Firm Productivity in Bangladesh Manufacturing Industries

Ana M. Fernandes*

The World Bank

January 22, 2008

Abstract This paper studies the correlates of firm total factor productivity (TFP) in Bangladesh using data from a recent survey of large manufacturing firms. TFP measures are obtained following Ackerberg, Caves and Frazer (2007) and using firm-specific deflators for output and inputs. Controlling for industry, location, and year fixed effects, we find that firm size and TFP are negatively correlated while firm age and TFP exhibit an inverse-U shaped relationship. We also find that managerial quality and global integration are positively associated with firm TFP. Finally, we find that power supply problems, heavy bureaucracy, and the presence of crime dampen firm TFP.

JEL Classification Numbers: D24, F23, L25 O33. Keywords: Total Factor Productivity, Simultaneity and Production Functions, Skilled Labor, Openness, Business Environment, Asia, Bangladesh.

∗ Development Research Group, The World Bank, 1818 H Street, N.W., Washington, DC 20433, USA, ph.: 202-4733983, email: afernandes@worldbank.org.

Acknowledgments: This paper draws on a background paper prepared for the World Bank 2006 report “Bangladesh: Strategy for Growth and Employment.” Four anonymous referees provided helpful comments that substantially improved the paper. We thank Thorsten Beck, Sandeep Mahajan, Luis Serven, and seminar participants at the Dhaka workshop on “Bangladesh: Strategy for Growth and Employment” and at the World Bank South Asia Brown Bag Lunch for comments. The findings and views expressed in this paper are those of the author and should not be attributed to the World Bank, its Executive Directors, or any of its members.

1

1. INTRODUCTION

A major stylized fact uncovered by the empirical industry evolution literature in

developed as well as developing countries is the enormous degree of heterogeneity in

productivity across firms even within narrowly defined manufacturing industries (Bartelsman

and Doms, 2000; Tybout, 2000). It has also been shown that long-term growth and

development across countries is driven to a large extent by productivity growth (Easterly and

Levine, 2001). From a policy perspective, it is therefore crucial to understand which factors

underlie the heterogeneity in firm productivity.

The literature has proposed various potential correlates of firm productivity, which

can provide micro foundations to several important findings obtained at the aggregate level in

cross-country growth studies. A strand of research has focused on the role of openness and

international integration for firm total factor productivity (TFP) following the theoretical

insights from the endogenous growth literature. A large number of studies show a beneficial

effect of exports (e.g., Blalock and Gertler, 2004; Van Biesebroeck, 2005b, and Wagner,

2007) or foreign ownership (e.g., Arnold and Javorcik, 2005; Kee, 2006) on firm TFP. A

different literature examines the impact of research and development (R&D) activities on

firm TFP (e.g., Griliches, 1998). Finally, a recent literature focuses on the role of the business

environment for firm TFP (e.g., Hallward-Driemeier, et al., 2006; Dollar, et al., 2005).1

Our paper contributes to the literature by combining into a single analytical

framework correlates of TFP which have been analyzed separately in previous studies:

managerial quality, integration into world markets, technology, business environment, and

firm size and age. Using data from a recent survey of large manufacturing firms in

Bangladesh, we estimate consistent TFP measures for the period 1999-2003 following the

Ackerberg, Caves, and Frazer (2007) methodology and examine the correlates of firm TFP.

2

An important feature of our study is the use of firm-specific output and input price deflators

which results in TFP measures capturing true firm efficiency, rather than a mix between

efficiency and market power, as in previous studies (see, e.g., Katayama et al., 2003).

Bangladesh is an interesting country to study for two reasons. First, the evidence on the

correlates of firm productivity in low-income countries is rare. Most productivity studies

focus on middle-income countries in Latin America or Eastern Europe due to data

availability. While Dollar et al. (2005) study firm TFP in low-income countries (Bangladesh

included), they focus exclusively on the role of the business environment. Second, the

manufacturing sector in Bangladesh is particularly dynamic having experienced very strong

growth in total value-added and exports since the 1990s liberalization, largely driven by the

ready-made garments (RMG) industry.2

Our econometric results identify several correlates of firm TFP, controlling for

industry, location, and year fixed effects. Smaller firms are significantly more productive

than firms in the largest size category. Firm age and TFP exhibit an inverse-U shaped

relationship. Firms with more educated or experienced managers and firms with foreign

ownership are more productive. Firms with more experience in export markets have

significantly higher TFP. Firms with quality certifications exhibit higher TFP. However,

firms engaged in R&D activities and firms using more computerized machinery do not.

While firms with an overdraft facility have higher TFP, firms with access to a bank loan have

lower TFP. Power supply problems and crime are negatively associated with firm TFP.

The paper is organized as follows. In Section 2, we describe the data and estimate

firm TFP. In Section 3, we examine the correlates of firm TFP. Section 4 concludes.

2. OBTAINING FIRM TFP MEASURES

(a) Data

3

Our analysis uses data from a firm survey conducted by the World Bank in Bangladesh

between November 2004 and September 2005 covering five industries: food,

leather/footwear, pharmaceuticals, RMG, and textiles. The survey collected information on

production variables, firm characteristics, and business environment aspects.3 The sample

used in the econometric analysis includes 575 firms. Each firm has at most five years of

production data. The majority of firms in the sample belong to the RMG industry (see

Appendix Table 1). This reflects the importance of the sub-sector in manufacturing in

Bangladesh, but also the sample design, described in the Appendix. Within RMG, 86% of

firms belong to the woven sub-industry, 21% to the knitwear sub-industry, and 13% to the

sweater sub-industry. Most firms in our sample have more than 50 workers, although the size

distribution varies significantly across industries.4 For example in the leather/footwear

industry about a quarter of the firms have less than 50 workers. About one-half of the firms

are located in Dhaka while 17% are located in Chittagong (excluding Export Processing

Zones (EPZs)). We finish by highlighting two important features of our sample: it covers

mostly (1) firms with more than 50 workers, and (2) young firms (about 80% of firms are less

than 20 years old). Hence our findings are representative only for the segment of larger and

relatively young manufacturing firms in Bangladesh.

(b) Production Function Estimation

Firm TFP measures are not observable but can be estimated as residuals from a

production function. For each of the five industries, we estimate the following Cobb-Douglas

production function, where value-added Yit is produced by a combination of three inputs:

labor lit, capital kit, (all in logarithms) and a measure of workforce human capital sit:5

itSitKitLitit sklAy βββ +++= (1)

Ait is TFP which represents the efficiency of the firm in transforming inputs into value-added.

4

Using ordinary least squares (OLS) to estimate the production function coefficients

( )SKL βββ ,, assumes that input choices are exogenous. However, input choices are

endogenous. For example, the number of workers hired by a firm and the quantity of

materials purchased may depend on unobserved managerial ability, which is part of TFP

known to the firm but not observable to the researcher. Since input choices and productivity

are correlated, OLS production function estimates are biased.6 This endogeneity bias could be

partly corrected using fixed effects estimation for the production function, which eliminates

unobserved fixed firm characteristics affecting simultaneously input choices and TFP.

However, unobserved time-varying firm characteristics are also important. We follow the

estimation methodology proposed by Ackerberg, Caves, and Frazer (2007) [henceforth ACF]

to correct for the simultaneity bias generated by firm time-varying unobservables. This

methodology modifies the Olley and Pakes (1996) technique in order to address collinearity

problems that prevent the identification of production function coefficients and is described

in detail in the Appendix. The main idea behind the methodology is that unobserved firm

productivity shocks can be approximated by a non-parametric function of an observable firm

characteristic – specifically investment - and as a result unbiased estimates of the production

function coefficients are obtained.

The output and input variables are defined as follows.7 Value-added is measured by

the difference between deflated sales and deflated material costs, which are obtained,

respectively, as the ratio between nominal sales or materials costs and corresponding firm-

specific deflators. The sales [materials] deflator is based on a survey question on annual

changes in the price of the firm’s main product [main materials]. The capital stock is obtained

by cumulating real investment flows using the perpetual inventory method formula with a

depreciation rate of 10%.8 For each firm, the initial capital stock entering the perpetual

5

inventory method formula is obtained by combining the book value of fixed assets 0iF and

the real investment flows 0iI in the firm’s first sample year as in )(2

1 000 d

IFK i

ii += (see

Kee, 2006).9 Labor is measured by the total number of workers and the workforce human

capital is measured by the share of skilled workers (professional, technical, administrative,

managerial, and skilled production workers).

Empirical work with micro data typically uses industry-specific deflators to obtain

revenue-based output and materials. While Foster et al. (2005) show that revenue-based TFP

measures are correlated with true efficiency, those measures confound the effects of

idiosyncratic demand and factor prices with efficiency differences. Large firms are likely to

have market power in product markets and monopsony power in input markets and thus

charge higher output prices and pay lower prices for materials, relative to the industry (see

Klette and Griliches, 1996).10 Deflating the sales and materials costs of large firms by

industry output and materials deflators will overestimate their value-added and thus

overestimate their TFP measures as efficiency and price-cost mark-ups are mixed. However,

we should note that since our firm-specific deflators are based on information on the

evolution of the prices faced by the firm for its main product and main material input, they

are still imperfect since firms are likely to produce multiple products and use multiple inputs.

Thus, while using firm-specific deflators allows our TFP measures to better capture true firm

efficiency, there is still scope for possible correlations between unobserved prices and inputs.

The production function coefficients are shown in Table 1 for OLS and ACF

estimation and their magnitudes are in line with those from previous studies. The ACF

coefficients on labor are lower than those obtained from OLS estimation in all but one

industry, while those on capital are higher than those obtained from OLS estimation and

significant in all industries, indicating a correction of the simultaneity bias. The ACF

6

coefficient on skilled workers is higher than that from OLS estimation in two of the five

industries.11 Returns to scale are increasing in all industries. Using the consistent ACF

production function coefficients in Table 1 ( )SKL βββ ,, , we compute time-varying TFP

measures for each firm as: itSitKitLitit sklyp βββ −−−≡ .

[Table 1 about here]

While an in-depth analysis of firm prices, markups, and productivity is beyond the

scope of this paper, we also show in Appendix Table 2 production function coefficients

obtained using measures of output and materials deflated by industry-specific, instead of

firm-specific, deflators. The ACF coefficients on labor, skilled workers, and capital in that

table are generally lower than those in Table 1. This finding suggests that for larger firms,

according to employment or capital, TFP based on the coefficients in Appendix Table 2

would indeed be overestimated relative to TFP based on the coefficients in Table 1, since the

former TFP measures mix true efficiency and market power as pointed out by Katayama et

al. (2003).

(c) Industry TFP Decompositions

The average TFP of an industry may grow (decline) because all firms become more

(less) productive or because output is reallocated towards more (less) productive firms. We

compute industry average TFP as a weighted average of firm TFP with firm shares of total

industry sales as weights. We follow Olley and Pakes (1996, p. 1290) in decomposing

industry average TFP into: (i) the unweighted average firm TFP (within-firm) and (ii) a term

measuring the covariance between firm market share and firm TFP (between-firm). The

between-firm component measures allocative efficiency, if positive, then the more productive

firms in the industry have higher market shares and the allocation of resources is efficient.12

7

Figure 1 shows the decomposition of industry TFP into the within-firm and between-firm

components for each industry, averaging the values of the components over the 1999-2003

period. The between-firm component is always positive but its contribution to average

industry TFP is small in the garments, leather, and textiles industries.

[Figure 1 about here]

3. CORRELATES OF FIRM TFP

(a) Empirical Framework and Measurement of Correlates

We consider a comprehensive set of policy-relevant potential correlates of firm TFP

proposed in different strands of the literature but generally not combined into a single

analytical framework. Specifically, we study how managerial ability ( j

irtX1 ), integration into

world markets ( j

irtX 2 ), technology ( j

irtX 3 ), and the business environment ( j

irtX 4 ) promote or

constrain firm TFP. In addition to these factors, we also focus on the role of firm size and

age. With i designating a firm, t a year, j an industry, and r a location, the empirical reduced-

form specifications which we estimate are given by:

jirt

rtjjirts

jirta

j

irt

j

irt

j

irt

j

irtit IIIsizeageXXXXp εββββββ +++++++++= 44332211 (2)

where jI , tI , and rI are industry, year, and location fixed effects, respectively.

We now describe the survey variables used to measure the potential correlates of firm

TFP.13 Managerial ability is captured by a dummy variable identifying managers with post-

graduate education and by the number of years of experience of the manager. Integration into

world markets is captured by a dummy variable for foreign ownership, dummy variables for

exporters and majority exporters (firms exporting more than 50% of their output), or by the

number of years of experience in export markets.14 Technology is captured by a dummy

variable for firms which were awarded one or more quality certifications (e.g., ISO), a

8

dummy variable for firms engaged in R&D activities, the percentage of machinery less than

five years old, and the percentage of computerized machinery. The business environment

corresponds to the institutional, policy, and regulatory environment in which firms operate.

While most previous studies relied on managerial opinions and perceptions about the

business environment, our survey has the advantage of including objective measures that

capture different dimensions of institutional and policy weaknesses affecting firms. Finance

is captured by a dummy variable for firms having an outstanding loan and a dummy variable

for firms with an overdraft facility or line of credit.15 The reliability of the public

infrastructure is captured by the number of power outages suffered and a dummy variable for

firms owning a generator. Potential crime is captured by protection payments made as a

percentage of firm sales.16 Bureaucracy and government efficiency in providing services are

captured by the number of days needed to clear customs for imports and by the percentage of

weekly time spent by managers dealing with regulation. Finally, corruption is captured by the

percentage of sales paid by firms in the industry as bribes to government officials to “get

things done”.17 Note that these business environment measures cover the most crucial

obstacles to growth and operations faced by manufacturing firms in Bangladesh, as revealed

by firm perceptions’ data.18

(b) Econometric Problems

The estimation of Eq. (2) suffers from several potential econometric problems and

consequently our results need to be interpreted with caution. First, there is a problem of

endogeneity and reverse causality for several TFP correlates. For example, if exporters are

found to be more productive, it may be the case that exporting leads to higher TFP but it may

also be the case that more productive firms are those able to self-select into export markets.

In theory, such problems could be solved if we had instrumental variables correlated with the

9

TFP correlates but not with firm TFP. In practice, such variables are not available,

particularly given the large number of factors considered in our analysis. Our approach to

deal with the endogeneity problem is to include in our regressions industry, location, and year

fixed effects, firm size and age. These variables control for potential unobserved factors

which may affect the correlates of TFP and TFP itself. The effects of TFP correlates

estimated when all control variables are included are less subject to endogeneity problems.

For business environment factors, we take an additional step: i.e., we include in the

regressions averages at the industry-location level of the business environment variables,

instead of including those variables at the firm level. The rationale for this approach is that

the business environment faced by all firms in a common industry and location is likely to be

similar. Moreover, it is plausible to assume that for an individual firm the business

environment in its industry and location is exogenous.19

Second, given the large number of potential correlates of TFP, our regressions may

suffer from a multicollinearity problem. If the correlates are related, the results from

regressions including many correlates can be difficult to interpret. Our approach to address

this problem is to also estimate regressions including a single correlate at a time along with

basic control variables (industry, location and year fixed effects, firm age and size). Such

regressions do not suffer from the multicollinearity problem but may suffer from an omitted

variables’ problem. If the effect of a given correlate on firm TFP is qualitatively similar in

the regressions including all correlates and in the regressions including only that correlate, we

have more confidence on the sign and significance of its association with TFP.

Third, many of the TFP correlates are available for each firm in a single year while

TFP is available in each of the firm’s sample years. For estimation purposes, we assume that

those correlates are constant over the sample period. While this is a relatively safe

assumption for business environment variables, it may be restrictive for other variables (e.g.,

10

technology variables). The lack of time variability in some correlates works against our

finding of significant effects, as the variability would help identify their association with firm

TFP. Moreover, this lack of time variability prevents us from using fixed effects estimation.

Instead, we estimate our main specifications by OLS including a rich set of control variables

to account for unobserved factors potentially influencing both TFP and its correlates.

Finally, our TFP measures may exhibit serial correlation within firms over the sample

period. In fact, the ACF production function estimation method builds this serial correlation

into the estimation procedure. Thus, we cluster the standard errors of our main TFP

regressions at the firm level.

(c) Main Results

In this section, we discuss our findings on the correlates of firm TFP shown in Table 2.

To address the aforementioned potential multicollinearity problem, we also show in

Appendix Table 3 the results from regressions of firm TFP on a single correlate at a time.

The sign and significance of the effects in those regressions are broadly similar to those in

Table 2.20 Thus, the concern of a multicollinearity problem is mitigated.

[Table 2 about here]

First, we focus on the role of firm size and age for TFP. Theoretical models of industrial

dynamics with firm heterogeneity predict that more productive firms are larger (Jovanovic,

1982). Also, several stylized facts on the impact of the life-cycle on manufacturing firms’

TFP have been established for developed countries (Bartelsman and Doms, 2000). Studies

based on U.S. data find that firms enter an industry at a small size with low productivity. The

firms that survive grow and converge quickly to the average size and productivity in the

industry. It is not clear whether in a low income country like Bangladesh, these stylized facts

hold. Our analysis is a first step in uncovering the effects of life-cycle on TFP for

11

manufacturing firms in Bangladesh. In Table 2, we find that, relative to the extremely large-

sized firms (more than 500 workers), firms of smaller sizes have higher TFP. Specifically,

medium-sized firms (10-50 workers) are the most productive - 84-87% more productive than

extremely large-sized firms - followed closely by relatively large-sized firms (50-150

workers) which are 66-70% more productive than extremely large-sized firms. The TFP of

small firms (1-10 workers) is not significantly different on average from that of extremely-

large sized firms. In robustness specifications where total employment is included as a

continuous variable or where size is measured by the firm’s capital stock, we obtain the same

finding, i.e., larger firms have lower TFP. Thus, in Bangladesh firms that are too large seem

to suffer from inefficiencies in terms of coordination, management, and supervision resulting

from poor corporate management and a lack of qualified middle managers.21 An important

remark should be made at this stage: our sample is skewed towards larger firms and includes

only a few small firms. Thus, the focus of our findings on size and TFP is on the comparison

across size categories for medium-sized and large-sized firms (i.e., more than 50 workers).

Overall, our findings are broadly in line with those for other developing countries for which

there is no evidence of a strong size disadvantage for firm productivity (Tybout, 2000). Van

Biesebroeck (2005a) finds that TFP increases monotonically with size for firms in nine

African countries. However, our results are not directly comparable to his since he classifies

all firms with more than 100 workers as large. In our sample such firms can belong to (i) the

relatively large size category, (ii) the very large size category, or (iii) the extremely large size

category, and firms of type (i) exhibit much higher TFP than firms of type (ii) or (iii).

Our findings suggest a very robust inverse-U shaped relationship between firm age

and TFP, conditional on our sample of relatively young firms. A clear ranking of firm TFP

across age categories can be established. The most productive firms are those 10-20 years

12

old, followed by firms 5-10 years old, next by firms 20-40 years old, then by firms more than

40 years old, and lastly by firms less than five years old. This inverted U-shaped life-cycle

pattern suggests that firms start at low TFP, then they learn, e.g., by doing, undertaking new

investments, participating in international markets, or realizing economies of scale as they

age, which increases TFP. After a certain age - around 40 years old in our sample - their

technology, modes of production, and operations likely become outdated and their TFP

advantage erodes. In robustness specifications that include age and age squared as continuous

variables, we find that TFP increases with firm age but at a decreasing rate. Our findings are

broadly consistent with the predictions from industrial evolution models of young firms

entering the industry at low productivity then growing and converging to the average

productivity in the industry. Our findings for the monotonically decreasing part of the

inverted U-shape relationship between age and TFP are consistent with evidence in Jensen et

al. (2001) and in Van Biesebroeck (2005a) which find that TFP is higher for younger firms

relative to older firms in the U.S. and in several African countries, respectively. However,

those studies provide no evidence of the increasing part of the inverted U-shape relationship.

Jensen et al. (2001) estimate a strictly linear relationship between age and TFP while Van

Biesebroeck (2005a) does not decompose his older age group (firms aged ‘20 or more years’)

into additional categories as our study does.

Next, we discuss the findings on the four different types of correlates of TFP. We

should note that these correlates are jointly significant in all TFP regressions. Firms with

more educated or more experienced managers are more productive than other firms. This

finding is qualitatively similar to those in Burki and Terrell (1998) and in Lall and Rodrigo

(2001) for firms in Pakistan and in India, despite methodological differences across studies in

the measurement of productivity. Moreover, we argue that the importance of managerial

quality for firm TFP actually strengthens our findings on the other correlates of TFP. Our

13

regressions suffer from the potential problem that better managers lead their firms to achieve

higher TFP but also influence some of the TFP correlates such as, e.g., exports or access to

finance. Since our regressions control for managerial education and experience, we believe

that the effects of other correlates on TFP are not driven by unobserved managerial ability.

The results in Table 2 show that firms with foreign ownership are more productive

than domestic firms, but the effect is weak. Since this finding is obtained in regressions that

control for industry, location, and year fixed effects, it is not driven by macroeconomic

fluctuations (i.e., business cycles in the FDI-sending countries that could make some years

more prone to FDI), by a composition effect (i.e., more productive industries are more prone

to receive FDI), or a location effect (i.e., FDI is more likely to be directed at certain regions

such as EPZs).22 While foreign-owned firms are likely to have an advantage relative to

domestic firms in terms of tangible assets (e.g., better technology) and intangible assets (e.g.,

better access to distribution and marketing channels and networks) as was shown by Arnold

and Javorcik (2005) for firms in Indonesia, that advantage is identified only imprecisely in

our sample of Bangladeshi firms.

Our findings also highlight a positive association between exports and firm TFP. The

regressions in columns (2) and (5) of Table 2 show that within industries and locations firms

majority exporters are 13% more productive than non-exporters. The TFP advantage of

exporters may be due to technological learning from foreign buyers but also to the possibility

that exporters improve their own technological capabilities in order to exploit profitable

opportunities in export markets.23 However, the positive association between TFP and

exports may reflect a self-selection of better firms into export markets, rather than the effect

of exporting on TFP.24 Self-selection and learning-by-exporting are not mutually exclusive

hypotheses though, as firms with high TFP which can afford the sunk costs of entry into

export markets may continue to improve TFP as a result of their exposure to exporting.

14

Exporters face several challenges that likely result in learning-by-exporting improvements in

TFP. Exporters need to solve new problems such as adopting stringent technical standards to

satisfy sophisticated consumers, or introducing more efficient machinery. Moreover, they are

pressured to meet orders in a timely fashion and ensure product quality for export markets

which are more competitive than the domestic market. As an attempt to assess the presence

of learning-by-exporting effects on firm TFP in Bangladesh, we include in columns (3) and

(6) of Table 2 a measure of export experience - the number of years that a firm has exported -

instead of export participation measures (see Fernandes and Isgut, 2007). Firms with longer

experience in export markets have significantly higher TFP. However, more productive firms

may be the first to enter export markets. Hence the correlation between export experience and

TFP may still reflect the self-selection of better firms into exporting.

Table 2 shows that quality certifications are positively, albeit weakly, associated with

firm TFP. Quality certifications such as ISO guarantee the use of internationally recognized

technical standards and are an important means for firms to acquire state of the art

technological know-how and raise their capability to compete in global markets. The results

in Table 2 suggest that there is no TFP advantage for firms engaged in R&D activities. It may

seem inadequate to talk about R&D activities for firms in Bangladesh given the low overall

R&D expenditures in the country.25 However, we do not interpret the R&D activities of

Bangladeshi firms as activities bringing breakthrough patentable innovations but rather as

low-level activities related to the adaptation of technology and production processes to local

conditions. In any case, such activities have not translated into TFP improvements for firms

in our sample. Similarly, we find that firms operating with a larger share of new machinery

have lower TFP than other firms. In contrast, our results suggest that the use of higher shares

of computerized machinery is associated with higher TFP for Bangladeshi firms. Overall, the

results on technology and TFP are mixed and generally weak, some being actually

15

counterintuitive in light of the widely accepted idea that the accumulation of knowledge is a

key determinant of TFP. We also estimate a variant of Eq. (2) where technology measures are

allowed to enter separately and interacted with the dummy variable for R&D. As Cohen and

Levinthal (1989) argue, R&D activities can perform two roles: (i) stimulate innovation but

also (ii) develop a firm’s ability to identify, assimilate, and exploit outside knowledge. The

unreported results suggest that firms engaged in R&D activities do not benefit more from the

use of advanced technology. We interpret our negative correlation between advanced

technology and firm TFP as reflecting the fact that firms using newer machinery are

undergoing a learning process. The firms’ absorptive capacity has not developed sufficiently

to allow them to exploit and benefit from the potential efficiency improvements associated

with more advanced technologies, although these technologies are likely to be productivity-

enhancing once their optimal use is reached. While there is no similar type of evidence for

developing countries, our findings are in line with those in Sakellaris (2004) of a productivity

decline associated with new technology adoption by U.S. manufacturing plants. Finally, we

should note that the lack of significance of our coefficients on technology-related factors in

Table 2 is not surprising given that the technology-related factors are fixed over time for a

given firm and the standard errors are clustered at the firm level.

Turning to business environment factors, Table 2 shows that firms with an overdraft

facility have higher TFP while firms with access to a bank loan have lower TFP. However,

both effects are not significant. In our sample, access to overdraft facilities and loans is more

common among relatively large-sized firms than among smaller firms.26 In the literature,

there is widespread evidence of a correlation between firm size and access to finance (see

Beck et al., 2005). However, we obtain similar results, i.e., firms with access to a bank loan

have lower TFP, when firm size is excluded from the regressions. Thus, the negative effect of

the loan dummy is not driven by the correlation between firm size and access to finance.

16

Finally, if there is a strong correlation between the overdraft facility and the loan dummies,

then the weak findings on access to finance and firm TFP could be driven by

multicollinearity. While there is some overlap between the firms with an overdraft facility

and those with a loan, we still find a positive correlation between the access to an overdraft

facility and a negative correlation between the access to a bank loan and firm TFP in

specifications that include only one of the finance variables.27 Interestingly, Van Biesebroeck

(2005a) finds a similar positive association between access to overdraft facilities and TFP

and a negative association between access to bank loans and TFP for firms in several African

countries. Overall, our findings suggest that access to short-term finance that addresses

working capital and day-to-day business needs has a positive link with TFP, while the access

to long-term finance that addresses investment needs does not. These findings may reflect

inefficiencies of the banking sector in Bangladesh and deserve further research. However,

they should by no means be taken as evidence that access to external finance is irrelevant for

firm performance.

Table 2 shows that firms in industries and locations experiencing more power outages

per year have lower TFP. The coefficients in columns (1)-(3) suggest that a firm belonging to

an industry and location with double the number of power outages (i.e., more 100%) than

another industry-location cell has about 11% lower TFP.28 It is possible that poor electricity

supply is less correlated with the TFP of firms owning a generator. In columns (4)-(6) of

Table 2, we show the results from estimating a variant of Eq. (2) where the number of power

outages is allowed to affect TFP differently for firms owning a generator. The results show

that the TFP of firms owning a generator suffers slightly less due to power outages than the

TFP of other firms. Our findings are qualitatively similar, though weaker than those obtained

by Dollar et al. (2005) on the TFP of ready-made garments firms in Bangladesh, China,

India, and Pakistan. The weakness of our findings on the association between electricity

17

infrastructure and firm TFP in Bangladesh is linked to the fact that the business environment

measures are not time-varying for a given firm but the standard errors are clustered at the

firm level. In fact, using robust standard errors as in Dollar et al. (2005), our negative effects

of power outages on firm TFP become significant. Our findings are close to those obtained

by Hallward-Driemeier et al. (2006) for firms in China, for which electricity infrastructure

has only a weak negative effect on TFP. This similarity in findings is actually surprising

given the difference in the two countries’ levels of development, namely the fact that

physical infrastructure may no longer be a bottleneck for the growth of Chinese firms, while

it is still a very important bottleneck hurting the growth of Bangladeshi firms.

The results in Table 2 also show a significant negative effect of crime on TFP. Firms

making larger protection payments are significantly less productive than other firms. We

assume that larger protection payments to be “spared” from organized crime proxy for an

environment with more potential crime. In unreported regressions, we also find a negative

effect of the ratio of security expenses to sales on firm TFP when that variable is the proxy

for crime.

The estimates in Table 2 show that the number of days taken to clear customs and the

percentage of time that firm managers must spend dealing with regulations are negatively

associated with firm TFP, although weakly. Thus, our results suggest a harmful effect in

productivity terms of heavier bureaucracy and red tape. However, the results in Table 2 show

that firm TFP is higher in industries and locations where firms pay on average a larger

percentage of their sales in bribes to get things done. Qualitatively similar results are

obtained corruption is proxied by the percentage of firms in each industry-location cell

paying bribes to get things done. We interpret this positive link between corruption and TFP

as reflecting reverse causality. As argued by Fisman and Svensson (2006) if government

18

officials have discretion in implementing or enforcing regulations, then they will customize

the amount of harassment on firms and try to extract as high a bribe as possible from firms

with a larger ability to pay. Better performing firms are likely to have a larger ability to pay

bribes to cut through bureaucratic hassles. Our positive coefficient on the proxy for

corruption suggests that indeed the industries and locations with better performing firms are

more targeted by government officials, thus the opportunities for bribe-seeking behavior are

larger and ultimately firms pay more bribes. While it may seem unlikely that better

performing firms can be identified by government when performance is measured by TFP, it

is plausible when performance is measured by profitability. For our sample of Bangladeshi

firms, there is a significant positive correlation between TFP and profit rates overall and

within industries.

[Table 3 about here]

In Table 3, we conduct several robustness checks to our main results. We estimate Eq.

(2) using random effects in column (1) and using OLS but restricting the sample to include a

single year per plant (its last sample year) in column (2). The results in both columns are

qualitatively similar to those in Table 2. In column (3) we consider an alternative - albeit

imperfect - way to address the serial correlation problem in firm TFP: estimating Eq. (2) by

OLS with non-clustered robust standard errors, but adding lagged TFP as a regressor. While

the signs of the coefficients are similar to those in Table 2 their significance is somewhat

higher. In columns (4) and (5), we allow the correlation between some correlates and TFP to

differ depending on the business environment in which the firm operates. Specifically, we

allow the effect of export experience and of technology-related factors to enter Eq. (2)

separately and interacted with a composite index representing the business environment.29

The results suggest that for firms operating in a better business environment the associations

19

between export experience or R&D and firm TFP are more positive and the associations

between the use of more advanced technology and firm TFP are less negative.

4. CONCLUSION

This paper uses data from a recent survey of manufacturing firms in Bangladesh to

obtain consistent firm-level time-varying TFP measures for the period 1999-2003 following

Ackerberg, Caves, and Frazer (2007) and empirically examine the correlates of TFP. Our

results suggest several important correlates of firm TFP, controlling for industry, location,

and year fixed effects. Smaller firms are significantly more productive than firms in the

largest size category. Firm age and TFP exhibit an inverse-U shaped relationship. Firms with

more educated or experienced managers and firms with foreign ownership are more

productive. Exporters have significantly higher TFP than firms selling in the domestic market

only. Firms with quality certifications have higher TFP. However, firms engaged in R&D

activities and firms using more computerized machinery do not. While firms with an

overdraft facility have higher TFP, firms with access to a bank loan have lower TFP. Power

supply problems have a negative effect on firm TFP. The presence of crime in industries and

locations hurts significantly firm TFP.

While further research would be needed to establish causality, our findings point to

some areas of policy relevance – e.g., electricity infrastructure - in which improvements may

be associated with TFP gains in Bangladesh. The international integration of firms through

exports and FDI is also an important correlate of firm TFP and may result in spillover effects

to other non-integrated domestic firms. Finally, we should mention that while this paper

conveys information on the correlates of performance for firms in five manufacturing

industries in Bangladesh, it is based on a survey that covers relatively large firms. Pursuing a

similar type of analysis using comprehensive manufacturing census data will be of great

20

interest. Such data collection efforts can have large payoffs for policy-makers as they enable

them to closely follow the trends in performance for manufacturing firms of all sizes and

generate appropriate policy responses where necessary.

REFERENCES

Ackerberg, D., Caves, K., and G. Frazer (2007). Structural identification of production

Functions. UCLA mimeo.

Arnold, J. and B. Javorcik (2005). Gifted kids or pushy parents? Foreign acquisitions and

plant productivity in Indonesia. World Bank Policy Research Working Paper 3597.

Bartelsman, E. and M. Doms (2000). Understanding productivity: lessons from longitudinal

microdata. Journal of Economic Literature 38 (3), 569-594.

Beck, T., Demirgüç-Kunt, A., and V. Maksimovic (2005). Financial and legal constraints to

firm growth: does firm size matter? Journal of Finance 60 (1), 137-177.

Blalock, G. and P. Gertler (2004). Learning from exporting revisited in a less developed

country setting. Journal of Development Economics 75 (2), 397-416.

Burki, A. and D. Terrell (1998) Measuring production efficiency of small firms in Pakistan.

World Development 26 (1), 155-169.

Cohen, W. and Levinthal, D. (1989). Innovation and learning: the two faces of R&D.

Economic Journal 99 (397), pp. 569-596.

Dollar, D., Hallward-Driemeier, M. and T. Mengistae (2005). Investment climate and firm

performance in developing economies. Economic Development and Cultural Change 54 (1),

1-31.

Easterly, W. and R. Levine (2001). It's not factor accumulation: stylized facts and growth

models. World Bank Economic Review 15 (2), 177-219.

Fernandes, A. (2006). Firm productivity in Bangladesh manufacturing industries. World

Bank Policy Research Working Paper 3988.

21

Fernandes, A. and A. Isgut (2007). Learning-by-exporting effects: are they for real? mimeo,

the World Bank.

Fisman, R. and J. Svensson (2007) Are corruption and taxation really harmful to growth?

Firm level evidence. Journal of Development Economics 83 (1), 63-75.

Foster, L., Haltiwanger, J., and C. Syverson (2005). Reallocation, firm turnover, and

efficiency: selection on productivity or profitability? NBER Working Paper 11555.

Griliches, Z. (1998). R&D and Productivity: The Econometric Evidence. Chicago: University

of Chicago Press.

Hallward-Driemeier, M., Wallsten, S., and C. Xu (2006). Ownership, investment climate, and

firm performance. The Economics of Transition 14 (4), 629-647.

Jovanovic, B. (1982). Selection and the evolution of industry. Econometrica 50 (3), 649-670.

Jensen, J., McGuckin, R., and J. Stiroh (2001). The Impact of vintage and survival on

productivity: evidence from cohorts of U.S. manufacturing plants. Review of Economics and

Statistics 83 (2), 323-332.

Katayama, H., Lu, S., and J. Tybout (2003). Why plant-level productivity studies are often

misleading, and an alternative approach to inference. NBER Working Paper 9617.

Kee, H., (2006). Foreign investment and domestic productivity. mimeo, The World Bank.

Lall, S. and G. Rodrigo (2001). Perspectives on the sources of heterogeneity in Indian

industry. World Development 29 (12), 2127-2143.

Levinsohn, J. and A. Petrin (2003). Estimating production functions using inputs to control

for unobservables. Review of Economic Studies 70 (2), 317-341.

Mahajan, S. (2005). Sources of productivity growth in Bangladesh. mimeo, The World Bank.

Olley, G. and A. Pakes (1996). The dynamics of productivity in the telecommunications

equipment industry. Econometrica 64 (6), 1263-1297.

22

Roberts, M. and J. Tybout (1997). The decision to export in Colombia: an empirical model of

entry with sunk costs. American Economic Review 87 (4), 545-564.

Sakellaris, P. (2004). Patterns of plant adjustment. Journal of Monetary Economics 51 (2),

425-450.

Syverson, C. (2004). Market structure and productivity: a concrete example. Journal of

Political Economy 112 (6), 1181-1222.

Tybout, J., (2000). Manufacturing firms in developing countries: how well do they do, and

why? Journal of Economic Literature 38 (1), 11-44.

Van Biesebroeck, J. (2005a). Firm size matters: growth and productivity growth in African

manufacturing. Economic Development and Cultural Change 53 (3), 545-583.

Van Biesebroeck, J. (2005b). Exporting raises productivity in Sub-Saharan African

manufacturing firms. Journal of International Economics 67 (2), 373-391.

Wagner, J. (2007). Exports and productivity: a survey of evidence from firm level data.

World Economy 30 (1), 60-82.

Westphal, L. (2002). Technology strategies for economic development in a fast changing

global economy. Economics of Innovation and New Technology 11 (4&5), 275-320.

World Bank (2005). End of MFA Quotas: Key Issues and Strategic Options for Bangladesh

Readymade Garment Industry.

1 While we focus on supply side factors (e.g., technology, skills), Syverson (2004) argues that demand side factors

(e.g., product substitutability) can also play a major role in explaining the heterogeneity in firm TFP.

2 Since 2000, manufacturing exports represent more than 90% of total exports in Bangladesh.

3 The survey questionnaire shares many questions in common with the World Bank’s Investment Climate Surveys

(http://www.entrprisesurveys.org) and is available upon request.

23

4 We adopt the size classification used in the Bangladesh Census of Manufacturing Industries: “small” firms with less

than 10 workers, “medium” firms with 10 to 50 workers, and “large” firms with more than 50 workers. We divide large

firms into 3 sub-categories: “relatively large” firms with 50 to 150 workers, “very large” firms with 150 to 500

workers, and “extremely large” firms with more than 500 workers.

5 For simplicity, we omit the industry superscript j from the variables.

6 See Levinsohn and Petrin (2003) for further discussion on the bias of OLS production function estimates.

7 Note that we eliminate from the estimation firms which are outliers in the production variables or exhibit other data

problems, as described in the Appendix. Summary statistics for the production variables are shown in Fernandes

(2006).

8 The real investment that enters the perpetual inventory method formula is obtained deflating nominal investment

flows by an aggregate investment deflator from World Development Indicators.

9 When firms have zero investment in their first sample year, we replace that formula by: 00 ii FK = .

10 These arguments assume that product and input markets do not operate under perfect competition.

11 Note that our findings in Table 1 (as well as those in Tables 2-3) are qualitatively similar when the quality of the

workforce is measured by the share of college-educated workers.

12 This positive statement has no normative content. There may be reasons why a reallocation of output to less

productive firms that are, e.g., more socially or environmentally responsible, could increase economic welfare. Here,

we consider only the efficient allocation of resources (output) to their more productive use.

13 Summary statistics for all correlates are shown in Fernandes (2006).

14 The export experience variable is based on information about the year when the firm first exported, whether and in

which year the firm interrupted exports, and the year when exports began again.

15 In robustness specifications, we also consider the percentages of working capital and of investments financed by

banks and related institutions (i.e., domestic commercial banks, international commercial banks, leasing arrangements,

special development financing, public financing (government agencies) or other public services).

16 In the questionnaire, protection payments are defined as payments to organized crime to prevent violence. We

include the ratio of protection payments to sales in the regressions.

17 In the questionnaire, “get things done” is defined as bribes needed for firms to resolve issues related to customs,

taxes, regulations, and services.

18 Firms were asked to rank 28 business environment issues according to the degree of obstacle they constitute to their

operations and growth. The major obstacles pointed out by firms are (a) corruption (65% of firms), (b) customs (63%),

24

(c) power from the public grid (44%), (d) cost of finance (35%), (e) business licensing and operating permits (22%), (f)

access to finance (20%), (g) frequent changes in government regulation and regulatory policy (16%), and crime (13%).

19 However, we include the finance variables at the firm level in the regressions.

20 In Tables 2-3 and Appendix Table 3, the number of observations in each regression (column) differs since it depends

on the number of firms with non-missing values for the corresponding correlates of TFP.

21 Anecdotal evidence from discussion with firm managers suggests that this problem is particularly acute in

Bangladesh.

22 This finding parallels that in Kee (2006) using the same data for the ready-made garments industry.

23 Westphal (2002) documents the latter possibility for firms in Taiwan.

24 The self-selection hypothesis is more likely to be verified if entry into export markets is characterized by

economically significant sunk costs, such that only the firms with higher TFP are able to export. These sunk costs have

been shown to be empirically relevant (Roberts and Tybout, 1997).

25 Mahajan (2005) documents that R&D expenditures represent only 0.03% of GDP in Bangladesh, compared to 0.7%

in China and India and 0.2% in the Philippines.

26 However, access to bank finance is more common among relatively large-sized firms than among very large-sized

firms (the largest size category).

27 The correlation between the overdraft dummy and loan dummy is 0.18, significant at the 1% confidence level.

28 We interpret the coefficient on the logarithm of the number of power outages as an elasticity since the dependent

variable in the regressions is the logarithm of TFP.

29 The composite index is obtained as the first principal component derived from factor analysis of the following

variables: dummy for firms with an overdraft facility or line of credit, dummy for firms having an outstanding loan, the

average number of power outages suffered, the share of protection payments made in firm sales, the average number of

days needed to clear imports’ customs, the average percentage of weekly time spent by managers dealing with

regulation, and the average share of firms in the industry-region paying bribes to government officials to “get things

done”.

25

Table 1: Production Function Estimates

LaborShare of Skilled

Workers Capital N. Obs. Labor

Share of Skilled

Workers Capital N. Obs.

Pharmaceuticals 1.229*** 0.831*** 0.149*** 0.794*** 0.143 0.407***(0.060) (0.167) (0.036) 236 (0.164) (0.135) (0.168) 203

Food 0.737*** 0.729*** 0.356*** 0.703*** 0.361* 0.408***(0.065) (0.167) (0.044) 379 (0.165) (0.210) (0.175) 305

Ready-Made Garments 1.032*** 0.237* 0.067*** 0.905*** 0.362*** 0.471***(0.031) (0.132) (0.017) 1170 (0.144) (0.138) (0.151) 852

Leather 0.809*** 0.749* 0.386*** 0.706*** 0.141 0.474***(0.079) (0.398) (0.095) 120 (0.177) (0.248) (0.187) 108

Textiles 0.683*** -0.076 0.287*** 0.907** 0.369** 0.471***(0.042) (0.218) (0.029) 573 (0.166) (0.186) (0.177) 412

OLS

Coefficient on:

Ackerberg, Caves, and Frazer (2007)

Coefficient on:

Notes: Robust standard errors in parentheses in the columns with OLS estimates. Bootstrap standard errors in parentheses in the columns with ACF estimates. ***, **, and * represent significance at 1%, 5%, and 10% confidence levels, respectively. The number of observations used for OLS estimation is higher since it includes firms with zero investment which are excluded from the Ackerberg, Caves, and Frazer (2007) estimation.

26

Table 2: Correlates of Firm TFP

(1) (2) (3) (4) (5) (6)

Very Large Size Dummy (150 to 500 Workers) 0.306*** 0.308*** 0.317*** 0.318*** 0.321*** 0.330***(0.077) (0.077) (0.078) (0.077) (0.077) (0.078)

Relatively Large Size Dummy (50 to 150 Workers) 0.669*** 0.660*** 0.670*** 0.693*** 0.684*** 0.696***(0.125) (0.125) (0.125) (0.126) (0.127) (0.127)

Medium Size Dummy (10 to 50 Workers) 0.864*** 0.843*** 0.845*** 0.886*** 0.866*** 0.870***(0.264) (0.262) (0.260) (0.263) (0.261) (0.259)

Small Size Dummy (Less than 10 Workers) -0.101 -0.12 -0.157 0.009 -0.005 -0.035(0.378) (0.378) (0.387) (0.382) (0.383) (0.392)

Dummy for Firms Aged 5 to 10 Years Old 0.623*** 0.621*** 0.590*** 0.610*** 0.608*** 0.577***(0.139) (0.139) (0.142) (0.139) (0.139) (0.142)

Dummy for Firms Aged 10 to 20 Years Old 0.701*** 0.698*** 0.617*** 0.689*** 0.686*** 0.604***(0.143) (0.144) (0.156) (0.144) (0.145) (0.157)

Dummy for Firms Aged 20 to 40 Years Old 0.555*** 0.567*** 0.456** 0.548*** 0.560*** 0.448**(0.162) (0.163) (0.179) (0.163) (0.163) (0.180)

Dummy for Firms Aged More than 40 Years Old 0.314 0.319 0.226 0.310 0.315 0.222(0.196) (0.197) (0.199) (0.196) (0.197) (0.199)

Dummy for Managers with Post-Graduate Education 0.013 0.012 0.019 0.014 0.013 0.019(0.072) (0.072) (0.072) (0.072) (0.073) (0.072)

Manager Years of Experience (log) 0.069 0.068 0.057 0.067 0.066 0.055(0.044) (0.044) (0.045) (0.044) (0.044) (0.045)

Foreign-Owned Dummy 0.094 0.092 0.071 0.091 0.089 0.068(0.135) (0.137) (0.138) (0.136) (0.138) (0.139)

Exporters Dummy 0.150 0.146(0.094) (0.094)

Majority Exporters Dummy 0.131 0.129(0.095) (0.095)

Years of Export Experience (log) 0.070* 0.070*(0.039) (0.039)

Quality Certification Dummy 0.087 0.096 0.083 0.091 0.100 0.086(0.083) (0.083) (0.083) (0.083) (0.083) (0.083)

Dummy for R&D Activities -0.009 -0.004 -0.011 -0.011 -0.006 -0.013(0.110) (0.110) (0.110) (0.112) (0.112) (0.112)

Perc. of Machinery Less than 5 Years Old -0.095 -0.095 -0.052 -0.104 -0.104 -0.061(0.150) (0.151) (0.149) (0.151) (0.152) (0.150)

Perc. of Computerized Machinery 0.185 0.207 0.173 0.180 0.201 0.167(0.146) (0.149) (0.145) (0.146) (0.149) (0.146)

Overdraft Dummy 0.070 0.071 0.071 0.065 0.066 0.066(0.076) (0.076) (0.076) (0.077) (0.077) (0.077)

Loan Dummy -0.027 -0.024 -0.028 -0.036 -0.034 -0.038(0.068) (0.069) (0.068) (0.069) (0.069) (0.069)

Number of Power Interruptions (log) -0.105 -0.109 -0.116 -0.127 -0.131 -0.14(0.214) (0.213) (0.211) (0.213) (0.212) (0.211)

Number of Power Interruptions (log)* Generator 0.017 0.017 0.018(0.015) (0.015) (0.015)

Protection Payments as Perc. Sales -85.460** -82.105** -85.759** -83.316** -80.039** -83.672**(35.561) (35.334) (35.306) (35.878) (35.557) (35.640)

Days to Clear Customs for Imports (log) -0.290 -0.286 -0.284 -0.293 -0.290 -0.288(0.209) (0.211) (0.205) (0.210) (0.211) (0.205)

Perc. Manag. Time Spent Dealing with Regulation -0.246 -0.230 -0.265 -0.239 -0.223 -0.258(0.293) (0.296) (0.294) (0.293) (0.296) (0.293)

Avg. Perc. of Sales Paid in Bribes to Get Things Done 1.398 1.245 1.409 1.358 1.206 1.365(0.958) (0.989) (0.966) (0.964) (0.995) (0.972)

Industry Fixed Effects Yes Yes Yes Yes Yes YesYear Fixed Effects Yes Yes Yes Yes Yes YesLocation Fixed Effects Yes Yes Yes Yes Yes Yes

N. Observations 2249 2249 2235 2244 2244 2230R-squared 0.710 0.710 0.710 0.710 0.710 0.710

Dependent Variable is Firm TFP

Notes: OLS estimation is used. Robust standard errors clustered at the firm level in parentheses. ***, **, and * indicate significance at 1%, 5%, and 10% confidence levels, respectively. The omitted size category is extremely large firms (more than 500 workers) and the omitted age category is firms less than five years old.

27

Table 3: Correlates of Firm TFP – Robustness Checks

Random Effects

OLS on Sample of Firms' Last

Year

OLS OLS OLS

(1) (2) (3) (4) (5)

Lagged TFP 0.868***(0.018)

Very Large Dummy (150 to 500 Workers) 0.335*** 0.396*** 0.079*** 0.316*** 0.310***(0.083) (0.083) (0.023) (0.078) (0.079)

Relatively Large Dummy (50 to 150 Workers) 0.632*** 0.680*** 0.137*** 0.669*** 0.655***(0.126) (0.132) (0.035) (0.125) (0.124)

Medium Size Dummy (10 to 50 Workers) 0.739*** 0.856*** 0.212*** 0.846*** 0.818***(0.210) (0.284) (0.066) (0.260) (0.257)

Small Size Dummy (Less than 10 Workers) -0.203 -0.181 0.023 -0.159 -0.147(0.478) (0.437) (0.073) (0.387) (0.379)

Dummy for Firms Aged 5 to 10 Years Old 0.621*** 0.600*** 0.037 0.589*** 0.598***(0.134) (0.151) (0.037) (0.142) (0.142)

Dummy for Firms Aged 10 to 20 Years Old 0.590*** 0.622*** 0.004 0.614*** 0.623***(0.145) (0.163) (0.040) (0.156) (0.156)

Dummy for Firms Aged 20 to 40 Years Old 0.382** 0.332* -0.057 0.452** 0.465***(0.175) (0.191) (0.048) (0.180) (0.179)

Dummy for Firms Aged More than 40 Years Old 0.205 0.175 -0.117* 0.221 0.215(0.226) (0.202) (0.062) (0.197) (0.203)

Dummy for Managers with Post-Graduate Education -0.035 -0.062 -0.011 0.019 0.021(0.075) (0.082) (0.018) (0.072) (0.072)

Manager Years of Experience (log) 0.077* 0.096* 0.016 0.058 0.057(0.044) (0.049) (0.010) (0.044) (0.045)

Foreign-Owned Dummy 0.03 0.016 0.086* 0.072 0.069(0.157) (0.174) (0.046) (0.139) (0.138)

Years of Export Experience (log) 0.126*** 0.088* 0.014 0.070* 0.074*(0.028) (0.045) (0.011) (0.039) (0.040)

Years of Export Experience (log) * Index of Business Environ. 0.006(0.026)

Quality Certification Dummy 0.097 0.165* 0.004 0.083 0.075(0.089) (0.091) (0.021) (0.083) (0.084)

Dummy for R&D Activities 0.062 -0.056 0.003 -0.011 -0.008(0.071) (0.114) (0.025) (0.110) (0.104)

Perc. of Machinery Less than 5 Years Old -0.06 -0.06 0.033 -0.051 -0.036(0.156) (0.167) (0.035) (0.149) (0.150)

Perc. of Computerized Machinery 0.194 0.105 -0.079* 0.175 0.177(0.140) (0.151) (0.043) (0.145) (0.159)

Dummy for R&D Activities * Index of Business Environ. 0.005(0.094)

Perc. of Machinery Less than 5 Years Old * Index of Business Environ. 0.149**(0.073)

Perc. of Computerized Machinery * Index of Business Environ. 0.010(0.118)

Overdraft Dummy 0.09 0.038 -0.005 0.073 0.092(0.078) (0.086) (0.019) (0.077) (0.076)

Loan Dummy -0.097 -0.067 0.026 -0.026 -0.030(0.077) (0.081) (0.019) (0.070) (0.069)

Number of Power Interruptions (log) -0.167 -0.085 -0.129 -0.129 -0.145(0.225) (0.236) (0.080) (0.215) (0.214)

Protection Payments as Perc. Sales -87.825 -121.283** -30.179** -87.348** -85.376**(69.442) (47.147) (15.197) (35.999) (35.146)

Days to Clear Customs for Imports (log) -0.343* -0.184 0.063 -0.267 -0.244(0.199) (0.214) (0.055) (0.224) (0.207)

Perc. Manag. Time Spent Dealing with Regulation -0.435 -0.553* 0.117 -0.290 -0.321(0.389) (0.317) (0.122) (0.331) (0.293)

Avg. Perc. of Sales Paid in Bribes to Get Things Done 1.705 1.632 0.118 1.353 0.684(1.099) (1.101) (0.315) (0.980) (1.051)

Industry Fixed Effects Yes Yes Yes Yes YesYear Fixed Effects Yes No Yes Yes YesLocation Fixed Effects Yes Yes Yes Yes Yes

N. Observations 2235 513 1722 2249 2235

Dependent Variable is Firm TFP

Notes: Robust standard errors clustered at the firm level in parentheses. ***, **, and * indicate significance at 1%, 5%, and 10% confidence levels, respectively. The omitted size category is extremely large firms (more than 500 workers) and the omitted age category is firms less than five years old.

28

0.2

.4.6

.81

Chemicals Food Garments Leather Textiles

Shares in Weighted Avg. Industry TFP

Unweighted Avg. Industry TFP Reallocation in Industry

Figure 1: Decomposition of Industry TFP

29

APPENDIX

(a) Data Issues A sample of 700 firms composed of 350 firms in the RMG industry and 350 firms in

other industries was drawn. Covering 10% of registered firms in the RMG industry was a

condition required by the study which financed the survey (World Bank, 2005). The sample

was drawn resorting to a different data source for each industry. For the RMG industry, a

stratified random sample of 350 firms was drawn based on the BGMEA directory of RMG

firms. For the other industries, random samples were drawn based on the most recent list of

firms from the corresponding business associations to cover 350 firms. However, data was

collected for only 332 firms in the other industries due to non-response by 18 firms.

While, strict quality control criteria were applied during the survey data collection and

processing phases, the final dataset of 682 firms includes some anomalies. We eliminated

from the estimating sample: (a) firms that report being subcontracted which do not purchase

materials themselves since they do not fit into our production function framework of an

optimizing firm making input and output choices; (b) firms that closed during the sample

period and then re-opened at the end of the sample period since they lack the lagged values

for inputs and output needed for production function estimation; (c) firms with year-to-year

growth rates in any of three ratios (real sales to total workers, real material costs to total

workers and capital to total workers) larger (smaller) than 150% (-150%); (d) firms with clear

data entry errors in the production variables. The final estimating sample includes 575 firms.

(b) Production Function Estimation

We estimate a Cobb-Douglas production function separately for each industry

following the ACF modification to the Olley and Pakes (1996) technique, which is based on a

dynamic profit maximization problem for the firm. The estimating equation is given by:

30

itititSitKitLit skly εωβββ ++++= , (A1)

where i designates a firm and t designates a year, yit is value added, lit is labor, kit is capital

(all in logarithms), sit is the share of skilled workers, itω is productivity known to the firm,

and itε is a mean-zero shock uncorrelated with input choices and unknown to the firm. While

itω and itε are unknown also to the researcher, itω is a state variable known to the firm to

which it adjusts its input choices. Capital for use at t is obtained following the accumulation

equation: 11)1( −− +−= ititit IKdK (d being the depreciation rate) and is thus assumed to be

known at t-1. Hence, capital is a state variable which can only be affected by the expected

value of productivity itω conditional on 1−itω . The main modification introduced by ACF to

the Olley and Pakes (1996) technique is the timing of labor input choices. Labor and the

share of skilled workers are not freely variable inputs but rather are assumed to be chosen at

sub-period t-b (0<b<1), after capital is known (at t-1), but before investment is chosen at t.

Thus, labor inputs can also be affected by the expected value of productivity itω conditional

on 1−itω . A rationale for this timing choice and for the fact that labor inputs are not freely

variable may be for example restrictions in hiring or firing of workers. ACF assume, as Olley

and Pakes (1996), that productivity itω follows a first-order Markov process:

( ) ( )11 // −− = itititit pIp ωωω where 1−itI is the information set available to the firm at t-1 or

itititit E μωωω += − ]/[ 1 where itμ is the unexpected part of productivity, which is mean

independent of all information known at t-1. Moreover, given the aforementioned timing

assumption for labor variables, ACF also assume that itω evolves according to a first-order

31

Markov process between the sub-periods t-1, t-b, and t: ( ) ( )bititbitit pIp −− = ωωω // and

( ) ( )11 // −−−− = itbititbit pIp ωωω .

The estimation procedure uses investment - an observable firm characteristic - to proxy

for the unobservable firm productivity. From the dynamic profit maximizing problem, firm

investment Iit depends on the state variables - capital and firm productivity - and from the

timing assumptions above it depends also on the labor inputs:

),,,( itititittit slkII ω= . (A2)

This investment function is assumed to monotonically increasing in productivity, conditional

on capital and on the labor inputs. Thus, it can be inverted to express the unobserved

productivity as a function of observables - capital, investment, and labor inputs:

),,,( itititittit slIkωω = . Inserting this expression for productivity into Eq. (A1) results in the

first stage semi-parametric equation:

itititititittit slIky εωφ += ),,,,( , (A3)

where ),,,(),,,,( itititittitSitLitKitititititt slIkslkslIk ωβββωφ +++= .

Eq. (A3) is estimated by OLS, approximating the unknown function (.)tφ by a third-degree

polynomial on ),,,( itititit slIk . While this stage of the estimation does not identify any input

coefficient it obtains an unbiased estimate for (.)tφ since the error term in Eq. (A3) is

uncorrelated with the regressors. The estimate for (.)tφ is the polynomial expression

evaluated at the estimated polynomial coefficients and represents value added net of the

untransmitted shock itε .

In the second stage of the estimation, we obtain the input coefficients on using three

independent moment conditions. Capital used at t is part of 1−itI and thus uncorrelated with

32

the unexpected part of productivity itμ , so the moment condition that identifies the

coefficient on capital is given by: 0]/[ =itit kE μ . Similar moment conditions for itl and its

do not identify their corresponding coefficients given that itl and its are chosen at t-b and

therefore can be correlated with part of itμ (in addition to being possibly correlated with the

conditional expected value of itω ). Lagged labor inputs 1−itl and 1−its are chosen at t-b-1 and

part of 1−itI and are therefore uncorrelated with the unexpected part of productivity itμ .

Hence the two moment conditions that identify the coefficients on labor and on the skilled

share of workers are 0]/[ 1 =−itit lE μ and 0]/[ 1 =−itit sE μ .

In order to operationalize these moment conditions we proceed as follows. First, for

any given set of candidate coefficients ( )KSL βββ ,, we use the first stage estimate for (.)tφ

to obtain an estimate for ( )KSLit βββω ,, as itSitLitKtKSLt slk βββφβββω −−−= (.)),,( .

Second, we regress non-parametrically ( )KSLit βββω ,, on ( )KSLit βββω ,,1− (obtained

similarly) and a constant. The residuals from this non-parametric regression are an estimate

for ( )KSLit βββμ ,, . Then we construct the sample analog to the three moment conditions as:

( )∑∑⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡∗

−

−t i

it

it

it

KSLit

s

l

k

NT1

1,,11 βββμ .

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients. The standard errors for the coefficients are obtained by block-

bootstrap (i.e., if a firm is randomly selected to be part of the bootstrap sample then all years

of the firm are included in the bootstrap sample).

33

The inversion of the investment function (Eq. (A2)) can be done only if investment is

positive in all years. Thus, for each industry, the production function coefficients are

estimated based on a sample that includes only years of positive investment. However,

following Olley and Pakes (1996), we compute TFP measures as the residual of the

production function Eq. (A1) for firms in all years even those with zero investment.

Appendix Table 1: Sample Composition

Small (<10 workers)

Medium (10-50

workers)

Relatively Large (50-150

workers)

Very Large (150-500 workers)

Extremely Large (> 500

workers)Pharmaceuticals 51 5.9% 15.7% 45.1% 33.3%Food 88 1.1% 12.5% 44.3% 33.0% 9.1%Ready-Made Garments 276 0.4% 0.7% 48.6% 50.4%Leather/Footwear 24 4.2% 20.8% 33.3% 29.2% 12.5%Textiles 136 2.2% 16.9% 44.1% 36.8%

Total 575 0.4% 4.0% 13.9% 44.0% 37.7%

< 5 Years Old

5-10 Years Old

10-20 Years Old

20-40 Years Old

> 40 Years Old

Pharmaceuticals 15.7% 11.8% 29.4% 29.4% 13.7%Food 26.4% 10.3% 35.6% 25.3% 2.3%Garments 27.5% 27.5% 35.9% 8.3% 0.7%Leather/Footwear 8.3% 25.0% 54.2% 12.5%Textiles 29.4% 27.2% 27.2% 11.0% 5.2%

Total 25.6% 22.7% 32.8% 15.3% 3.7%

DhakaDhaka Export

Processing Zone

Chittagong

Chittagong Export

Processing Zone

Khulna Other

Pharmaceuticals 72.6% 5.9% 21.6%Food 28.4% 39.8% 4.6% 27.3%Garments 62.3% 4.7% 15.9% 9.8% 7.3%Leather/Footwear 87.5% 4.2% 4.2% 4.2%Textiles 29.4% 1.5% 9.6% 3.7% 55.9%

Total 51.3% 2.8% 16.7% 5.6% 0.7% 23.0%

Size Distribution (% of Firms)

Age Distribution (% of Firms)

Location (% of Firms)

IndustryNumber of

Firms

Industry

Industry

Notes: One firm in Food industry does not report its year of establishment needed to calculate its age. One firm in the garments industry does not report its location.

34

Appendix Table 2: Production Function Estimates using Industry-Specific Deflators

LaborShare of Skilled

Workers Capital N. Obs. Labor

Share of Skilled

Workers Capital N. Obs.

Pharmaceuticals 1.237*** 0.855*** 0.137*** 0.785*** 0.103 0.367**(0.059) (0.168) (0.035) 236 (0.244) (0.168) (0.163) 203

Food 0.753*** 0.589*** 0.364*** 0.644*** 0.370** 0.301**(0.062) (0.156) (0.043) 381 (0.181) (0.169) (0.146) 307

Ready-Made Garments 1.034*** 0.262** 0.064*** 0.815*** 0.314* 0.276*(0.032) (0.130) (0.017) 1170 (0.167) (0.175) (0.187) 854

Leather 0.797*** 0.679* 0.508*** 0.533*** 0.104 0.579***(0.067) (0.345) -0.072 120 (0.180) (0.355) (0.195) 108

Textiles 0.684*** -0.093 0.288*** 0.810*** 0.311 0.277*(0.042) (0.222) (0.028) 573 (0.173) (0.220) (0.177) 413

Coefficient on:

OLS Ackerberg, Caves, and Frazer (2007)

Coefficient on:

Notes: Robust standard errors in parentheses in the columns with OLS estimates. Bootstrap standard errors in parentheses in the columns with ACF estimates. ***, **, and * represent significance at 1%, 5%, and 10% confidence levels, respectively. The number of observations used for OLS estimation is higher since it includes firms with zero investment which are excluded from the Ackerberg, Caves, and Frazer (2007) estimation.

35

Appendix Table 3: Regressions with a Single Correlate of Firm TFP

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Very Large Dummy (150 to 500 Workers) 0.213*** 0.253*** 0.273*** 0.253*** 0.253*** 0.251*** 0.264*** 0.252*** 0.262***

(0.071) (0.070) (0.070) (0.069) (0.069) (0.069) (0.069) (0.069) (0.069)Relatively Large Dummy (50 to 150 Workers) 0.520*** 0.557*** 0.575*** 0.558*** 0.565*** 0.550*** 0.566*** 0.558*** 0.580***

(0.115) (0.113) (0.114) (0.113) (0.112) (0.111) (0.111) (0.113) (0.116)Medium Size Dummy (10 to 50 Workers) 0.436* 0.519** 0.684*** 0.518** 0.565** 0.544** 0.567** 0.517** 0.543**

(0.246) (0.237) (0.251) (0.238) (0.242) (0.238) (0.237) (0.237) (0.242)Small Size Dummy (Less than 10 Workers) -0.439 -0.438 -0.393 -0.44 -0.356 -0.371 -0.406 -0.442 -0.412

(0.374) (0.362) (0.345) (0.366) (0.362) (0.359) (0.370) (0.366) (0.369)Dummy for Firms Aged 5 to 10 Years Old 0.678*** 0.729*** 0.694*** 0.731*** 0.732*** 0.728*** 0.665*** 0.730*** 0.732***

(0.106) (0.106) (0.113) (0.106) (0.107) (0.106) (0.111) (0.107) (0.106)Dummy for Firms Aged 10 to 20 Years Old 0.824*** 0.849*** 0.806*** 0.853*** 0.845*** 0.838*** 0.704*** 0.852*** 0.846***

(0.097) (0.096) (0.106) (0.097) (0.096) (0.096) (0.116) (0.096) (0.096)Dummy for Firms Aged 20 to 40 Years Old 0.643*** 0.706*** 0.629*** 0.706*** 0.690*** 0.694*** 0.523*** 0.708*** 0.711***

(0.116) (0.116) (0.129) (0.116) (0.115) (0.115) (0.135) (0.116) (0.116)Dummy for Firms Aged More than 40 Years Old 0.362** 0.455*** 0.376** 0.457*** 0.454*** 0.452*** 0.310* 0.452*** 0.443***

(0.157) (0.156) (0.171) (0.157) (0.155) (0.158) (0.162) (0.157) (0.156)Dummy for Managers with Post-Graduate Educ. 0.027

(0.068)Manager Years of Experience (log) 0.064

(0.044)Foreign-Owned Dummy 0.067

(0.122)Exporters Dummy 0.133

(0.087)Majority Exporters Dummy 0.134

(0.086)Years of Export Experience (log) 0.083**

(0.037)Dummy for R&D Activities 0.039

(0.100)Quality Certification Dummy 0.103

(0.075)

N. Observations 2470 2465 2465 2362 2465 2465 2465 2443 2465 2465R-squared 0.65 0.68 0.69 0.7 0.69 0.69 0.69 0.69 0.69 0.69

(11) (12) (13) (14) (15) (16) (17) (18) (19)Very Large Dummy (150 to 500 Workers) 0.249*** 0.267*** 0.252*** 0.248*** 0.249*** 0.249*** 0.250*** 0.256*** 0.253***

(0.069) (0.070) (0.070) (0.070) (0.069) (0.069) (0.069) (0.069) (0.072)Relatively Large Dummy (50 to 150 Workers) 0.552*** 0.576*** 0.554*** 0.550*** 0.541*** 0.549*** 0.553*** 0.548*** 0.586***

(0.112) (0.113) (0.112) (0.113) (0.112) (0.112) (0.112) (0.113) (0.113)Medium Size Dummy (10 to 50 Workers) 0.510** 0.632*** 0.520** 0.504** 0.506** 0.502** 0.509** 0.528** 0.518**

(0.238) (0.223) (0.236) (0.237) (0.234) (0.239) (0.239) (0.246) (0.237)Small Size Dummy (Less than 10 Workers) -0.454 -0.400 -0.424 -0.459 -0.400 -0.472 -0.453 -0.502 -0.446

(0.366) (0.369) (0.370) (0.364) (0.384) (0.363) (0.365) (0.360) (0.366)Dummy for Firms Aged 5 to 10 Years Old 0.724*** 0.741*** 0.730*** 0.731*** 0.732*** 0.726*** 0.732*** 0.735*** 0.718***

(0.127) (0.106) (0.107) (0.107) (0.105) (0.108) (0.107) (0.106) (0.112)Dummy for Firms Aged 10 to 20 Years Old 0.843*** 0.887*** 0.849*** 0.852*** 0.835*** 0.851*** 0.852*** 0.850*** 0.831***

(0.126) (0.098) (0.097) (0.096) (0.095) (0.096) (0.096) (0.096) (0.100)Dummy for Firms Aged 20 to 40 Years Old 0.697*** 0.742*** 0.707*** 0.708*** 0.677*** 0.708*** 0.708*** 0.713*** 0.715***

(0.144) (0.118) (0.116) (0.116) (0.115) (0.116) (0.116) (0.115) (0.120)Dummy for Firms Aged More than 40 Years Old 0.446** 0.473*** 0.463*** 0.454*** 0.451*** 0.444*** 0.455*** 0.426*** 0.444***

(0.179) (0.155) (0.159) (0.157) (0.153) (0.157) (0.156) (0.153) (0.158)Perc. of Machinery Less than 5 Years Old -0.017

(0.139)Perc. of Computerized Machinery 0.200

(0.127)Overdraft Dummy 0.032

(0.069)Loan Dummy -0.019

(0.064)Days to Clear Customs for Imports (log) -0.391**

(0.189)Number of Power Interruptions (log) 0.131

(0.207)Perc. Manag. Time Spent Dealing with Regulation -0.030

(0.327)Perc. of Firms Paying Bribes to Get Things Done -0.159*

(0.092)Protection Payments as Perc. Sales -72.373**

(31.373)

N. Observations 2465 2455 2470 2470 2465 2465 2465 2442 2360R-squared 0.69 0.7 0.98 0.98 0.69 0.69 0.69 0.69 0.69

Dependent Variable is Firm TFP

Dependent Variable is Firm TFP

Notes: All regressions include industry fixed effects, year fixed effects, and location fixed effects. Robust standard errors in parentheses. ***, **, and * represent significance at 1%, 5%, and 10% confidence levels, respectively.