Research in International Business and Finance 45 (2018) 62–71

Contents lists available at ScienceDirect

Research in International Business and Finance

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

The influence of cash flow on the speed of adjustment to the

T
optimal capital structure
⁎
Dominique Dufour , Philippe Luu, Pierre Teller
University of Nice, GRM Lab, 24 Avenue des diables bleus, 06300 Nice, France

AR TI CLE I NF O AB S T R A CT

JEL classifications: The aim of this paper is to examine the influence of cash flow on French SMEs’ speed of ad-
C23 justment (SOA) to their capital structure targets. Adjusting a firm’s financial structure results in
G32 transaction costs, including information costs, bargaining costs, and monitoring costs.
Keywords: Transaction costs have a strong influence on the SOA of a company’s financial structure to its
Target leverage target leverage (TL). Furthermore, transaction costs are considered higher for SMEs than for
Speed of adjustment listed companies. In this context, studying cash flows is particularly relevant because cash flow is
Cash flows a resource that involves low transaction costs. We apply a two-step model to French panel data
Transaction costs
collected during the period 2005–2014. In the first step, the TL is estimated. We considered two
SME
leverages: a short-term leverage and a long-term leverage. In the second step, the target is used to
estimate adjustment speeds by distinguishing between over-levered companies and under-lev-
ered companies. There are two main contributions of this study. This study’s first contribution
was that we found a significant difference in the SOA between over-levered and under-levered
firms for short-term leverage but not for long-term leverage. This study’s second contribution was
to highlight SMEs’ behaviour while adjusting their financial structure. For over-levered firms, the
statistical results showed that the speed of firms with a positive cash flow is higher than the speed
of firms with a negative cash flow. Contrary to findings related to listed companies, for under-
levered firms, our results do not show that a negative cash flow implies a faster adjustment to TL.

1. Introduction

Ever since Modigliani and Miller’s 1958 contribution, many works have been devoted to the study of capital structure. The trade-
off theory holds that each firm has an optimum capital structure. This optimum is the result of a trade-off between the benefits (i.e.,
the tax benefits of interest payments) and costs (i.e., the costs of financial distress) of debt. Classic research on financial structure has
focused on listed companies. The literature has raised the question of whether SMEs’ financial management can be analysed using
tools built for listed companies (Michaelas et al., 1999). Ang (1992) stresses that agency, information, failure costs, taxes, and
transaction costs matter not only for listed companies but also for SMEs. For these reasons, Psillaki and Daskalakis (2009) have found
that the theory initially built to study the capital structure of listed companies can be used to study SMEs’ capital structure. And in the
conclusion of a study dedicated to UK SMEs, Michaelas et al. (1999) have shown that most of the determinants of capital structure
used in financial theory seem to be relevant for small businesses in the U.K. It is thus legitimate to consider the existence of a target
debt structure for SMEs. Michaelas et al. (1999), Ozkan (2001), Cassar and Holmes (2003), Psillaki and Daskalakis (2009), Heyman
et al. (2008), Aybar-Arias et al. (2012), Palacin-Sânchez et al. (2013), and Mateev et al. (2013) are among those who have studied

⁎
corresponding author.
E-mail addresses: [email protected] (D. Dufour), [email protected] (P. Luu), [email protected] (P. Teller).

https://doi.org/10.1016/j.ribaf.2017.07.132
Received 19 June 2017; Accepted 5 July 2017
Available online 12 July 2017
0275-5319/ © 2017 Elsevier B.V. All rights reserved.
D. Dufour et al. Research in International Business and Finance 45 (2018) 62–71

SMEs’ financial structure. The most commonly used determinants include size, age, sector, profitability, tax shields, growth and asset
structure. When a firm’s financial structure moves away from the TL, the company experiences deviation costs and therefore has an
incentive to move closer to its target. For an over-levered firm, deviation costs are an increase of the cost of failure, whereas for an
under-levered firm, deviations costs are a shortfall caused by the loss of tax savings.
Nevertheless, firms also experience adjustment costs when they adjust their financial structures. If there were no adjustment costs,
the leverage observed should equal the optimal leverage. These costs are transaction costs and include information costs, bargaining
costs, and monitoring costs. A firm incurs these costs when finalizing a contract that affects its financial resources. The leverage
moves closer to the target only when the transaction costs are lower than the deviation costs. These costs are the reason that a firm’s
adjustment to its optimal financial structure might only be partial; they are the reason that the notion of SOA arose. Speed of
adjustment is defined as the ratio between the variation of the observed leverage and the deviation from the target.
The literature describes several variables that moderate the SOA. All of these studies revolve around the same key point: the
importance of adjustment costs. High adjustment costs have a negative impact on the SOA. The literature reviews the elements that
are likely to affect these costs.
The first factor is the distance to TL. Mukherjee and Wang (2013) and Welch (2004) show a positive relationship between this
distance and the speed of adjustment. Two things can explain these results. First, the costs of not adjusting—especially bankruptcy
costs in the case of an over-levered firm—increase with this distance and thus, managers are more likely to adjust. Second, the
presence of fixed costs decreases the adjustment speed if the distance is small.
The impact of market valuation and equity mispricing on adjustment speed has also been studied (Warr et al., 2012). An over-
valuation of equity reduces the issuing cost and the adjustment cost when the firm is over-levered. Consequently when equity is
undervalued, issuing equity is more costly and adjustment costs are higher. The authors found that for over-levered firms, the speed
of adjustment is higher when equity is over-valued. They provide evidence based on a sample of US listed companies.
Lockhart (2014) shows that the existence of a credit line provides the firm with a low marginal cost that leads to a more rapid
adjustment.
Drobetz et al. (2015) study both the differences in adjustment speed across financial systems and the impact of recession and
sector. They note that these differences are attributable to different adjustment costs.
Company size is also likely to impact adjustment costs. Mukherjee and Wang (2013) find that large firms adjust more quickly than
small firms.
The moderating effect of cash flows on the speed of adjustment has also been studied by Byoun (2008) and Faulkender et al.
(2012). These studies of listed companies conclude that cash flows have a strong influence on the SOA. Both use the same argument in
their work: positive CF enables financial structure adjustment at low transaction costs.
Almost all of these studies are dedicated to listed companies. Aybar-Arias et al. (2012) are an interesting exception. In this work,
we study the SOA to TL in the context of French SMEs. We examine cash flow’s impact on this speed by distinguishing between long-
term and short-term debt leverage. Questioning the SOA in the context of SMEs appears well founded because financial structure
management is particularly important for small companies. First, SMEs experience a higher risk of failure than listed companies.
Therefore, the issue of debt is a major issue. Second, SMEs have fewer financing options than listed companies because SMEs cannot
access capital markets. Third, SMEs experience higher transaction costs than listed companies when raising funds and do not have the
same fundraising skills as big companies (Holmes and Kent, 1991). It is also more difficult for SMEs to manage the complexity of
administrative procedures. In a 2015 inquiry by the European Commission dedicated to the financing of European enterprises, 7% of
respondents cited paperwork as the main factor limiting their debt (European Commission, 2015).
Informational asymmetry in the relationship between SMEs and their financial environment—mainly banks—also plays an im-
portant role. First, SMEs disclose less financial information than do listed companies. Second, this information is subject to less-
stringent audit standards, leading to an increase in informational asymmetry. This asymmetry generates higher transaction costs. For
example, when obtaining a bank loan, asymmetry can lead the lender to require additional guarantees. In the European Commission
study cited above, it appears that most SMEs are experiencing an increase in the non-interest costs of financing. The lender must also
support monitoring costs after the loan has been granted to ensure, for example, that there is no opportunistic use of the loan. This
issue is more acute in the French context. France’s legal structure is considered to offer poor protection to investors and creditors. On
a creditor-rights scale, a score of 0 (weak creditor rights) has been assigned to France by Oztekin and Flannery (2012). Additionally,
Psillaki and Daskalakis (2009) note that transaction costs tend to be particularly acute in France.
Studies of the SOA have adopted two empirical approaches. Some rely on a two-step procedure: in the first step, the TL is obtained
and in the second step, the SOA is estimated (Byoun 2008; Faulkender et al., 2012). We choose to follow this option. Other works
develop an integrated model in which determinants of TL and SOA are estimated together in a single model (Aybar-Arias et al., 2012).
This paper’s aim is to study the impact of cash flows on SMEs’ SOA. This question has been addressed in prior studies; however,
those studies examine listed companies (Byoun 2008; Faulkender et al., 2012). This question is even more important to SMEs: cash
flows involve low transaction costs and are SMEs’ primary source of funding. We choose two measures of debt ratio: short-term debt
leverage and long-term debt leverage. The targets and adjustment speeds associated with these leverages may differ. Long-term debts
are usually contracted to fund fixed assets. Their negotiation is often long and includes the acceptance of guarantees. Short-term debts
are generally used to fund the operating cycle. Their management is easier and the transaction costs involved are a priori lower. For
all of these reasons, it appears relevant to distinguish between the two types of debt.
We obtained several interesting results. First, we compared the SOA of over-levered and under-levered firms. Studies of listed
companies note that over-levered firms show a greater SOA to their debt leverage target. The SMEs in our French sample behave

63
D. Dufour et al. Research in International Business and Finance 45 (2018) 62–71

differently. Even if the SOA of short-term debts is higher for over-levered companies, our findings show that there is no difference
between adjustment speeds in the case of long-term debts.
Second, we found evidence that the SOA is influenced by cash flows. On the one hand, the adjustment speed is higher when firms
have above-target debt with a positive cash flow than when they have above-target debt with a negative cash flow. This result is
consistent with the findings of Byoun (2008). On the other hand, for under-levered companies, the SOA of firms with a negative CF is
not significantly greater than the SOA of firms with a positive CF. The nature of the relationship between SOA and cash flow in our
sample of French SMEs is therefore different from that found by Byoun (2008) and Faulkender et al. (2012) for listed companies.
This paper is organized as follows. Section 2 presents the model, the variables and the data. Section 3 presents the estimation
methods and discusses the empirical results. Next, we conclude.

2. Model, variables and data

Our methodology follows a two-stage approach, as in Byoun (2008) and Warr et al. (2012). In the first stage, the TL is estimated
using panel data methodology. In the second stage, the adjustment speed is estimated based on TL obtained in the first stage.

2.1. Target leverage

Over-levered firms’ SOA should be higher than under-levered firms’ SOA. We consider two approaches to estimate the TL.
If the TL linearly depends on a set of firm characteristics, one way to perform a direct estimation would be to use the following
model:

LEVit = βXit + ηi (1)

where Xit is a vector of determinants commonly used in the literature and ηi is the latent time-invariant firm i effect. Our first TL
estimation is the predicted value TLit = βXˆ it where β̂ is the vector of coefficients estimated from Eq. (1).
Following the literature (Warr et al., 2012; Aybar-Arias et al., 2012; Mateev et al., 2013), a second specification can be used to
estimate the TL. Our work is based on a classic adjustment speed equation:
LEVit − LEVit − 1 = λ (TLit − LEVit − 1) + εit (2)

where λ is the SOA, LEVit is the debt-to-asset ratio, TLit represents the TL for firm i at time t and εit is the error term. Rewriting Eq. (2)
to isolate LEVit on the left side of the equation and using Formula (1) to replace TLit, we obtain the following dynamic panel model:
LEVit = (1 − λ ) LEVit − 1 + λβXit + ληi + εit (3)
ˆ it and the actual lagged leverage LEVit−1, our second TL
Considering the estimated coefficient λ̂ , the predicted leverage LEV
1 ˆ it − (1 − λˆ ) LEVit − 1) .
estimation is TLit' = ˆ (LEV
λ

2.2. Speed of adjustment model

Based on Eq. (2) and by noting ΔLEVit = LEVit − LEVit−1, DEVit = TLit − LEVit−1, we specify the following model:

ΔLEVit = λ 0 + λ1 δitDEV −DEVit + λ2 δitDEV +DEVit + εit (4)

where δitDEV −is a dummy variable equal to one if the deviation DEVit is negative for firm i at time t (leverage above the target) and
δitDEV + is a dummy variable equal to one if the deviation DEVit is positive (leverage below the target). The model associated with Eq.
(4) allows the adjustment speed to vary depending on the debt-ratio position from the TL.
We also further permit the speed to vary according to the amount of cash flow available. We consider the model that includes
interactions between dummy variables:

ΔLEVit = (β1 δitCF + + β2 δitCF −) + (β3 δitCF + + β4 δitCF −) δitDEV −DEVit + (β5 δitCF + + β6 δitCF −) δitDEV +DEVit + εit (5)
where δitCF − is a dummy variable equal to one if cash flow is negative and δitCF + is a dummy variable equal to one if cash flow is
positive.
To check the robustness of our results, we also estimate the TL with lagged values of determinants (Xit is replaced by Xit−1 in Eqs.
(1) and (3)). The conclusions are comparable to those presented below.

2.3. Variables

2.3.1. Leverage
In this study, we use two measures of leverage: Total Bank Borrowings, excluding short-term debt to Total Assets (DR_LONG), and
Total Bank Borrowings, excluding long-term debt to Total Assets (DR_SHORT). The use of these two leverages is commonly found in
the literature on SMEs (Sogorb-Mira 1995; Palacin-Sânchez et al., 2013). Our variables are based on book values because the SMEs in
the sample are non-listed firms. Commercial debt was not included in the leverages. It is important to distinguish between short-term
and long-term leverages. The purposes and trading conditions associated with long-term debt are different from those associated with

64
D. Dufour et al. Research in International Business and Finance 45 (2018) 62–71

Table 1
List of variables.

Variable Denomination Description (Diane Database variable denomination in brackets)

Debt Ratios
DR_LONG_ini Long-term debt ratio Global debt ratio minus short debt ratio
DR_SHORT_ini Short-term debt ratio Ratio of short-term financial debts (Incl. short-term fin. debts) to total assets (Total assets: net figure)

Dependant variables
DR_LONG Sector adjusted long debt ratio Long-term debt ratio minus sector median ratio
DR_SHORT Sector adjusted short debt Short-term debt ratio minus sector median ratio
ratio

Leverage's explanatory variables

SIZE Size Logarithm of sales (Net Sales). Larger firms are supposed to have more leverage because their
information is more transparent and because borrowing costs are smaller. However, there are
conflicting results on the relationship between size and leverage.
RISK Risk of default Ratio of the standard deviation of the difference between operating cash flow and working capital needs
variation over the period starting in 2005 until the current year to the mean of total assets (total assets:
gross figure) over the same period. The variability of the firm's future income can affect the firm’s ability
to manage debt-related fixed charges. One might anticipate a negative link between risk and leverage.
PROF Profitability Ratio of operating profit (operating profit) to assets (total assets: gross figure). On the one hand, a firm
with high earnings could prefer a low leverage using retained earnings as source of funds. On the other
hand, a highly profitable firm could increase its leverage to benefit from tax reduction induced by
interest paid.
TAXSH Non-debt tax shield Ratio of depreciation (total fixed assets: depreciation, provisions) to total fixed assets (FIXASSETS).
Firms with higher depreciation expenses are less interested in a tax deduction associated with debt
financing. They can use non-debt tax benefits as a substitute for leverage tax benefits.
TANG Tangibility Ratio of net tangible fixed assets (net tangible fixed assets) to assets (total assets: net figure). Firms with
greater tangible assets have a higher debt capacity. They can offer more valuable assets as collateral.
GROWTH Growth Ratio of cash flow standard deviation over the period starting in 2005 to the mean of total assets over
the period starting in 2005. A firm with important growth opportunities may be considered risky and be
confronted with difficulties in improving its leverage. Myers (1984) notes a negative link between
leverage and growth opportunities.

Other variables
FIXASSETS Fixed Assets Sum of the following Diane items:
- Land: gross figure
- Buildings: gross figure
- Plant and equipment: gross figure
- Prop., plant and eq. in progress: gross figure
- Prepaym. to tangible assets: gross figure
- Other tangible assets: gross figure
INVEST Investment Purchases (total fixed assets: purchases, contributions) minus financial assets sales (financial assets:
sales), minus extra revenue from capital transaction (extr. rev. from capital transact.).
CF Cash Flow Operating cash flow (operating cash flow) minus variation of working capital needs (working capital
needs) between t and t-1, minus investment (INVEST).

short-term debt. The transaction costs of these two forms of financing are not a priori equal: the transaction costs associated with
short-term debt might be considered lower.

2.3.2. Leverage determinants

The following determinants of TL are considered: size, profitability, risk, non-debt tax shield, tangibility and growth (details and
calculations using information from the Diane database can be found in Table 1). These firm characteristics have been widely used in
capital structure studies (Michaelas et al., 1999; Bhaird and Lucey 2010; Psillaki and Daskalakis 2009; Aybar-Arias et al., 2012;
Palacin-Sânchez et al., 2013).

2.4. Data

Our sample is derived from the DIANE database, developed by Bureau Van Dijk. This database contains accounting and financial
information about French firms obtained from the annual financial statements deposited at the clerk's office of the Commercial Court.
We selected firms that meet the European Commission’s definition of an SME (Recommendation 2003/361/EC, May 6th, 2003).
More specifically, the firms in our sample met the following criteria: (1) fewer than 250 employees; (2) turnover of less than €50
million; and (3) total assets between €0 and €43 million. All sectors are covered with the exception of finance, insurance and utilities.
Variables defined as positive were required to be positive (namely, assets, net assets, fixed assets, equity, short and long debts).
Cases with errors or missing values were then eliminated. Following prior studies (Aybar-Arias et al., 2012), firms showing extreme
figures in any of the variables were excluded from the sample to minimize the impact of outliers (firms falling outside the interval
based on the variable mean ± 2 times the standard deviation were removed).
To control for common determinants of leverage among firms in the same industry, debt leverages are adjusted by the industry

65
D. Dufour et al. Research in International Business and Finance 45 (2018) 62–71

Table 2
Sample breakdown by firm size and sector.

N Firms N Obs sales to assets DR_LONG_ini DR_SHORT_ini

Panel A: Size
Very Small (employees < 10) 755 3818 1.826 0.141 0.038
Small (9 < employees < 50) 1175 6059 1.718 0.123 0.036
Medium (49 < employees < 250) 204 1012 1.633 0.109 0.036

Panel B: Sector (French classification of activities NAF Rev 2, 2008)

A – Agriculture, forestry and fishing 28 137 0.880 0.221 0.045
B – Mining and quarrying 4 21 0.851 0.080 0.037
C – Manufacturing 398 2031 1.332 0.131 0.033
F – Construction 415 2100 1.727 0.096 0.032
G – Wholesale, retail trade and repair of motor vehicles and motorcycles 805 4139 2.064 0.142 0.045
H – Transportation and storage 125 662 1.942 0.115 0.034
I – Accommodation and food service activities 47 228 1.371 0.200 0.027
M – Professional, scientific and technical activities 138 689 1.325 0.113 0.026
N – Administrative and support service activities 102 526 1.910 0.100 0.024
P – Education 13 60 1.342 0.102 0.032
Q – Human health and social work activities 27 132 1.381 0.197 0.042
R – Arts, entertainment and recreation 8 35 1.899 0.095 0.022
S – Other service activities 24 129 1.357 0.138 0.042

Total Sample 2134 10889 1.748 0.128 0.036

median. Adjusted debt leverage is defined as the difference between a firm’s leverage and the median leverage of the industry to
which the firm belongs.
After refining the information obtained, the final sample contains 2134 firms. Because of the estimation methods, we kept firms
with at least four consecutive years of data available. This represents an unbalanced panel comprising 10,889 observations. Although
data were collected from 2005 to 2014, because of the definition of our variables, our period of analysis starts in 2007 and runs until
2014.
Companies in the total sample have an average long-term (short-term) debt ratio of 12.8% (3.6%) (Table 2). Thirty-seven point
one percent of the companies are very small firms with a mean long-term (short-term) debt ratio of 14.1% (3.8%). Small companies
account for half of the total sample (52.6%), with a mean long-term (short-term) debt ratio of 12.3% (3.6%). Medium companies
represent 10.3% of the sample, with a mean long-term (short-term) debt ratio of 10.9% (3.6%). The dataset represents the diversity of
French SMEs and covers all sectors. Economic sectors with the highest total leverage mean values are agriculture, forestry and fishing
(26.6%); human health and social work activities (23.9%); accommodation and food service activities (22.7%); and wholesale, retail
trade and repair of motor vehicles and motorbikes (18.7%).
A summary of univariate descriptive statistics for the dependent and explanatory variables is given in Table 3. French SMEs have
average assets of €5.036 million, average net sales of €3.951 million and average cash holdings of €142.366 thousand.
It appears from reading the table that the long-term debt ratio is far higher than the short-term debt ratio. To simplify the
presentation of the results, estimates relating to the global debt ratio (i.e., the sum of the long-term and short-term debt ratios) are not
presented. These results are comparable to those obtained for the long-term debt ratio.

Table 3
Descriptive statistics.

Mean Std Min Median Max

ASSETS (K€) 5036.323 103736.444 85.036 1411.574 4961542.00

NET SALES (K€) 3951.102 4530.942 253.401 2105.352 24988.710
CF (K€) 142.366 3157.809 −86957.000 40.969 282323.000
DR_LONG_ini 0.128 0.116 0.000 0.100 0.497
DR_SHORT_ini 0.036 0.049 0.000 0.010 0.200
DR_LONG 0.019 0.102 −0.350 0.000 0.491
DR_SHORT 0.019 0.047 −0.131 0.000 0.198
SIZE 7.780 0.977 5.535 7.652 10.126
RISK 0.085 0.054 0.000 0.074 0.289
PROF 0.045 0.058 −0.134 0.038 0.235
TAXSH 0.804 1.081 0.022 0.742 43.423
TANG 0.108 0.098 −0.009 0.077 0.447
GROWTH 0.213 3.737 −0.427 0.046 257.633

66
D. Dufour et al. Research in International Business and Finance 45 (2018) 62–71

3. Estimation methods and results

3.1. Estimation method

Our methodology adopts a two-stage approach. In the first stage, the TL is estimated using panel data methodology. To take into
account the panel structure of our data, Eq. (1) is estimated using a panel regression with fixed or random individual effects according
to the outcome of the Hausman test. Fixed-effects model (FE) permits regressors to be endogeneous provided they are correlated only
with ηi. The random-effects model (RE) assumes that regressors are completely exogenous. The Hausman test assumes that the RE
model is fully efficient under the null hypothesis. A significant test leads to a strong rejection of the null hypothesis that the RE model
provides consistent estimates; thus, fixed-effects are preferred. In both specifications, fixed year effects are included.
Ordinary least squares (OLS) and FE estimators are inappropriate for estimating (5). OLS results are biased and inconsistent
because fixed effects are omitted and the lagged dependent variable is correlated with the fixed effect ηi. Although the FE estimator
removes the fixed effects through the within-group transformation, parameter estimates remain biased because the lagged trans-
formed dependent variable continues to correlate with the transformed error term. That bias vanishes asymptotically (when the
number of years increases in available data) but is significant in short panels.
To correct this short panel bias, econometric methods based on the instrumental variable approach have been developed. Arellano
and Bond (1991) use a generalized method of moments framework and propose the first-difference GMM estimator. This estimator is
based on a first-difference transformation in which the lagged endogenous variable constitutes an instrument for the first-differenced
variable. However, if the lagged levels are serially correlated, the correlation between the differenced variable and the lagged levels
might be important if the number of years present in the panel is small (this correlation decreases as the number of years increases).
Lagged levels might then be weak instruments for the first-differenced variable, and the difference GMM estimator might present non-
negligible finite sample biases (Bond 2002). To overcome this finite sample bias, Blundell and Bond (1998) improve the efficiency of
first-difference GMM and propose the GMM system estimator: in addition to the first-difference transformation equation (in-
strumented using lagged levels), the method jointly estimates the equation in levels (instrumented using lagged first differences).
We estimate (3) with the GMM system estimator, which has become the estimator of choice in many panel data settings (Warr
et al., 2012; Aybar-Arias et al., 2012; Mateev et al., 2013; Faulkender et al., 2012). More specifically, we use the two-step robust
version, which seems modestly superior to the one-step version (Roodman 2006). Two diagnostic tests are performed to assess the
validity of the GMM system model. First, the methodology assumes that there is no second-order serial correlation in the first
differenced residuals. Rejection of the null hypothesis of an AR(2) test (H0: no second-order autocorrelation) implies that the moment
conditions are not valid and the estimations are not consistent. Second, because the GMM system estimator is an instrumental
variable method, the Sargan-Hansen test of over-identifying restrictions is employed. A significant p-value implies that the instru-
ments are not valid.
In the second stage, the adjustment speed in Eqs. (2), (4) and (5) is estimated based on TL obtained in the first stage and using OLS
regression with correction for heteroscedasticity and firm-level clustering.

3.2. Results

3.2.1. Target leverage estimation

We begin by estimating the determinants of capital structure to highlight a target financial structure. Table 4 shows the results of
panel regressions on the determinants of TL.
The TL estimation shows several interesting results.
Size is statistically significant and negatively related to long-term leverage. Conversely, there is no statistically significant relation
between size and short-term leverage. Ozkan (2001) writes “there are conflicting results on the relationship between size and
leverage”. Indeed, periods, methods, variables and the definitions of leverage are numerous. Thus, comparison with other studies can
be difficult. Most studies show a positive relationship between leverage and size (Bhaird and Lucey 2010; Aybar-Arias et al., 2012).
However, Palacin-Sânchez et al. (2013), for example, find a positive relation on their total sample, while in the same study finding
conflicting results on sub-samples.
We do not find a statistically significant relationship between risk and debt. A negative relationship has frequently been docu-
mented (Psillaki and Daskalakis 2009; Ozkan 2001). However, different results have sometimes been highlighted. Using a British
sample, Michaelas et al. (1999) have revealed a positive relationship between risk and debt.
We find a negative relation between leverage and profitability. This result is consistent with those of Sogorb-Mira (1995), Psillaki
and Daskalakis (2009) and Palacin-Sânchez et al. (2013). High profitability is associated with low leverage.
We find evidence that there is no link between non-debt tax shield and leverage. Non-debt tax benefits have no influence on
leverage.
Leverage is positively correlated with tangibility for long-term debt. This result is common in the literature. Tangible assets have a
high collateral value and therefore, firms with fixed assets have easier access to long-term debt. In contrast, there is a negative link
between tangibility and short-term leverage. This second result is in line with Sogorb-Mira (1995) and Palacin-Sânchez et al. (2013).
We do not find any relation between growth and leverage. This result is in line with Psillaki and Daskalakis (2009).
Within the GMM estimations, the one-period-lagged-value of the dependent variable is the only variable in the regression that is
always statistically significant. The adjustment coefficient is equal to 0.678 for long-term leverage and 0.572 for short-term leverage.

67
D. Dufour et al. Research in International Business and Finance 45 (2018) 62–71

Table 4
Parameter estimates on the determinants of the target debt-ratio. The dependent variables are sector-adjusted debt ratios. Two debt ratios are considered: long-term
debt ratio (DR_LONG) and short-term debt ratio (DR_SHORT). The independent variables are logarithm of sales (SIZE), ratio of the standard deviation of the difference
between operating cash flow and working capital needs variation to the mean of total assets over the same period (RISK), ratio of operating profit to assets (PROF),
ratio of depreciation to total fixed assets (TAXSH), ratio of net tangible fixed assets to assets (TANG) and ratio of cash flow standard deviation over the period to the
mean of total assets over the same period (GROWTH). For each dependent variable, several models are considered: the panel regression with fixed/random firm effects
(FE/RE) model or the GMM system model (GMM). All of the regressions include year dummies. The R2 presented is the R2 within. The F-test (in a FE panel model) and
Wald test (in a RE model) null is that the regressors are all jointly zero. In an FE model, the fixed-effects F-test null is that the individual dummy variables are all jointly
zero. In an RE model, the Breusch and Pagan Lagrange (BP) multiplier test null is the absence of random effects. The Hausman test null hypothesis is that RE provides
consistent estimates: if the test p-value is too low, FE is preferred. We used the two-step robust version of the GMM system estimator. We considered year dummies as
strictly exogenous, the dependent variable as predetermined but not strictly exogenous and the independent variables as endogenous. Two diagnostic tests are
performed to assess the validity of the GMM system model: (1) the AR(2) test of no second-order serial correlation in the first differenced residuals; and (2) the Sargan-
Hansen test of over-identifying restrictions. To obtain consistent estimates and valid instruments, both tests need to be non-significant.

Dependent variable: DR_LONG DR_SHORT

TL estimation: FE GMM RE GMM

L.Y. 0.678*** 0.572***

(12.342) (5.963)
SIZE −0.014*** 0.014· −0.001· −0.003
(−4.223) (1.658) (−1.898) (−0.830)
RISK 0.042· 0.136 −0.001 −0.035
(1.888) (1.054) (−0.110) (−0.434)
PROF −0.090*** −0.218* −0.105*** −0.093
(−6.372) (−2.163) (−14.399) (−1.574)
TAXSH −0.001 −0.003 −0.000 −0.001
(−0.951) (−1.175) (−0.041) (−1.374)
TANG 0.527*** 0.085 −0.021*** 0.011
(35.697) (1.150) (−3.441) (0.468)
GROWTH 0.001 −0.000 −0.000 0.000
(1.601) (−0.387) (−1.024) (0.417)
Constant 0.053* −0.118· 0.034*** 0.034
(1.966) (−1.892) (4.811) (1.047)

Year Fixed Effects Yes Yes Yes Yes

R2 within 0.153 0.022
F-test p-value 0.000
F-test Fixed effects p-value 0.000
Wald test p-value 0.000
BP test p-value 0.000
Hausman test p-value 0.000 0.212
N instruments 126 126
AR(2) test p-value 0.860 0.143
Sargan/Hansen test p-value 0.106 0.201
N obs 10889 8755 10889 8755
N groups 2134 2134 2134 2134

· p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001 (t-statistic in parentheses).

3.2.2. Adjustment speed

Table 5 shows the results of three target adjustment models.
Panel A reports the estimation results for the classic partial adjustment model specified by Eq. (2). The coefficient estimates range
between 0.172 and 0.365. These speeds of adjustment are statistically significant (p-values inferior to 0.1%). Because estimation
methods and variable definitions are not identical, comparisons with previous works are uneasy. Mukherjee and Wang (2013) obtains
an average speed of 0.164. Smith et al. (2004) find that firms move towards target ratios at speeds ranging from 0.248 to 0.255,
whereas Faulkender et al. (2012) estimate adjustment speed at 0.219. Getzmann et al. (2014) obtain an annual adjustment speed of
between 0.24 and 0.45 using an Asian sample. Oztekin and Flannery (2012) obtained speeds ranging from 0.276 to 0.333 on a French
sample.
These studies use data on listed companies. There are fewer works on SMEs than on large companies. Aybar-Arias et al. (2012)
examined adjustment speed on a Spanish SMEs data. Using an integrated approach, they obtained an average speed of 0.2628.
Panel B tests whether there is a difference between the adjustment speed of over-levered firms and the adjustment speed of under-
levered firms (Eq. (4)). The speed of adjustment is a trade-off between deviation costs and adjustment costs. SMEs, like listed
companies, face deviation costs when their leverage is non-optimal. The literature considers the characteristics of these costs to vary
according to whether the company is over-levered or under-levered. Deviation costs are considered to be higher in the first case than
in the second case. In the case of over-levered firms, high financial distress costs provide strong incentives to adjust their capital
structures, although the associated transaction costs are high (Dang et al., 2014). Mukherjee and Wang (2013, p. 609) assert that the
manager’s incentive to adjust is higher for an over-levered firm than for an under-levered firm. They provide an interesting reason:
“The firm’s ultimate cost of being over-levered is the total loss of value of its shares, while the ultimate cost of being under-levered is the
opportunity cost associated with the lost tax shield. Thus, the managers’ incentive to adjust in an over-levered situation is likely to be higher

68
D. Dufour et al. Research in International Business and Finance 45 (2018) 62–71

Table 5
Target adjustment model. This table shows the results of a regression model with correction for heteroscedasticity and firm-level clustering. Two dependent variables
are considered: long-term debt ratio (DR_LONG) and short-term debt ratio (DR_SHORT). DEV represents the deviation of the leverage from its target. The target debt
level is estimated by the predicted value from a panel regression with fixed/random firm effects (FE/RE) or a GMM system model (GMM). δDEV− (δDEV+) is a dummy
variable equal to one if DEV is negative (positive). Test 1 is for the null hypothesis that the parameter estimate above the target deviation (λ1) is smaller than the
parameter estimate below the target deviation (λ2). Test 2 is for the null hypothesis that the parameter estimate above the target deviation with positive cash flow (β3)
is smaller than the parameter estimate above the target deviation with negative cash flow (β4). Test 3 is for the null hypothesis that the parameter estimate below the
target deviation with negative cash flow (β6) is smaller than the parameter estimate below the target deviation with positive cash flow (β5).

Panel A: Target adjustment model 1 (Eq. (2))

DR_LONG DR_SHORT

FE GMM RE GMM

DEV 0.206*** 0.172*** 0.365*** 0.348***

(22.652) (24.069) (28.519) (28.296)
Constant −0.002*** −0.001 0.001 0.001*
(−3.566) (−1.200) (1.395) (2.479)
R2 0.117 0.106 0.178 0.175
N obs 8755 8755 8755 8755

Panel B: Target adjustment model 2 (Eq. (4))

DR_LONG DR_SHORT

FE GMM RE GMM

DEV−
DEV.δ (λ1) 0.215*** 0.180*** 0.425*** 0.426***
(15.888) (16.898) (21.139) (21.691)
DEV.δDEV+ (λ2) 0.193*** 0.161*** 0.234*** 0.191***
(9.967) (9.549) (8.115) (7.963)
Constant −0.002 0.000 0.004*** 0.005***
(−1.399) (0.038) (4.993) (7.422)
R2 0.117 0.106 0.181 0.181
Test 1 t-statistic (H0: λ1 ≤ λ2) 0.789 0.798 4.621 6.506
Test 1 p-value 0.215 0.212 0.000 0.000
N obs 8755 8755 8755 8755

Panel C: Target adjustment model 3 (Eq. (5))

DR_LONG DR_SHORT

FE GMM RE GMM

δFCF+
(β1) −0.010*** −0.007*** −0.002** −0.001
(−8.805) (−5.837) (−3.002) (−1.255)
δFCF−
(β2) 0.017*** 0.015*** 0.020*** 0.020***
(8.224) (7.525) (12.451) (14.038)
DEV.δDEV−.δCF+ (β3) 0.231*** 0.186*** 0.453*** 0.456***
(15.127) (15.908) (20.572) (21.288)
DEV.δDEV−.δCF− (β4) 0.136*** 0.093*** 0.285*** 0.273***
(5.534) (4.152) (7.838) (8.036)
DEV.δDEV+.δCF+ (β5) 0.188*** 0.139*** 0.217*** 0.169***
(9.467) (7.769) (8.198) (7.880)
DEV.δDEV+.δCF− (β6) 0.204*** 0.150*** 0.120* 0.123**
(5.378) (5.330) (2.336) (2.737)
R2 0.173 0.148 0.256 0.257
Test 2 t-statistic (H0: β3 ≤ β4) 3.339 3.654 4.173 4.806
Test 2 p-value 0.000 0.000 0.000 0.000
Test 3 t-statistic (H0: β6 ≤ β5) 0.381 0.367 −1.792 −0.995
Test 3 p-value 0.352 0.357 0.963 0.840
N obs 8755 8755 8755 8755

· p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001 (t-statistic in parentheses).

than when the firm is in an under-levered situation, implying a faster speed of adjustment when the firm is over-levered”. The SOA of over-
levered firms should be higher than the SOA of under-levered firms. Literature often highlights the existence of this asymmetry.
Byoun (2008), Lockhart (2014), Faulkender et al. (2012), Mukherjee and Wang (2013) and Dang et al. (2014) provide evidence that
the SOA is greater for over-levered firms.
Panel B of Table 5 shows the following findings.

69
D. Dufour et al. Research in International Business and Finance 45 (2018) 62–71

For the short-term debt ratio and when using system GMM to estimate the TL, the coefficient estimate corresponding to the SOA of
an over-levered SME (λ1) is 0.426 (p < 0.001) and the coefficient estimate corresponding to the SOA of an under-levered firm (λ2) is
0.191 (p < 0.001). The difference in the estimates is highly significant: we performed a t-test (Test 1) that led to a strong rejection of
the null hypothesis that λ1 is smaller than λ2 (p < 0.001). When we use RE to estimate the TL, the same conclusion holds. The
parameter estimate λ1 is equal to 0.425 (p < 0.001), λ2 is equal to 0.234 (p < 0.001) and the difference is very significant (Test 1 p-
value is below 0.1%). This result is in line with evidence from the literature.
Conversely, when we take interest in the long-term debt ratio, the asymmetric adjustment leads to a different and interesting
result: in our sample of SMEs, we do not observe significant differences in adjustment speeds. Indeed, for the long-term debt ratio and
when using system GMM to estimate the TL, the coefficient estimate λ1 is 0.180 (p < 0.001) and the coefficient estimate λ2 is 0.161
(p < 0.001). The SOA appears greater for over-levered SMEs than for under-levered SMEs, but this difference is not statistically
significant. A t-test does not lead to the rejection of the null hypothesis that λ1 is smaller than λ2 (p = 0.212). When we use FE to
estimate the TL, λ1 is equal to 0.215 (p < 0.001), λ2 is equal to 0.193 (p < 0.001) and the difference is still insignificant (the Test 1
t-test’s p-value is equal to 0.215).
How can we explain this specific behaviour related to long-term leverage adjustment? Transaction costs related to long-term debt
can be considered higher than transaction costs related to short-term debt (Hall et al., 2004). One might also say that SMEs incur
higher transaction costs related to long-term leverage than listed companies because SMEs do not have the same skills and bargaining
power to renegotiate debts. Consequently, the reduction of long-term debt looks much more difficult than the reduction of short-term
debt (Aybar-Arias et al., 2012). These remarks are even more relevant in the French context, which is characterized by high
transaction costs (Psillaki and Daskalakis 2009). All of these reasons can explain why we found no significant difference in the SOA
for long-term leverage between over-levered and under-levered firms.
Panel C reports the estimation results for the adjustment model specified by Eq. (5). The aim of this section is to test the influence
of the sign of the cash flow on the adjustment speeds of over-levered firms and under-levered firms.
Let us first consider the over-levered situation. The coefficient estimates on the above-target debt with a positive cash flow (β3)
are between 0.186 and 0.456, whereas the coefficient estimates on the above-target debt with a negative cash flow (β4) are between
0.093 and 0.285. A t-test with the null hypothesis that the parameter estimate β3 is smaller than the parameter estimate β4 was
conducted. For all of our models, we found evidence that β4 is significantly smaller than β3 (in each case, p-values associated with
Test 2’s t-test are lower than 0.1%). When a firm is over-levered, cash flow realizations influence adjustment speed: SOA is higher for
firms with a positive cash flow than for firms with a negative cash flow. It must be noted that this finding is validated for both
leverages. This result is in line with Byoun (2008).
In Byoun’s study, as in ours, transaction costs play a central role. A positive CF incurs low transaction costs. This statement is true
not only in the case of listed companies but also in the case of SMEs (Psillaki and Daskalakis 2009). A positive CF allows an over-
levered firm to adjust its leverage at low costs by using part of this CF to refund debts. This possibility is even more important for
SMEs, which incur higher transaction costs than do listed companies. We found evidence of a greater SOA for over-levered firms
experiencing positive CF.
Next, we consider under-levered firms. The coefficient estimates on the below-target debt with a positive cash flow (β5) range
between 0.139 and 0.217, whereas the coefficient estimates on the below-target debt with a negative cash flow (β6) are between
0.120 and 0.204. The adjustment speed of firms with a negative cash flow is not higher than adjustment speed of firms with a positive
cash flow (all of the p-values associated with a t-test of difference in Test 3 are above 35%). Our results show that a negative CF does
not lead to a faster adjustment to the TL. Here again, both leverages lead to the same conclusion. These results are different from
those highlighted in samples of listed companies (Byoun 2008; Faulkender et al., 2012).
Byoun posits that an under-levered firm will adjust its leverage more rapidly when facing a negative cash flow than when facing a
positive cash flow: a listed company will prefer to adjust its leverage by increasing its debt instead of by reducing its equity. Although
a negative CF does not reduce its transaction costs, it might enhance a firm’s motivation to increase its leverage to reduce its deviation
costs. It is reasonable to think that SMEs are less concerned than listed companies with deviation cost when they are under-levered.
That may explain why we do not observe a faster SOA in our sample.

4. Summary and conclusions

In this paper, we analyse the speed of SMEs’ adjustment to their optimal capital structure. We emphasize the influence of cash
flow realizations on the speed of adjustment of both under-levered firms and of over-levered firms using a panel consisting of French
SMEs. To our knowledge, our work is the first to address this influence in the context of SMEs. The French setting is characterized by
high transaction costs related to funding given the insufficient protection offered to creditors and equity investors. Like Aybar-Arias
et al. (2012), our work notes that transaction costs significantly affect SMEs’ speed of adjustment to their target structure.
A study of the impact of cash flow appears relevant because a positive cash flow result in a low transaction cost.
We drove this analysis by distinguishing between short-term leverage and long-term leverage. Transaction costs are supposed to
be lower for short-term debt in the case of SMEs. We used two methods to estimate TL. We compared the adjustment speeds of over-
levered firms and under-levered firms. Several interesting results were obtained.
For short-term leverage, our results are in line with the evidence in the literature. Over-levered firms adjust more quickly than
under-levered firms. Conversely, we find no difference between the speeds of adjustment to long-term leverage. This result is different
from those obtained from samples of listed companies. SMEs’ transaction costs related to long-term debt can be considered higher
than those incurred through short-term leverage management. This difference makes the management of such costs more difficult for

70
D. Dufour et al. Research in International Business and Finance 45 (2018) 62–71

SMEs because they do not enjoy the same level of expertise as listed companies in this particular area. This could explain the results
obtained in the context of French SMEs and could encourage policy-makers to facilitate access to and the renegotiation of long-term
debt.
We also studied the effect of cash flow on adjustment speed by distinguishing four situations depending on whether cash flow is
positive or negative and whether the company is over-levered or under-levered. For over-levered firms, our statistical results showed
that the speed of firms with a positive cash flow is higher than the speed of firms with a negative cash flow. This result is in line with
evidence from the literature.
For under-levered firms, our results are different from the results observed among samples of listed companies. Our results do not
show that negative cash-flow firms adjust faster than positive cash-flow firms. For over-levered firms, transaction costs can be
reduced by a positive CF. In the case of under-levered firms, a negative CF does not seem to result in a decrease in transaction costs.
Many avenues of research remain open. The presence of a large number of companies with no leverage could be investigated. This
issue has begun to be considered in the context of listed companies (Bessler et al., 2013) and it would be interesting to attempt to
address it in the context of SMEs. It would also be possible to attempt to integrate other explanatory factors such as the desire to
maintain financial flexibility or even the desire of many SME owners to refuse intrusion in their business. The challenge is to integrate
these explanatory factors into standard analytical frameworks. It might be possible to adopt a broader view of financial debt by
considering it in its net dimension, in other words, by withdrawing cash holdings present in the company’s assets. It must be noted
that cash management is global and concerns both cash holdings and short-term debt.

References

Ang, J., 1992. On the theory of finance for privately held firms. J. Small Bus. Finance 3, 185–203.
Arellano, M., Bond, S., 1991. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Rev. Econ. Stud. 58 (2),
277–297.
Aybar-Arias, C., Casino-Martınez, A., Lopez-Gracia, J., 2012. On the adjustment speed of SMEs to their optimal capital structure. Small Bus. Econ. 39, 977–996.
Bessler, W., Drobetz, W., Haller, R., Meier, I., 2013. The international zero-leverage phenomenon. J. Corp. Finance 23, 196–221.
Bhaird, C., Lucey, B., 2010. Determinants of capital structure in Irish SMEs. Small Bus. Econ. 35, 357–375.
Blundell, R., Bond, S., 1998. Initial conditions and moment restrictions in dynamic panel data models. J. Econom. 87 (1), 115–143.
Bond, S.R., 2002. Dynamic panel data models: a guide to micro data methods and practice. Port. Econ. J. 1 (2), 141–162.
Byoun, S., 2008. How and when do firms adjust their capital structures toward targets? J. Finance LXIII (6).
Cassar, G., Holmes, S., 2003. Capital structure and financing of SMEs: Australian evidence. Account. Finance 43 (2), 123–147.
Dang, V.A., Kim, M., Shin, Y., 2014. Asymmetric adjustment toward optimal capital structure: evidence from a crisis. Int. Rev. Financial Anal. 33, 226–242.
Drobetz, W., Schilling, D.C., Schröder, H., 2015. Heterogeneity in the speed of capital structure adjustment across countries and over the business cycle. Eur. Financial
Manage. 21, 936–973.
European Commission, 2015. Survey on the access to finance of enterprises (SAFE). Analytical Report 2015. p.172.
Faulkender, M.W., Flannery, M.J., Hankins, K.W., Smith, J.M., 2012. Cash flows and leverage adjustments. J. Financ. Econ. 103, 632–646.
Getzmann, A., Lang, S., Spremann, K., 2014. Target capital structure and adjustment speed in asia. Asia-Pac. J. Financ. Stud. 43, 1–30.
Hall, G., Hutchinson, P., Michaelas, N., 2004. Determinants of the capital structures of European SMEs. J. Busi. Finance Account. 31 (5–6), 711–728.
Heyman, D., Deloof, M., Ooghe, H., 2008. The financial structure of private held Belgian firms. Small Bus. Econ. 30, 301–313.
Holmes, S., Kent, P., 1991. An empirical analysis of the financial structure of small and large Australian manufacturing enterprises. J. Small Bus. Finance 1 (2),
141–154.
Lockhart, G.B.1, 2014. Credit lines and leverage adjustments. J. Corp. Finance 25, 274–288.
Mateev, M., Poutziouris, L., Ivanov, K., 2013. On the determinants of SME capital structure in central and eastern Europe: a dynamic panel analysis. Res. Int. Bus.
Finance 27 (1), 28–51.
Michaelas, N., Chittenden, F., Poutziouris, P., 1999. Financial policy and capital structure choice in U.K. SMEs: empirical evidence from company panel data. Small
Bus. Econ. 12, 113–130.
Mukherjee, T., Wang, W., 2013. Capital structure deviation and speed of adjustment. Financ. Rev. 48 (4), 597–615.
Myers, S.C., 1984. The capital structure puzzle. J. Finance 39 (3), 575–592.
Ozkan, A., 2001. Determinants of capital structure and adjustment to long run target: evidence from UK company panel data. J. Bus. Finance Account. 2 (1–2).
Oztekin, O., Flannery, M.J., 2012. Institutional determinants of capital structure adjustment speeds. J. Financ. Econ. 103, 88–112.
Palacin-Sânchez, M., Ramirez-Herrera, L., Di Pietro, F., 2013. Capital structure of SMEs in Spanish regions. Small Bus. Econ. 41, 503–519.
Psillaki, M., Daskalakis, N., 2009. Are the determinants of capital structure country or firm specific? Small Bus. Econ. 33, 319–333.
Roodman, D., 2006. How to Do Xtabond2: An Introduction to Difference and System GMM in Stata. Center for Global Development working paper (103).
Sogorb-Mira, F., 1995. How SME uniqueness affects capital structure: evidence from a 1994–1998 Spanish data panel. Small Bus. Econ. 25, 447–457.
Warr, R., Elliott, W., Koeter-Kant, J., Öztekin, O., 2012. Equity mispricing and leverage adjustment costs. J. Financ. Quant. Anal. 47 (3), 589–616.
Welch, I., 2004. Capital structure and stock returns. J. Polit. Econ. 112, 106–131.

71