18 views

Uploaded by Anonymous pB9NV18F

financial

- PSYCHOLOGY TEST - SELF ACCEPTANCE
- error in Chemical Analysis
- Descriptive Statistics
- Output Spss Dan Eviews
- ppaers
- MAT LAB NEW
- Weighted Regression of Calibration Curve
- Solution HW3
- 2 klmpk
- Semi-detailed-lesson-plan.docx
- L11 Regression
- Agricultural Functions in Purnia
- November 30 Time Series Cross Sectional Analysis
- Quantitative Methods 1
- Statistics
- Regression Analysis Complete SAS DEMO (2)
- Jpe12793 Sup 0001 AppendixS1
- Simple Linear Regression Collection
- 02
- Determinants Od MF_1

You are on page 1of 29

29 (2015) 129

Management

journal homepage: www.elsevier.com/locate/econbase

leverageperformance relationship during the

nancial crisis of 20072009

Chaiporn Vithessonthi a,, Jittima Tongurai b,1

a

Department of Accountancy and Finance, School of Business, University of Otago, PO Box 56,

Dunedin 9054, New Zealand

b

Miyazaki International College, 1405 Kano, Kiyotake, Miyazaki 889-1605, Japan

a r t i c l e

i n f o

Article history:

Received 12 September 2014

Accepted 5 November 2014

Available online 13 November 2014

JEL classication:

E2

E4

E5

G1

G3

G32

Keywords:

Financial crisis

Financial leverage

Firm size

Private rms

Thailand

a b s t r a c t

We draw on a comprehensive set of data of all registered rms in

Thailand to examine whether rm size affects the relation between

leverage and operating performance during the global nancial

crisis of 20072009. From a data set of 496,430 rm-year observations of a sample of 170,013 mostly private rms, we nd that

the magnitude of the effect of leverage on operating performance

is non-monotonic and conditional on rm size. While our panel

regression results indicate that leverage has a negative effect on

performance across rm size subsamples, our year-by-year crosssectional regression results show that the effect of leverage on

performance is positive for small rms and is negative for large

rms. Our ndings show that about 75% of Thai rms in our sample

appear to have managed to get through the global nancial crisis on

the basis that they do not have to simultaneously deleverage and

liquidate their assets.

2014 Elsevier B.V. All rights reserved.

Corresponding author. Tel.: +64 3 479 9047; fax: +64 3 479 8171.

E-mail addresses: chaiporn.vithessonthi@otago.ac.nz (C. Vithessonthi), jtongurai@sky.miyazaki-mic.ac.jp (J. Tongurai).

1

Tel.: +81 0985 85 5931; fax: +81 0985 84 3396.

http://dx.doi.org/10.1016/j.muln.2014.11.001

1042-444X/ 2014 Elsevier B.V. All rights reserved.

1. Introduction

Motivated by recent studies on the leverageperformance relationship, such as that of Margaritis

and Psillaki (2010), who show that nancial leverage has a positive effect on rm performance,

Cai and Zhang (2011), who show that a change in nancial leverage is related to future operating

performance, Xu (2012), who nd that protability negatively affects leverage, and Faulkender and

Petersen (2006), who report that larger rms tend to be more leveraged, we attempt to examine

whether the leverageperformance linkage is inuenced by rm size.2 While there is a growing body

of empirical evidence for the leverageperformance linkage for publicly traded rms, there is limited

understanding of such a relationship for small and medium-sized private rms. However, small and

medium-sized rms in almost all countries contribute signicantly to economic growth. We use a

unique set of data on all rms in Thailand to examine the linkages between leverage and operating

performance. As publicly listed rms constitute an extremely small proportion of the sample (i.e., the

average number of listed rms in Thailand is smaller than 600 rms during the sample period; our

sample consists of 170,013 rms), this study can alternatively be considered as an analysis of private

rms in an emerging market country, thereby making this study potentially one of the rst studies on

the leverageperformance relation using a very large sample of private rms. And as we will show,

rm size affects the leverageperformance relation.

Several scholars, such as Paulson and Townsend (2004), show that small rms are typically very

small, that similarities between small entrepreneurial rms in the US and Thailand are very striking,

and that small, entrepreneurial rms are a key source of jobs and economic growth. They show that

small rms in Thailand are more likely to pass up a protable investment opportunity because they do

not have sufcient funds to exploit them, suggesting that nancial constraints are a key determinant

of entrepreneurial activities of small rms. Traditionally, it would be argued that nancial constraints

affect future investment of small rms by limiting the extent to which the small rms nance potentially protable investments through borrowing when internal and/or equity nancing is limited. Prior

studies, such as Denis and Sibilkov (2010), Li (2011), and Chen and Chen (2012), show that nancial

constraints affect investment (e.g., capital expenditure and/or R&D investment).

This paper focuses on the main idea that rm size3 moderates the relation between leverage and

performance. In particular, we conjecture that as rm size increases (i.e., moving from a small rm

category to a medium-sized rm category), an additional debt capacity provided by nancial intermediaries (e.g., banks) allows rms to invest more in new projects that should subsequently result

in an improvement in performance, which in turn increases the rms additional debt capacity. Thus,

for small and medium-sized rms, leverage has a positive effect on performance and vice versa. It is

important to note that while our prediction of a positive effect of leverage on performance for small

and medium-sized rms is consistent with the agency cost hypothesis as in, for example, Margaritis

and Psillaki (2010), our prediction is rather based on the dynamic performanceleverageperformance

effect that channels through rm size. Due to (1) past jumps in leverage ratios potentially above the

long-run optimal target level, (2) limited investment opportunities for large rms, and/or (3) organizational complexity (as discussed by several scholars such as Miller and Friesen (1983), Fredrickson

and Iaquinto (1989), Huff et al. (1992), and Audia et al. (2000) that large rms have to encounter with

greater demands for well-developed administrative systems to maintain creativity, innovativeness

and entrepreneurship to overcome an impediment to growth), good rm performance should subsequently lead to a reduction in leverage ratios. Accordingly, for large rms the effect of performance

on leverage is negative, while the effect of leverage on performance is either negative or insignicant.

2

While prior studies, such as Antoniou et al. (2008), examine the effect of rm size on leverage or performance, they generally

do not investigate the possible interactive effect between leverage and rm size on operating performance.

3

It should be noted that rm size has been often used as a proxy for nancial constraints and could be correlated with other

rm characteristics, such as rm age; we nd that the correlation between rm size and rm age is statistically signicant but

small in magnitude (r = 0.30, p-value <0.001, N = 496,430) in our panel sample. As we include rm age as a control variable in

our regressions, we test the inuence of rm size on the relation between leverage and performance, after controlling for rm

age.

Our results are generally consistent with our predictions; that is, the effect of leverage on performance

is positive for small rms and is negative for large rms.

From a data set of 496,430 rm-year observations of a sample of 170,013 mostly private rms in

Thailand during 20072009, we attempt to answer our fundamental question of what we potentially

know more about the leverageperformance relation of private rms from this sample than what we

previously knew from the literature. Given that prior studies on the leverageperformance relation

in private rms are very limited, and that prior insights into this relation for the private rms are

indirectly derived from the initial public offering literature (e.g., when past nancial statements of

newly listed rms become available to the public), in this study we show several key ndings that

have not been systematically documented in the literature with respect to private rms. First, in our

panel regression analyses, leverage is negatively associated with operating performance (ROA), and

the effect of leverage on operating performance is non-monotonic and tends to be stronger for larger

rms than for small rms. Second, our year-by-year cross-sectional regression results consistently

show that the effect of leverage on operating performance is positive for small rms and is negative

for large rms across years. These results appear to provide a plausible explanation for the mixed

results for the sign of the relation between leverage and performance reported in the literature. Third,

past operating performance has a negative effect on a change in leverage; however, for both very small

and very large rms, past operating performance does not affect the change in leverage. Fourth, over

the global nancial crisis period, smaller rms substantially expand their asset base, while larger rms

reduce theirs.

In terms of methodology, we generally use an approach that is similar to the work of Firth et al.

(2008) and Cai and Zhang (2011). Our approach to the leverageperformance modeling follows the

literature. However, contrary to most models, we condition the leverageperformance relation on rm

size. Furthermore, we account for the possible non-linear relation between leverage and performance.

To address potential endogeneity related to the linkage between leverage and operating performance,

we test the robustness of our results by using (1) the instrumental variable (IV) and two-stage least

square (2SLS) estimation and (2) two-step system GMM estimation. We show that our results are

robust to different specications. We further show that our results are relatively stable across years,

as our year-by-year cross-sectional IV 2SLS regression results show a similar pattern over the sample

period, suggesting that our ndings are stable across time and are less likely to be driven by the

inuence of the global nancial crisis.4

In this paper, we provide new evidence on the linkages between a change in capital structure and

operating performance and extend the existing literature in two important ways. First, in complementary to prior studies that examine these linkages in publicly listed rms, we examine a large data

set that includes both private and public rms in Thailand. Previous studies, such as Aivazian et al.

(2005) and Korajczyk and Levy (2003), have been mainly restricted to examining publicly traded rms

in developed countries (e.g., the US). Similar studies have also been conducted in emerging market

countries but mainly limited to examining publicly traded rms. For instance, Drifeld and Pal (2001)

examine the effect of nancial leverage on corporate investments of non-nancial rms in Indonesia,

Korea, Malaysia, and Thailand during 19891997. In this respect, our paper is perhaps one of the rst

to examine the relation between leverage and operating performance for all private and public rms

in Thailand (or in any country in Asia). Second, we are able to show that rm size affects the relation

between leverage and operating performance. This nding is important because while prior studies

typically examine the direct effect of rm size on leverage (e.g., Strebulaev and Yang, 2013; Xu, 2012)

or rm performance (e.g., Giroud et al., 2012; Margaritis and Psillaki, 2010), they largely ignore the

indirect effect of rm size on the relation between leverage and rm performance. In this study, we

highlight the moderating effect of rm size on the effect of leverage on performance.

4

Due to the limitation of our sample data set, we cannot empirically ascertain whether the pattern of our results is similar to

that of the pre- and post-crisis periods. If we are to assume that the negative ramications of the global nancial crisis gradually

increases from 2007 to 2009 (with the peak of the crisis in the nancial markets in mid-2008), the stability of our year-by-year

regression results would suggest that the effect of the global nancial crisis on the observed leverageperformance relations in

our study is not substantial.

In summary, we present the evidence indicating that the leverageperformance relation is conditional on rm size, and, more importantly, that the sign of the impact of leverage on performance

and of the impact of operating performance on leverage change depends on rm size. Our ndings

contribute directly to the literature on the relation between leverage and performance by showing

that the mixed results observed in the literature could, to some extent, be explained by rm size. As

several scholars, such as Erickson and Whited (2000), Duchin et al. (2010b), and Moyen and Platikanov

(2013), use rm size as a rough proxy for the extent to which rms are nancially constrained, then

our results indicating that rm size affects the inuence of performance on leverage change are in

line with Korajczyk and Levy (2003), who show that the magnitude of the impact of operating performance on leverage is contingent on the level of nancial constraints. However, the fact that only small

rms expanded their asset base over the nancial crisis, but larger rms reduced theirs implies that

small rms are not as nancially constrained as generally expected. While this result is in contrast

with Paulson and Townsend (2004) who suggest that nancial constraints for very small rms appear

to limit the realization of investment opportunities as these rms cannot obtain external nancing

although they might be protable, it does not necessarily imply that small rms do not encounter

nancial constraints. With no or limited external nancing through a traditional banking channel, very

small rms tend to remain small as they cannot obtain necessary funds to expand their businesses.

2. Data and descriptive statistics

Our raw data set consists of all rms registered at the Ministry of Commerce in Thailand

over the period 20072009. As our primary objective is to test whether rm size affects the

leverageperformance relation in a cross-section analysis and potentially over the global nancial crisis period, rms that did not exist during the entire three-year period are removed from the sample.5

After checking and screening for apparent miscoding, we also exclude rms that did not have complete

records on total assets, total liabilities, and EBIT. Accordingly, we have 540,591 rm-year observations

involving 184,980 rms.6 In an ideal setting, we would wish to examine whether the pattern of our

results observed during the nancial crisis differs from the tranquil period; however, we do not have

a data set for periods prior to 2007 and for periods after 2009.

Consistent with Faulkender and Petersen (2006), we nd some rms with leverage ratios larger

than 1. To reduce noise in our data set and to be conservative, we follow the approach of Bena and

Ondko (2012) by removing rms with (1) very small total assets (i.e., less than 50,000 THB at the

beginning year), (2) very high leverage (i.e., the ratio of total liabilities to total assets of two for any

given year), and/or (3) very large prot/loss (i.e., the absolute value of the EBIT-to-total-assets ratio

(EBIT/TA) of 10 for any given year)7 from the sample. Therefore, the nal sample of our study consists

of 170,013 rms and 496,430 rm-year observations.

Table 1 provides descriptive statistics for our sample. In light of the objective of this paper and

as suggested by Faulkender and Petersen (2006) that a rms leverage should depend on its characteristics, we divide our sample into two groups: (1) the large rm size subsample, which includes

all rms with total assets larger than the median value at the end of each year, and (2) the small

rm size subsample, which includes all rms with total assets equal to or smaller than the median

value. Consistent with Faulkender and Petersen (2006), who report that larger rms tend to be more

leveraged, we nd that larger rms have higher leverage ratios. Large rms have signicantly greater

leverage than small rms (see Panels B and C of Table 1). The large rm subsample has an average

leverage ratio of 0.53, whereas the small rm subsample has an average leverage ratio of 0.25. The

difference is 0.39 (p-value <0.01), suggesting that larger rms leverage ratios are on average 157%

The original list of rms at the end of 2009 consists of 296,046 rms.

With this data set, all new and exit rms in the three-year period are excluded from the sample, which naturally removes

the concern of a composition effect because our sample becomes cross-sectional. See Xu (2012), for example, for a discussion

of the issue of compositional effect.

7

Bena and Ondko (2012) use (1) EUR1000 as a cutting point for rm size, (2) the ratio of the total long-term debt to the total

assets of two as a criterion for leverage, and (3) the ratio of the absolute value of prot/loss to the total assets of 10 as a criterion

for protability.

6

Table 1

Descriptive statistics on key variables. Panel A of this table presents summary descriptive statistics on key variables for the

full sample. Panel B presents summary statistics on the variables for the large rm subsample that includes all rms with total

assets larger than the median value. Panel C presents summary statistics on the variables for the small rm subsample that

includes all rms with total assets equal to or smaller than the median value.

Mean

Median

Minimum

Maximum

S.D.

Interest

GDP growth

Age

TA (in millions)

EBIT (in millions)

ROA

Leverage

Log age

Log TA

ROA

LEV

TA

6.47

1.87

10.44

153.72

6.54

0.02

0.39

1.96

1.79

0.00

0.00

0.38

Age

TA (in millions)

EBIT (in millions)

ROA

Leverage

Log age

Log TA

ROA

LEV

TA

Panel C: Small rms

Age

TA (in millions)

EBIT (in millions)

ROA

Leverage

Log age

Log TA

ROA

LEV

TA

Observations

6.75

1.60

7.00

4.60

0.12

0.03

0.28

1.95

1.53

0.00

0.00

0.01

5.85

1.10

1.00

0.00

43,915.55

9.83

0.49

0.00

7.57

10.25

1.97

1.00

6.85

5.40

94.00

1,740,192.42

63,226.20

9.52

1.99

4.54

14.37

10.25

1.95

6213.98

0.45

2.64

9.06

7083.90

259.62

0.26

0.39

0.93

1.91

0.33

0.18

16.35

496,430

496,430

496,430

496,430

496,430

496,430

496,430

496,430

496,430

340,026

340,026

340,026

12.50

300.28

12.80

0.04

0.53

2.22

3.23

0.00

0.00

0.62

11.00

16.97

0.51

0.03

0.54

2.40

2.83

0.00

0.00

0.02

1.00

4.05

43,915.55

8.84

0.49

0.00

1.40

8.84

1.97

0.99

94.00

1,740,192.42

63,226.20

9.52

1.99

4.54

14.37

10.25

1.87

6213.98

9.41

9926.79

363.78

0.15

0.37

0.85

1.49

0.17

0.16

23.10

252,693

252,693

252,693

252,693

252,693

252,693

252,693

170,012

170,012

170,012

8.30

1.77

0.06

0.00

0.25

1.69

0.29

0.00

0.00

0.13

5.00

1.51

0.02

0.02

0.07

1.61

0.41

0.00

0.00

0.01

1.00

0.00

20.58

9.83

0.02

0.00

7.57

10.25

1.96

1.00

89.00

4.71

12.68

8.86

1.99

4.49

1.55

9.95

1.95

99.65

8.14

1.15

0.44

0.34

0.35

0.94

0.85

0.43

0.20

0.98

243,737

243,737

243,737

243,737

243,737

243,737

243,737

170,014

170,014

170,014

[(0.5312 0.2488)/0.2488] larger than smaller rms. One plausible interpretation of this nding is

that small rms are nancially constrained (in a narrow sense with respect to bank credit), perhaps,

as a result of information asymmetry between rms and nancial intermediaries (e.g., banks). It is

possible that small rms can only raise the level of debt when they accompany the additional debt

with equity issuance. A competing interpretation is that in comparison with larger rms, small rms

are more nancially conservative and do not use up their debt capacity, thereby leading to potentially

lower-than-optimal nancial leverage ratios.

To gain insights on the inuence of rm size on the relation between leverage and performance,

we divide the sample into six groups based on total assets at the end of each year: (1) TA1 very small

(TA < 1 million THB); (2) TA2 small (1 million THB TA < 5 million THB); (3) TA3 medium (5 million

THB TA < 10 million THB); (4) TA4 upper medium (10 million THB TA < 100 million THB); (5) TA5

large (100 million THB TA < 1000 million THB); and (6) TA6 very large (TA 1000 million THB).8

In Table 2, we nd that an average leverage ratio for rms with total assets smaller than 1 million

8

While we arbitrarily set the size classication, rms that are considered large (i.e., with total asset of 100 million THB or

more) are likely to be sufciently large to meet one of criteria for listing on the MAI (the second stock market in Thailand). That

Table 2

Descriptive statistics on key variables, six rm size subsamples. The full sample is divided into six groups, based on total assets

(TA) at the end of each year: Panels AF of this table present summary statistics on key variables for six rm size subsamples:

TA1 (TA < 1 million THB), TA2 (1 million THB TA < 5 million THB), TA3 (5 million THB TA < 10 million THB), TA4 (10 million

THB TA < 100 million THB), TA5 (100 million THB TA < 1000 million THB), and TA6 (TA 1000 million THB), respectively.

Mean

Median

Minimum

Maximum

S.D.

Observations

Panel A: TA1

Age

TA (in millions)

EBIT (in millions)

ROA

Leverage

Log age

Log TA

ROA

LEV

TA

7.81

0.58

0.02

0.08

0.20

1.60

0.72

0.02

0.00

0.03

5.00

0.59

0.00

0.01

0.04

1.61

0.53

0.00

0.00

0.01

1.00

0.00

7.32

9.83

0.00

0.00

7.57

10.25

1.96

1.00

66.00

1.00

6.24

8.86

1.99

4.19

0.00

9.78

1.95

44.15

8.07

0.29

0.21

0.52

0.35

0.96

0.69

0.65

0.23

0.59

73,214

73,214

73,214

73,214

73,214

73,214

73,214

50,039

50,039

50,039

Panel B: TA2

Age

TA (in millions)

EBIT (in millions)

ROA

Leverage

Log age

Log TA

ROA

LEV

TA

8.62

2.46

0.11

0.04

0.28

1.75

0.79

0.00

0.00

0.18

6.00

2.20

0.08

0.04

0.12

1.79

0.79

0.00

0.00

0.02

1.00

1.00

20.58

9.56

0.02

0.00

0.00

9.47

1.86

0.99

89.00

5.00

13.15

5.39

1.99

4.49

1.61

9.95

1.94

99.65

8.19

1.14

0.53

0.22

0.35

0.93

0.47

0.28

0.18

1.11

184,379

184,379

184,379

184,379

184,379

184,379

184,379

126,446

126,446

126,446

Panel C: TA3

Age

TA (in millions)

EBIT (in millions)

ROA

Leverage

Log age

Log TA

ROA

LEV

TA

10.94

7.12

0.32

0.05

0.44

2.07

1.94

0.00

0.00

0.27

9.00

6.92

0.31

0.05

0.37

2.20

1.94

0.00

0.00

0.03

1.00

5.00

58.31

8.84

0.00

0.00

1.61

8.84

1.52

0.98

80.00

10.00

70.11

7.63

1.99

4.38

2.30

5.99

1.87

102.23

8.65

1.44

1.23

0.17

0.38

0.85

0.20

0.20

0.18

1.82

69,916

69,916

69,916

69,916

69,916

69,916

69,916

48,043

48,043

48,043

Panel D: TA4

Age

TA (in millions)

EBIT (in millions)

ROA

Leverage

Log age

Log TA

ROA

LEV

TA

12.65

31.59

0.88

0.03

0.58

2.25

3.24

0.00

0.00

0.49

11.00

23.09

0.61

0.03

0.60

2.40

3.14

0.00

0.00

0.01

1.00

10.00

391.70

7.21

0.49

0.00

2.30

7.20

1.81

0.99

88.00

100.00

325.49

9.52

1.99

4.48

4.61

10.25

1.67

830.99

9.20

22.11

5.19

0.14

0.37

0.83

0.64

0.16

0.16

7.73

126,949

126,949

126,949

126,949

126,949

126,949

126,949

86,830

86,830

86,830

Panel E: TA5

Age

TA (in millions)

EBIT (in millions)

ROA

Leverage

Log age

Log TA

ROA

LEV

TA

15.18

292.88

10.96

0.03

0.60

2.44

5.47

0.01

0.00

1.49

14.00

210.52

3.68

0.02

0.64

2.64

5.35

0.00

0.00

0.01

1.00

100.00

942.98

3.32

0.00

0.00

4.61

6.96

1.97

0.95

90.00

999.89

3058.66

6.95

1.99

4.50

6.91

3.10

1.44

6213.98

10.34

209.07

48.92

0.13

0.34

0.83

0.62

0.15

0.15

52.00

34,936

34,936

34,936

34,936

34,936

34,936

34,936

23,884

23,884

23,884

Table 2 (Continued)

Panel F: TA6

Age

TA (in millions)

EBIT (in millions)

ROA

Leverage

Log age

Log TA

ROA

LEV

TA

Mean

Median

Minimum

Maximum

S.D.

Observations

17.05

8680.25

385.68

0.06

0.55

2.56

7.94

0.01

0.01

2.68

14.00

2101.51

90.37

0.04

0.56

2.64

7.65

0.00

0.01

0.03

1.00

1000.38

43,915.55

1.37

0.00

0.00

6.91

3.66

1.69

0.78

94.00

1,740,192.42

63,226.20

3.83

1.99

4.54

14.37

1.53

0.99

2603.77

12.15

58,879.02

2144.20

0.12

0.31

0.80

0.99

0.12

0.13

65.82

7036

7036

7036

7036

7036

7036

7036

4784

4784

4784

Fig. 1. Leverage and operating performance. This gure shows leverage (total liabilities/total assets) and returns on assets

(EBIT/TA in percent) for six rm size subsamples: TA1 (TA < 1 million THB), TA2 (1 million THB TA < 5 million THB), TA3 (5

million THB TA < 10 million THB), TA4 (10 million THB TA < 100 million THB), TA5 (100 million THB TA < 1000 million THB),

TA6 (TA 1000 million THB), respectively.

THB (the TA1 subsample) is 0.20 versus 0.55 for the sample of rms with total assets larger than 1000

million THB (the TA6 subsample). In comparison with rms with total assets smaller than 1 million

THB, rms with total assets between 5 million THB and 10 million THB (the TA3 subsample) has an

increase in the rms leverage by 120% [0.44 0.20/0.20].

In Fig. 1, we depict the mean of leverage ratios and EBIT/TA ratios (in percent) for six rm size

subsamples, as also shown in Table 2. Fig. 1 illustrates the non-linear relation between leverage and

rm size (excluding nancial rms (i.e., rms with Thailand Standard Industrial Classication (TSIC)

codes from J65 to J67) does not change the pattern of the plot). We nd evidence to suggest that as rm

size increases, leverage ratios increase. The fact that very large rms (the TA6 subsample) have lower

leverage ratios than large rms (the TA5 subsample) indicates that the relation between leverage and

is, the paid-up capital in common shares must be not less than 20 million THB after the IPO. The equivalent requirement for

listing on the Stock Exchange of Thailand (SET) is the paid-up capital in common shares of 300 million THB after the IPO. Firms

in the very large subsample are considered large enough to be listed on the SET.

Fig. 2. Leverage and changes in leverage by year and rm size. Panel A shows the mean leverage (total liabilities/total assets)

by year for the full sample and six rm size subsamples: TA1 (TA < 1 million THB), TA2 (1 million THB TA < 5 million THB),

TA3 (5 million THB TA < 10 million THB), TA4 (10 million THB TA < 100 million THB), TA5 (100 million THB TA < 1000

million THB), TA6 (TA 1000 million THB), respectively. Panel B shows the mean change in leverage by year (i.e., LEV Change

20072008 = leverage in 2008leverage in 2007; LEV Change 20082009 = leverage in 2009leverage in 2008) for the full sample

and six rm size subsamples.

rm size is inverted U-shaped. In addition, Thai rms appear to have a higher leverage than rms in

developed countries. For comparison, the median leverage ratio for large Thai rms (0.64 and 0.56 for

the TA5 and TA6 subsamples; see Panels E and F of Table 2) is substantially larger than the median

leverage ratio of 0.28 for Japanese rms reported by Antoniou et al. (2008).9

Panel A of Fig. 2 shows the mean leverage by year during 20072009 for the full sample and six rm

size subsamples. The mean for leverage increases as rm size becomes larger, but becomes smaller

in the largest rm size subsample (the TA6 subsample), indicating an inverted U-shaped relation

between leverage and rm size. Overall, there is a small decrease in the mean leverage over the

sample period for almost all rm size subsamples. In Panel B of Fig. 2, we show how year-on-year

changes in leverage evolve over the sample period for the full sample and six rm size subsamples.

For small rms (the TA1 and TA2 subsamples), the mean year-on-year change in leverage is positive for

both periods (i.e., from 2007 to 2008 and from 2008 to 2009), indicating that smaller rms increased

their leverage ratios during the nancial crisis. For the medium-sized and large rms (the TA3, TA4,

TA5, and TA6 subsamples), the mean year-on-year change in leverage is negative for both periods,

indicating that medium-sized and large rms decreased their leverage ratios during the nancial

When we exclude nancial rms from the sample in the robustness tests, leverage ratios decrease slightly.

crisis. However, these changes appear to be small in magnitude. Suppose that rm size is an accurate

proxy for nancial constraints; these results would suggest that larger rms were more nancially

constrained than smaller rms during the nancial crisis. This interpretation is not consistent with

the literature (e.g., Duchin et al., 2010b; Korajczyk and Levy, 2003), conjecturing that larger rms are

less nancially constrained than smaller rms because large rms presumably have better access to

nancial markets than do small rms.

3. The relationship between leverage and operating performance

3.1. The effect of leverage on operating performance

In this subsection, we examine the effect of leverage on operating performance, which is measured by the return on assets (ROA). Consistent with the corporate nance literature, we estimate the

following baseline regression:

ROAi,t = 0 + 1 INTt + 2 GDPt + 3 AGEi,t + 4 SIZEi,t + 5 EBITi,t + 6 LEVi,t + vi + i,t ,

(1)

where the return on assets, ROAi,t , is measured as the EBIT/TA ratio for rm i in year t and is used as

a proxy for operating performance; INTt denotes an average of the 12 month-end minimum prime

interest rates for year t, and GDPt denotes the GDP growth rate for year t. We obtain the interest rate

and GDP growth rate series from the Bank of Thailand. Consistent with the literature (Ahn et al., 2006;

Aivazian et al., 2005; Faulkender and Petersen, 2006; Mizen and Tsoukas, 2012), leverage, LEVi,t , is

measured as the ratio of the book value of total liabilities to the book value of total assets for rm i in

year t.10 As rms in our sample do not report the amount of bank loans in their nancial statements,

we cannot use the ratio of the book value of total bank loans to the book value of total assets as a

measure of leverage. Furthermore, almost all rms in our sample are not publicly traded; therefore,

we cannot use market leverage. Similar to Firth et al. (2008), we include rm size and rm age as

control variables. As in Firth et al. (2008) and Zheng et al. (2012), rm size, SIZEi,t , is measured as the

natural logarithm of total assets for rm i in year t, while, as in Jimnez et al. (2009), rm age, AGEi,t , is

measured as the natural logarithm of the number of years since the rm was established for rm i in

year t, i is a rm-specic xed effect, and I,t is the error term. To capture more information from our

panel data set that is limited with respect to short time series (i.e., three years), we rst run a panel

regression analysis with no lagged explanatory variables. While this approach allows us to bring all

the data to bear at once to generate the best estimation, it does not address the possible endogeneity

concern related to the relation between leverage and rm performance, which will be addressed more

formally in subsequent sections.

We analyze the non-monotonic effect of leverage on operating performance by adding a squared

leverage in the model specication as follows:

2

ROAi,t = 0 + 1 INTt + 2 GDPt + 3 AGEi,t + 4 SIZEi,t + 5 EBITi,t + 6 LEVi,t + 7 LEVi,t

+ vi + i,t .

(2)

We estimate all panel OLS regressions using the xed effects models with Whites

heteroscedasticity-consistent cross-section standard errors and covariance.11 Table 3 presents the

estimation results of Eqs. (1) and (2) using ROA as the dependent variable. In Panel A of Table 3, the

results show that leverage is negatively related to ROA in the full sample. Columns 24 of Panel A

show a similar pattern of results for all of the industry subsamples (i.e., agricultural, manufacturing,

and service industries).12

10

The ratio of the book value of long-term debt to the book value of total assets is used in some studies, such as Aivazian et al.

(2005) and Hall (2012).

11

Several studies, such as Akbar et al. (2013), also use the xed effects models. We undertake Hausman tests, which suggest

that the xed effects models are preferred. MacKay and Phillips (2005) also note that rm characteristics seem to change slowly.

12

Financial intermediaries, insurance companies, pension funds, and nancial services rms are included in the service

industry subsample.

10

The coefcient on GDP growth is positive and signicant in all models, suggesting that higher

economic growth offers business opportunities, which in turn have a positive effect on ROA. Stated

differently, it suggests that recessions have a negative effect on ROA. There is also a positive relation

between rm size and ROA. While interest rate has no effect on the ROA in the full sample, it has a

negative effect on ROA for the manufacturing subsample. One explanation is that interest rates may

also indicate business conditions. During the global nancial crisis, as higher interest rates reect a

fall in economic activities (e.g., due to expected recessions) and a hike in risk premium (e.g., due to

loan supply shocks), manufacturing rms are likely to suffer from economic recessions more severely

than rms in agricultural and service industries. Intuitively, we use the two macroeconomic variables

(i.e., INT and GDP) to control for unobserved year effects (time-specic factors) that may also inuence

ROA. Alternatively, we include a time-xed effect into Eqs. (1) and (2) to control for the unobserved

year effects (and remove the two macroeconomic variables from both equations). We nd that our

(untabulated) results from the panel regression analysis with rm-xed effects and time-xed effects

are almost identical to those reported in Tables 35.

Panel B of Table 3 provides evidence to support the notion that the leverageperformance relation

is non-linear, given that in the manufacturing subsample, the coefcient on leverage is negative and

statistically signicant, and the coefcient on the squared term of leverage is positive and statistically signicant. This nding suggests that the relation between leverage and ROA is U-shaped for

manufacturing rms.

To test whether the relation between nancial leverage and operating performance is stronger for

large rms than for small rms, we rst add the interaction term between leverage and a size dummy

variable, which takes a value of one when total assets are larger than the median value and zero

otherwise, in Eq. (1). In the investment literature, rm size is used as a proxy for rm-level nancial

constraints. For instance, Duchin et al. (2010b) classify nancially constrained rms and nancially

unconstrained rms based on the median of rm size (i.e., total assets). Column 1 of Table 4 shows that

the estimated coefcient on the interactive term involving leverage and the size dummy variable is

negative and signicant, which may suggest that the level of nancial constraints negatively moderates

the effect of leverage on performance. Next, we divide the sample into two subsamples: the small and

the large rm subsamples. We estimate Eq. (1) separately for the two subsamples. The estimated

coefcient on leverage is negative and signicant for both small and large rm subsamples, while

the coefcient on leverage for the large rm subsample is slightly smaller than that of the small rm

subsample.

To better understand the inuence of rm size on the effect of leverage on operating performance,

we estimate Eq. (1) and (2) for the full sample and each of the six rm size subsamples. Table 5 reports

the estimation results for Eq. (1). In column 1 of Table 5, we show that all of ve coefcients on

the interaction term between leverage and rm size dummy variables are positive and statistically

signicant at the 1% level or better. These results indicate that by using the TA6 sample (i.e., the largest

rm size subsample) as a benchmark, rms in all other rm size groups exhibit the smaller effect of

leverage on ROA, given that the coefcient on LEV is negative and the coefcient on the interaction

term is positive. In columns 27 of Table 5, we show that the negative effect of leverage on performance

is evident in all rm size subsamples. The key nding is that the magnitude of the effect of leverage

on performance varies across different rm size subsamples, suggesting that the effect of leverage on

performance is non-monotonic.

We now examine whether a change in leverage is associated with a change in operating performance. In particular, we test the effect of leverage change on operating performance by estimating

the following regression:

ROAi,t = 0 + 1 AGEi,t1 + 2 SIZEi,t1 + 3 EBITi,t1 + 4 ROAi,t1 + 5 LEVi,t1

+ 6 LEVi,t + vi + i,t ,

(3)

11

Table 3

Panel OLS regressions on the return on assets. This table presents the results of unbalanced panel OLS regressions examining

the effect of rm age, rm size, leverage, interest rate, and GDP growth on the return on assets (ROA). Dependent variable, ROA,

is the EBIT/TA ratio. In columns 24, the sample is divided into three industry groups: agricultural (AGR), manufacturing (MAN),

and services (SER) subsamples. All estimates are based on rm xed-effect model with Whites heteroscedasticity-consistent

cross-section standard errors and covariance. Robust standard errors are reported in parentheses.

Variable

Panel A

Constant

Interest rate

GDP growth

Log age

Log TA

EBIT

LEV

Adjusted R2

F-statistic

Panel B

Constant

Interest rate

GDP growth

Log age

Log TA

EBIT

LEV

LEV2

Adjusted R2

F-statistic

Firms

Observations

*

Full sample

Industry sample

AGR

MAN

SER

0.144***

(0.025)

0.002

(0.002)

0.004***

(0.001)

0.030***

(0.009)

0.235***

(0.043)

0.000***

(0.000)

0.486***

(0.090)

0.138***

(0.013)

0.001

(0.001)

0.003***

(0.001)

0.022***

(0.008)

0.234***

(0.040)

0.000***

(0.000)

0.476***

(0.085)

0.195***

(0.056)

0.009***

(0.001)

0.003***

(0.001)

0.024***

(0.006)

0.199***

(0.036)

0.000***

(0.000)

0.413***

(0.084)

0.131***

(0.022)

0.001

(0.002)

0.004***

(0.001)

0.033***

(0.010)

0.242***

(0.046)

0.000***

(0.000)

0.508***

(0.094)

0.337

2.481***

0.350

2.566***

0.340

2.511***

0.332

2.448***

0.144***

(0.025)

0.002

(0.001)

0.004***

(0.001)

0.030***

(0.009)

0.234***

(0.044)

0.000***

(0.000)

0.485***

(0.099)

0.001

(0.019)

0.138***

(0.013)

0.001

(0.001)

0.003***

(0.001)

0.022***

(0.008)

0.235***

(0.041)

0.000***

(0.000)

0.483***

(0.096)

0.005

(0.031)

0.194***

(0.057)

0.008***

(0.001)

0.004***

(0.001)

0.025***

(0.007)

0.205***

(0.035)

0.000***

(0.000)

0.470***

(0.075)

0.039***

(0.011)

0.144***

(0.025)

0.002

(0.001)

0.004***

(0.001)

0.030***

(0.009)

0.234***

(0.044)

0.000***

(0.000)

0.485***

(0.099)

0.001

(0.019)

0.337

2.481***

170,013

496,430

0.350

2.566***

45,579

132,804

0.340

2.513***

26,825

78,814

0.337

2.481***

170,013

496,430

Signicance at the 5% level.

***

Signicance at the 1% level.

**

where the change in ROA, ROAi,t , is measured as the EBIT/TA ratio for rm i in year t minus the

EBIT/TA ratio in year t 1; the change in leverage, LEVi,t , is measured as the leverage ratio for

rm i in year t minus the leverage ratio in year t 1; and all other variables are dened as before.

As we use the one-period lagged leverage in the model as the explanatory variable, this approach

should account for the possible endogeneity problem related to the linkage between leverage and

operating performance. In Section 3.4, we run several tests to formally address the endogeneity

issue.

If managers do not adjust leverage ratios in preparation for a change in expected future operating

performance, we should not observe the relation between the change in leverage and the change in

12

Table 4

Panel OLS regressions on the return on assets: the interactive effect of leverage and rm size. This table presents the results of

unbalanced panel OLS regressions examining the effect of rm age, rm size, revenue, and leverage on the return on assets (ROA).

Dependent variable, ROA, is the EBIT/TA ratio. A size dummy variable takes a value of one when total asset is larger than the

median value and zero otherwise. All estimates are based on rm xed-effect model with Whites heteroscedasticity-consistent

cross-section standard errors and covariance. Robust standard errors are reported in parentheses.

Variable

Full sample

Small rm sample

(Size dummy = 0)

Large rm sample

(Size dummy = 1)

Constant

0.149***

(0.027)

0.002

(0.002)

0.004***

(0.001)

0.029***

(0.009)

0.240***

(0.045)

0.000***

(0.000)

0.471***

(0.086)

0.040***

(0.012)

0.003

(0.018)

0.002***

(0.001)

0.005***

(0.000)

0.026***

(0.007)

0.247***

(0.050)

0.500***

(0.011)

0.327***

(0.068)

0.019

(0.033)

0.013***

(0.001)

0.004***

(0.001)

0.015***

(0.002)

0.104***

(0.021)

0.000***

(0.000)

0.318***

(0.063)

0.337

2.483***

170,013

496,430

0.585

4.722***

92,145

243,737

0.425

3.010***

93,065

252,693

Interest rate

GDP growth

Log age

Log TA

EBIT

LEV

LEV size dummy

Adjusted R2

F-statistic

Firms

Observations

*

**

Signicance at the 5% level.

***

operating performance, after controlling for other factors that affect operating performance. Results

in Table 6 show that the estimated coefcients on lagged leverage and leverage change are negative

and signicant in the full sample, in all industry subsamples, and in all rm size subsamples. These

ndings indicate that after controlling for past performance (ROA and EBIT) and past leverage, the

change in leverage is negatively associated with the change in ROA.13 More specically, the results

suggest that a hypothetical reduction in leverage results in the improvement in ROA, and that rms

with higher leverage ratios tend to have smaller changes in ROA. Looking at columns 510, we notice

that the magnitude of the impact of the change in leverage on the change in ROA tends to decrease,

as rm size increases.14 For comparison, the coefcient on leverage change for the TA1 subsample

is 0.993 versus 0.389 for the TA6 subsample. The evidence in Table 6 suggests that the effect of

leverage on operating performance is affected by rm size. Overall, our nding that the positive change

in leverage is associated with the negative change in ROA is consistent with previous studies, such as

Giroud et al. (2012), who show that in a sample of Austrian ski hotels, by reducing a debt overhang,

rms are likely to be able to improve protability, but are at odds with Cai and Zhang (2011), who

argue that the increase in nancial leverage is not intended to prepare for deteriorating performance

in the future.

13

To test the robustness of these results, we drop the LEV variable from Eq. (3) and nd that the estimation results are

qualitatively similar to those in Tables 7 and 8.

14

To test the robustness of our results, we estimate Regression (3) again using the level of EBIT as the dependent variable for

the full sample, three industry subsamples, and six rm size subsamples, similar to Tables 7 and 8. We nd that the negative

effect of leverage change on EBIT remains evident in all specications.

13

Table 5

Panel OLS regressions on the return on assets, six rm size subsamples. This table presents the results of unbalanced panel

OLS regressions examining the effect of rm age, rm size, revenue, and leverage on the return on assets (ROA). Dependent

variable, ROA, is the EBIT/TA ratio. The sample is divided into six subsamples, based on total assets at the end of each year: (1)

TA1 (TA < 1 million THB); (2) TA2 (1 million THB TA < 5 million THB); (3) TA3 (5 million THB TA < 10 million THB); (4) TA4 (10

million THB TA < 100 million THB); (5) TA5 (100 million THB TA < 1000 million THB); and (6) TA6 (TA 1000 million THB).

All estimates are based on rm xed-effect model with Whites heteroscedasticity-consistent cross-section standard errors and

covariance. Robust standard errors are reported in parentheses.

Variable

Full sample

Firm size

TA1

Constant

Interest rate

GDP growth

Log age

Log TA

EBIT

LEV

LEV TA1 dummy

LEV TA2 dummy

LEV TA3 dummy

LEV TA4 dummy

LEV TA5 dummy

Adjusted R2

F-statistic

Firms

Observations

*

**

***

TA2

TA3

TA5

TA6

0.039

(0.015)

0.003***

(0.000)

0.002***

(0.000)

0.002***

(0.001)

0.023**

(0.010)

0.021***

(0.000)

0.142***

(0.032)

0.016

(0.024)

0.004***

(0.001)

0.002***

(0.000)

0.011***

(0.002)

0.010**

(0.004)

0.002***

(0.000)

0.107***

(0.020)

0.064*

(0.038)

0.020***

(0.001)

0.003***

(0.000)

0.022***

(0.004)

0.025***

(0.006)

0.000***

(0.000)

0.280***

(0.040)

0.806

11.506***

50,056

126,949

0.853

15.921***

13,609

34,936

0.577

4.623***

2646

7036

TA4

0.147

(0.028)

0.002

(0.001)

0.003***

(0.001)

0.029***

(0.008)

0.236***

(0.046)

0.000***

(0.000)

0.657***

(0.120)

0.103**

(0.044)

0.244***

(0.032)

0.207***

(0.022)

0.142***

(0.014)

0.055***

(0.008)

***

0.395

(0.063)

0.011***

(0.002)

0.006***

(0.000)

0.030***

(0.006)

0.370***

(0.073)

1.878***

(0.038)

0.270***

(0.044)

0.002

(0.014)

0.007***

(0.003)

0.001**

(0.001)

0.018***

(0.004)

0.015

(0.015)

0.385***

(0.003)

0.099***

(0.025)

0.083

(0.013)

0.002

(0.001)

0.001***

(0.000)

0.005***

(0.001)

0.036***

(0.006)

0.135***

(0.002)

0.038***

(0.011)

0.340

2.504***

170,013

496,430

0.713

6.741***

31,685

73,214

0.886

19.898***

76,185

184,379

0.978

90.770***

35,013

69,916

***

***

***

Signicance at the 5% level.

Signicance at the 1% level.

Because a rms performance updates current and potential creditors (e.g., banks) about the quality of its assets, we expect protable rms to be more leveraged. To investigate whether operating

performance affects leverage change, we estimate the following regression:

LEVi,t = 0 + 1 AGEi,t1 + 2 SIZEi,t1 + 3 EBITi,t1 + 4 ROAi,t1 + 5 LEVi,t1 + vi + i,t ,

(4)

where a change in leverage, LEVi,t , is measured as the leverage ratio for rm i in year t minus the

leverage ratio in year t 1; and all other variables are dened as before.15 As we use the one-period

lagged value for all explanatory variables in Eq. (4), possible endogeneity related to the relationship

between performance and leverage change should not pose major concerns.

Table 7 presents the estimation of Eq. (4) using leverage change (LEV) as the dependent variable

for the full sample and the industry subsamples. We nd that the adjusted R2 of all models in Table 7

15

Consistent with Antoniou et al. (2008), we add the one-period lagged leverage term in the model.

14

Variable

Constant

Log agei,t1

Log TAi,t1

EBITi,t1

ROAi,t1

LEVi,t1

LEVi,t1

Adjusted R2

F-statistic

Firms

Observations

*

**

***

Full sample

Industry

Firm size

AGR

MAN

SER

TA1

TA2

TA3

TA4

TA5

TA6

0.283***

(0.009)

0.012***

(0.003)

0.104***

(0.005)

0.000

(0.000)

1.426***

(0.020)

0.237***

(0.018)

0.511***

(0.014)

0.225***

(0.018)

0.018***

(0.006)

0.082***

(0.011)

0.000

(0.000)

1.478***

(0.044)

0.251***

(0.035)

0.471***

(0.026)

0.353***

(0.023)

0.018*

(0.008)

0.107***

(0.011)

0.000

(0.000)

1.387***

(0.060)

0.205***

(0.036)

0.481***

(0.023)

0.287***

(0.012)

0.007*

(0.005)

0.112***

(0.007)

0.000

(0.000)

1.414***

(0.025)

0.240***

(0.024)

0.537***

(0.019)

0.207***

(0.028)

0.022*

(0.011)

0.234***

(0.024)

0.128

(0.102)

1.555***

(0.047)

0.517***

(0.059)

0.993***

(0.047)

0.188***

(0.007)

0.007**

(0.003)

0.152***

(0.008)

0.028*

(0.015)

1.199***

(0.047)

0.147***

(0.019)

0.491***

(0.014)

0.303***

(0.017)

0.002

(0.006)

0.083***

(0.013)

0.049***

(0.016)

0.993***

(0.061)

0.220***

(0.038)

0.479***

(0.026)

0.195***

(0.015)

0.039***

(0.004)

0.044***

(0.005)

0.001***

(0.000)

1.337***

(0.066)

0.167***

(0.023)

0.337***

(0.017)

0.173***

(0.034)

0.060***

(0.007)

0.029***

(0.007)

0.000

(0.000)

1.378***

(0.076)

0.195***

(0.044)

0.343***

(0.034)

0.087

(0.067)

0.071***

(0.013)

0.006

(0.008)

0.000

(0.000)

1.225***

(0.127)

0.275***

(0.072)

0.389***

(0.058)

0.619

4.124***

170,013

326,417

0.620

4.118***

45,579

87,225

0.656

4.691***

26,825

51,989

0.613

4.043***

97,607

187,199

0.735

5.663***

27,364

46,071

0.794

7.664***

69,237

119,713

0.825

8.380***

29,921

46,959

0.736

6.027***

47,242

85,354

0.668

4.665***

12,927

23,570

0.657

4.555***

2551

4750

Signicance at the 5% level.

Signicance at the 1% level.

Table 6

Panel OLS regressions on the change in return on assets. This table presents the results of unbalanced panel OLS regressions examining the effect of lagged rm age, lagged rm size,

lagged revenue, lagged leverage, and leverage change on the change in ROA. Dependent variable is ROA, which is the rst difference in the EBIT/TA ratios. In columns 24, the sample is

divided into three industry groups: agricultural (AGR), manufacturing (MAN), and services (SER) subsamples. In columns 510, the sample is divided into six subsamples based on total

assets at the end of each year: (1) TA1 (TA < 1 million THB); (2) TA2 (1 million THB TA < 5 million THB); (3) TA3 (5 million THB TA < 10 million THB); (4) TA4 (10 million THB TA < 100

million THB); (5) TA5 (100 million THB TA < 1000 million THB); and (6) TA6 (TA 1000 million THB). All estimates are based on rm xed-effect model with panel corrected standard

errors (PCSE). Robust standard errors are reported in parentheses.

Variable

Full sample

AGR

MAN

SER

TA1

TA2

TA3

TA4

TA5

TA6

Constant

0.422***

(0.004)

0.026***

(0.002)

0.040***

(0.002)

0.000

(0.000)

0.016***

(0.004)

1.019***

(0.006)

0.460***

(0.008)

0.031***

(0.004)

0.052***

(0.004)

0.000

(0.000)

0.019**

(0.007)

1.015***

(0.011)

0.494***

(0.014)

0.033***

(0.005)

0.046***

(0.006)

0.000

(0.000)

0.007

(0.009)

0.989***

(0.013)

0.382

(0.005)

0.022***

(0.003)

0.033***

(0.003)

0.000

(0.000)

0.018***

(0.005)

1.028***

(0.008)

0.108***

(0.013)

0.035***

(0.005)

0.060***

(0.009)

0.004

(0.005)

0.010

(0.008)

0.999***

(0.020)

0.286***

(0.005)

0.013***

(0.003)

0.054***

(0.004)

0.017***

(0.006)

0.025**

(0.011)

0.972***

(0.010)

0.523***

(0.012)

0.017***

(0.006)

0.045***

(0.006)

0.015***

(0.005)

0.070***

(0.033)

0.956***

(0.013)

0.661***

(0.010)

0.004

(0.004)

0.027***

(0.003)

0.000

(0.000)

0.017**

(0.008)

0.994***

(0.008)

0.749***

(0.028)

0.027***

(0.008)

0.013**

(0.006)

0.000

(0.000)

0.017

(0.036)

1.027***

(0.016)

0.757***

(0.069)

0.052***

(0.016)

0.008

(0.008)

0.000

(0.000)

0.037

(0.029)

1.032***

(0.027)

0.530

3.164***

170,013

326,417

0.539

3.240***

45,579

87,225

0.539

3.267***

26,825

51,989

0.523

3.104***

97,607

187,199

0.545

3.018***

27,364

46,071

0.653

4.246***

69,237

119,713

0.776

6.452***

29,921

46,959

0.670

4.665***

47,242

85,354

0.640

4.238***

12,927

23,570

0.580

3.571***

2551

4750

Log agei,t1

Log TAi,t1

EBITi,t1

ROAi,t1

LEVi,t1

Adjusted R2

F-statistic

Firms

Observations

*

Industry

Firm size

Table 7

Panel OLS regressions on the change in leverage, industry subsamples. This table presents the results of unbalanced panel OLS regressions examining the effect of lagged rm age, lagged

rm size, lagged revenue, lagged EBIT, lagged leverage and lagged ROA on leverage change. Dependent variable, LEV, is leverage change, which is the rst difference of leverage ratios.

In columns 24, the sample is divided into three industry groups: agricultural (AGR), manufacturing (MAN), and services (SER) subsamples. In columns 510, the sample is divided into

six subsamples based on total assets at the end of each year: (1) TA1 (TA < 1 million THB); (2) TA2 (1 million THB TA < 5 million THB); (3) TA3 (5 million THB TA < 10 million THB); (4)

TA4 (10 million THB TA < 100 million THB); (5) TA5 (100 million THB TA < 1000 million THB); and (6) TA6 (TA 1000 million THB). All estimates are based on rm xed-effect model

with panel corrected standard errors (PCSE). Robust standard errors are reported in parentheses.

**

***

Signicance at the 1% level.

15

16

is between 0.52 and 0.54 are largely consistent with prior studies, such as Bris et al. (2004). We show

that the estimated coefcient on the one-period lagged ROA is negative and signicant in the full

sample, after controlling for the one-period lagged rm size, the one-period lagged leverage, and the

one-period lagged EBIT. The negative effect of the lagged ROA on leverage change is also evident in two

industry subsamples (the agricultural and service subsamples). The negative coefcient on past ROA

indicates that more protable rms tend to have a smaller change in leverage. The observed negative

relation between past ROA and leverage change is in line with Rajan and Zingales (1995), Baker and

Wurgler (2002), Antoniou et al. (2008), Bris et al. (2004), and Xu (2012), who nd that protability

has a negative effect on leverage.

Results in columns 510 of Table 7 show that the linkage between past operating performance

and changes in leverage varies across rm size subsamples. That is, the negative effect of the lagged

ROA on a change in leverage is not symmetric across rm size. More importantly, for very small

(the TA1 subsample) and very large rms (the TA5 and TA6 subsamples), lagged ROA is not associated with leverage change, whereas for medium-sized rms (rms in the TA2, TA3, and TA4

subsamples), the relation between lagged ROA and leverage change exists. A likely explanation for

these results is that for very small rms, past operating performance does not necessarily induce

the change in leverage because, for these rms, loan supply constraints from banks might limit

the extent to which they can invest in new projects. Thus, past performance may not induce a

change in leverage in the short-run for the very small rms. Similarly, for very large rms, past

performance appears to be an insufcient condition for altering leverage because their leverage

may be at the optimal level (or at the target level). This nding is inconsistent with Antoniou

et al. (2008) and Bris et al. (2004), who report the positive effect of rm size on leverage. Furthermore, results in Table 7 indicate that the one-period lagged leverage has a negative effect on

leverage change, which suggests that highly leveraged rms tend to exhibit a smaller change in

leverage.

As the subgroup of very large rms has a relatively high leverage ratio (the median value of 0.56,

see Panel F of Table 2), we argue that the absence of an observed linkage between lagged ROA and

leverage changes for the very large rms may be due to the fact that leverage may have reached the

upper bound. Therefore, lagged ROA has no effect on leverage change for the very large rms. The fact

that lagged ROA has a positive effect on leverage change for the TA2 and TA3 subsamples, but it has

a negative effect on leverage change for the TA4 subsample further suggests that the direction of the

effect of performance on leverage depends on rm size.

The positive relation between lagged ROA and leverage change for the TA2 and TA3 subsamples

may be an indication that information asymmetry between rms and banks reduces as banks better

observe asset quality or investment opportunities for these rms and thus are more willing to provide

additional loans to them (the median value of the leverage ratio is 0.12 and 0.37 for the TA2 and TA3

subsamples, see Panels B and C of Table 2). Another likely explanation is that smaller rms (i.e., the

TA1 subsample) might be nancially conservative and thus do not take on additional debt although

they might have unused debt capacity. When a rm grows to a certain point, its leverage may reach

a limit allowed for by banks (the median value of leverage ratio is 0.60, see Panel D of Table 2), the

relation between operating performance and leverage becomes negative (i.e., for the TA4 subsample). Overall, our ndings suggest that rm size inuences the effect of operating performance on

leverage.

3.4. Robustness tests

In this subsection, we perform three sets of robustness tests to examine whether our basic ndings

are robust after addressing four potential issues: nancial rms, outliers, endogeneity, and estimation

stability (e.g., consistency across time) issues.

First, because our sample includes nancial rms (TSIC code: J65-J67), it is possible that our results

are biased due to the inclusion of nancial rms in our analyses. To address this issue and make our

results comparable with prior studies, we drop 3655 nancial rms from the sample and repeat the

analysis of Tables 37. In our untabulated results, we nd that removing nancial rms from the

sample does not alter our main results.

17

Second, we address an issue of whether our results are driven by observations with very high

leverage ratios (i.e., 1 < leverage ratios < 2). To address this issue, we exclude all observations with

leverage larger than one from our sample (6.78% of the sample), which in essence removes rms with

debt overhang problems. Our restricted sample size reduces to 462,719 rm-year observations. As

the restricted sample is still very large and the proportion of excluded observations is small, we do

not recode all leverage larger than one to one, as in Faulkender and Petersen (2006). Repeating the

analysis of Tables 37, we nd that the pattern of our ndings remains unchanged,16 conrming the

robustness of our results to the exclusion of the outliers (i.e., highly leveraged rms).

Third, there is a question of whether our observed negative relation between leverage and ROA is

driven by an endogeneity problem because, as suggested, for example, by Aivazian et al. (2005) and

Firth et al. (2008), rms, rather than external factors, determine leverage. Firms may increase their

leverage in anticipation of weaker operating performance ex ante so that we observe the negative

relation between leverage and ROA ex post. However, we argue that during nancial crises, such as the

global nancial crisis, the fact that external factors (i.e., the liquidity shocks in the nancial markets),

rather than rms, are likely to play a major role in determining leverage should mitigate endogeneity

concerns in our regressions. During the global nancial crisis, the upper bound of leverage ratios is

likely to be exogenously given (or constrained) by nancial institutions (e.g., through a forced reduction

in leverage ratios due to renancing (rollover) failures) rather than endogenously determined by rms.

To address the endogeneity issue, we use the instrumental variable (IV) and two-stage least square

(2SLS) estimation approach. Following Aivazian et al. (2005), Firth et al. (2008) and Xu (2012), we

use a rms asset tangibility, dened as the ratio of net xed assets to total assets, as an instrumental

variable. We argue that asset tangibility is viable choice as an instrumental variable because the level

of asset tangibility is generally correlated with nancial leverage (because rms use tangible assets

as a collateral for a bank loan) but is not highly correlated with operating performance. Our empirical

ndings provide support to these arguments. For instance, the correlation coefcient between leverage

and asset tangibility in 2007, 2008, and 2009 is 0.28, 0.29, and 0.29 (p-value <0.01), whereas the

correlation coefcient between asset tangibility and ROA in 2007, 2008, and 2009 is 0.07, 0.00,

and 0.00 and is statistically insignicant (p-value <0.01 for 2007; p-value >0.10 for 2008 and 2009),

suggesting that asset tangibility is an appropriate instrumental variable for leverage. In the rst-stage

regression, we regress leverage on asset tangibility and ve control variables (i.e., interest rates, GDP

growth, rm age, rm size, and EBIT) as follows:

LEVi,t = 0 + 1 INTt + 2 GDPt + 3 AGEi,t + 4 SIZEi,t + 5 EBITi,t + 6 TANGIBILITYi,t + vi + i,t ,

(5)

where TANGIBILITYi,t is the ratio of net xed assets to total assets for rm i in year t. We use the

predicted values of leverage as our generated instrument for leverage in the second-stage regressions.

We estimate our performance models as follows:

ROAi,t = 0 + 1 INTt + 2 GDPt + 3 AGEi,t + 4 SIZEi,t + 5 EBITi,t + 6 IVLEVi,t + vi + i,t ,

(6)

where IVLEVi,t is the predicted value for LEVi,t for rm i in year t from the rst-stage Eq. (5). We estimate

Eqs. (5) and (6) using the rm xed effect model with Whites heteroscedasticity-consistent crosssection standard errors and covariance. In the rst-stage regression, the coefcient on asset tangibility,

TANGIBILITY, is 0.027 and signicant at the 1% level. In the second-stage regression, the coefcient on

leverage, IVLEV, is negative and statistically signicant at the 1% level in the full sample (see column 2

of Table 8). In addition, we nd that the coefcient on leverage is negative and statistically signicant

in ve of the six rm size subsamples (see columns 38 of Table 8), and that the size of the coefcients

appears to decrease as rm size increases (i.e., from 0.87 for the TA1 subsample to 0.17 for the TA5

subsample).

As there is still a concern that the IV-2SLS estimations may not fully address the endogeneity

issue, we repeat the analysis of Table 8 by estimating Eqs. (5) and (6) using the two-step system

16

For brevity, these estimates are not reported but are available from the authors upon request.

18

Table 8

Robustness Test: The effect of leverage on the return on assets, the instrumental variable and two-stage least square estimation.

This table presents the results of instrumental variable (IV) two-stage least square (2SLS) regressions. Column 1 presents rststage regression results for the full sample, and columns 28 present second-stage regression results using asset tangibility,

calculated as the net xed assets divided by total assets, as an instrument, for the full sample and six rm size subsamples. In

the rst-stage regression, leverage (LEV) is the dependent variable. In the second-stage regression, the dependent variable is

ROA. IVLEV is the predicted value of leverage from the rst-stage regression. All estimates are based on rm xed-effect model

with Whites heteroscedasticity-consistent standard errors and covariance. The full sample is divided into subsamples based on

total assets at the end of each year: (1) TA1 (TA < 1 million THB); (2) TA2 (1 million THB TA < 5 million THB); (3) TA3 (5 million

THB TA < 10 million THB); (4) TA4 (10 million THB TA < 100 million THB); (5) TA5 (100 million THB TA < 1000 million THB);

and (6) TA6 (TA 1000 million THB). Robust standard errors are reported in parentheses.

Variable

First-stage

regression

Second-stage regression

Full sample

Full sample

Firm size

TA1

Constant

Interest rate

GDP growth

Log age

Log TA

EBIT

TANGIBILITY

***

0.189

(0.015)

0.009***

(0.000)

0.006***

(0.000)

0.024***

(0.004)

0.109***

(0.015)

0.000***

(0.000)

0.027***

(0.004)

IVLEV

Adjusted R2

F-statistic

Firms

Observations

*

**

***

0.880

22.446***

170,013

496,422

***

TA3

TA2

***

**

TA5

TA4

***

***

TA6

**

0.374

(0.021)

0.025***

(0.001)

0.018***

(0.001)

0.036***

(0.011)

0.519***

(0.035)

0.000**

(0.000)

0.571

(0.088)

0.012**

(0.005)

0.009***

(0.002)

0.019

(0.012)

0.471***

(0.062)

1.941***

(0.042)

0.048

(0.019)

0.005*

(0.003)

0.003***

(0.001)

0.01***

(0.002)

0.033**

(0.015)

0.394***

(0.003)

0.113

(0.003)

0.001

(0.001)

0.002***

(0.000)

0.000

(0.001)

0.025***

(0.008)

0.137***

(0.002)

0.144

(0.018)

0.007***

(0.001)

0.004***

(0.000)

0.011***

(0.001)

0.06***

(0.003)

0.022***

(0.000)

0.036

(0.016)

0.006***

(0.002)

0.002***

(0.000)

0.012***

(0.000)

0.015*

(0.008)

0.002***

(0.000)

0.172

(0.121)

0.026***

(0.004)

0.003

(0.002)

0.018

(0.013)

0.008

(0.044)

0.000***

(0.000)

3.108***

(0.112)

0.872***

(0.325)

0.335***

(0.070)

0.167***

(0.047)

0.546***

(0.021)

0.17**

(0.082)

0.133

(0.432)

0.276

2.114***

170,013

496,422

0.706

6.550***

31,685

73,213

0.884

19.406***

76,185

184,375

0.978

89.170***

35,013

69,915

0.796

10.866***

50,056

126,948

0.848

15.283***

13,609

34,935

0.542

4.135***

2646

7036

Signicance at the 5% level.

Signicance at the 1% level.

GMM specication. The untabulated results (available upon request) remain qualitatively similar as

the estimated coefcient on leverage is negative and statistically signicant for the full sample and

for some rm size subsamples.

Thus far, we examine the relation between leverage and performance using a panel regression

approach; we now address our main research question again using the year-by-year regression

approach to examine whether the pattern of our results remains relatively stable across years. We

estimate year-by-year regressions of a simplied version of Eqs. (5) and (6). Table 9 reports the yearby-year IV-2SLS estimation results. For brevity, we do not report the rst-stage regression results

for the rm size subsamples. It is important to note that the estimated coefcient on TANGIBILITY is

positive and is statistically signicant at the 5% level or better for all rm size subsamples and for all

years (with one exception: the TA6 subsample for 2008). As can be seen in Panel AC of Table 9, the

pattern of the effect of leverage on performance is largely identical across years. That is, for small rms

(i.e., TA1, TA2, and TA3 subsamples), the effect of leverage on performance is positive, but the effect is

negative for large rms (i.e., TA4, TA5, and TA6 subsamples). In unreported results, we nd evidence

of the non-linear effect of leverage on performance for all rm size subsamples and all years (i.e., the

estimated coefcient on the squared term of TANGIBILITY is statistically signicant at the 5% level or

better for all but for TA1 and TA6 subsamples in 2009), thereby providing further evidence that the

non-linear relation between leverage and performance is evident across years.

Table 9

Robustness test: the effect of leverage on the return on assets, the instrumental variable and two-stage least square estimation. This table presents the year-by-year regression results of

instrumental variable (IV) two-stage least square (2SLS) regressions. Column 1 presents rst-stage regression results for the full sample, and columns 28 present second-stage regression

results using asset tangibility, calculated as the net xed assets divided by total assets, as an instrument, for the full sample and six rm size subsamples. In the rst-stage regression, leverage

(LEV) is the dependent variable. In the second-stage regression, the dependent variable is ROA. IVLEV is the predicted value of leverage from the rst-stage regression. The full sample is

divided into subsamples based on total assets at the end of each year: (1) TA1 (TA < 1 million THB); (2) TA2 (1 million THB TA < 5 million THB); (3) TA3 (5 million THB TA < 10 million

THB); (4) TA4 (10 million THB TA < 100 million THB); (5) TA5 (100 million THB TA < 1000 million THB); and (6) TA6 (TA 1000 million THB). Whites heteroscedasticity-consistent

standard errors are reported in parentheses.

Variable

Second-stage regression

Full sample

Full sample

0.350***

Constant

(0.002)

Log agei,2007

0.022***

(0.001)

0.080***

Log TAi,2007

(0.001)

EBITi,2007

0.000*

(0.000)

0.000*

TANGIBILITYi,2007

(0.000)

IVLEVi,2007

Firm size

TA1

TA2

TA3

TA4

TA5

TA6

0.250***

(0.027)

0.017***

(0.002)

0.047***

(0.006)

0.000***

(0.000)

0.009

(0.005)

0.005***

(0.002)

0.064***

(0.002)

1.803***

(0.009)

0.016***

(0.002)

0.001

(0.000)

0.033***

(0.001)

0.386***

(0.001)

0.081***

(0.002)

0.000

(0.000)

0.043***

(0.001)

0.139***

(0.000)

0.093***

(0.010)

0.001

(0.001)

0.019***

(0.002)

0.021***

(0.000)

0.232***

(0.055)

0.003

(0.003)

0.024***

(0.001)

0.002***

(0.000)

1.435***

(0.170)

0.065***

(0.010)

0.016***

(0.004)

0.000

(0.000)

0.056**

(0.022)

Yes

0.024***

(0.006)

Yes

0.007**

(0.003)

Yes

0.029

(0.018)

Yes

0.128*

(0.066)

Yes

1.977***

(0.249)

Yes

Industry dummies

Yes

0.700***

(0.076)

Yes

R2

Adj. R2

F-statistic

Observations

0.156

0.156

4825.149***

156,086

0.008

0.008

214.514***

156,086

0.682

0.682

8244.145***

23,049

0.801

0.801

38,710.919***

57,764

0.959

0.959

86,085.787***

21,864

0.654

0.654

12,641.656***

40,105

0.737

0.736

5148.311***

11,052

0.143

0.141

62.600***

2252

0.144***

(0.023)

0.007***

(0.002)

0.023***

(0.005)

0.000***

(0.000)

0.038***

(0.007)

0.009***

(0.002)

0.127***

(0.003)

1.911***

(0.013)

0.024***

(0.001)

0.002***

(0.000)

0.033***

(0.001)

0.377***

(0.001)

0.080***

(0.002)

0.000

(0.000)

0.042***

(0.001)

0.138***

(0.000)

0.088***

(0.009)

0.002

(0.001)

0.017***

(0.001)

0.021***

(0.000)

0.208***

(0.060)

0.003

(0.004)

0.021***

(0.001)

0.002***

(0.000)

3.172***

(0.685)

0.201***

(0.047)

0.111***

(0.026)

0.000***

(0.000)

19

0.337***

Constant

(0.002)

Log agei,2008

0.025***

(0.001)

0.082***

Log TAi,2008

(0.001)

EBITi,2008

0.000**

(0.000)

0.000***

TANGIBILITYi,2008

(0.000)

First-stage regression

20

Table 9 (Continued)

Variable

First-stage regression

Second-stage regression

Full sample

Full sample

TA1

TA2

TA3

TA4

TA5

TA6

Yes

0.433***

(0.067)

Yes

0.029

(0.029)

Yes

0.017***

(0.005)

Yes

0.006**

(0.003)

Yes

0.031**

(0.015)

Yes

0.121*

(0.073)

Yes

5.183***

(1.143)

Yes

0.156

0.156

5224.438***

170,013

0.008

0.008

227.385***

170,013

0.601

0.601

6422.056***

25,579

0.815

0.815

46,553.122***

63,573

0.960

0.960

95,708.128***

23,710

0.627

0.627

12,024.752***

42,948

0.712

0.712

4867.181***

11,830

0.099

0.096

43.217***

2373

2.257***

(0.241)

0.203***

(0.022)

0.431***

(0.049)

0.000***

(0.000)

0.048***

(0.008)

0.013***

(0.003)

0.170***

(0.004)

1.883***

(0.015)

0.014***

(0.002)

0.000

(0.000)

0.022***

(0.001)

0.365***

(0.001)

0.085***

(0.002)

0.002***

(0.000)

0.040***

(0.001)

0.134***

(0.000)

0.124***

(0.007)

0.003**

(0.001)

0.026***

(0.001)

0.021***

(0.000)

0.263***

(0.047)

0.005

(0.004)

0.032***

(0.001)

0.002***

(0.000)

1.037***

(0.105)

0.066***

(0.009)

0.024***

(0.004)

0.000

(0.000)

Yes

6.023***

(0.647)

Yes

0.046

(0.028)

Yes

0.004

(0.005)

Yes

0.001

(0.002)

Yes

0.030**

(0.012)

Yes

0.096*

(0.053)

Yes

1.465***

(0.158)

Yes

0.121

0.121

3906.100***

170,013

0.021

0.021

612.534***

170,013

0.608

0.608

6311.575***

24,460

0.822

0.822

48,423.208***

62,873

0.964

0.964

108,321.235***

24,333

0.658

0.658

14,056.743***

43,882

0.670

0.670

4084.257***

12,054

0.097

0.095

42.964***

2411

Industry dummies

2

R

Adj. R2

F-statistic

Observations

0.372***

Constant

(0.003)

Log agei,2009

0.034***

(0.001)

0.075***

Log TAi,2009

(0.001)

EBITi,2009

0.000***

(0.000)

0.000

TANGIBILITYi,2009

(0.000)

IVLEVi,2009

Industry dummies

2

R

Adj. R2

F-statistic

Observations

*

**

***

Signicance at the 5% level.

Signicance at the 1% level.

IVLEVi,2008

Firm size

21

4. The effect of the global nancial crisis on leverage and operating performance

The previous section shows that the relation between leverage and performance is inuenced

by rm size in panel and year-by-year cross-sectional regression analyses. The goal of this section

is to expand our understanding of whether a global liquidity shock indeed has a negative effect on

leverage of rms, regardless of size, and to investigate whether our main results are robust to different

specications. Similar to Section 3.4, we use a restricted sample in which rms with leverage in 2007

larger than one are excluded from the sample to reduce noise in the data set; we have 146,599 rms

in the restricted sample for further analyses in this section.

The global nancial crisis of 20072009 provides us a unique and natural setting to examine the

linkage between leverage and performance for two important reasons. First, the nancial crisis had

led to global nancial constraints, thereby limiting the extent to which new external nancing could

be obtained (e.g., due to credit contractions for rms). Small and less credit worthy rms may be cut

off from credit markets altogether in the aftermath of the nancial crisis. For example, Campello et al.

(2010), Kannan (2012), and Akbar et al. (2013) provide evidence of the effect of credit conditions on

investment and growth. Second, the negative effect of the nancial crisis on economic growth, as

shown by Berkmen et al. (2012) and Sentance et al. (2012), indicates that investment opportunities at

the aggregated level during the crisis had been limited.17 Given a combination of limited investment

opportunities (e.g., due to recessions in the US and European countries) and nancial constraints (e.g.,

due to loan supply shocks) during the global nancial crisis, it is reasonable to assume that there is

no new investment for most rms; therefore, a change in operating performance is related to assetin-place rather than to new assets. Without new investment, any change in leverage should reect a

rms capital structure adjustments conditional on its asset-in-place and other internal and external

factors.

If the global liquidity shocks18 effectively force rms to deleverage, we should observe a fall in leverage during the nancial crisis. That is, in our sample, the leverage ratio in 2009 should be signicantly

lower than the leverage ratio in 2007. We nd that the leverage change (leverage in 2009 leverage

in 2007) is 0.007 and statistically signicant at the 1% level. For the TA1 and TA2 subsamples (we use

total assets in 2007 to divide our sample into six rm size subsamples as dened earlier.), the leverage change is positive and statistically signicant at the 1% level, but for the TA3, TA4, TA5, and TA6

subsamples, the leverage change is negative and statistically signicant at the 1% level. We nd that

about 55.02% of rms in the sample have a lower level of leverage in 2009 than in 2007 (the leverage

change is negative), and that about 44.78% of rms in the sample have a higher level of leverage in

2009 than in 2007 (the leverage change is positive).

Fig. 3 illustrates the average change in leverage (leverage in 2009 leverage in 2007) and average

percentage changes in total assets (100 (total assets in 2009 total assets in 2007)/total assets in

2007) during the global nancial crisis of 20072009 for the full sample and for the rm size subsamples. In Panel A of Fig. 3, we observe that rms with total assets less than 5 million THB in 2007 (the

TA1 and TA2 subsamples) on average exhibit a large increase in rm size over the three-year period

and a small increase in leverage. On the other hand, medium-sized and large rms (the TA3, TA4,

TA5 and TA6 subsamples) have a moderate increase in total assets and a small reduction in leverage.

One likely explanation is that given that the change in leverage is small relative to the change in total

assets, small rms primarily nance their expansion through new equity nancing (e.g., new shares

that might be bought by owners, family relatives and friends) during the crisis as in tranquil periods.

Larger rms appear to nance the increase in assets through retained earnings during the crisis. In

Panel B of Fig. 3, we exclude nancial rms from the sample to address the concern of whether the

inclusion of nancial rms in the sample drives leverage upwards. While we notice that in comparison

with Panel A, leverage ratios in Panel B are smaller, the pattern of the relation between leverage and

17

In the context of Thailand, we argue that the business conditions had been poor during 20072009, based on the business

sentiment index (BSI) published by the Bank of Thailand. In particular, the indexes were smaller than 50 throughout the period,

indicating that business sentiment had worsened.

18

See, e.g., Covitz et al. (2013) for a discussion on a collapse of the asset-backed commercial paper market, Hristov et al. (2012)

for a discussion on loan supply shocks in the EU area, and Dick-Nielsen et al. (2012) for a discussion of the liquidity of bonds.

22

Fig. 3. Leverage change and percentage change in total assets during the global nancial crisis of 20072009. These gures

present the mean of the change in leverage (i.e., leverage in 2009leverage in 2007) and the mean of the change in total assets in

percent (i.e., 100 (total assets in 2009 total assets in 2007)/total assets in 2007) during the global nancial crisis of 20072009

for rms with leverage in 2007 not greater than one for the full sample and six rm size subsamples: TA1 (TA < 1 million THB),

TA2 (1 million THB TA < 5 million THB), TA3 (5 million THB TA < 10 million THB), TA4 (10 million THB TA < 100 million

THB), TA5 (100 million THB TA < 1000 million THB), TA6 (TA 1000 million THB), respectively. In Panel A, nancial rms are

included in the sample. In Panel B, nancial rms are excluded from the sample.

rm size in Fig. 3 is similar to that of Fig. 1. Our results raise a fundamental question of whether small

rms are more nancially constrained than large rms as commonly assumed in the literature. Our

ndings certainly indicate otherwise.

If the global liquidity crisis results in renancing failures of rms (i.e., rms are unable to

rollover maturing debts), which in turn force rms to liquidate assets (e.g., after exhausting liquidity provisions19 ), we should observe a fall in leverage and a fall in total assets simultaneously for

most rms in the sample. We nd that 24.41% of rms have a negative change in leverage (i.e., leverage in 2009 leverage in 2007 < 0) and a fall in total assets (i.e., total assets in 2009 total assets in

2007 < 0).20 This nding suggests that these rms deleverage and downsize during the global nancial

crisis.

If rms are forced to downsize due to economic downturns but still manage to rollover their debts

(assuming that the amount of total liabilities remains constant; i.e., total liabilities in 2009 total

liabilities in 2007 0), a change in total assets should be negative (total assets in 2009 total assets

19

See Campello et al. (2011) for a discussion on liquidity management during the 20082009 nancial crisis.

For the sample of rms with the fall in leverage and the fall in total assets, the mean leverage change is 0.16 and the mean

change in total assets (scaled by total assets in 2007) is 20.55%.

20

23

in 2007 < 0) and a change in leverage should be positive (leverage in 2009 leverage in 2007 > 0). We

nd that about 11.27% of the rms have, a non-negative change in total liabilities, a positive change

in leverage and a negative change in total assets, suggesting that these rms are likely to have access

to bank loans during the crisis.

Furthermore, we observe that 27.70% of the rms have a positive change in leverage (leverage

in 2009 leverage in 2007 > 0) and an increase in total assets (total assets in 2009 total assets in

2007 > 0),21 implying that during the nancial crisis, these rms manage to issue new debts to nance

the expansion in total assets.

Overall, these observations suggest that the negative effect of the global nancial crisis of

20072009 on rms in Thailand does not seem to be as severe as in the US and in Europe, at least

in terms of shocks to the balance sheet of rms. A question arises whether Thai rms actually invested

in long-term assets (capital expenditure) during the nancial crisis, as indicated by the increase in

total assets. To answer this question, we calculate the growth in xed assets during 20072009 ([xed

assets in 2009 xed assets in 2007]/total assets in 2007). In untabulated results, the growth in xed

assets for the full sample is 22.38% and is statistically signicant at the 1% level, suggesting that during

the nancial crisis period, the increase in xed assets accounts for a small proportion of the increase in

total assets. For medium- and large-sized rms (from the TA3 subsample to the TA6 subsample), the

increase in xed assets is positive and statistically signicant at the 1% level but small in magnitude,

ranging from 2.24% for the TA6 subsample to 5.76% for the TA3 subsample. Because the growth rate

in xed assets appears to be positively skewed, we test whether the median of the growth rate is

different from zero using the Wilcoxon signed rank test. We nd that the median for the full sample

is 0.03% and statistically signicant at the 1% level, and that the median for the six rm size subsamples is either zero (for the TA1 and TA2 subsamples) or negative (for the TA3, TA4, TA5, and TA6

subsamples) and statistically signicant at the 1% level. These ndings support the notion that rms

did not make substantial investments in long-term assets during the global nancial crisis and suggest that the increase in rm size was due to an expansion in current assets, rather than an increase

in long-term investments.

We test whether the change in leverage during 20072009 (leverage in 2009 leverage in 2007)

is related to the change in operating performance during 20072009 (ROA in 2009 ROA in 2007) by

estimating the following regression22 :

ROAi,20092007 = 0 + 1 AGEi,2007 + 2 SIZEi,2007 + 3 EBITi,2007 + 4 ROAi,2007 + 5 LEVi,2007

+ 6 LEVi,20092007 + 7 INDi,2007 + i,t ,

(7)

where the change in ROA, ROAi,20092007 , is calculated as the EBIT/TA ratio for rm i in 2009 minus

the EBIT/TA ratio in 2007 (ROAi,2009 ROAi,2007 ); the change in leverage, LEVi,20092007 , is calculated

as the leverage ratio for rm i in 2009 minus the leverage ratio in 2007 (LEVi,2009 LEVi,2007 ); and all

other variables are dened as before and two-period lagged (i.e., values in 2007).

Table 10 reports the cross-sectional regression results for Eq. (7) for the full sample and six rm

size subsamples. We nd that after controlling for past leverage, operating prot (EBIT), ROA and size,

the change in leverage has a negative effect on the change in ROA, and that this pattern of results

remains evident in all six rm size subsamples. Conditional on the leverage level in 2007, rms with

a larger change in leverage during 20072009 tend to experience a smaller change in ROA over the

same period. To test the asymmetric effect of the change in leverage on the change in ROA with respect

to the direction of the change in leverage, we add an interaction term involving leverage change and

a binary variable, which takes a value of one when leverage change is positive, and zero otherwise, in

Eq. (7). We nd that the coefcient on the interaction term is negative and signicant at the 1% level

21

For the sample of rms with the hike in leverage and the hike in total assets, the mean leverage change is 0.17 and the mean

change in total assets (scaled by total assets in 2007) is 280.03%.

22

Our approach is similar to Duchin et al. (2010a), who suggest that reducing the sample size to fewer observations has an

advantage of mitigating a potentially serial correlation problem that occurs when estimating a regression using long time-series

in differences-in-differences. However, the severe serial correlation problem is unlikely to be a major concern in our study, given

that we have a short time-series (i.e., three observations per rm) sample.

24

Table 10

OLS regressions on the change in ROA over the 20072009 global nancial crisis. This table presents the results of OLS regressions

of the change in ROA on rm age, rm size, EBIT, EBIT/TA, leverage, and leverage during the 20072009 global nancial crisis.

Dependent variable is the change in ROA (ROA), which is the EBIT/TA ratio in 2009 the EBIT/TA ratio in 2007. Leverage

change (LEV) is the leverage ratio in 2009 minus the leverage ratio in 2007, and all other explanatory variables are in 2007.

Firms with leverage in 2007 larger than one are excluded from the sample. The sample is divided into six subsamples based on

total assets at the end of each year: (1) TA1 (TA < 1 million THB); (2) TA2 (1 million THB TA < 5 million THB); (3) TA3 (5 million

THB TA < 10 million THB); (4) TA4 (10 million THB TA < 100 million THB); (5) TA5 (100 million THB TA < 1000 million THB);

and (6) TA6 (TA 1000 million THB). Whites heteroscedasticity-consistent standard errors are reported in parentheses.

Variable

Full sample

Firm size

TA1

TA3

TA4

TA5

TA6

Industry dummies

0.038

(0.006)

0.011***

(0.002)

0.008**

(0.004)

0.150***

(0.043)

0.947***

(0.021)

0.068***

(0.012)

0.490***

(0.022)

Yes

0.024

(0.004)

0.000

(0.001)

0.007**

(0.003)

0.035***

(0.013)

0.882***

(0.036)

0.066***

(0.005)

0.397***

(0.015)

Yes

0.006

(0.042)

0.007***

(0.002)

0.003

(0.024)

0.060

(0.064)

1.033**

(0.412)

0.032***

(0.012)

0.290***

(0.020)

Yes

0.013

(0.009)

0.006***

(0.001)

0.005**

(0.003)

0.001

(0.002)

0.693***

(0.057)

0.015***

(0.004)

0.236***

(0.015)

Yes

0.035

(0.026)

0.008***

(0.002)

0.009**

(0.004)

0.000

(0.000)

0.640***

(0.064)

0.012

(0.012)

0.197***

(0.025)

Yes

0.029

(0.044)

0.001

(0.004)

0.012***

(0.003)

0.000

(0.000)

0.712***

(0.129)

0.053**

(0.027)

0.344***

(0.065)

Yes

Adjusted R2

F-statistic

Observations

0.321

8644.909***

146,599

0.439

2173.294***

22,204

0.301

2976.649***

55,279

0.156

4,69.949***

20,370

0.160

871.561***

36,471

0.217

351.250***

10,129

0.279

104.924

2146

Log agei,2007

Log TAi,2007

EBITi,2007

ROAi,2007

LEVi,2007

LEVi,2009 2007

TA2

0.017

(0.002)

0.000

(0.001)

0.006***

(0.000)

0.000

(0.000)

0.821***

(0.012)

0.053***

(0.003)

0.369***

(0.009)

Yes

Constant

***

***

***

**

Signicance at the 5% level.

***

Signicance at the 1% level.

for the full sample as well as for ve of the six rm size subsamples (with the exception of the TA6

subsample), suggesting that the effect of leverage change on the change in ROA is stronger for the

increase in leverage than the decrease in leverage.

Our last analysis examines the effect of the change in ROA on the change in leverage during the

nancial crisis. We repeat our analysis of Table 9 by estimating Eq. (7) using the change in leverage as

our dependent variable. Table 11 provides evidence of the impact of the change in ROA on the change

in leverage in cross-sectional regressions during the nancial crisis. Column 1 of Table 11 shows the

results for the full sample, indicating that the change in ROA has a negative effect on leverage change.

Columns 27 indicate that the negative impact of the change in ROA on leverage change is evident

across all rm size subsamples.

5. Discussion

In this paper we documented several key ndings. First, leverage is negatively associated with operating performance (ROA). Second, the effect of leverage on operating performance is non-monotonic,

thereby providing a plausible explanation for the mixed results for the sign of the relation between

leverage and performance. It is possible that given the non-linearity (e.g., the U-shaped form) of

the linkage between leverage and performance, some studies capture the positive relation because

observed leverage ratios in their sample are below the reecting point (i.e., the optimal leverage),

whereas some studies capture the negative relation because observed leverage ratios in their sample

are above the reecting point. Our results provide further support to Coricelli et al. (2012), who show

that there is an inverted U-shaped relation between leverage and total factor productivity growth.

Unfortunately, our limited data set cannot, however, address the question of whether the observed

non-monotonic relationship is a result of the nancial crisis.

25

Table 11

OLS regressions on the change in leverage over the 20072009 global nancial crisis. This table presents the results of OLS

regressions of the change in leverage (LEV) on rm age, rm size, leverage, EBIT, ROA, and ROA during the 20072009

global nancial crisis. Dependent variable is leverage change (LEV), which is the leverage ratio in 2009 minus the leverage

ratio in 2007. The change in ROA, ROA, is measured as the EBIT/TA ratio in 2009 minus the EBIT/TA ratio in 2007. All other

explanatory variables are in 2007. Firms with leverage in 2007 larger than one are excluded from the sample. The sample

is divided into six subsamples based on total assets at the end of each year: (1) TA1 (TA < 1 million THB); (2) TA2 (1 million

THB TA < 5 million THB); (3) TA3 (5 million THB TA < 10 million THB); (4) TA4 (10 million THB TA < 100 million THB); (5)

TA5 (100 million THB TA < 1000 million THB); and (6) TA6 (TA 1000 million THB). Whites heteroscedasticity-consistent

standard errors are reported in parentheses.

Variable

Full sample

TA1

TA2

TA3

TA4

TA5

TA6

Constant

Industry dummies

0.076***

(0.002)

0.014***

(0.001)

0.004***

(0.000)

0.000

(0.000)

0.306***

(0.009)

0.148***

(0.002)

0.217***

(0.008)

Yes

0.095***

(0.004)

0.019***

(0.002)

0.013***

(0.003)

0.121***

(0.030)

0.227***

(0.019)

0.234***

(0.011)

0.200***

(0.013)

Yes

0.087***

(0.003)

0.015***

(0.001)

0.002

(0.002)

0.044***

(0.008)

0.219***

(0.021)

0.179***

(0.005)

0.231***

(0.012)

Yes

0.056**

(0.025)

0.008***

(0.002)

0.005

(0.014)

0.004

(0.034)

0.385*

(0.212)

0.145***

(0.007)

0.208***

(0.031)

Yes

0.044***

(0.007)

0.010***

(0.001)

0.006***

(0.002)

0.002*

(0.001)

0.401***

(0.039)

0.112***

(0.004)

0.213***

(0.028)

Yes

0.096***

(0.018)

0.010***

(0.002)

0.007**

(0.003)

0.000*

(0.000)

0.316***

(0.047)

0.101***

(0.006)

0.158***

(0.054)

Yes

0.048*

(0.028)

0.011***

(0.004)

0.005*

(0.003)

0.000

(0.000)

0.295***

(0.045)

0.114***

(0.012)

0.243***

(0.046)

Yes

Adjusted R2

F-statistic

Observations

0.135

2851.651***

146,599

0.157

517.284***

22,204

0.145

1177.118***

55,279

0.114

327.401***

20,370

0.089

449.009***

36,471

0.070

96.779***

10,129

0.126

39.493***

2146

Log agei,2007

Log TAi,2007

EBITi,2007

ROAi,2007

LEVi,2007

ROAi,2009 2007

*

**

***

Firm size

Signicance at the 5% level.

Signicance at the 1% level.

Third, our year-by-year cross-sectional regression results show that the negative effect of leverage on performance is evident for large rms, while the positive effect of leverage on performance

is observed for small rms. The negative effect of leverage on protability is larger in magnitude

for larger rms than for medium-sized rms, while the positive effect of leverage on protability is

larger in magnitude for very small rms than for small rms. In addition, we show that the impact

of leverage on performance is non-linear across rm size subsamples and across years. In terms of

methodologies, these results demonstrate one of the limitations of the panel regression approach

when the time-series dimension of the sample size is relatively short (i.e., three years in our sample). While panel regressions provide an opportunity to examine dynamic behaviors of rms, they

require variation in explanatory variables within each cross-sectional unit (i.e., a rm in this study)

over time. However, we effectively have only a maximum of three time-period observations (or two

observations when using lagged explanatory variables in regressions) for each rm in our panel data

set. As the Hausman tests (see footnote 10) suggest that xed-effects are preferred to random effects,

small or minimal time-series variation in explanatory variables within cross-sectional units is a cause

for concern for potentially poor estimations of panel OLS regressions with rm xed-effects in this

study.23 Therefore, comparing estimations of year-on-year cross-sectional regressions with those of

panel OLS regressions with rm xed effects should be made with caution. For instance, any difference

in estimated coefcients between the panel regression and the cross-sectional regression observed

23

The inclusion of xed effects in panel OLS regressions removes cross-section specic means from the variables, and hence

parameters of panel OLS regressions are estimated from the demeaned data. For a detailed discussion of advantages and

limitations of panel regressions, please see, e.g., Farkas (2005), Hsiao (2007), Baltagi (2008), Wooldridge (2010), and Saradis

and Wansbeek (2011).

26

in this study could be due to minimal or no variation in the variable within rms, especially with the

panel data set with many cross-sectional units over a very short time period.

Fourth, after controlling for past leverage and past operating performance, a change in leverage is

negatively associated with a change in operating performance across all rm sizes. This nding is consistent with Giroud et al. (2012), who report that a reduction in a debt overhang improves operating

performance. We nd that past leverage is negatively associated with the change in operating performance (as well as levels of prots), and that a positive change in leverage is associated with a negative

change in operating performance. In addition, the magnitude of the effect of leverage change on the

change in operating performance decreases with rm size. That is, larger rms tend to experience a

smaller impact of leverage change on the change in operating performance than smaller rms.

Fifth, past operating performance negatively affects a change in leverage. This nding is consistent

with Bris et al. (2004) and Xu (2012), who nd that operating performance has a negative effect on

leverage. However, for both very small and very large rms, past operating performance does not

affect the change in leverage. One plausible explanation for this observation is that because small

rms may not be able to obtain external nancing from banks (as evidenced by the median leverage of 0.04 for rms with total assets less than 1 million THB), past operating performance thus has

no effect on leverage. Consistent with our prediction, as rm size increases, the effect of operating

performance on leverage becomes positive and signicant, suggesting that nancial constraints on

medium-sized rms are not as severe as those of very small rms, leading to the observed positive linkage between past operating performance and the change in leverage. This result is also

consistent with past studies, such as Mehrotra et al. (2003), showing a positive relation between

protability and leverage. When a rm becomes larger, the effect of operating performance on leverage becomes negative and signicant. For very large rms, leverage may be at the optimal level

and thus is no longer inuenced by operating performance. Another potential explanation is that

the very large rms have reached the upper bound of leverage that maximizes rm value; therefore, any improvement in operating performance does not induce any signicant change in leverage.

If our sample of rms are nancially constrained during a nancial crisis, our ndings of the negative impact of operating performance on leverage for a large rm subsample (rms with total

assets between 10 and 100 million baht) and of the positive impact of operating performance on

leverage for medium-size rm subsamples (rms with total assets between 1 and 10 million baht)

seem to be in line with Korajczyk and Levy (2003), who suggest that nancially constrained

rms tend to increase leverage during periods of good performance than nancially unconstrained

rms.

Sixth, over the global nancial crisis period, smaller rms manage to substantially expand their

asset base (see Fig. 3) potentially through new equity issuances to existing shareholders, given that

leverage change is very slight for these rms. The fact that most of the rms in our sample are private

suggests that they are less likely to be subject to information asymmetries between insiders and

owners, especially when owners are in charge of managing the rm. This empirical result is relatively

new to the literature and is not entirely consistent with the pecking order models of capital structure

(e.g., Leary and Roberts, 2010; Lemmon and Zender, 2010; Morellec and Schrhoff, 2011; ShyamSunder and Myers, 1999) that primarily attempt to describe a nancing behavior of public rms rather

than private rms. Over the same period, larger rms appear to have a small reduction in leverage

and a small increase in assets, suggesting that these rms have not severely suffered from loan supply

shocks as compared to those larger rms in the US. We nd that about 24.41% of the rms in our sample

have a negative change in leverage during 20072009 (leverage in 2009 leverage in 2007 < 0) and a

fall in total assets during 20072009 (total assets in 2009 total assets in 2007 < 0). This observation

implies that about 75% of Thai rms in our sample appear to have managed to get through the global

nancial crisis on the basis that they do not have to simultaneously deleverage and liquidate their

assets. As noted by Denis and McKeon (2012) that after an initial hike in leverage ratios, a subsequent

downward change in leverage ratios toward the estimated target leverage is usually slow and that

nancial exibility permits rms to keep high leverage ratios (i.e., above the estimated target level)

over several years, it could be that our sampled rms might have unused debt capacity (i.e., a source of

nancial exibility) that allowed them to maintain their leverage ratios during the nancial crisis. Our

ndings are generally consistent with macroeconomic indicators, showing that the economic growth

27

in Thailand contracted only about 1.1% in 2009. For comparison, according to the World Bank, the GDP

growth in 2009 was 3.4% and 4.0% for the U.S. and the U.K., respectively.

One of key caveats of our study is that we are unable to ascertain whether the global nancial

crisis indeed has a long-term adverse effect on Thai rms, through for example substantial contractions in demand for Thai products and services in international markets (especially in the US and

European markets) after 2009, which may affect the moderating effect of rm size on the relation

between leverage and performance. But we do, however, show that during the global nancial crisis,

the leverageperformance relation is inuenced by rm size. The nding that operating performance

has no effect on leverage for very small and very large rms (see Table 9) can be seen as a sign that

these rms are not severely affected by the global liquidity shocks caused by the global nancial

crisis, suggesting that the very small and very large rms are not extremely constrained nancially.

It is possible that when owners of very small rms are not nancially constrained, small rms may

take advantage of the global nancial crisis by making further investments (e.g., acquiring potentially undervalued rms/business units that are severely suffered from liquidity shocks), which are

nanced by new equity issuance to existing or new shareholders. As information asymmetry issues

are less severe for rms that are managed by owners than rms that are managed by non-owner

managers, external nancing for the former is less problematic than for the latter. Another plausible

explanation for the insignicant impact of operating performance on leverage for very large rms is

that they may try to maintain their leverage target. Suppose that initial variations from the leverage

target are very small for these rms during the global nancial crisis, they are unlikely to signicantly

change leverage ratios if they do not suffer from loan supply shocks. This interpretation is based on

Harford et al. (2009), who nd evidence to support the notion that rms have a target level of capital

structure by showing that when rms nance acquisitions with debt, they subsequently lower the

acquisitions leverage effect within ve years.

Our ndings have some important repercussions for policy makers. The ndings suggest that nancial constraints for very small rms limit the realization of investment opportunities because these

rms cannot obtain nancing from banks although they might be protable. To overcome this issue,

rms will have to rely more on equity nancing, which raises an important question of whether small

rms are nancially constrained as normally assumed in the literature. We argue that the pecking

order models that are mainly developed to explain the nancing behavior of public rms may not be

necessarily applicable to the nancing behavior of private rms, especially during periods of nancial

turmoil. In this respect, our results stress the importance of micronance activities. With no or limited

external nancing through a traditional banking channel, small rms are likely to have remained small

simply because they cannot obtain necessary funds to expand their businesses, which will in turn limit

economic growth. Our results appear to support the view of Paulson and Townsend (2004), in the sense

that for very small rms (i.e., rms with total assets less than 1 million baht (THB) or approximately

32,250 USD at the exchange rate of 31 THB/USD) the median leverage is 0.04, which indicates almost

zero-leverage for the very small rms (see Strebulaev and Yang (2013) for a detailed discussion on

zero-leverage rms). As discussed earlier, a competing argument for low leverage ratios for small rms

is that they might be nancially conservative. Unfortunately, we cannot empirically address this issue.

6. Concluding remarks

In this paper, we attempt to ll the gap in the literature by examining the relation between nancial

leverage and operating performance, measured as ROA, with a special focus on a question of whether

rm size affects the linkage between leverage and performance. Using a large set of data on all Thai

rms during 20072009, we show several key ndings that help address mixed results documented in

the literature. Leverage has a negative and non-monotonic effect on operating performance in a panel

regression analysis. Past operating performance positively and negatively affects changes in leverage

for small rms and medium-sized rms, respectively. For both very small and very large rms, past

operating performance has no effect on the change in leverage. However, using a year-on-year crosssectional regression approach, we show that (1) in the full sample, the leverage has positive effect on

performance, (2) for large rms the effect of leverage on performance is negative, and (3) for small

rms the effect is positive. This pattern of cross-sectional results is stable across the study period. In

28

lights of these results, we show that the direction of the leverageperformance relation depends on

not only rm size but also the methodological approach.

With the extensive dataset of rms in Thailand, our study provides new empirical evidence on the

leverageperformance relation for rms in emerging markets economies. Given limited investment

opportunities (due to the global recessions) and nancial constraints (especially, liquidity shocks)

during the nancial crisis, rm size appears to moderate the relation between leverage and operating

performance, providing potential explanations for mixed results observed in the literature.

References

Ahn, S., Denis, D.J., Denis, D.K., 2006. Leverage and investment in diversied rms. J. Financ. Econ. 79, 317337.

Aivazian, V.A., Ge, Y., Qiu, J., 2005. The impact of leverage on rm investment: Canadian evidence. J. Corp. Financ. 11, 277291.

Akbar, S., Rehman, S., Ormrod, P.U., 2013. The impact of recent nancial shocks on the nancing and investment policies of UK

private rms. Int. Rev. Financ. Anal. 26, 5970.

Antoniou, A., Guney, Y., Paudyal, K., 2008. The determinants of capital structure: capital market-oriented versus bank-oriented

institutions. J. Financ. Quant. Anal. 43, 5992.

Audia, P.G., Locke, E.A., Smith, K.G., 2000. The paradox of success: an archival and a laboratory study of strategic persistence

following radical environmental change. Acad. Manage. J. 43, 837853.

Baker, M., Wurgler, J., 2002. Market timing and capital structure. J. Financ. 57, 132.

Baltagi, B.H., 2008. Econometric Analysis of Panel Data, 4th ed. John Wiley & Sons, Chichester, UK.

Bena, J., Ondko, P., 2012. Financial development and the allocation of external nance. J. Empir. Financ. 19, 125.

Berkmen, S.P., Gelos, G., Rennhack, R., Walsh, J.P., 2012. The global nancial crisis: explaining cross-country differences in the

output impact. J. Int. Money Financ. 31, 4259.

Bris, A., Koskinen, Y., Pons, V., 2004. Corporate nancial policies and performance around currency crises. J. Bus. 77, 749796.

Cai, J., Zhang, Z., 2011. Leverage change, debt overhang, and stock prices. J. Corp. Financ. 17, 391402.

Campello, M., Giambona, E., Graham, J.R., Harvey, C.R., 2011. Liquidity management and corporate investment during a nancial

crisis. Rev. Financ. Stud. 24, 19441979.

Campello, M., Graham, J.R., Harvey, C.R., 2010. The real effects of nancial constraints: evidence from a nancial crisis. J. Financ.

Econ. 97, 470487.

Chen, H., Chen, S., 2012. Investment-cash ow sensitivity cannot be a good measure of nancial constraints: evidence from the

time series. J. Financ. Econ. 103, 393410.

Coricelli, F., Drifeld, N., Pal, S., Roland, I., 2012. When does leverage hurt productivity growth? A rm-level analysis. J. Int.

Money Financ. 31, 16741694.

Covitz, D., Liang, N., Suarez, G.A., 2013. The evolution of a nancial crisis: collapse of the asset-backed commercial paper market.

J. Financ. 68, 815848.

Denis, D.J., McKeon, S.B., 2012. Debt nancing and nancial exibility evidence from proactive leverage increases. Rev. Financ.

Stud. 25, 18971929.

Denis, D.J., Sibilkov, V., 2010. Financial constraints, investment, and the value of cash holdings. Rev. Financ. Stud. 23, 247269.

Dick-Nielsen, J., Feldhtter, P., Lando, D., 2012. Corporate bond liquidity before and after the onset of the subprime crisis. J.

Financ. Econ. 103, 471492.

Drifeld, N., Pal, S., 2001. The East Asian crisis and nancing corporate investment: is there a cause for concern? J. Asian Econ.

12, 507527.

Duchin, R., Matsusaka, J.G., Ozbas, O., 2010a. When are outside directors effective? J. Financ. Econ. 96, 195214.

Duchin, R., Ozbas, O., Sensoy, B.A., 2010b. Costly external nance, corporate investment, and the subprime mortgage credit

crisis. J. Financ. Econ. 97, 418435.

Erickson, T., Whited, T.M., 2000. Measurement error and the relationship between investment and q. J. Polit. Econ. 108,

10271057.

Farkas, G., 2005. Fixed-effects models. In: Kempf-Leonard, K. (Ed.), Encyclopedia of Social Measurement. Elsevier, New York, pp.

4550.

Faulkender, M., Petersen, M.A., 2006. Does the source of capital affect capital structure? Rev. Financ. Stud. 19, 4579.

Firth, M., Lin, C., Wong, S.M.L., 2008. Leverage and investment under a state-owned bank lending environment: evidence from

China. J. Corp. Financ. 14, 642653.

Fredrickson, J.W., Iaquinto, A.L., 1989. Inertia and creeping rationality in strategic decision processes. Acad. Manage. J. 32,

516542.

Giroud, X., Mueller, H.M., Stomper, A., Westerkamp, A., 2012. Snow and leverage. Rev. Financ. Stud. 25, 680710.

Hall, T.W., 2012. The collateral channel: evidence on leverage and asset tangibility. J. Corp. Financ. 18, 570583.

Harford, J., Klasa, S., Walcott, N., 2009. Do rms have leverage targets? Evidence from acquisitions. J. Financ. Econ. 93, 114.

Hristov, N., Hlsewig, O., Wollmershuser, T., 2012. Loan supply shocks during the nancial crisis: evidence for the Euro area.

J. Int. Money Financ. 31, 569592.

Hsiao, C., 2007. Panel data analysisadvantages and challenges. TEST 16, 122.

Huff, J.O., Huff, A.S., Thomas, H., 1992. Strategic renewal and the interaction of cumulative stress and inertia. Strateg. Manage.

J. 13, 5575.

Jimnez, G., Lopez, J.A., Saurina, J., 2009. Empirical analysis of corporate credit lines. Rev. Financ. Stud. 22, 50695098.

Kannan, P., 2012. Credit conditions and recoveries from nancial crises. J. Int. Money Financ. 31, 930947.

Korajczyk, R.A., Levy, A., 2003. Capital structure choice: macroeconomic conditions and nancial constraints. J. Financ. Econ. 68,

75109.

Leary, M.T., Roberts, M.R., 2010. The pecking order, debt capacity, and information asymmetry. J. Financ. Econ. 95, 332355.

29

Lemmon, M.L., Zender, J.F., 2010. Debt capacity and tests of capital structure theories. J. Financ. Quant. Anal. 45, 11611187.

Li, D., 2011. Financial constraints, R&D investment, and stock returns. Rev. Financ. Stud. 24, 29743007.

MacKay, P., Phillips, G.M., 2005. How does industry affect rm nancial structure? Rev. Financ. Stud. 18, 14331466.

Margaritis, D., Psillaki, M., 2010. Capital structure, equity ownership and rm performance. J. Bank. Financ. 34, 621632.

Mehrotra, V., Mikkelson, W., Partch, M., 2003. The design of nancial policies in corporate spin-offs. Rev. Financ. Stud. 16,

13591388.

Miller, D., Friesen, P.H., 1983. Successful and unsuccessful phases of the corporate life cycle. Organ. Stud. 4, 339356.

Mizen, P., Tsoukas, S., 2012. The response of the external nance premium in Asian corporate bond markets to nancial

characteristics, nancial constraints and two nancial crises. J. Bank. Financ. 36, 30483059.

Morellec, E., Schrhoff, N., 2011. Corporate investment and nancing under asymmetric information. J. Financ. Econ. 99,

262288.

Moyen, N., Platikanov, S., 2013. Corporate investments and learning. Rev. Financ. 17, 14371488.

Paulson, A.L., Townsend, R., 2004. Entrepreneurship and nancial constraints in Thailand. J. Corp. Financ. 10, 229262.

Rajan, R.G., Zingales, L., 1995. What do we know about capital structure? Some evidence from international data. J. Financ. 50,

14211460.

Saradis, V., Wansbeek, T., 2011. Cross-sectional dependence in panel data analysis. Econ. Rev. 31, 483531.

Sentance, A., Taylor, M.P., Wieladek, T., 2012. How the UK economy weathered the nancial storm. J. Int. Money Financ. 31,

102123.

Shyam-Sunder, L., Myers, C.S., 1999. Testing static tradeoff against pecking order models of capital structure. J. Financ. Econ. 51,

219244.

Strebulaev, I.A., Yang, B., 2013. The mystery of zero-leverage rms. J. Financ. Econ. 109, 123.

Wooldridge, J.M., 2010. Econometric Analysis of Cross Section and Panel Data, 2nd ed. The MIT Press, Cambridge, Massachusetts.

Xu, J., 2012. Protability and capital structure: evidence from import penetration. J. Financ. Econ. 106, 427446.

Zheng, X., El Ghoul, S., Guedhami, O., Kwok, C.C.Y., 2012. National culture and corporate debt maturity. J. Bank. Financ. 36,

468488.

- PSYCHOLOGY TEST - SELF ACCEPTANCEUploaded byVidhya Sarath
- error in Chemical AnalysisUploaded byAhmed Atef
- Descriptive StatisticsUploaded byNurul Intan Fairuz
- Output Spss Dan EviewsUploaded byuki rinawati
- ppaersUploaded byRavi Gupta
- MAT LAB NEWUploaded byDayanand Gowda Kr
- Weighted Regression of Calibration CurveUploaded byBaharnarenj11
- Solution HW3Uploaded bymrezza
- 2 klmpkUploaded bydellajuliatum
- Semi-detailed-lesson-plan.docxUploaded byJaymaymaMadrid
- L11 RegressionUploaded byDonald Yum
- Agricultural Functions in PurniaUploaded byMahmood Ansari
- November 30 Time Series Cross Sectional AnalysisUploaded byzebatif
- Quantitative Methods 1Uploaded byRizwan Hussain
- StatisticsUploaded byDiobino Mantac
- Regression Analysis Complete SAS DEMO (2)Uploaded byBba Rockze
- Jpe12793 Sup 0001 AppendixS1Uploaded byCatalin Roibu
- Simple Linear Regression CollectionUploaded bySoniyaKanwalG
- 02Uploaded byElfrid BAEL
- Determinants Od MF_1Uploaded byPritiVandana
- Data Pre Processing - NGUploaded byVatesh_Pasrija_9573
- Optimization of Esterification of Maleic acid with Ethanol over Homogeneous Catalyst using Box-Behnken Experimental DesignUploaded byTI Journals Publishing
- GridDataReport-DATA1Uploaded bySamsulRizal
- GridDataReport JUT SEMPE2Uploaded byyudhi
- MB Lecture 2Uploaded byrui
- Paper Kopsidas OdysseasUploaded byΟδυσσεας Κοψιδας
- Sample Test 3Uploaded byJayanthi Loganathan
- sample medium term planUploaded byapi-268683759
- Effectiviness of Business Innovation and R&DUploaded bydnlkaba
- Quantitative TechniquesUploaded byMohamed Kurshid

- Individual Differences in Servant leadership: the roles of values and personalityUploaded byAnonymous pB9NV18F
- Determinan Profitability bankUploaded byAnonymous pB9NV18F
- Corporate Governance UndipUploaded byDonny Iskandarsyah
- Bentuk Soal Tes TOEFL StructureUploaded byAnonymous pB9NV18F
- 1 paperUploaded byAnonymous pB9NV18F
- jurnalUploaded byAnonymous pB9NV18F
- 1-s2.0-S0305750X12000666-mainUploaded byAnonymous pB9NV18F
- Earnings Management and Voluntary DisclosureUploaded byAnonymous pB9NV18F
- kelompokjurnal internasionalUploaded byAnonymous pB9NV18F
- Jurnal 1Uploaded byFitria Amar Mufidah
- PPT Jones ch01Uploaded bysumihosa

- MediationUploaded byHalim Mukti Wibowo
- 10s 12 Linear RegressionUploaded byalkmind
- Nguyen Grad.msu 0128D 10733Uploaded byali
- GulfPropUploaded bySagar Gogia
- CFA Level 2 Formula Sheets SampleUploaded byBernard Bortey
- 44630_UAS FiskomUploaded bySoffi Lutfia Dewi Trizana
- -Multiple Linear RegressionUploaded byCyn Syjuco
- Flotation Circuit Design Using a Geometallurgical ApproachUploaded byNataniel Linares
- Econometrics Assignment 1Uploaded byPeter Chenza
- Becker and Toutkoushian - Measuring Gender Bias in the Salaries (1)Uploaded byMohd Aizuddin
- Casio Fx 82 Ms StatsUploaded byEightball IskandarJrs
- ch08.docxUploaded bySaied Aly Salamah
- Statistical InferenceUploaded bySangkar Roy
- ps8_fall2015_shorter+versionUploaded byLuisSanchez
- Questions & Answers Chapter- 7 Set 1Uploaded byFaridOraha
- 203605060515Econometrics Ch12 Gujarati EditedUploaded byMutia Wahyuni
- TWO_133-Business Statistics (1)Uploaded byRocky Rider
- StochasticVolatilityUsingTheKalmanUploaded byPaola Andrea
- 2slsUploaded byRobson Glasscock
- Uncertainty Slope Intercept of Least Squares FitUploaded byPerfectKey21
- lab4-stats originalUploaded byapi-362845526
- Pastor StambaughUploaded bymirando93
- me envUploaded byGaja A
- 27.M.E. Applied EleUploaded bykalai
- p5130 Lec09 Qualitative IvsUploaded byAhmedAlhosani
- Test MediationUploaded byPorntep Sukhonwimolmal
- Time Series STATA Manual.pdfUploaded byPrashant Poddar
- dummy-19Uploaded byIla Agarwal
- garcia-berthouUploaded byBill Jungers
- psg collegeUploaded bychandhan reddy