Moreover, Ouyang (2009) has pointed out that crises may undermine

long-run productivity growth by destroying potentially productive firms during their infancy.

In view of the competing predictions generated by the cleansing and scarring hypotheses,

empirical evidence becomes all the more important in determining which view is correct. Cross-

country studies on the impact of crises and shocks on long-run growth using aggregate data

furthermore suggest that (policy induced) distortions and market imperfections are a crucial

determinant of both the short- and long-run impact of crises on productivity. Bergoeing et al.

3 Whether policy reforms that strengthen competitive pressures also serve to improve resource allocation has also been debated (Hallward-Driemeier and Thompson, 2009).

Page 7: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

7

(2004) have demonstrated that countries with more distortionary regulations4 have experienced more

severe recessions. Collier and Goderis (2009) assess how developing countries‘ ability to cope with

aggregate shocks varies depends on which structural policies they have implemented. They consider

a wide variety of policies—including those pertaining to trade, financial deepening, and labor

markets-and conclude that regulations that delay the speed of firm closure are the most important

determinant of short-term growth losses from adverse price-shocks in mineral exporting countries,

while labor market regulations play an important role in mitigating shocks due to natural disasters.

In addition, while not based on crisis periods, there is a burgeoning literature on the impact of the

investment climate on firm performance that provides empirical support for the idea that market

imperfections hamper growth. Hallward-Driemeier (2009), for example, shows that red tape,

corruption, cronyism and weak property rights may attenuate the link between productivity and exit

and thereby undermine the Schumpetarian process of creative destruction.

While firm-level evidence on the impact of crises on resource allocation is scant, there are a

few studies that shed light on the debate. Liu and Tybout (1996) compare the performance of

continuing plants and exiting plants in Chile and Colombia from 1980-1985 but find no evidence for

systematic covariance of the efficiency gap between these two groups of firms over the business

cycle in either country, even though Chile suffered a recession in 1992. Exit rates of Chilean firms

only increased modestly during the recession. Casacuberta and Gandelsman (2009) examine the

impact of the 2002 banking crises in Uruguay on resource allocation. Even during the crisis,

productivity was negatively correlated with exit. Although it is not the direct focus of their paper,

their estimates suggest that the crisis attenuated the link between productivity and exit somewhat.

However, their estimation strategy makes it difficult to examine whether or not the link between

productivity and exit during the crisis was significantly different from what it was before the crisis.

In addition, they do not examine the relationship between productivity and employment

reallocation. Nishimura, et al. (2005) use a cohort analysis comparing productivity of entrants and

survivors to examine the impact of the Japanese recession on productivity growth and find that the

1996/97 banking crisis induced the exit of some relatively efficient firms among the youngest

cohorts. However, since their analysis essentially amounts to a comparison of means, it could suffer

from omitted variable bias; it is not clear whether the recession attenuated the link between

4 In terms of financial restrictions, trade barriers, firm entry costs, inefficient bankruptcy procedures, bureaucratic red tape, tax burden and labor regulations.

Page 8: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

8

productivity and exit, per se, or whether the above average productivity of young exiting firms was

associated with other factors. Nevertheless, their result is indicative the crises may not be cleansing.

A plausible explanation for the findings of Nishimura, et al. (2005) is that, consistent with

the scarring hypothesis, crises exacerbate credit market imperfections. Peek and Rosengren (2005)

demonstrate that Japanese banks levied additional credit to the weakest firms in order to avoid

balance sheet losses (also see Okada and Horioka, 2007). Empirical evidence supporting the idea

that credit market imperfections are important determinants of the reallocation process during

economic crises is provided by Gallego and Tessada (2009) who analyze job flows in Latin America

in response to sudden stops and demonstrate that sectors more dependent on external finance are

more heavily affected by the creation margin, while liquidity needs affect the destruction margin. In

addition, they found a negative correlation between firing and dismissal costs and labor destruction.

Moreover, using the same dataset as the one considered in this paper, Blalock, Gertler and Levine

(2008) compare the performance of foreign owned and domestically owned exporters in order to

test for the existence of liquidity constraints. The superior performance of foreign-owned firms,

which are likely to have had better access to finance, suggests that they were better capable of taking

advantage of the depreciation of the rupiah. This evidence of the existence of credit constraints

further raises the possibility that the potential for reallocation in post-crisis Indonesia could be

limited. However, crises need not always exacerbate credit market imperfections; Borensztein and

Lee (2002) find evidence that in response to the East Asian crisis Korean banks reallocated credit

from conglomerate (chaebol) firms to relatively more efficient firms.

Further evidence for the idea that market imperfections have particularly pernicious

consequences during times of crisis, which is the crucial assumption underlying the scarring

hypothesis, is that jobs created during recessions are usually less productive and less likely to last (see

Bowlus, 1993; Barlevy, 1998, Davis, Haltiwanger and Schuh, 1996). Fallon and Lucas (2002)

furthermore argue that crises are typically associated with excess churning of workers and jobs.

In short, theoretical models yield competing predictions regarding the impact of crises on

resource allocation, whilst the empirical evidence on the impact of crises on firm dynamics is limited

and ambiguous. Discriminating between the cleansing and scarring hypotheses requires firm-level

data as it requires one to assess the link between productivity and exit. There are only a handful of

studies that have used firm-level data to examine how resource allocation evolves during a crisis, yet

these studies have not tackled the scarring hypothesis directly, nor offered a comprehensive

Page 9: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

9

assessment that relates microeconometric-evidence on the link between productivity, employment

growth and survival to macro-trends in productivity growth.

2.2 Hypotheses and Approach

The cleansing and scarring hypotheses yield competing testable predictions at both the macro- and

the micro-level. If the cleansing view is correct, one would expect crises to accelerate the weeding

out of unproductive firms, resulting in a stronger association between productivity and survival at

the micro-level; in other words, one would expect unproductive firms to be disproportionately

affected. One might furthermore anticipate the correlation between firm‘ productivity and

employment growth to strengthen, as one would expect less productive firms to contract more in

response to shocks. At the macro-level, one would expect to see a corresponding increase in the

contribution of exit to aggregate productivity growth as well as stronger correlations between

productivity and changes in market share. By contrast, if crises are scarring, one would anticipate the

efficiency of resource allocation to deteriorate, and the link between productivity, exit and

employment growth to attenuate, undermining aggregate allocative efficiency. To the extent that

these scarring effects arise because of increased credit market imperfections, one might expect firms

more reliant on finance to be more severely affected by the crisis.

To test these competing hypotheses we combine analysis of aggregate reallocation patterns

with firm-level analysis and examine how reallocation dynamics evolved over time. Macro-level

decompositions of aggregate productivity growth are used to examine to what extent aggregate

productivity growth was due to adjustment within firms, reallocation between surviving firms and

changes in average productivity due to entry and exit. The decompositions help us assess whether

crises catalyze or retard efficiency enhancing re-allocation, and to analyze how industry dynamics

during crises differ from pre- and post crisis dynamics. Of particular interest are the contributions

due to exit and reallocation between surviving firms, both of which the cleansing hypothesis predicts

will rise.

These decompositions are complemented with firm-level analysis of survival and

employment growth. Such analysis i) helps us understand what is driving aggregate reallocation

patterns ii) enables us to document the heterogeneity in firm behavior (and what is driving this) and

to isolate the role of productivity from the impact of other firms characteristics and iii) helps explore

different dimensions of adjustment. Using a difference-in-difference approach, we offer a direct test

Page 10: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

10

of the cleansing hypothesis by examining whether or not the crisis induced changes in the

relationship between productivity, survival and employment growth. Moreover, we distinguish

between the short-run and medium-run impacts of crises on the reallocation process and examine

whether crisis induced changes in the reallocation process persist during the recovery. In addition,

we offer an indirect test of the scarring hypothesis by examining whether firms more reliant on

finance were hit harder by the crisis and whether exit and employment growth were less strongly

correlated with productivity for such firms; we compare the performance of firms across sectors that

vary in their dependence on external credit, and within sectors, we exploit information on the

composition of investment finance to examine whether there are differential impacts of the crisis.

3. Data and Context

3.1 Data

The Indonesian Manufacturing Census (1991-2001) provides the empirical basis for our analysis. It

is uniquely well-suited for a detailed examination of the impact of the East Asian crisis on firm

behavior since it contains information on all Indonesian manufacturing establishments with more

than 20 employees and spans the pre- and post- crisis periods, as well as the crisis itself. Moreover, it

has very detailed information on employment, inputs and outputs, industrial classification, exporting,

ownership and investment behavior. To examine how dynamics differ across sectors, we augment

the data with industry-level measures of financial dependence, employment turnover, the natural rate

of establishment entry and other salient sector characteristics.

As the data is an annual census of manufacturing establishments,6 we can not only track

performance over time, but also identify entry and exit.5 Our definitions of entry and exit are

affected by the fact that only establishments with 20 or more employees are included in the survey.

Entry is defined as entry into the survey. it is when establishments cross the 20 employee threshold,

not necessarily when they began operations. As establishments are not required to report their

closure, exit is inferred from establishments ceasing to file reports to BPS. We do not count as exits

5 Since our data track establishment, rather than firm behavior, we are unable to tell whether the behavior of one establishment relative to another represents a within-, rather than between-, firm reallocation. BPS submits a separate questionnaire to the head office for multi-establishment firms, but these data were unavailable for this study. A recent BPS study has suggested that less than 5% of establishments in the Manufacturing Census are owned by a multi-establishment firm (see Blalock, Gertler, Levine 2008 for discussion). For this reason, we believe that our results for establishments are likely to generalize to firms and we will use the terms ―firms‖ and ―establishments‘ interchangeably in the remainder of the text.

Page 11: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

11

temporary lapses in reporting. Exit is defined as exit from the survey; establishments that exit need

not discontinue their operations; they may simply be dropping below the 20 employee threshold.

This paper considers the period 1991-2001. Additional years could be available covering the

1980s. However, the sampling procedures changed during this period, with procedures more

rigorous and comparable over time beginning in 1991. As we are interested in understanding the

crisis period, this still provides adequate time pre-crisis for our comparisons. We extended the data

through 2001 to include a sufficient period post-crisis.6

There are two particular issues regarding response rates that need to be addressed. The first

regards a spike in exits in 2001, which coincides with the year in which administrative boundaries of

provinces changed as the decentralization laws passed after Suharto was removed from power came

into effect.7 Aggregate data on manufacturing GDP growth presented in figure 1 suggest that there

was a small downturn in GDP growth in 2001. However, this minor downturn is unlikely to account

for the jump we see. Coupled with lower than average exit rates in the preceding two years, it is

possible that some firms were ‗zombies,‘ that were carried forward for a year or two, with such a

practice then stopped in 2001. Discussions with officials in BPS indicate this may be part of the

explanation. The issue is whether the timing for some of these firms‘ exit should have been in 1999

or 2000 – and how this may affect our results.

The potential presence of such zombie firms in 1999 and 2000 likely makes it harder for us

to reject the cleansing hypothesis. There are two potentialtypes of zombie firms. The first type,

which we identify, are firms that only partially report their performance, effectively dropping out of

our sample as it is not possible to calculate their productivity. However, if we move up their date of

exit to the last year of complete reporting, it turns out these were relatively productive firms (see the

Appendix). Not including these firms as exiters would thus make it easier to find evidence consistent

with the cleansing hypothesis. The second possible type, which we cannot identify but may exist, do

report performance indicators but they may really only be achieving these on paper. To the extent

they are reporting productivity higher than their performance, they too would bias results in favor of

the cleansing hypothesis. Their inclusion among incumbents overestimates the productivity of

incumbents during the crisis and their existence during the recovery would make it appear that more

6 We also have data post 2001, but using these to compare firm dynamics with the crisis period is difficult because of changes in firm identifiers, changing definitions of key explanatory variables, and province splits that results in all the incumbents firms ‗exiting‘ from the old province and ‗entering‘ the new one. 7 We have worked with BPS to acquire the bridging concordance of id numbers. A number of provinces also divided then and it is possible that some firms are erroneously marked as exiting in one province and entering in another. This is a problem in later years, but seems unlikely to be of major significance in 2001. See the appendix for more details.

Page 12: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

12

productive firms are exiting then – the opposite of what we find. To test the robustness of our

results we also expand the window for analyzing firm survival so that the specific timing of exit is

not an issue, e.g. look at the probability of firms that operated in 1996 surviving to 2001.

A second issue regarding response rates pertains to the reporting of information on the

capital stock, for which our preferred proxy is the replacement value of machinery and equipment.

In 1996, the year immediately prior to the onset of the crisis, information on the capital stock was

not collected. In other years, measures of capital were included, although it is missing for some

firms in some years.8 We attempt to redress this issue by imputing the capital stock for firms for

which it is missing. To examine the robustness of our results, we present results with and without

controlling for the capital stock and examine how sensitive our results are to different proxies for

capital.

Our preferred measure of productivity is value-added per worker, which is not sensitive to the

imputation of the capital stock and robust to the endogeneity concerns that plague Total Factor

Productivity (TFP) estimation.9 Nevertheless, we examine how robust our results are to using

measures of TFP as proxies for productivity instead and use a variety of methods, including the

Solow method and the recent procedure proposed by Ackerberg, Caves and Frazer (2005) to

compute TFP. See the Appendices for more information on the construction of our data, key

explanatory variables and our productivity estimates.

3.2 Context: the Indonesian Crisis

The early 1990s saw Indonesia at the height of an extended period of industrialization and economic

growth under the ―New Order‖ government of Suharto, with labor-intensive exports being an

important driver of economic growth. A substantial reduction in trade protection and barriers to

foreign investment combined with effective macroeconomic and exchange rate management and a

flexible labor market helped fuel the growth of the manufacturing sector (Dwor-Frecault et al.,

1999).

In the wake of the devaluation of the Thai Baht, flows of foreign capital shrank drastically

and investors pulled resources out of the country as speculation against the rupiah mounted.

8 By contrast, all firms report labor, output and value added without breaks. 9 Since we do not observe firm-level prices, our productivity measures are revenue, rather than output based (see Foster, Haltiwanger, and Syverson (2008) for an illuminating comparison of revenue-based and output-based productivity measures and a discussion of the possible caveats of using revenue-based productivity measures).

Page 13: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

13

Between July 1997 and December 1998, the rupiah fell from 2400 to the dollar to a low of 217,000

during the political violence in 1998. Interest rates were raised to defend the currency, which

exacerbated the decline in demand. Inflation rates, which had been 12% shot up to close to 100%.

Figure 1 plots GDP over time, showing that it contracted severely in 1997, and fell by over 13% in

1998. The evolution of manufacturing output followed a similar pattern, though manufacturing

growth had already suffered a mild dip in 1996. The drop in manufacturing output was larger than

the corresponding dip in GDP as manufacturing was one of the hardest hit sectors due to its greater

reliance on imported inputs, exposure to changes in foreign demand (particularly given the

importance of intra-regional trade) and greater reliance on external financing, often in foreign

currency. Indeed, the growing trend of borrowing in foreign currency became an enormous burden

for firms post-devaluation. As sources for new credit froze, the inability to service loans helped

drive firms into bankruptcy, even ones that were otherwise productive. Unemployment rates also

increased, though they mask substantial employment reallocation across sectors (see e.g. Fallon and

Lucas, 2002).

The ensuing political crisis precipitated the end of the Suharto regime and led to a dramatic

change in the political and business environments. Large firms that had benefitted from ownership

connections to the Suharto family lost their privileged status and were subjected to greater

competitive pressures (Fisman, 2001; Mobarak and Purbasari, 2008). Policy priorities also shifted,

with greater labor protections put into place, and a move to a more flexible exchange rate regime

(that implied greater exchange rate volatility in the short run). While the new governments were

more populist, the increase in minimum wages and the severance payments regulations they

introduced to protect workers may have had the unintended consequence of discouraging hiring

workers. post-crisis there has been a lack of a resurgence of the labor-intensive sectors of the

economy which had driven growth prior to the crisis (see Aswicahyono et al. (2008) for a detailed

discussion). In short, post-crisis, labor markets were much less flexible than they had been in the

Suharto era.

4. A Bird’s Eye View of Reallocation: Job Flows and Aggregate Productivity Dynamics

We first present a bird‘s eye view of aggregate reallocation patterns and employment and

productivity growth prior to turning to a firm-level analysis of adjustment dynamics.

4.1 Job Flows, Entry and Exit

Page 14: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

14

Table 4.1 and Figures 2 and 3 present aggregate entry, exit and employment growth statistics.

Average exit over the entire period is 8.8% while average entry is 11.1%. Firm exit spiked during the

crisis - in 1997 10.8% of firms exited, while in 1998, 11.2% of all firms exited - and dropped

precipitously during the recovery in 1999 and 2000, to peak again in 2001 (see Section 3 and the

Appendix). Employment growth followed a similar trend, but did not spike in 2001; before the

crisis, manufacturing employment grew quite rapidly. The crises induced substantial job losses; on

average firms shrank employment by 3.30% in 1997 and by 0.93% in 1998. Employment growth

recovered in 1999 and 2000, but dropped in 2001. A possible explanation for the decline in

manufacturing employment growth is that more stringent labor regulations (e.g. the Manpower Act)

increased the cost of labor. These aggregate numbers mask heterogeneity among firms: some firms

cut employment more than others, and surprisingly many firms reported expanding employment: in

1997, 24% of firms grew, while in 1998, 38% of all firms grew. Nevertheless, a striking feature of the

data is how persistent employment is. Over the entire period considered in each year on average,

about a quarter of all firms did not change their labor force and firms that did change their labor

force, typically did so by minimal amounts (the inter-quartile range is 4%).

To examine patterns of aggregate employment reallocation, we examine gross employment

flows following the method proposed by Dunne et al. (1992). Job creation is measured as the sum of

employment gains at new and expanding plants, while job destruction is measured as the sum of

employment losses at contracting and closing plants. Gross job reallocation is the sum of

employment changes at contracting and closing plants. Excess job reallocation is defined as the

difference between gross job reallocation and net job reallocation (job creation minus job

destruction).

Figure 3 presents evidence on aggregate manufacturing employment growth and shows that

the crisis resulted in large employment losses, driven by both a slowdown in job creation as well as a

spike in job destruction. Interestingly, however, a substantial of number of firms expanded

employment during the crisis, pointing towards heterogeneity in firm performance. As a result the

amount of gross reallocation taking place far exceeded the amount required to achieve net

employment adjustment, leading to enormous excess churning.

The figure also shows a longer trend towards lower job creation, both in aggregate and net

creation. This has raised concerns about a ‗jobless recovery‘ (World Bank, 2009b). The firm-level

analysis presented in section 5 examines which are the firms that are expanding and contracting.

Page 15: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

15

Figure 4 decomposes gross job flows by incumbents, entrants and exiters while Figure

presents gross entry and exit rates. While the crisis was associated with decreased entry and exit,

most of the employment adjustment was accounted for by changes in labor usage by existing plants.

Thus, in looking at overall adjustment, exit is clearly not the only relevant dimension to be

examined. In the crisis and immediate recovery, relatively more destruction and creation was

undertaken by incumbents.

4.2 Decomposing Aggregate productivity growth

To examine how crises and recovery affect industrial dynamics and firm behavior, we decompose

the evolution of aggregate productivity growth into adjustment within firms, reallocation between

surviving firms and changes in average productivity due to entry and exit. This enables us to assess

whether crises catalyze or retard efficiency enhancing re-allocation, and to analyze how industry

dynamics during crises differ from pre- and post crisis dynamics.

Our preferred decomposition methodology is due to Foster, Haltiwanger and Krizan

(1998)10, who demonstrate that the evolution of aggregate productivity, P, can be decomposed as:

(1)

Where Pt represents average productivity at time t, is a difference operator, itp represents

the productivity of firm i at time t, and it is the market share of firm i at time t. C denotes the set

of incumbent firms surviving from period t-1 to period t, N denotes the set of new entrants and X

the set of firms that exited. The first term in this decomposition represents the ―within‖ effect, the

contribution of within-establishment productivity growth of surviving firms, weighted by initial

market share. The second term reflects the ―between-effect‖, the contribution of market share

reallocation to productivity growth. This effect is of particular interest as it indicates the extent to

which relatively more productive firms are expanding relatively faster, hence improving allocative

efficiency. The third term represents the ―cross-effect‖ (i.e. the covariance between the within and

the between effect). The last two terms capture the contributions of entry and exit to average

productivity growth. The main advantage of the FHK decomposition is that it provides an

10

Additional decompositions have also been proposed (Basu and Fernald 2002, Petrin and Levinsohn 2008, Basu et al 2009.). However, we cannot compare revenue and output productivity. Our data does not contain the information on prices required to calculate measures of mark-ups which are necessary to implement these decompositions.

1 1 1 1 1 1 1( ) ( ) ( )it it it it t it it it it t it it t

i C i C i C i N i X

P p p P p p P p P

Page 16: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

16

integrated treatment of entry and exit and enables us to separate out ―between‖ and ―within‖ effects

from ―cross‖ effects.

Note that if crises catalyze creative destruction, one would expect within-firm adjustment to

be relatively smaller than during less turbulent times. Rather, one would expect to see an increase in

the ‗between‘ term - relatively more productive firms gaining market share - as well as higher

contributions from exit and entry and potential increases in the ‗cross term‘ - firms that improve

productivity also growing relatively faster.

Figure 3 shows very strikingly the dominant effect of the ‗within‘ adjustment over the crisis

and recovery periods. Incumbent firms (especially large ones) experienced large declines in labor

productivity that was then (partially) regained during the recovery. In contrast, there is not much

evidence that either the ‗between‘ or ‗cross‘ terms contribution grew. Thus resources do not appear

to have been reallocated toward more productive firms.

In terms of exit, the cost to aggregate productivity was greater in 1997, consistent with

relatively more productive firms exiting during the crisis. As pointed out by Liu and Tybout (1996),

however, the small spike in exit rates does not mean that turnover has a minor long-run impact on

productivity. The rise in the contribution of entering firms is the one piece of evidence that points

to improved productivity.

A potential problem with the FHK decomposition is that it is rather sensitive to

measurement error. Suppose, for example, that output is measured with error. This will yield a

positive covariance between changes in productivity and changes in market share, thus resulting in a

spuriously low within-establishment effect and a spuriously high cross-effect. To examine how

robust our results are to measurement error, the Griliches and Regev (1995) decomposition is used:

(2)

Where horizontal bars denote time-averages. (e.g. 1i it it ). While the disadvantage of this is

that it is harder to interpret since it does not allow one to distinguish ‗within‘ and ‗between‘ effects

from ‗cross‘ effects, it has the advantage of being less sensitive to measurement error since the time-

averaging of market shares and productivity will mitigate the influence of classical measurement

error.

1 1( ) ( ) ( ))i it it i it it it it

i C i C i N i X

P p p P p P p P

Page 17: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

17

Given the large contribution of ‗within‘ adjustment and small contribution of the ‗cross

effect,‘ it is not surprising that Figure 6 does not show the GR decomposition as being that

different. The ‗within‘ adjustment still clearly dominates. There is some increase in the between

effect after the crisis, while the increase in entry‘s contribution declines somewhat.

Three additional robustness checks are conducted. First, we use sales rather than

employment to determine market share. As Figure 7 shows, the effects are very similar. The ‗within

effect‘ still dominates. The effect of entry is again positive and rising. Exit‘s contribution falls

during the crisis and again in recent years.

A second robustness test comes from using other measures of firm productivity.

Decompositions based on TFP estimates (see the Appendix for details on how TFP was estimated)

yield very similar patterns.

The final robustness check is to expand the time window used to calculate the effects.11 So

far the decompositions have been calculated on an annual basis. This can increase the variability of

the components over time and can have the effect of underestimating the contributions of entry and

exit. If young firms face a steep learning curve, those that survive will likely be more productive a

few years from their start-up. On the exit side, if firms decline as they head toward closure, looking

only at the most recent data may underestimate the larger costs of losing that firm. We find that

lengthening the window to three years does smooth out the series, but there is not a dramatic

change in the contribution of entry (see Figure 8). For exit, the extent of the positive contribution

declines in the earlier years. This would be consistent with a ‗shadow of death,‘ but one where the

decline is relatively rapid. Rather than the contributions of entry or exit changing significantly, the

more striking change is that the cross effect becomes stronger and more positive. Allowing longer

adjustments to productivity and market share do show a positive correlation. This is one of the

more encouraging pieces of evidence that allocation is improving post-crisis.

5. Firm-level Analysis

While decompositions of productivity growth are useful to understand what is happening in

aggregate, examining the impact of crises on firm exit and employment growth requires firm-level

11

In addition, to examine whether the dominance of the within effect was related the data only covering firms with

more than 20 employees, we re-ran the decompositions separately for firms with fewer than 100 employees and

firms with more than 100 employees. The results are robust to splitting the sample this way; results are not presented

to conserve space, but available from the authors upon request.

Page 18: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

18

analysis. Moreover, firm-level analysis helps us understand the remarkable heterogeneity in

enterprise performance observed in the aggregate data. We are especially interested in testing

whether or not productivity became a more important determinant of firm success as a result of the

crisis.

5.1 Firm Survival

5.1.1 Descriptive Statistics

To get a first idea of whether or not firms that exited were indeed the least productive ones, Table

5.1 presents summary statistics for survivors and exiting firms, disaggregated by time period. Over

the entire period considered, firms that exit are on average younger, smaller, less capital intensive,

employ more unskilled workers, less likely to be government owned, more likely to be foreign

owned, less likely to export and less productive than firms that survive (see column 1).

Comparing across columns to examine how the differences between surviving firms and

exiting firms over time. During the crisis, younger firms seem to have been especially vulnerable,

while exporters were less likely to exit (compared to other years), perhaps because of favorable

exchange rate movements. In addition, firms that have a high proportion of women in their

workforce were more likely to survive. During the crisis, survivors were not, on average, more likely

to operate in sectors highly dependent on external finance, whereas pre- and post-crisis they were,

suggesting that the crisis hit sectors more dependent on external finance relatively harder. The

productivity gap between exiting and surviving firms seems to have narrowed during the crisis; while

before the crisis the difference in the average log value-added of surviving firms compared to exiting

firms was about .31, it narrowed to .21 during the crisis, to increase again in the recovery.

Demeaning value-added per worker by sector or by sector-year yields a similar, but slightly less

dramatic, pattern. Thus, the aggregate attenuation effect is due to a combination of more productive

sectors being more severely impacted by the crisis, as well as a decrease in the productivity

differential between continuing and surviving firms within sectors. Bear in mind, however, that these

statistics do not condition on other firm characteristics; it need not be the case that productivity

offered relatively less protection during the crisis. For example, firm size is typically positively

associated with productivity and the typical firm that exited during the crisis was smaller than in

other years. It could thus be the case that it was not productivity, but size that accounts for the

Page 19: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

19

diminution of the productivity gap between firms that exit and firms that survive. The next test

provides a more formal test of the cleansing hypothesis which controls for other firm characteristics.

5.1.2 Econometric Framework

To model firm survival a discrete-time proportional hazards survival model is used. In discrete time,

the survival function 𝑆 𝑡𝑖 𝒙 , which measures the probability that for a firm with a vector of

characteristics 𝒙 survival time is at least T, can be expressed as the product of the complements of

period specific hazard rates: 𝑆 𝑡𝑖 𝑥 = Pr(𝑇 ≥ 𝑡𝑖 |𝒙) = (1 − 𝜆 𝑡𝑖|𝒙 )𝑡𝑖=1 . The 𝜆 𝑡𝑖|𝒙 refer to

the period specific hazard rates, which measure the conditional probability of exiting at time 𝑡𝑖

conditional on surviving until time 𝑡𝑖−1 and are defined as 𝜆 𝑡𝑖|𝒙 = Pr(𝑡𝑖−1 ≤ 𝑇 < 𝑡𝑖|𝑇 ≥

𝑡𝑖−1, 𝑥𝑖−1) = 1 −𝑓(𝑡|𝒙)

𝐹(𝑡|𝒙)=

𝑓(𝑡|𝒙)

𝑆(𝑡|𝒙), where 𝐹 𝑡 𝒙 is the cumulative density function of time at risk

given the vector of characteristics 𝒙𝒊.12 Following Cox (1972) we assume that the hazard rate of a

firm characterized by covariates 𝑥𝑖 , is proportional to a baseline hazard 𝜆0:

𝜆𝑡 𝑡𝑖

𝜆0 𝑡𝑖 = exp(𝛽𝒙𝒊

′𝒙𝒊) <=> 𝑙𝑜𝑔 𝜆𝑡 𝑡𝑖 = 𝑙𝑜𝑔 𝜆0 𝑡𝑖 + 𝛽𝒙𝒊

′𝒙𝒊

This choice of the hazard function implies that each regression coefficient 𝛽𝑥𝑖 measures the

proportionate change in hazards associated with absolute changes in the corresponding covariate

𝑥𝑖 ∈ 𝒙𝒊. Consider the impact of productivity as an example. If productivity does not affect the

hazard rate then, exp 𝛽𝑃 = 1, whereas it increases the likelihood of exit (survival) if 𝑒𝑥𝑝𝛽𝑃 > 1 ,

(𝑒𝑥𝑝𝛽𝑃 < 1).

To examine whether crises catalyze creative destruction and to examine how the

determinants of firm-survival varied over time, we interact these explanatory variables with dummies

for the crisis and the recovery:

𝑙𝑜𝑔 𝜆𝑡 𝑡𝑖 = 𝑙𝑜𝑔 𝜆0 𝑡𝑖 + 𝛽𝒙𝒊

′𝒙𝒊

+𝛽𝑪𝒓𝒊𝒔𝒊𝒔∗𝒙𝒊

′𝐶𝑟𝑖𝑠𝑖𝑠 ∗ 𝒙𝒊 + 𝛽𝑹𝒆𝒄𝒐𝒗𝒆𝒓𝒚∗𝒙𝒊

′𝑅𝑒𝑐𝑜𝑣𝑒𝑟𝑦 ∗ 𝒙𝒊+ 𝛽𝑪𝒓𝒊𝒔𝒊𝒔𝒊

′𝐶𝑟𝑖𝑠𝑖𝑠 + 𝛽𝑹𝒆𝒄𝒐𝒗𝒆𝒓𝒚𝒊

′𝑅𝑒𝑐𝑜𝑣𝑒𝑟𝑦

Where Crisis is a dummy variable for 1997 and 1998 and 𝑅𝑒𝑐𝑜𝑣𝑒𝑟𝑦 a dummy for the period 1999-

12 Note the last equality follows from the definition of the survivor function: 𝑆 𝑡 𝑥 ≡ 1 − 𝐹(𝑡|𝑥)

Page 20: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

20

2001. The years leading up to the crisis will be referred to as the pre-crisis period. Effectively we are

testing how during the crisis and the subsequent recovery period the relationship between exit and

covariates differed from the pre-crisis process. The ―crisis‖ interactions examine the short-term

impact of the crisis, while the ―recovery‖ interactions examine its medium-term impact. This testing

strategy is very general as we allow all parameters of the hazard function to vary, thus minimizing

parameter constancy restrictions. The proportional hazard specification is very convenient since it

enables us to test whether firms with certain characteristics were disproportionately more or less likely

to exit in certain periods.

Under the null hypothesis of no short-run differential effect of crises on creative destruction,

𝑒𝑥𝑝𝛽𝑪𝑹𝑰𝑺𝑰𝑺∗𝑷 = 1. If crises catalyze creative destruction, 𝑒𝑥𝑝𝛽𝐶𝑅𝐼𝑆𝐼𝑆∗𝑃 > 1, while 𝑒𝑥𝑝𝛽𝐶𝑅𝐼𝑆𝐼𝑆∗𝑃 <

1 if they hamper it. At the risk of belaboring the point if 𝑒𝑥𝑝𝛽𝑪𝑹𝑰𝑺𝑰𝑺∗𝑷 = 1, this does not mean that

crises are not weeding out productive firms; whether this is happening depends of course also on

𝛽𝑷. The interaction term tells us whether productive firms were overrepresented amongst the exiters

relative to other periods. Following Rajan and Zingales (1998) this strategy, which essentially

constitutes a difference-in-difference approach, can also be used to examine the importance of

sector characteristics such as financial dependence and the natural rate of employment turnover.

Testing whether the association between sector-level measures of financial dependence and exit

varies over time will help us assess whether firms dependent on external finance are especially (less)

likely to survive during crises.

5.1.3 Results

Our baseline specification models firm-survival as a function of the log of the age of the firm, its

size, the proportion of blue collar workers in the total workforce (the ―unskilled ratio‖), the

proportion of women, foreign and government ownership, whether or not the firm exports and

productivity, proxied by value added per worker. In addition, industry, year, and province dummies

are included to eliminate time, industry and location effects. Results are presented in Table 5.2. The

first column pools all years, while columns two, three and four estimate the model separately for the

pre-crisis13, crisis, and recovery periods. The shaded areas indicate that the coefficient estimate is

significantly different from the pre-crisis one at the 10% level, while a bolded coefficient indicates

that the coefficient is significant at the 5% level (based on pooling over all periods and interacting

13 Note that these specifications only cover the period from 1993 onwards since the share of women in the workforce is not available prior to 1993.

Page 21: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

21

the time dummies with all parameters of the hazard function).

The results in column 1 are consistent with the descriptive statistics presented in the

previous section and other studies of firm survival in developing countries and this dataset (see

Blalock, Gertler and Levine, 2008, Harrison and Scorse, 2009, Frazer, 2005, Söderbom and Teal,

2006); size, age, and productivity all increase the probability of survival. Interestingly, foreign owned

firms are less likely to exit while exporters are more likely to exit, ceteris paribus.

However, these effects are not stable over time, as is evidenced by a comparison of columns

2-4. Starting with the result of focal interest, the link between productivity and crisis seems to be

attenuated by the crisis; the odds ratio associated with value added per worker is always significantly

below 1, yet moves much closer to 1 during the crisis, suggesting that the crisis attenuated the link

between productivity and exit. This is not consistent with the cleansing hypothesis; while more

productive firms remain less likely to exit – ceteris paribus – the protective impact of productivity is

significantly weaker than it was pre-crisis. This difference between the pre-crisis and crisis effect is

statistically significant (as indicated by the fact that the coefficient is bolded); Post-crisis the

correlation between productivity and survival strengthened. Again, the coefficient on value-added is

significantly different from its pre-crisis counterpart.

The crisis has benefited exporters, who may have benefited from increased international

competitiveness due to the depreciation of the rupiah; while exporting is associated with a higher

propensity to exit during other periods, exporters were not ceteris paribus more likely to exit during

the crisis; the association between exporting and exit during the crisis is significantly different from

the pre-crisis association. Somewhat surprisingly, exit during the crisis is also negatively correlated

with the share of female employees and again this association is statistically significantly different

from the association between the gender composition of the workforce and exit pre-crisis. This

result is a sector composition effect; sectors which employed proportionately more women were hit

harder.14

During the recovery, some of these effects were reversed; firm-age was less strongly

correlated with exit than it had been before the crisis, while exports became – ceteris paribus – more

likely to exit than non-exporters. Also, government owned firms were much less likely to exit during

the recovery. In addition, firms employing proportionally more blue collar employees were more

likely to exit.

14 Result are omitted to conserve space, but available from the authors upon request.

Page 22: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

22

5.1.3.1 Robustness

Table 5.3 presents alternative specifications and robustness test. The set of explanatory variables is

identical to those reported in table 5.2, except that it also includes industry dummies. Again, shaded

areas indicate that the coefficient is significantly different from the pre-crisis one at the 10% level,

while bolded coefficients indicate the effect is significantly different at the 5% level. To conserve

space, we only report the coefficients on value-added per worker, the explanatory variable of focal

interest. To start with, the result is not driven by differences across sectors or how different sectors

were impacted by the crisis as we can see in rows 1 and 2 of table 5.3, which control for industry

dummies and industry-year dummies respectively. The attenuation of the link between productivity

and exit is thus not simply due to the fact that more productive sectors were more severely impacted

by the crisis (recall that the descriptive statistics presented in Table 5.1 showed that firms in more

productive sectors were more likely to exit during the crisis), but operates within as well as across

sectors. Nevertheless, sectors differ in their vulnerability to the crisis; a point which we will return to

below.

Second, the pattern of results is robust to including the anomalous observations (see

Appendix B) that we have excluded from our sample and to dropping the top and bottom 10% of

most productive firms by sector-year from our analysis. If anything, excluding these extreme values

strengthens our results. The result is thus not driven by outliers.

Third, to examine the sensitivity of our result to our choice of the hazard function, the

baseline specification was re-estimated using a complementary log-logistic hazard function (i.e.

𝜆𝑡 𝑡𝑖 = 1 − (1 − 𝑙𝑜𝑔 𝜆0 𝑡𝑖 )𝐸𝑥𝑝 (𝛽 ′𝒙𝒊), or log(−log(1 − 𝜆𝑡 𝑡𝑖 )) = 𝑎𝑖 + 𝛽′𝒙𝒊,

where 𝑎𝑖 = log 1 − (1 − log 𝜆0 𝑡𝑖 . The findings are robust to using this alternative

specification of the hazard function.

Fourth, we re-estimated the models using deeper lags of value-added per worker to alleviate

concerns that the weakened association between productivity and exit is an artifact of the difficulties

of measuring productivity during turbulent times. Since productivity is very strongly correlated over

time (the autocorrelation coefficient for value added per worker is .84 over a one year window15 and

.63 over a five year period), lagged productivity is a good proxy for current productivity.16 While the

differences in the protective impact of productivity become somewhat smaller, the pattern of results

15 During the crisis-years, the correlations remain roughly constant; over a one-year window the correlation between value-added and lagged value-added is .85, while it is .66 over a five-year period. 16 This is also true in the crisis years, when the correlation between contemporaneous productivity and lagged productivity is .8.

Page 23: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

23

is robust to using lagged productivity measures. We interpret this as strong evidence that the

attenuation effect is genuine and not driven by measurement error.

Fifth, one might worry that our results are driven by the fact that we do not observe firms

with fewer than 20 employees. If firms shrink and drop below this threshold, we will not observe

them in our data and define them as having exited. While we obviously cannot trace which of the

firms that exit from our sample stop operating altogether and which ones continue, but operate at a

smaller scale, we can investigate whether and how the exclusion of this latter group of firms might

affect our results. We do this by setting arbitrary size thresholds for exit and examining how our

results change. We exclude firms with fewer than 30 (40 or 50) employees and define a firm as

having exited when it either disappears from the data altogether or fails to report having more than

29 (39 or 49) employees in any of the subsequent years. Although using higher thresholds for exit

reduces the attenuation effect, the results are robust to using different exit thresholds.

Sixth, the regressions thus far have not controlled for capital intensity since capital might be

inaccurately measured.17 This is likely to introduce omitted variable bias. However, controlling for

capital intensity does not alter the pattern of results. On the contrary, once capital intensity is

controlled for, productivity loses its protective power altogether during the crisis – regardless of

whether we use the observed capital stock, or a combination of observed and imputed values.

Seventh, one may be concerned that exit is a protracted process and that by focusing on short-

term effects, we are missing the action. To address this concern, we assess how robust the results

are to using longer time windows.18 The bottom row presents estimates of the probability of long-

run survival, defined as being in business 5 or years later, using our baseline specification, with the

exception of the share of women in the total workforce, as this variable only became available in

1993. Column 1 models the probability of remaining in business until 1996 for firms operating in

1991, column 2 models the probability of survival until at least 2001 for firms operating in 1996.

Column 3 models for the probability of staying in business until 1996 for firms active in 1993 while

column 4 models the likelihood that firms that were active in 1996 were still in existence in 1999.

Overall, the pattern of results does not change when longer time-horizons are considered;

productivity offers less protection during the crisis than it did pre-crisis, ceteris paribus.

These results also provide a robustness check on the concern that the spike in exits in 2001

includes firms that actually exited in the year or years prior to 2001 but remained on the books until 17 One concern is that the response rate is lower for measures of the capital stock than for other variables. It is also missing in the 1996 data, requiring it to be estimated for that year (see Appendix). 18 Note that the spike in exit rates during crisis years suggests that the crisis did have an immediate impact on firm exit.

Page 24: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

24

then, i.e. the issue of ―zombie firms‖. With longer time windows, the exact timing of these firms‘

exit is not an issue, the results remain.

Finally, since value-added per worker is a partial productivity measure we examine how

robust our results are to using a multitude of measures of multifactor productivity. Results are

presented in Table 5.4. The estimated attenuation effect becomes stronger.19

Past productivity is no

longer a significant predictor of productivity during the crisis, regardless of whether we use

productivity proxies computed by means of OLS, fixed effects estimation, the Solow residual or the

Ackerberg-Caves-Frazer procedure.

5.1.3.2 Dependence on credit and exit: indirect evidence

The evidence for the attenuation effect discussed above thus seems robust and begs the question

what is driving this attenuation effect. The scarring hypothesis postulates that the attenuation effect

arises because of market imperfections. Distortions in the credit market feature particularly

prominently as a potential explanation for the attenuation of the link between productivity and exit.

However, since we cannot directly observe which firms are credit constrained and which are not,

testing them is difficult with the data at hand. Instead, we try to probe whether such imperfections

might help explain the attenuation effect by comparing the performance of firms that are likely to

differ in their exposure to changing credit market conditions, both across and within sectors. If

changing loan conditions are the explanation for the attenuation effect, then one would expect the

attenuation effect to be especially severe amongst firms that were most exposed to these changing

conditions. Examining the impact of changing loan conditions is of interest in its own right, since

together with the Mexican crisis in the early nineties the East Asian crisis was one of the first ―third

generation‖ crises, in which financial fragility, rather than fiscal imbalances or weaknesses in real

activity, appears to have played a big role (Chang and Velasco, 2001).

To identify firms‘ exposure to changing credit market conditions we exploit information on

differences in dependence of finance across sectors, as well as information on differences in the

composition of investment finance controlling for sector. To start with, we examine whether or not

firms operating in sectors more dependent on credit (as proxied by the Rajan-Zingales measures of

dependence on external finance - see Appendix B20) were more likely to exit during the crisis –

19 Note that this result is not driven by controlling for the capital stock (see Row 2 in Table 5.2). 20 We also experimented with using the inventories-to-sales ratio as proposed by Raddatz (2006) as proxy for liquidity needs. Using this indicator, qualitatively the same pattern of results obtains. Results are omitted to conserve space, but available upon request from the authors.

Page 25: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

25

controlling for firm characteristics and other sector characteristics. Moreover, if the scarring

hypothesis is correct and changing loan conditions are driving the attenuation effect,21 one would

expect that productivity would offer less protection against exit for firms in sectors highly dependent

on external credit. For firms in such sectors, the ability to access or repay loans might become a

relatively more important determinant of survival than productivity per se during the crisis.

Secondly, we complement this analysis – which essentially compares firms in different

sectors – by exploiting information on firms‘ investment financing and examining how exit patterns

vary with exposure to credit market changes after controlling for sector characteristics. Comparing

how the exit propensity of firms that took out loans to invest evolved relative to that of firms that

financed their investment by other means helps us assess whether changing credit market conditions

might explain the attenuation effect. Again, if more stringent loan conditions explain the attenuation

effect, then one would expect the protective impact of productivity to be especially weak for firms

that took out loans to invest.

The results of our regressions are presented in Table 5.5. The top half of the table presents

our results on the impact of credit constraints across sectors. Firms‘ survival is modeled as a

function of firm characteristics included in the baseline survival regressions presented in Table 5.2

In addition, to minimize omitted variable bias, we not only control for a sector‘s financial

dependence, but also its competitiveness and contestability, proxied by the Herfindahl index, the

―natural‖ rate of entry and of employment turnover. Turning to the results, before and after the

crisis firms operating in industries more dependent on access to credit are less likely to exit – as is

evidenced by the odds ratio associated with financial dependence in column 1 that is significantly

below 1. However, during the crisis, they became much more likely to exit suggesting that the crisis

hit firms in sectors more dependent on external finance relatively harder. In addition, firms in

sectors with higher levels of employment turnover and – somewhat surprisingly – industries with a

higher concentration ratio, had a higher exit propensity. Note, however, that controlling for these

sector characteristics does not explain the attenuation effect, which is robust to including sector

controls.

When we interact measures of financial dependence with value-added per worker (see row

2), we find that, generally, productivity is a less important determinant of survival in industries with

high levels of financial dependence. During the crisis, however, this effect was reversed; while firms 21 Our proxy for financial dependence is essentially an indicator of how much finance firms in the sector of interest manage to obtain from external sources. It can thus roughly be interpreted as a measure of excess to credit. Paradoxically, sectors and firms that lack access to credit may be more insulated from the shock.

Page 26: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

26

in industries more dependent on external finance were more likely to exit, this was especially true for

relatively unproductive firms. That is, the crisis hit firms in sectors highly dependent on credit

harder, but relatively productive firms in such sectors where better capable of weathering the shock.

Rather than explaining the attenuation of the link between productivity and exit, this effect is

consistent with cleansing in sectors with better access to credit. Overall, however, the attenuation

effect remains. Productivity was a less important determinant of survival for firms in sectors that

have fewer outside financing opportunities.

To examine whether a similar mechanism is at work within sectors, Table 5.6 presents

regressions that include sector dummies and include dummies that indicate whether or not a firm

took out loans to finance any of the investments made during the past three years or not. The

omitted category is the group of firms that did not invest at all. Firms that invest are generally less

likely to exit and this is especially true for firms that took out loans to finance their investment.22

Yet, during the crisis, the negative association between investing and exit became weaker.

Interestingly, this weakening was more pronounced for firms that used loans to invest than for firms

that financed investment by other means, suggesting the former group of firms was hit harder by the

crisis.23 Note that the attenuation of the link between productivity and exit is robust to inclusion of

these investment proxies.

The bottom part of table 5.6 presents regressions in which we interact the investment

dummies with the log of value-added per worker. The odds ratio on the dummy for having invested

and taking out a loan to do so is not significantly different from one before or after the crisis, but

very high and significant during the crisis. The odds ratio associated with the interaction between

productivity and having financed investments by means of a loan is also not significantly different

from one either pre- or post-crisis, but much lower than one during the crisis. Thus, firms that took

out loans to invest were severely affected by the crisis, yet, amongst them, the most productive firms

were better capable of weathering the shock. Again, these findings are at odds with the idea that the

exit of firms more exposed to credit market conditions caused the attenuation effect and consistent

with the pattern we observed across sectors. The odds ratio associated with the interaction between

productivity and investing but not using loans for this purpose is always below 1, but especially low

during the crisis and the recovery, while the odds ratio associated with the dummy or having

22 This is reflected in a negative correlation between investing and exit, which is consistent with conventional economic theory; investing is a strong signal of future profitability. 23 These results are robust to controlling for the amount of investment spending; results are available upon request from the authors but omitted to conserve space.

Page 27: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

27

invested without borrowing is always higher than 1, but only significantly different from 1 before the

crisis. The link between productivity and exit was thus somewhat attenuated for this category of

firms. The attenuation was strongest for firms that did not invest at all.

In sum, these results are consistent with the hypothesis that firms more exposed to

fluctuations in credit market conditions were hit harder by the crisis, yet these effects cannot

account for the attenuation of the link between productivity and exit during the crisis. If anything,

the crisis induced a cleansing effect amongst firms more exposed to the volatilities of the credit

market, as the negative association between productivity and exit strengthened for these firms. The

attenuation effect was thus most pronounced amongst firms that were less likely to have a loan. At

the risk of belaboring the point, this does not rule out the possibility that credit constraints may be

an important part of the explanation as to why we are observing an attenuation effect in the

aggregate. However, it seems unlikely that changing loan conditions for firms that had taken out

loans can account for the overall attenuation effect observed.

5.1 Employment Growth

5.2.1 Econometric Framework

To examine which firms grow fastest and to assess whether or not employment growth became

more strongly associated with productivity during the crisis, or whether, as is the case with survival,

the link between employment and productivity was attenuated, we estimate employment growth

models of the form; ∆𝐿𝑡𝑖 = 𝛽𝒙𝒕𝒊−𝟏

′𝒙𝒕𝒊−𝟏, where 𝒙𝒕𝒊−𝟏 is a vector of firm and sector characteristics,

including size, ownership, age, productivity. We follow a similar approach as proposed in section

5.1.2 and interact period dummies with this vector of explanatory variables to assess how the firm

growth process evolved over time. That is, we estimate an equation that enables us to examine

whether the crisis induced short- and medium-run changes in the process governing employment

growth;

∆𝐿𝑖𝑡 = 𝛽𝒙𝒕𝒊−𝟏

′𝒙𝒕𝒊−𝟏+𝛽𝑪𝒓𝒊𝒔𝒊𝒔∗𝒊𝒙𝒕𝒊−𝟏

′𝐶𝑟𝑖𝑠𝑖𝑠 ∗ 𝒙𝒕𝒊−𝟏 + 𝛽𝑹𝒆𝒄𝒐𝒗𝒆𝒓𝒚𝒊

′𝑅𝑒𝑐𝑜𝑣𝑒𝑟𝑦 ∗ 𝒙𝒕𝒊−𝟏 + 𝛽𝑪𝒓𝒊𝒔𝒊𝒔𝒊

′𝐶𝑟𝑖𝑠𝑖𝑠

+ 𝛽𝑹𝒆𝒄𝒐𝒗𝒆𝒓𝒚𝒊

′𝑅𝑒𝑐𝑜𝑣𝑒𝑟𝑦

Page 28: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

28

Thus, we use a difference-in-difference approach. Under the null hypothesis that the crisis

did not improve the allocative efficiency of employment re-allocation amongst continuing firms

𝛽𝑪𝑹𝑰𝑺𝑰𝑺∗𝑷 = 0, whereas 𝛽𝑪𝑹𝑰𝑺𝑰𝑺∗𝑷 > 0 (𝛽𝑪𝑹𝑰𝑺𝑰𝑺∗𝑷 <0) under the alternative hypothesis that the

crisis enhanced (diminished) the importance of productivity as a key determinant of firm growth.

Serial correlation in the error term, in conjunction with the presence of lagged size as an

explanatory variable would render OLS estimates of the employment growth equation biased. To

address this concern, we also estimate the employment growth equation using a Fixed Effects

estimator, thus removing time-invariant unobserved heterogeneity. Unfortunately, the fixed effects

transformation leads to a correlation between the transformed error and the transformed

explanatory variables, resulting in the well-known Nickell bias (1981). As pointed out by Bond

(2002), the OLS and Fixed Effects estimators are biased in opposite directions, thus providing us

with a confidence interval within which we might expect the true parameters to lie.24

Our growth regressions may suffer from selection bias, since we only observe employment

growth for surviving firms. While in principle it would be feasible to correct for selection bias using

a control function or instrumental variable approach, it is difficult to think of variables that affect

employment growth but not exit or vice versa. As a result, selection corrections would have to rely

on stringent functional form assumptions. Instead, we follow the approach suggested by Dunne, et

al. (1989) and try to gauge the magnitude of survivor bias by comparing the results obtained by

estimating our models on the entire sample with those obtained by estimating the models on a

sample of firms that remained in the data until 2001 only. Comparison of the two sets of results is

informative about the likelihood of survivor bias, as estimates obtained from the latter sample suffer

from maximum survivor bias.

5.2.2 Results

Firm survival and employment growth are determined by the same data generating process,

notably the one determining firm-size, and we therefore use the same explanatory variables as in the

firm survival models. That is, we model employment growth from period t-1 to period t as a

function of the age of the firm, its size and size squared to allow for non-linear size effects,

24 While the Difference and Systems GMM estimators developed by Arellano and Bond (1991) are in principle capable of yielding unbiased estimates, these estimators are not well suited for our data; the difference GMM estimator is likely to result in poorly behaved estimates when variables are highly persistent as is the case with our data (for surviving firms the correlation between lnLt and lnLt-1 is 0.98 – both in crisis and non-crisis years), while the Systems GMM estimators relies on a mean stationarity assumption that is palpably undesirable in the context of a crisis (See Roodman, 2006, for a discussion). We therefore eschew this approach.

Page 29: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

29

productivity proxied by value-added per worker, the proportion of women it employed, whether it

exported, and ownership structure at time t-1. In addition, industry, year, and province dummies are

included to eliminate time, industry and location effects.

The results of this baseline specification are presented in Table 5.7. The first column pools

all years, while columns two, three and four present model estimates for the pre-crisis,25 crisis, and

recovery periods. Shaded areas indicate that the coefficient estimates are significantly different from

the corresponding pre-crisis estimate at the 5% significance level. It is very difficult to predict

employment growth as evidenced by the consistently low R2s, even though we are including a rich

set of firm characteristics and dummy variables. The results are generally consistent with the

literature on firm growth: younger firms, smaller firms and more productive firms grow faster (see

for example, Bigsten and Gebreeyesus, 2007). Productivity is also correlated with faster employment

growth. Interestingly, firms with a higher proportion of women are also ceteris paribus more likely

to grow over the entire period, perhaps because women are paid lower wages.

Examining differences in these effects over time we find that the crisis did not only attenuate

the link between productivity and exit, but that the association between productivity and

employment growth was even weaker after the crisis. The finding that more productive firms that

survived were not less likely to shed labor, ceteris paribus, suggests that employment reallocation

amongst surviving firms was not especially efficiency enhancing during the crisis.

Of course, our estimates exclude exiting firms. If we include these, see Table 5.8, then we

again find evidence of an attenuation of the link between productivity and employment growth

during the crisis and recovery of this link after the crisis – consistent with our survival models.

We also find evidence that firms in sectors more dependent on external finance contracted

relatively faster (see Table 5.9). Firms that invested did not shed fewer jobs than firms that did not

(see Table 5.10), while before the crisis they had been growing faster, especially if they had taken out

loans to finance investment; these firms suffered a slightly higher reduction in growth rates

compared to the pre-crisis period than firms that financed investment by other means, though the

differences are marginal. Post-crisis firms that took out loans to invest grew, ceteris paribus, faster

than firms that did not invest. However, the growth rates of firms that invested but did not finance

such investments by taking out loans remained lower than that of firms that did not invest.

25 Note that these specifications only cover the period from 1993 onwards since the share of women in the workforce is not available prior to 1993.

Page 30: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

30

Table 5.6 also presents a host of robustness checks. The observed pattern is robust to

controlling for sector and sector-year effects. Survivor bias is unlikely to be a major issue as the

estimates on the sample of firms that survive until 2001 are not radically different from those

obtained for the sample that includes firms that exit at one point or another. The results are

furthermore robust to using TFP instead of value-added per worker, to using lags of value-added per

worker, to controlling for the capital stock, to using longer growth windows and to controlling for

fixed effects (see Table 5.11).

6 Conclusion

While crises are recognized to be periods of intensified adjustment and aggregate studies suggest that

firm dynamics are a key determinant of the depth and duration of crises, firm-level evidence on their

impact on resource allocation is scant. Perhaps because of the paucity of the empirical evidence,

there is an active debate as to whether or not crises have a silver lining by improving resource

allocation. On the one hand, a host of macroeconomic models are predicated on the idea that the

additional competitive pressure induced by the crisis will hurt inefficient producers

disproportionately. They predict that the crisis will ―cleanse‖ out unproductive firms and reallocate

resources towards more efficient producers. On the other hand, a series of recent papers point out

that, in the presence of market imperfection, these conclusions may be overturned and that crises

may ―scar‖ the economy by driving productive firms out of business.

Using the Indonesian manufacturing census data from 1991-2001 to examine the impact of

the East Asian crisis on resource allocation, this paper finds strong evidence against the cleansing

hypothesis. Decompositions of aggregate productivity growth reveal that rather than the crisis

resulting in massive reallocation between firms, widespread within firm adjustment dominated.

Moreover, the link between productivity and exit was attenuated during the crisis and there was less

regard for prior productivity in terms of which firms adjusted labor. These findings are especially

concerning given the spike in exit and the excessive amount of job reallocation that took place

during the crisis.

Market imperfections (and possible aggravation thereof) provide a potential explanation for

the attenuation of the link between firm productivity, survival and employment growth. Consistent

with this explanation, firms more exposed to changing credit market conditions were more severely

impacted. Firms in sectors more dependent on external finance were more likely to exit compared

Page 31: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

31

to other periods and shrank more during the crisis. In addition, firms that used loans to finance their

investments were hit harder by the crisis than firms that financed investment by other means. Yet,

amongst firms more exposed to changing credit market conditions, productive firms did better. The

attenuation effect was thus most severe for firms that had worse access to credit to start with. Our

data do not allow us to rule out the possibility that credit market imperfections account for the

attenuation effect, yet do suggest that it was not purely driven by firms with pre-existing loans being

hammered. To what extent our results regarding the importance of access credit generalize to other

crises is uncertain, since the East Asian crisis was one of the first crises in which financial fragility

appears to have played a dominant role.

An important question, particularly for policy makers deciding whether or not to support

firms in the short-run, is whether the efficiency of the resource allocation improved over time.

Fortunately, crisis-induced changes in the reallocative process did not last; post-crisis the link

between firm survival and productivity was restored suggesting that the crisis did not permanently

scar the Schumpetarian process of creative destruction. The attenuation of the link between

productivity and employment growth had greater persistence post-crisis. We speculate that this

might be due to the introduction of more stringent labor-market regulations in the immediate

aftermath of the crisis. Investigating whether this is indeed the case, assessing the role of specific

market imperfections and examining to what extent our findings generalize to other crisis-episodes

are promising areas for further research.

Page 32: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

32

References

Ackerberg, D., Caves, K. and G. Frazer (2005). ―Structural identification of production functions,‖ Mimeo., UCLA.

Arellano, M., and S. Bond. (1991) ―Some tests of specification for panel data: Monte Carlo evidence

and an application to employment equations‖. Review of Economic Studies 58: 277(97). Aswicahyono, H. (2008) ―A Survey of Micro-data Analyses in Indonesia.‖ Ch. 12 in eds. J. Corbett

and S. Umezaki Deepening East Asian Economic Integration, ERIA Research Project Report No. 1. Economic Research Institute for ASEAN and East Asia. Jakarta, Indonesia.

Baily, M., Hulten, C. and D. Campbell (1992) ―The Distribution of Productivity in Manufacturing

Plants,‖ Brookings Papers: Microeconomics Brookings Papers: Microeconomics, 187–249. Balasubramanian, N. and Sivadasan, J. (2009) ―What happens when firms patent? New evidence

from US Economic Census data‖ Mimeo. University of Michigan. Barlevy, G. (2002). The sullying effect of recessions‖ Review of Economic Studies 69: 65–96. Barlevy, G. ―(2003) ―Credit market frictions and the allocation of resources over the business cycle.‖

Journal of Monetary Economics 50( 8): 1795-1818 Bartelsman, E., J. Haltiwanger and S. Scarpetta. (2004). ―Microeconomic Evidence of Creative

Destruction in Industrial and Developing Countries.‖ World Bank Policy Research Working Paper, #3464. World Bank: Washington, DC.

Bartelsman, E. and M. Doms (2000). ―Understanding Productivity: Lessons from Longitudinal

Microdata.‖ Journal of Economic Literature 38(3): 569-95. Basu, S. and J. Fernald. (2002) "Aggregate Productivity and Aggregate Technology," European

Economic Review, 46, 963-991. Basu, Sustano Luigi Pascali, Fabio Schiantarelli, Luis Serven, 2009 "Productivity, Welfare and

Reallocation: Theory and Firm-Level Evidence, " NBER Working Paper No. 15579 Beegle, K. E. Frankenberg, D. Thomas. (1999) ―Measuring Change in Indonesia.‖ RAND Working

Paper No. 99-07. RAND Corp.: Los Angeles, CA. Bergoeing, R., N. Loayza and A. Repetto (2004). ―Slow Recoveries.‖ Journal of Development Economics,

75(2): 473-506. Bigsten, A. and Gebreeyesus, M. (2007) ―The Small, the Young, and the Productive: Determinants

of Manufacturing Firm Growth in Ethiopia‖ Economic Development and Cultural Change. 55(4):

813-840.

Page 33: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

33

Blalock, G., P. Gertler and D. Levine (2008). ―Financial Constraints on Investment in an Emerging Market Crisis‖ Journal of Monetary Economics. 55: 568-591.

Blundell, R., and S. Bond. (1998). ―Initial Conditions and Moment Restrictions in Dynamic Panel

Data Models.‖ Journal of Econometrics 87: 113-143. Bond, S. (2002). ―Dynamic panel data models: A guide to micro data methods and practice.‖

Working Paper 09/02. Institute for Fiscal Studies: London. Borensztein, E. and J. Lee. (2002) ―Financial Crisis and Credit Crunch in Korea: Evidence from

Firm-Level Data.‖ Journal of Monetary Economics. 49(4): 853-875. Bowlus, A. (1993) ―Job match quality over the business cycle‖ Panel Data and Labour Market

Dynamics: 21-41. Caballero R. and M. Hammour (1994). ―The Cleansing Effect of Creative Destruction.‖

American Economic Review 84(5): 1350-68. Caballero R. and M. Hammour., (1999). Improper Churn: Social Costs and Macroeconomic

Consequences. Mimeo, MIT. Casacuberta, C. and N. Gandelman. (2009) ―Protection, openness, and factor adjustment : evidence

from the manufacturing sector in Uruguay‖ IDB Working Paper. Interamerican Development Bank: Washington, D.C.

Caves, R.(1998). ―Industrial Organization and New Findings on the Turnover and Mobility of

Firms.‖ Journal of Economic Literature 36(4): 1947-82. Chan, R. (2009)_ ''A Study on the Effects of Financial Constraints on Firms in Developing

Countries'', Mimeo. University of Michigan. Chang, R. and A. Velasco (2001), ―A model of currency crises in emerging markets.‖ Quarterly Journal

of Economics 116 (2):489-517. Collier, P. and B. Goderis (2009) ―Structural Policies for Shock-Prone Developing Countries"

CSAE Working Paper WPS 2009-03. Centre for the Study of African Economies, Oxford Univ. Oxford.

Cox, R. (1972). ―Regression Models and Life Tables‖ Journal of the Royal Statistic Society 34: 187-202. Davis, S., and J. Haltiwanger (1992) 'Gross job creation, gross job destruction and employment

reallocation.' Quarterly Journal of Economics CVII, 819-63. Davis, S., J. Haltiwanger, and S. Schuh (1996) Job Creation and Destruction. MIT Press: Cambridge. Dunne, T., Roberts, M. and L. Samuelson (1989). ―The Growth and Failure of U.S. Manufacturing

Plants.‖ Quarterly Journal of Economics 104: 671-98.

Page 34: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

34

Dwor-Frecault, D., Colaco, F. and M. Hallward-Driemeier. (1999) Asian Corporate Recovery. World Bank: Washington DC.

Fallon, P. and R. Lucas (2002). ―The Impact of Financial Crises on Labor Markets, Household

Incomes, and Poverty: A Review of Evidence.‖ World Bank Research Observer. 17(1): 21-45. Fisman, R. (2001). ―Estimating the Value of Political Connections.‖ American Economic Review. 91(4):

1095-1102. Foster, L., Haltiwanger,J. and C. Krizan, (1998) ―Aggregate Productivity Growth: Lessons from

Microeconomic Evidence,‖ NBER Working Paper no. 6803. National Bureau of Economic Analysis: Cambridge.

Foster, L., Haltiwanger, J. and C. Syverson, "Reallocation, Firm Turnover, and Efficiency: Selection on Productivity or Profitability?" American Economic Review 2008, 98:1, 394–425,

Frazer, G. (2005), ―Which Firms Die? A Look at Manufacturing Firm Exit in Ghana,‖ Economic

Development and Cultural Change 53: 585-617. Gallego, F. and J. Tessada (2009) Sudden Stops and Reallocation: Evidence from Job Flows in Latin

America.‖ Mimeo. Gomes, J., Greenwood, J. and S. Rebelo, (1997) ―Equilibrium Unemployment.‖ Mimeo, University

of Rochester. Griliches, Z. and Regev, H. (1995) ―Firm productivity in Israeli industry: 1979–1988.‖ Journal of

Econometrics. 65: 175–203. Griliches, Z. and J. Mairesse. (1998). ―Production Functions: The Search for Identification.‖:1-53. in

Econometrics and Economic Theory in the Twentieth Century: The Ragnar Frisch Centennial Symposium,

ed. S. Strøm. Cambridge, UK: Cambridge University Press. Hall R. E. (1990). Invariance

Properties of Solow‘s Productivity Residual. In Diamond, P. (ed.). Growth, Productivity,

Employment. Cambridge, MIT Press,

Hall, R. (1995) ―Lost jobs‖ Brookings Papers on Economic Activity, 1: 221-256.

Hallward-Driemeier, M. (2009) ―Who Survives? The Impact of Corruption, Competition and

Property Rights across Firms‖ World Bank Policy Research Working Paper, No.5084. World

Bank: Washington, D.C.

Hallward-Driemeier, M. and F. Thompson. (2009) ―Creative Destruction and Policy Reforms

Changing Productivity Effects of Firm Turnover in Moroccan Manufacturing.‖ World Bank Policy Research Working Paper, No.5085. World Bank: Washington, D.C.

Hallward-Driemeier, M., B. Rijkers and A. Waxman. (2010) ―Firm Adjustment During Crisis:

Wages vs. Employment‖. Mimeo, World Bank: Washington, D.C..

Page 35: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

35

Harrison, A. and J. Scorse. (2009) "Do foreign-owned firms pay more? Evidence from the

Indonesian manufacturing sector,", in Labor Markets and Economic Development (eds.) R. Kanbur and J. Svejnar, Routledge, Taylor & Francis Group, London and New York.

Harrison, A. and J. Scorse. Forthcoming. ―Moving Up or Moving Out? Anti-Sweatshop Activism and

Labor Market Outcomes. American Economic Review. Hsieh, C. and P. Klenow (2009) "Misallocation and Manufacturing TFP in China and India"

Quarterly Journal of Economics 124(4): 1403-1448. Levinsohn, J. and A. Petrin, (2003). ―Estimating Production Functions Using Inputs to Control for

Unobservables,‖ Review of Economic Studies 70: 317-341. Liu, L. and Tybout, J. (1996) ―Productivity Growth in Chile and Colombia: The Role of Entry, Exit,

and Learning.‖ pp. 73-103 in eds. M. Roberts and J. Tybout. Industrial Evolution in Developing Countries. Oxford University Press: Oxford, UK.

Loayza, N., G. Perry, L. Serven. (2005) ―Schumpeter in the Tropics: Regulation and Macro

Performance.‖ Mimeo, World Bank. Manning, C. (2000) ―The Impact of Financial Crises on Labor Markets, Household Incomes, and

Poverty: A Review of Evidence.‖ Bulletin of Indonesian Economic Studies 36(1): 105-136. McKenzie, D. (2003) ―How do Households Cope with Aggregate Shocks? Evidence from the

Mexican Peso Crisis.‖ World Development 31 ( 7): 1179–1199 McKenzie, D. (2004) ―Aggregate Shocks and Urban Labor Market Responses: Evidence from

Argentina's Financial Crisis‖ Economic Development and Cultural Change, 52:719–758. Mobarak, A. and D. Purbasari. (2008) ―Protection For Sale to Firms: Evidence From Indonesia‖

Mimeo., Yale Univ. Micco, A. and Pages, C. (2004) "Employment Protection and Gross Job Flows: A Differences-in -Differences Approach," RES Working Papers 4365, Inter-American Development Bank, Research Department: Washington, D.C. Nickell, S. (1981) ―Biases in Dynamic Models with Fixed Effects.‖ Econometrica. 49(6): 1417-1426. Nishimura, K., T. Nakajima and K. Kiyota (2005) ―Does the natural selection mechanism still work

in severe recessions? Examination of the Japanese economy in the 1990‖ Journal of Economic Behavior & Organization. 58: 53-78.

Okada, T. and Horioka, C. (2008). "A comment on Nishimura, Nakajima, and Kiyota's "Does the natural selection mechanism still work in severe recessions? Examination of the Japanese economy in the 1990s"," Journal of Economic Behavior & Organization, 67(2): 517-520.

Page 36: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

36

Olley, S. and A. Pakes (1996) ―The dynamics of productivity in the telecommunications equipment industry,‖ Econometrica, 64, 1263-1297.

Organisation for Economic Co-operation and Development (1995, 2002) ―Input-Output Tables.‖

Directorate for Science, Technology and Industry, OECD. Paris.

http://www.oecd.org/document/3/ 0,3343,en_2649_34445_38071427_1_1_1_1,00.html.

Last accessed 01/15/2010.

Ouyang, M. (2009) ―The Scarring Effect of Recessions,‖ The Journal of Monetary Economics. 56(2): 184-199.

Paci, P, Revenga, A. and B. Rijkers (2009) ―Coping with Crises: How to Protect Employment and

Earnings‖ World Bank Policy Research Working Paper 5094. World Bank: Washington, D.C. Peek, J. and E. Rosengren. (2004). ―Unnatural Selection: Perverse Incentives and the Misallocation

of Credit in Japan‖ American Economic Review. 95(4). 1144-1166. Petrin, A. and Levinsohn, J. (2008) "Measuring Aggregate Productivity Growth Using Plant-Level

Data", NBER Working Paper No. 11887.

Raddatz, C. (2006) "Liquidity Needs and Vulnerability to Financial Underdevelopment," Journal of Financial Economics, Elsevier, 80(3):677-722.

Rama, M. (2001) ―The Consequences of Doubling the Minimum Wage.‖ Industrial and Labor Relations

Review. 54(4): 864-81. Rajan, R and Zingales, L. (1998) ―Financial Dependence and Growth‖ The American Economic Review

88(3): 59-586. Restuccia, D and Rogerson, R. (2008) ―Policy distortions and aggregate productivity with

heterogeneous establishments.‖ Review of Economic Dynamics. 11(4): 707-720. Roodman, D. (2006) ―How to Do xtabond2: An Introduction to ―Difference‖ and ―Sysyem‖ GMM

in Stata. Center for Global Development Working Paper No. 103: Washington, D.C. Schumpeter, J. (1939) Business Cycles. McGraw-Hill: New York. Söderbom, M., F. Teal, and A. Harding. (2006) "The Determinants of Survival Among African

Manufacturing Firms." Economic Development and Cultural Change 54(3): 533–56. Syverson, C. (2010) ―What Determines Productivity?‖ NBER Working Paper # 15712. National

Bureau of Economic Research: Cambridge, MA.

World Bank (2009a). World Development Indicators 2009. Washington, D.C.

World Bank (2009b). Indonesia Jobs Report Poverty Team, World Bank Office, Jakarta.

Page 37: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

37

Figure 1 : Growth in Indonesia (1991-2001)

Figure 2: Firm Entry and Exit

-20

-15

-10

-5

0

5

10

15

1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001

% C

han

ge, A

nn

ual

Growth in Production for total manufacturing, sa GDP growthSources: OECD.stat, WDI

0

0.05

0.1

0.15

0.2

1992 1993 1994 1995 1996 1997 1998 1999 2000 2001

Exit rate

Entry rate

Page 38: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

38

Figure 3: Aggregate Job Flows

Figure 4: Aggregate Job Flows Incumbents, Entrants and Exiters

-500000

-250000

0

250000

500000

750000

1000000

1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001

Job Creation

Job Destruction

Net Job Creation

Excess Churning

-400000

-300000

-200000

-100000

0

100000

200000

300000

400000

500000

19

91

19

92

19

93

19

94

19

95

19

96

19

97

19

98

19

99

20

00

20

01

Job Creation IncumbentsJob Creation - Entry

Job Destruction IncumbentsJob Destructions Exiters

Page 39: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

39

Figure 5: FHK Decomposition of the growth of real value added per worker, employment market share

Figure 6: Griliches-Regev decomposition of the growth of real value added per worker, employment market share

Figure 7: FHK Decomposition of Total Factor Productivity, using employment market

share

-0.150

-0.100

-0.050

0.000

0.050

0.100

0.150

0.200

1992 1993 1994 1995 1996 1997 1998 1999 2000 2001

within

between

cross

entry

exit

-0.150

-0.100

-0.050

0.000

0.050

0.100

0.150

1992 1993 1994 1995 1996 1997 1998 1999 2000 2001

within

between

entry

exit

-0.030

-0.020

-0.010

0.000

0.010

0.020

0.030

0.040

0.050

0.060

0.070

1992 1993 1994 1995 1996 1997 1998 1999 2000 2001

within

between

cross

entry

exit

Page 40: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

40

Figure 8: FHK Decomposition, using sales to define market shares

Figure 9: FHK Decomposition, using 3 year window

-0.150

-0.100

-0.050

0.000

0.050

0.100

0.150

1992 1993 1994 1995 1996 1997 1998 1999 2000 2001

within

between

cross

entry

exit

-0.200

-0.150

-0.100

-0.050

0.000

0.050

0.100

0.150

0.200

0.250

0.300

1994 1995 1996 1997 1998 1999 2000 2001

within

between

cross

entry

exit

Page 41: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

41

Table 3.1: Average Entry, Exit and employment growth

Entry, Exit and Employment Growth

Entry Exit

Employment Growth (Survivors)

Employment Growth (Survivors)

mean sd % negative emp change

No emp change

% positive emp change

1992 14.54% 9.74% 2.69% 24.41% 24.88% 14.64% 36.97%

1993 10.51% 9.02% 6.61% 25.38% 20.10% 7.08% 52.18%

1994 11.70% 7.41% 2.66% 24.53% 25.92% 16.88% 37.18%

1995 17.96% 6.19% 1.32% 22.38% 24.49% 17.50% 33.39%

1996 14.41% 8.97% 0.72% 22.56% 28.42% 19.29% 32.26%

1997 8.07% 10.78% -3.30% 22.24% 46.05% 13.59% 23.81%

1998 7.93% 11.21% -0.93% 22.94% 32.30% 10.61% 38.03%

1999 7.61% 5.06% 0.41% 21.19% 27.09% 23.00% 28.87%

2000 5.65% 5.20% 1.11% 19.85% 23.27% 31.38% 27.85%

2001 9.36% 13.66% -0.23% 22.05% 27.32% 26.04% 23.21%

Note: entry and exit are defined as entry into the survey and exit from the survey.

Page 42: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

42

Table 5.1.A Descriptive Statistics, by period

Period precrisis crisis recovery ’91-‘95 ’96-‘97 ’98-‘01

Firmage

All 11.41 11.68 13.32 Surviving(1) 11.56 11.97 13.15 Exiting(2) 9.82 9.29 12.39 Difference(1-2) 1.74*** 2.68*** 0.77***

lnL

All 4.22 4.15 4.19 Surviving(1) 4.25 4.21 4.23 Exiting(2) 3.80 3.64 3.65 Difference(1-2) 0.46*** 0.57*** 0.58***

Ln(K/L)

All 6.84 6.47 6.15 Surviving(1) 6.86 6.51 6.21 Exiting(2) 6.64 6.17 5.53 Difference(1-2) 0.22*** 0.34*** 0.69***

Unskilled ratio

All 0.86 0.86 0.86 Surviving(1) 0.86 0.86 0.86 Exiting(2) 0.88 0.88 0.89 Difference(1-2) -0.02*** -0.02*** -0.03***

%Women

All 0.39 0.39 0.40 Surviving(1) 0.39 0.39 0.39 Exiting(2) 0.40 0.36 0.40 Difference(1-2) -0.01*** 0.03*** -0.01***

Foreign-owned

All 0.05 0.06 0.07 Surviving(1) 0.05 0.06 0.07 Exiting(2) 0.03 0.03 0.04 Difference(1-2) 0.03*** 0.03*** 0.04***

Government Owned

All 0.03 0.03 0.04 Surviving(1) 0.03 0.03 0.04 Exiting(2) 0.02 0.01 0.05 Difference(1-2) 0.01*** 0.01*** -0.01***

Exporter

All 0.17 0.16 0.14 Surviving(1) 0.17 0.17 0.13 Exiting(2) 0.15 0.10 0.10 Difference(1-2) 0.02*** 0.06*** 0.04***

ln (V/L)

All 7.88 7.90 7.95 Surviving(1) 7.91 7.93 7.91 Exiting(2) 7.60 7.72 7.54 Difference(1-2) 0.31*** 0.20*** 0.37***

ln (V/L) (demeaned by sector)

All -0.02 -0.01 0.03 Surviving(1) 0.00 0.01 -0.02 Exiting(2) -0.25 -0.19 -0.29 Difference(1-2) 0.25*** 0.20*** 0.28***

ln (V/L) (demeaned by sector-year)

All -0.00 0.00 0.00 Surviving(1) 0.02 0.02 0.03 Exiting(2) -0.23 -0.16 -0.29 Difference(1-2) 0.25*** 0.18*** 0.32***

*=90%confidence interval, **=95% confidence interval, ***=99% confidence interval for two-tailed t-test. Shaded differences indicate 90% confidence level in rejecting that difference in indicated period is not different from pre-crisis. Bold and underline indicates to the 95% confidence level.

Page 43: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

43

Table 5.1.B Descriptive Statistics (continued), by period

Period precrisis crisis recovery ’91-‘95 ’96-‘97 ’98-‘01

TFP

All 2.69 2.71 2.67 Surviving(1) 2.68 2.70 2.58 Exiting(2)

2.80 2.83 2.63 Difference(1-2)

-0.11*** -0.14*** -0.05***

TFP (demeaned by sector)

All -0.00 0.02 -0.01 Surviving(1) -0.00 0.02 -0.10 Exiting(2)

-0.01 0.03 -0.06 Difference(1-2)

0.01 -0.01 -0.03***

TFP (demeaned by sector-year)

All 0.00 -0.00 -0.00 Surviving(1) 0.00 -0.00 0.00 Exiting(2)

-0.01 0.02 -0.03 Difference(1-2)

0.01 -0.02** 0.03***

turnover

All 19.27 19.30 19.25

Surviving(1) 19.20 19.19 19.20 Exiting(2)

20.02 20.17 19.75 Difference(1-2)

-0.82*** -0.97*** -0.56***

Rajan & Zingales – financial dependence

All 0.16 0.17 0.23 Surviving(1) 0.16 0.17 0.23 Exiting(2)

0.15 0.17 0.19

Difference(1-2) 0.02*** -0.00 0.04***

Natural rate of entry

All 6.05 6.02 5.95 Surviving(1) 6.04 6.01 5.94 Exiting(2)

6.20 6.12 6.06 Difference(1-2)

-0.16*** -0.11*** -0.12***

Herfindahl index

All 0.03 0.04 0.04 Surviving(1) 0.03 0.04 0.04 Exiting(2)

0.03 0.04 0.04 Difference(1-2)

-0.001*** 0.00 0.00***

Invested (since t -2) - loan

All 0.27 0.27 0.21 Surviving(1) 0.27 0.27 0.24 Exiting(2)

0.20 0.21 0.16 Difference(1-2)

0.07*** 0.06*** 0.08***

Invested (since t -2) - no loan

All 0.34 0.35 0.33

Surviving(1) 0.34 0.35 0.32 Exiting(2)

0.34 0.34 0.34 Difference(1-2)

0.00 0.00 -0.01** *=90%confidence interval, **=95% confidence interval, ***=99% confidence interval for two-tailed t-test. Shaded differences indicate 90% confidence level in rejecting that difference in indicated period is not different from precrisis. Bold and underline indicates to the 95% confidence level. TFP estimates based on a four-factor sector-specific gross output production function estimated by OLS (see Appendix A)

Page 44: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

44

Table 5.2 : Survival model

Logistic Survival model -Odds Ratios: Relative Probability of Exit -

Period all precrisis crisis recovery

‘93-‘01 ’93-‘96 ’97-‘98 ’99-‘01

coef/se coef/se coef/se coef/se

Firmage (log) 0.793*** 0.791*** 0.697*** 0.932***

(0.009) (0.015) (0.013) (0.020)

lnL 0.250*** 0.288*** 0.237*** 0.209***

(0.016) (0.029) (0.028) (0.023)

lnL2 1.099*** 1.087*** 1.101*** 1.115***

(0.007) (0.012) (0.014) (0.013)

Unskilled ratio 1.123 0.941 1.100 1.460***

(0.081) (0.118) (0.137) (0.187)

%Women 1.014 1.096 0.903* 1.087

(0.036) (0.069) (0.056) (0.067)

Foreign-owned 0.988 0.817* 0.970 1.275***

(0.057) (0.089) (0.098) (0.118)

Government Owned 0.793*** 1.060 1.123 0.718***

(0.050) (0.132) (0.154) (0.062)

Exporter 1.214*** 1.498*** 1.047 1.109*

Ln Value-added per worker

(0.041) (0.082) (0.063) (0.069)

0.863*** 0.801*** 0.953*** 0.844***

(0.009) (0.015) (0.017) (0.014)

Province dummies Yes Yes Yes Yes

Year dummies Yes Yes Yes Yes Number of observations 145,882 51,797 39,034 54,986

chi2 (df) 5,792 (41) 1,493 (36) 1,804(34) 2,831 (36)

Log-Likelihood -39,448 -13,190 -12,482 -13,422

ll_0 -42,344 -13,936 -13,384 -14,838

Pseudo R2 0.068 0.054 0.067 0.095

note: - *** p<0.01, ** p<0.05, * p<0.1 - Shaded differences indicate 90% confidence level in rejecting that difference in indicated period is not different from pre-crisis. Bold and underline indicates to the 95% confidence level.

Page 45: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

45

Table 5.3: Robustness Checks and Alternative Specifications

Logistic Survival model -Odds Ratios: Relative Probability of Exit -

Robustness checks and alternative specifications

‘93-‘01 ’93-‘96 ’97-‘98 ’99-‘01 Odds ratio/se Odds ratio/se Odds ratio/se Odds ratio/se

SECTOR AND SECTOR-TIME EFFECTS Including industry & industry*year 0.885*** 0.828*** 0.964* 0.870***

dummies (0.010) (0.016) (0.019) (0.016)

ln(V/L) normalized by sector 0.890*** 0.831*** 0.964* 0.869***

and year (0.010) (0.016) (0.019) (0.016)

OUTLIERS AND ANOMALOUS OBSERVATIONS

Including anomalous observations 0.867*** 0.805*** 0.941*** 0.859***

(0.009) (0.015) (0.017) (0.015)

Excluding top and bottom 10% of ln(V/L)

0.943*** 0.926** 0.996 0.929**

10% of ln(V/L) by sector-year (0.017) (0.031) (0.033) (0.028)

ALTERNATIVE SPECIFICATION: C-LOGLOG MODEL C-loglog Specification 0.867*** 0.805*** 0.941*** 0.859***

(0.009) (0.015) (0.017) (0.015)

LAGGED VALUE ADDED PER WORKER ln(V/L)- t-2 0.879*** 0.837*** 0.929*** 0.859***

(0.011) (0.019) (0.020) (0.017)

N 129,709 44,427 34,491 50,732

ln(V/L)- t-3 0.904*** 0.900*** 0.936*** 0.873***

(0.012) (0.023) (0.024) (0.018)

N 115,918 39,262 29,318 47,285

DIFFERENT EXIT TRESHHOLDS Exit- 30 people treshhold 0.862*** 0.814*** 0.928*** 0.847***

(0.008) (0.014) (0.016) (0.013)

Exit-40 people treshhold 0.858*** 0.809*** 0.922*** 0.850***

(0.008) (0.014) (0.016) (0.013)

Exit 50 people treshhold 0.867*** 0.822*** 0.924*** 0.863***

(0.008) (0.014) (0.016) (0.014)

CONTROLLING FOR CAPITAL INTENSITY Ln (K/L) - Observed capital 1.011 1.060*** 0.973 0.987

(0.009) (0.015) (0.017) (0.017)

ln(V/L) 0.862*** 0.790*** 0.967 0.823***

(0.013) (0.020) (0.025) (0.024)

N 103,708 40,804 29,777 33,075

Ln (K/L) - Imputed capital 1.012 1.060*** 0.988 0.980

(0.009) (0.014) (0.017) (0.013)

ln(V/L) 0.875*** 0.799*** 0.968 0.877***

(0.011) (0.020) (0.024) (0.022)

N 145,882 51,797 39,034 54,986

LONG-RUN SURVIVAL Period 1991-1996 1996-2001 1993-1996 1996-1999 Survival time 5 year 5 years 3 years 3 years ln(V/L) 0.756*** 0.868*** 0.728*** 0.953***

(0.015) (0.015) (0.017) (0.018) - *** p<0.01, ** p<0.05, * p<0.1 - unless otherwise indicated all specifications include the following controls: firmage (log), lnL, lnL2, Ln(K/L), Unskilled ratio, %Women, Foreign-owned, Government Owned, Exporter, Province dummies and Year dummies -Shaded differences indicate 90% confidence level in rejecting that difference in indicated period is not different from pre-crisis. Bold and underline indicates to the 95% confidence level. - Standard errors for regressions including imputed capital are bootstrapped.

Page 46: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

46

Table 5.4: TFP and survival

Estimated impact of TFP on survival

Logistic Survival model -Odds Ratios: Relative Probability of Exit -

all precrisis crisis recovery

‘93-‘01 ’93-‘96 ’97-‘98 ’99-‘01

Odds ratio/se Odds ratio/se Odds ratio/se Odds ratio/se

TFP – OLS 0.841*** 0.780*** 1.008 0.755***

(0.038) (0.074) (0.077) (0.070)

TFP-OLS 0.842*** 0.769*** 1.008 0.762***

No controls for capital (0.042) (0.072) (0.078) (0.071)

TFP – OLS 1.096*** 1.138*** 1.237*** 0.938**

Without industry dummies (0.023) (0.033) (0.046) (0.030)

TFP – OLS 0.873*** 0.780*** 1.028 0.791***

normalized by sector-year (0.046) (0.074) (0.089) (0.068)

TFP – FE 0.821*** 0.728*** 1.012 0.728***

(0.038) (0.061) (0.076) (0.064)

TFP – Solow procedure 0.984 0.829* 1.206** 0.929

(0.059) (0.088) (0.093) (0.115)

TFP- ACF 0.743*** 0.649*** 0.890* 0.704***

(0.032) (0.048) (0.062) (0.060)

TFP – VA - OLS 0.875*** 0.797*** 0.988 0.869***

(0.022) (0.030) (0.038) (0.038)

N 75,593 29,734 20,424 25,380

Note: - *** p<0.01, ** p<0.05, * p<0.1 -unless otherwise indicated all specifications include the following controls: firmage (log), lnL, lnL2, Ln(K/L), Unskilled ratio, %Women, Foreign-owned, Government Owned, Exporter, Industry Dummies, Province dummies and Year dummies -Shaded differences indicate 90% confidence level in rejecting that difference in indicated period is not different from pre-crisis. Bold and underline indicates to the 95% confidence level. - Standard errors are bootstrapped

Page 47: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

47

Table 5.5: Industry Characteristics and Exit

Logistic Survival model -Odds Ratios: Relative Probability of Exit -

all precrisis crisis recovery

‘93-‘01 ’93-‘96 ’97-‘98 ’99-‘01

Odds

ratio/se Odds

ratio/se Odds

ratio/se Odds ratio/se

Industry Characteristics RZ_fin_dependence 0.898** 0.803*** 1.319*** 0.805***

(0.040) (0.068) (0.109) (0.057)

turnover 1.005** 1.002 1.021*** 1.002

(0.003) (0.005) (0.005) (0.004)

nat_rate_entry 1.090*** 1.122*** 1.050** 1.063***

(0.013) (0.025) (0.022) (0.022)

Herfindahl 0.829 1.237 3.179*** 0.536***

(0.129) (0.455) (1.197) (0.111)

ln(V/L) 0.873*** 0.820*** 0.960** 0.853***

(0.009) (0.016) (0.018) (0.015)

Number of observations 145,882 51,797 39,034 54,986

Industry Characteristics Interacted with ln(V/L) RZ_fin_dep*ln(V/L) 1.054* 1.135** 0.902* 1.144***

(0.032) (0.064) (0.052) (0.053)

RZ_fin_dependence 0.611** 0.310*** 2.923** 0.289***

(0.142) (0.135) (1.297) (0.104)

turnover 1.008*** 1.004 1.018*** 1.003

(0.003) (0.005) (0.005) (0.005)

nat_rate_entry 1.078*** 1.115*** 1.062*** 1.056**

(0.014) (0.025) (0.023) (0.023)

Herfindahl 1.022 0.887 3.592*** 0.317**

(0.249) (0.366) (1.471) (0.155)

ln(V/L) 0.871*** 0.808*** 0.975 0.840***

(0.010) (0.017) (0.020) (0.016)

Number of observations 141,903 50,333 37,888 53,626 Shaded differences indicate 90% confidence level in rejecting that difference in indicated period is not different from pre-crisis. Bold and underline indicates to the 95% confidence level. All specifications include the following controls: firmage (log), lnL, lnL2, Ln(K/L), Unskilled ratio, %Women, Foreign-owned, Government Owned, Exporter, Industry Dummies, Province dummies and Year dummies

Page 48: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

48

Table 5.6: Dependence on Credit and Exit

Logistic Survival model -Odds Ratios: Relative Probability of Exit -

Investment Invested (since t-2) -used loans 0.815*** 0.789*** 0.889*** 0.743***

(0.022) (0.036) (0.040) (0.035)

Invested (since t-2) - no loans 0.871*** 0.893*** 0.876*** 0.837***

(0.020) (0.036) (0.034) (0.033)

ln(V/L) 0.884*** 0.833*** 0.962* 0.867***

(0.010) (0.016) (0.019) (0.016)

Number of observations 145,851 51,785 39,026 54,975 Financial Characteristics Interacted with ln(V/L)

Invested (since t-2) -used loans 1.172 0.829 3.376*** 0.829

(0.204) (0.254) (1.056) (0.248)

Invested (since t-2) - no loans 1.473** 2.309*** 1.436 1.306

(0.229) (0.662) (0.405) (0.328)

Invested (since t-2) –loans * ln(V/L) 0.954** 0.993 0.843*** 0.985

(0.021) (0.039) (0.033) (0.038)

Invested (since t-2) - no loans * ln(V/L) 0.933*** 0.882*** 0.938* 0.942*

(0.019) (0.033) (0.034) (0.031)

ln(V/L) 0.916*** 0.873*** 1.025 0.888***

(0.014) (0.025) (0.028) (0.022)

Number of observations 145,851 51,785 39,026 54,975 Note: Shaded differences indicate 90% confidence level in rejecting that difference in indicated period is not different from pre-crisis. Bold and underline indicates to the 95% confidence level.

Page 49: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

49

Table 5.7: Employment Growth

Employment Growth Model – OLS -

all precrisis crisis recovery

‘93-‘01 ’93-‘96 ’97-‘98 ’99-‘01

coef/se coef/se coef/se coef/se

Firmage (log) -0.018*** -0.018*** -0.020*** -0.016***

(0.001) (0.001) (0.002) (0.001)

lnL -0.155*** -0.214*** -0.163*** -0.089***

(0.010) (0.023) (0.009) (0.008)

lnL2 0.014*** 0.019*** 0.014*** 0.007***

(0.001) (0.002) (0.001) (0.001)

Unskilled ratio 0.025*** 0.052*** 0.020** 0.006

(0.006) (0.010) (0.010) (0.008)

%Women 0.029*** 0.046*** 0.038*** 0.009**

(0.003) (0.005) (0.005) (0.004)

Foreign-owned 0.014*** 0.026*** 0.020*** 0.003

(0.003) (0.006) (0.007) (0.005)

Government Owned -0.009* -0.013 0.007 -0.010

(0.005) (0.009) (0.010) (0.007)

Exporter 0.014*** 0.004 0.036*** 0.004

(0.002) (0.004) (0.005) (0.004)

Ln (V/L) 0.024*** 0.034*** 0.021*** 0.019***

(0.001) (0.002) (0.001) (0.001)

Province dummies Yes Yes Yes Yes

Year dummies Yes Yes Yes Yes

Number of observations 132,586 47,520 34,464 50,602

R2 0.038 0.054 0.042 0.020

Adjusted R2 0.038 0.053 0.041 0.019

Note: -*** p<0.01, ** p<0.05, * p<0.1 -Shaded differences indicate 90% confidence level in rejecting that difference in indicated period is not different from pre-crisis. Bold and underline indicates to the 95% confidence

level.

Page 50: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

50

Table 5.8: Employment Growth – Alternatives and Robustness

-*** p<0.01, ** p<0.05, * p<0.1Shaded differences indicate 90% confidence level in rejecting that difference in indicated period is not different from pre-crisis. Bold and underline indicates to the 95% confidence level. -Specifications include the following controls: firmage (log), lnL, lnL2, Ln(K/L), Unskilled ratio, %Women, Foreign-owned, Government Owned, Exporter, and Province, year and industry dummies. - Standard errors for regressions including controls for TFP or imputed capital are bootstrapped

Employment Growth Models - OLS Robustness checks and alternative specifications all precrisis crisis recovery

‘93-‘01 ’93-‘96 ’97-‘98 ’99-‘01 coef/se coef/se coef/se coef/se

INCLUDING EXITING FIRMS (e.g set dlnL=--log(Lt-1-20) if exited) Min ( set dlnL=--log(Lt20) if

exited

0.047*** 0.065*** 0.033*** 0.042***

If exited (0.003) (0.005) (0.006) (0.004)

Max set dlnL=--log(Lt) iexited 0.063*** 0.088*** 0.040*** 0.056***

If exited (0.003) (0.006) (0.007) (0.005)

SECTOR AND SECTOR-TIME EFFECTS

Controlling for industry & 0.025*** 0.034*** 0.022*** 0.019***

Industry-time effects (0.001) (0.002) (0.002) (0.001)

ln(V/L) normalized by sector

and year

0.025*** 0.034*** 0.022*** 0.019***

(0.001) (0.002) (0.002) (0.001)

SURVIVOR BIAS – LONG-RUN SURVIVORS ONLY

ln(V/L) 0.025*** 0.039*** 0.022*** 0.019***

(0.001) (0.003) (0.002) (0.001)

N 103,318 30,999 26,466 45,853

TFP

TFP 0.023*** 0.030*** 0.024*** 0.014***

(0.003) (0.005) (0.006) (0.005)

LAGGED VALUE-ADDED PER WORKER

Lag ln(V/L) 0.017*** 0.022*** 0.017*** 0.014***

(0.001) (0.002) (0.002) (0.001)

CONTROLLING FOR CAPITAL INTENSITY

Ln (K/L) - Observed 0.008*** 0.009*** 0.003** 0.010***

(0.001) (0.001) (0.001) (0.001)

ln(V/L) 0.019*** 0.026*** 0.019*** 0.014***

(0.001) (0.002) (0.002) (0.002)

N 95,061 37,727 26,543 30,791

Ln (K/L) -Imputed 0.008*** 0.010*** 0.004*** 0.010***

(0.001) (0.001) (0.001) (0.001)

ln(V/L) 0.018*** 0.027*** 0.018*** 0.012***

(0.001) (0.002) (0.002) (0.001)

N 132,586 47,520 34,464 50,602

LONGER WINDOWS

Period 1991-1996 1996-2001 1996-1999 1999-2003 Survival time 5 year 5 years 3 years 3 years

ln(V/L) 0.065*** 0.053*** 0.070*** 0.035***

(0.007) (0.005) (0.006) (0.004)

N 9,947 13,251 12,735 15,398

Page 51: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

51

Table 5.9: Dependence of Employment Growth on Industry Characteristics and Exit

Employment Growth Model – OLS -

all precrisis crisis recovery

‘93-‘01 ’93-‘96 ’97-‘98 ’99-‘01

coef/se coef/se coef/se coef/se

Industry Characteristics RZ_fin_dependence 0.001 -0.013** -0.025*** 0.015***

(0.003) (0.006) (0.007) (0.004)

turnover -0.001*** -0.002*** -0.001** -0.000

(0.000) (0.000) (0.000) (0.000)

nat_rate_entry 0.001 0.002 0.000 0.001

(0.001) (0.002) (0.002) (0.001)

Herfindahl 0.084*** 0.121*** -0.055* 0.116***

(0.016) (0.027) (0.030) (0.026)

ln(V/L) 0.025*** 0.035*** 0.021*** 0.019***

(0.001) (0.002) (0.002) (0.001)

Number of observations 132,556 47,508 34,456 50,592 Shaded differences indicate 90% confidence level in rejecting that difference in indicated period is not different from pre-crisis. Bold and underline indicates to the 95% confidence level. specifications include the following controls: firmage (log), lnL, lnL2, Ln(K/L), Unskilled ratio, %Women, Foreign-owned, Government Owned, Exporter, and Province, year and industry dummies.

Table 5.10: Dependence of Employment Growth on Credit and Exit

Employment Growth Model – OLS -

all precrisis crisis recovery

‘93-‘01 ’93-‘96 ’97-‘98 ’99-‘01

coef/se coef/se coef/se coef/se

Investment Invested (since t-2) -used loans 0.004** 0.008*** -0.002 0.004*

(0.002) (0.003) (0.003) (0.003)

Invested (since t-2) - no loans 0.001 0.006** -0.002 -0.002

(0.001) (0.003) (0.003) (0.002)

ln(V/L) 0.025*** 0.034*** 0.022*** 0.019***

(0.001) (0.002) (0.002) (0.001)

Number of observations 132,556 47,508 34,456 50,592 Shaded differences indicate 90% confidence level in rejecting that difference in indicated period is not different from pre-crisis. Bold and underline indicates to the 95% confidence level. Note: specifications include the following controls: firmage (log), lnL, lnL2, Ln(K/L), Unskilled ratio, %Women, Foreign-owned, Government Owned, Exporter, and Province, year and industry dummies.

Page 52: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

52

Table 5.11: Fixed Effects Regression with Period Interactions

Employment Growth – Fixed Effects

Coef/se Coef/se

lnfirmage 0.023*** Crisis*ln(V/L) 0.003

(0.005) (0.002)

lnL -0.715*** Recovery*lnfirmage -0.011***

(0.027) (0.004)

lnL2 0.030*** Recovery*lnL 0.015

(0.003) (0.020)

Unskilled ratio 0.022

Recovery*lnL2 -0.002

(0.014) (0.002)

%Women 0.031*** Recovery*skratio -0.003

(0.011) (0.015)

Foreign-Ownership -0.005 Recovery*%Women -0.027***

(0.011) (0.008)

Government-owned -0.017 Recovery*Foreign-

Ownership 0.005

(0.013) (0.009)

exporter -0.009* Recovery*Government-

owned 0.004

(0.005) (0.015)

ln(V/L) 0.014*** Recovery*exporter 0.015**

(0.002) (0.007)

Crisis*lnfirmage -0.007*** Recovery*ln(V/L) 0.001

(0.003) (0.003)

Crisis*lnL -0.018 Year dummies Yes

(0.016)

Crisis*lnL2 0.001 Number of

observations 132,586

(0.002) R2 0.261

Crisis*skratio -0.017 Adjusted R2 0.261

(0.014)

Crisis*%Women -0.008

(0.008)

Crisis*Foreign-Ownership 0.015*

(0.009)

Crisis*Government-owned 0.020

(0.013)

note: *** p<0.01, ** p<0.05, * p<0.1

Page 53: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

53

Appendix A: Estimating Productivity

A.1 Econometric Framework

Our starting point is a four-factor industry-specific gross-output Cobb-Douglas production

function:

𝑦𝑖𝑗𝑡 = 𝛽𝑘𝑗 𝑘𝑖𝑗𝑡 + 𝛽𝑚𝑗 𝑚𝑖𝑗𝑡 + 𝛽𝑏𝑗 𝑏𝑖𝑗𝑡 + 𝛽𝑤𝑗 𝑤𝑖𝑗𝑡 + 𝜔𝑖𝑗𝑡 + 𝑣𝑖𝑗𝑡

Where 𝑦𝑖𝑗𝑡 is the log of output of firm i operating in sector j at time t, 𝑘𝑖𝑗𝑡 is the log of capital, 𝑏𝑖𝑗𝑡

the log of blue-collar workers, 𝑤𝑖𝑗𝑡 is the log of white-collar labour and 𝜔𝑖𝑗𝑡 denotes the log of

TFP, and 𝑣𝑖𝑗𝑡 is a random error term. To account for differences in technology across sectors,

factor-coefficients are allowed to vary by sector, as indicated by the sector subscripts j.

A well known problem is that unobserved productivity 𝜔𝑖𝑗𝑡 may be correlated with inputs,

causing OLS estimates of the production function to be biased. To assess how robust our

productivity estimates are to the resulting endogeneity bias, we contrast OLS estimates with those

obtained by means of Fixed Effects (FE) estimation. In addition, we compute input coefficients

using the Solow method and estimate productivity using the Ackerberg-Caves-Frazer (2006)

(henceforth ACF) procedure.

The Fixed-Effect estimator eliminates bias that may arise because of a correlation between

time-invariant unobservable productivity and input choices. Unfortunately, Fixed effects estimators

suffer from the Nickell bias, arising because the transformed explanatory variables and the

transformed error term are negatively correlated by construction. In addition, the empirical

performance of Fixed Effects Estimator has been disappointing, often resulting in implausibly low

estimates of returns to scale (see e.g. Gricliches and Mairesse, 1996).26 As pointed out by Bond

(2002), OLS and Fixed Effects estimators are biased in opposite directions and thus provide a

confidence interval within which we may expect our productivity-estimates to lie.27

Factor shares are also computed using the Solow method, a non-parametric procedure to

obtain output elasticities. Under the assumption of constant returns to scale and perfectly

competitive markets, 𝛽𝑘𝑗 , 𝛽𝑚𝑗 , 𝛽𝑏𝑗 and 𝛽𝑤𝑗 are equal to the shares of capital costs, materials costs

and the wagebills for blue-collar and white collar workers, respectively, in total costs (see Hall,

26 Estimating the production function in first-differences would circumvent the Nickell bias and eliminate fixed effects, yet exacerbates measurement error. 27 The Difference and Systems GMM methods (see Arelano and Bond, 1991, Arellano and Bover, 1995, Blundell and Bond, 1999) are in principle very well-suited to tackle potential endogeneity problems. Unfortunately, however, we were not able to identify a suitable specification; none of our instrumenting sets were statistically satisfactory and we therefore eschew this approach.

Page 54: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

54

1990). That is:

𝑆𝐾 =rK

rK + wbB + ww W + mM= 𝛽𝑘𝑗

𝑆𝑀 =mM

rK + wb B + ww W + mM= 𝛽𝑚𝑗

𝑆𝑊 =ww W

rK + wbB + ww W + mM= 𝛽𝑤𝑗

𝑆𝐵 =wbB

rK + wb B + ww W + mM= 𝛽𝑏𝑗

where 𝑆𝐾 is the capital cost share, 𝑆𝑀 is the material inputs cost share, while 𝑆𝑊 and 𝑆𝐵 are

the share of the wagebill for white and blue collar workers in total cost, respectively, r is the rental

cost of capital, m is the unit price of material inputs ww is the average wage of white collar workers

and wb is the average wage for blue collar workers. If the assumptions of perfect competition and

constant returns to scale hold the average of these capital-input and labor shares should not differ

from the capital- and labor-output elasticities estimated by means of OLS.28 If they differ, this is a

sign that endogeneity may be a problem, although it might also be that the assumptions of constant

returns and perfect competition are false. Since the data lack reliable information on expenditure on

capital, in our empirical implementation, the value of total output is used as a measure of rK +

wbB + ww W + mM; information on expenditure on labor and material inputs enables us to

retrieve an estimate of rK.

In addition, we follow the estimation procedure outlined by Ackerberg, Caves and Frazer

(2006).29 Assuming that materials are a perfectly flexible input, chosen at time t, (when productivity

is observed) and that the prices of raw-materials are constant in the cross-section enables us to

obtain a value-added production function:30

𝑣𝑎𝑖𝑗𝑡 = 𝛽𝑘𝑗 𝑘𝑖𝑗𝑡 + 𝛽𝑏𝑗 𝑏𝑖𝑗𝑡 + 𝛽𝑤𝑗 𝑤𝑖𝑗𝑡 + 𝜔 ∗𝑖𝑗𝑡 + 𝑣𝑖𝑗𝑡

If we are furthermore willing to assume that capital, 𝑘𝑖𝑗𝑡 k, is chosen at period t-1 (since

28 Of course, it is quite restrictive to assume that the production function exhibits constant returns to scale and that markets are competitive. These assumptions are not needed if the production function is estimated by means of regression. 29 Note: the exposition here borrows heavily from Ackerberg et al. (2006), Chan (2009) and Balasubramanian and Sivadasan (2009). We are grateful to the latter Jagadeesh Sivadasan for sharing the code used to implement the ACF procedure. 30 Note that these are very strong assumption and that, in general, returns to scale estimated on the basis of value-added production functions are biased unless i) competition is perfect ii) the substitutability between material inputs and other inputs is zero (see Fernald and Basu, 1996) .

Page 55: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

55

investment takes time to materialize; capital is assumed to follow the law of motion 𝑘𝑖𝑗𝑡 =

(1 − 𝛿)𝑘𝑖𝑗𝑡 −1 + 𝑖𝑖𝑗𝑡 −1 , where 𝛿 represents depreciation and 𝑖 represents investment), and that

white and blue collar labour are also chosen at time t-1), that is, before material inputs are chosen,

and that productivity 𝜔 ∗𝑖𝑗𝑡 evolves according to a first-order Markov process between periods t-1

and t: i.e 𝑝(𝜔 ∗𝑖𝑗𝑡 𝐼𝑡−1 = 𝑝(𝜔 ∗𝑖𝑗𝑡 𝜔 ∗𝑖𝑗𝑡 −1 , material input demand will be a function of

productivity, capital and labour inputs; 𝑚𝑖𝑗𝑡 𝑘𝑖𝑗𝑡 , 𝑏𝑖𝑗𝑡 , 𝑤𝑖𝑗𝑡 , 𝜔 ∗𝑖𝑗𝑡 , that is monotonically increasing

in 𝜔 ∗𝑖𝑗𝑡 . Inverting this demand function enables us to rewrite the production function as an

equation of the form:

𝛷𝑖𝑗𝑡 = 𝛽𝑘𝑗𝑘𝑖𝑗𝑡 + 𝛽𝑏𝑗 𝑏𝑖𝑗𝑡 + 𝛽𝑤𝑗 𝑤𝑖𝑗𝑡 + 𝑓𝑗𝑡−1 𝑘𝑖𝑗𝑡 , 𝑏𝑖𝑗𝑡 , 𝑤𝑖𝑗𝑡 , 𝜔 ∗𝑖𝑗𝑡 + 𝑣𝑖𝑗𝑡

ACF propose estimating this equation in two stages. In the first stage, output is estimated as

polynomial of capital, blue-collar labor, white-collar labor, raw materials and output, enabling us to

get rid of 𝑣𝑖𝑗𝑡 .

In the second stage, estimates of 𝛽𝑘𝑗 , 𝛽𝑤𝑗 and 𝛽𝑏𝑗 can be retrieved. The timing assumptions

on capital and labour input choices in conjunction with the supposition that productivity follows a

Markov process – i.e. that 𝜔 ∗𝑖𝑗𝑡 = 𝐸𝑡−1 𝜔 ∗𝑖𝑗𝑡 + 𝝃𝒊𝒋𝒕 = 𝐸 𝜔 ∗𝑖𝑗𝑡 |𝜔 ∗𝑖𝑗𝑡 −1 + 𝝃𝒊𝒋𝒕 - enable us to

formulate three orthogonality conditions:

𝐸 𝑘𝑖𝑗𝑡 𝝃𝒊𝒋𝒕 = 0

𝐸 𝑏𝑖𝑗𝑡 1𝝃𝒊𝒋𝒕 = 0

𝐸 𝑤𝑖𝑗𝑡 𝝃𝒊𝒋𝒕 = 0

where 𝝃𝒊𝒋𝒕 is the productivity innovation. Appealing to the analogy principle allows us to replace

these population moments by their sample counterparts and to retrieve consistent estimates of 𝛽𝑘𝑗

and 𝛽𝑙𝑗 by minimizing the criterion function:

𝝃𝒊𝒋𝒕

𝑵

𝒊=𝟏

𝑻

𝒕=𝟏

𝑘𝑖𝑗𝑡

𝑏𝑖𝑗𝑡

𝑤𝑖𝑗𝑡

𝑻

𝑪 𝝃𝒊𝒋𝒕

𝑵

𝒊=𝟏

𝑻

𝒕=𝟏

𝑘𝑖𝑗𝑡

𝑏𝑖𝑗𝑡

𝑤𝑖𝑗𝑡

Empirical estimates of the productivity innovation 𝝃𝒊𝒋𝒕 can be retrieved by non-parametrically

regressing 𝜔 ∗𝑖𝑗𝑡 on 𝜔 ∗𝑖𝑗𝑡 −1 , and 𝜔 ∗𝑖𝑗𝑡 can be obtained by computing 𝛷𝑖𝑗𝑡 -𝛽𝑘𝑗

𝐺𝑘𝑖𝑗𝑡 − 𝛽𝑤𝑗𝐺𝑤 −

𝛽𝑏𝑗𝐺𝑏𝑖𝑗𝑡 , where 𝛽𝑘𝑗

𝐺𝑚, 𝛽𝑤𝑗𝐺

and 𝛽𝑏𝑗𝐺

denote initial values for 𝛽𝑘𝑗 , 𝛽𝑤𝑗 and 𝛽𝑏𝑗 . We use OLS

estimates as initial values.

Page 56: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

56

A.2 Results31

Table A.1: Correlation matrix between different productivity measures

TFP-Gross output-OLS

TFP-Gross output-FE

TFP-VA-OLS

TFP-VA-ACF

TFP-SOLOW ln(Y/L) ln(VA/L)

TFP-Gross output-OLS 1 TFP-Gross output-FE 0.940 1

TFP-VA-OLS 0.538 0.588 1 TFP-VA-ACF 0.403 0.424 0.679 1

TFP-SOLOW 0.719 0.686 0.559 0.286 1 ln(Y/L) 0.039 0.243 0.521 0.515 0.184 1

ln(VA/L) 0.242 0.434 0.728 0.357 0.322 0.863 1 - All correlation are significant at the 0.001 significance level.

Table A.1 presents a correlation matrix between the different demeaned TFP measures. All

measures are positively correlated with each other and with output and value-added per worker.

Overall, then, the relative ranking of firms in terms of productivity is rather robust to using different

methods of estimating productivity.

31

Sector-specific production function estimates are available from the authors upon request, but suppressed to

conserve space.

Page 57: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

57

Appendix B: Data Appendix

B.1 Construction of panel

Survei Industri (SI) from Indonesia‘s statistical bureau (Badan Pusat Statistik—BPS) provides the

plant-level data used in this paper. The SI is an annual census of all manufacturing establishments with more

than 20 employees. While the annual survey has been conducted going back to 1975, field procedures for

identifying new firms were dramatically improved over 1985-1990, making the accuracy of measuring entry

prior to 1990 problematic (see Ascwicahyono, 2008). As a result, this paper only considers the data post

1990. We use data up until 2001, since data from later years are not immediately comparable due to changes

in firm identifiers, changing definitions of key explanatory variables, and province splits.1

B.2 Construction of Key Explanatory Variables

Sector

Sector of main product In order to classify establishments by industry, BPS records the five-digit

International Standard Industrial Classification (ISIC) for firms based on the product with the largest

production value in any given year. In 2001, BPS changed the classification of plants from the second

revision of the ISIC to the third. A consistent bridge between these different coding systems was constructed

based upon the inclusion of both codings in the dataset for the 2000 survey database. This bridge was

corroborated using a bridge provided by BPS. In many cases, the industry code provided in the dataset was

truncated to four or fewer digits. Where possible, if an adjacent year‘s reported industry for the same firm

was available that was used to fill in the truncated digits. Where plants produce multiple products or the

production processes allow changing from one product to another, we may expect to see the coded industry

of production change from year to year. In such cases, the mode sector code is used (with ties going to the

initial sector code reported).

Labor

Total labor for an establishment was defined as the sum of all paid and non-paid workers, whether

production or non-production workers.32 Production workers were defined by BPS as all workers who work

directly in the production process or activities connected with production process, and non-production

workers were defined to be all other workers. These definitions roughly correspond to traditional definitions

of blue- and white-collar workers, respectively (see below).

Blue collar workers the sum of all production and unpaid workers

White collar workers: the sum of all non-production workers

Unskilled ratio. The unskilled ratio was constructed as the share of blue collar workers (including unpaid

production workers) as a share of the total establishment employment.

The establishment-level total wage bill was constructed as the sum of cash wages/salary and in-kind benefits

for production and non-production workers deflated to 1993 rupiah using the national consumer price index

obtained from the World Development Indicators.

32

Although BPS does not explicitly define it, this is understood to include both permanent and temporary workers.

Page 58: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

58

Capital and materials

Capital was defined using the provided estimated value of machinery and equipment for each establishment

in a given year. Where the estimated value was not available, the book value was used as provided. These

values were deflated to 1993 rupiah values on an annual basis using the reported GDP deflator for machinery

for 1993-2001 provided by BPS and using the manufacturing sector deflator for 1990-1992. The

manufacturing survey does not include data on the value of the capital stock for 1996, so these values were

interpolated using predicted values constructed from a regression of the capital stock on contemporaneous

output, material inputs, labour usage, ownership characteristics, whether or not the firm exports, province

and lagged capital for 1991-1995. In a separate imputation procedure, we also imputed capital for firms that

did not report their capital stocks.

Materials were defined using the total reported value of raw materials used by the firm. These were deflated

to 1993 rupiah using a deflator constructed from 2-digit industry GDP deflators weighted by the shares of

input indicated from input-output tables obtained from the OECD Input-Output Table Database for 1995

and 2000.

Value-added and gross output

Value-added and gross output were calculated by BPS on the basis of the total value of production (and

net of total expenditure for value added). Both variables were deflated to 1993 rupiah using a 2-digit industry

level GDP deflator.

Firm age

Firm age was constructed using the difference between the survey year and the year the firm reported the

start of production in that province

Sector characteristics: Financial Dependence, turnover and the natural rate of entry

3-digit ISIC revision 3 industry-level measures for financial dependence, turnover, the natural rate of entry are

based on those presented in Rajan and Zingales (1998) and Micco and Pages (2004). Note that these proxies

are based on US industries.

The financial dependence measure is equal to the industry-level median of the ratio of capital expenditures

minus cash flow over capital expenditures. The numerator and denominator are summed over all years for

each firm before dividing collected from Standard and Poor‘s Compustat database. Cash flow is defined as the

sum of funds from operations, decreases in inventories, decreases in receivables, and increases in payables.

Capital expenditures include net acquisitions of fixed assets.

Natural rate of entry is measured as the mean number of new firms between periods t-1 and t , divided by

number of firms in period t, by two digit SIC code for the United States over the period 1963-1982. Source:

Table 5 in Dunne et al (1988).

Turnover (Excess employment reallocation) is defined as the sum of job creation and job destruction at 4

digit SIC code, minus the absolute value of net job creation. As with job reallocation, we compute the time

average of the annual measures. Time average by sector for period 1973-1993. Source: Job Flows data Davis,

Haltiwanger and Schuh (1996)

Page 59: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

59

Export Status and ownership

Exporters were identified based upon reporting that the share of output exported was nonzero and

nonmissing, and foreign owned firms based upon whether foreign ownership was reported to be nonzero

and was nonmissing. Government owned firms were defined the same way based on both local and

national government.

Entry and exit

Entry is defined as the first year an establishment is observed in the data. Firms may have been in operation

for several years prior to crossing the minimum threshold of 20 employees to be included in the survey.

Firm exit is defined as permanent exit from the dataset. Note that, since the survey only includes firms with

20 or more employees, firms whose employment falls below this treshhold will be defined as having exited in

case they do not reappear in the data

B. 3 Cleaning of outliers and nonsensical values

As mentioned earlier, the problem of non-persistent extreme values for many variables is widely

discussed in published work using the SI. In some cases, these values may be the result of key punch errors,

where, for example, ownership share is recorded as 340 percent rather than 34 percent. Where shares

(exports and ownership) were reported, these were easily corrected, but for balance sheet variables, more

extensive cleaning was required.

These likely key punch errors and other highly volatile trends were corrected in the data by

identifying firms with implausibly large non-persistent changes over one or two years and replacing values

with those of preceding and succeeding years (or just one or the other of the observation was at beginning or

end of the series). Based on close observation of normal variation of these variables, the thresholds for

identifying large nonpersistent shifts were a 100% change in Labor, a 200% in real value added and real

output, a 150% change in real inputs, and 100% change in capital and average wages. Next, firms were

dropped from the data when they had such a significant number of non-persistent jumps that it was

impossible to interpolate (this constituted less than 1% of the sample). Finally remaining outliers were

identified by eyeballing plots of key relationships (e.g. inputs per worker) and spot interpolating obvious

remaining outliers.

Page 60: Do Crises Catalyze Creative Destruction€¦ · Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia Mary Hallward-Driemeier, DECRG Bob Rijkers, DECRG1 April

60

Appendix C: Assessing the robustness of our results against excluding zombie firms.

The SI dataset shows a noticeable spike in exit in 2001.33 A large number of firms that exited in 2001

reported exactly identical values on key explanatory variables (i.e. having the same, nonmissing values for

labor, output, value-added, capital, materials, and wages) for the period 1998-2000 prior to exiting. Since this

is unlikely to be an accurate reflection of their true performance, it is likely that these firms may only be

achieving these results on paper.

To examine how these ―zombie‖ firms might affect our estimates of entry and exit, we distinguish between two different types of ―zombie‖ firms: (1) zombie1: firms that exit the year after next and have the same values next year as this year (2) zombie2: firms that exit in 3 years and have the same values for intervening years, Table C.1 documents the annual prevalence of each type as well as adjusted exit rates that define firms as

exiting when they become ―zombies‖ rather than when they exit from the data. Each year there are a small

but nontrivial number of zombie firms. However, the number of zombie firms spikes during the crisis. When

we correct exit rates for this problem, we see quite a different pattern in exit rates, not just for 2001.

EXIT ADJUSTED FOR ZOMBIES

year zombie1 zombie2 exit exit (corrected

for zombie 1) exit (corrected for

zombie 1 and 2) 1992 3.7% 2.8% 8.6% 12.2% 11.4% 1993 3.1% 2.1% 7.8% 7.4% 9.9% 1994 2.7% 3.4% 6.8% 6.5% 7.7% 1995 5.8% 4.0% 5.8% 8.2% 7.9% 1996 8.7% 4.7% 8.7% 6.4% 10.4% 1997 2.0% 1.0% 10.0% 10.0% 10.0% 1998 1.9% 2.3% 10.5% 10.5% 10.3% 1999 2.7% 6.5% 4.5% 4.5% 5.3% 2000 8.2%

4.7% 4.7% 10.3%

2001

12.4% 12.4% 4.2% Total 2.8% 2.4% 8.0% 8.0% 8.0%

To examine how exclusion of such zombie firms might affect our results, we re-estimated our model using

alternative definitions of exit; the attenuation of the link between productivity and exit turns out to be robust

to exclusion of these zombie firms.

Results with uncorrected exit measure:

all precrisis Crisis recovery

Odds

ratio/se

Odds

ratio/se Odds

ratio/se Odds

ratio/se

Exit (uncorrected)

0.881*** 0.827*** 0.961** 0.866***

(0.010) (0.016) (0.019) (0.016) Exit-corrected for zombie1 0.897*** 0.847*** 0.975 0.875***

(0.010) (0.017) (0.019) (0.017)

Exit-corrected for zombie1 and zombie2

0.897*** 0.857*** 0.972 0.884*** (0.010) (0.015) (0.021) (0.017) (0.010) (0.015) (0.018) (0.020)

33

For this table, exit in 2001 means that 2000 was the last year the firm appeared in the data.