You are on page 1of 32

The European Journal of Finance

ISSN: 1351-847X (Print) 1466-4364 (Online) Journal homepage: http://www.tandfonline.com/loi/rejf20

Financial constraints and asset pricing:


comprehensive evidence from London Stock
Exchange

Nikolaos Balafas & Alexandros Kostakis

To cite this article: Nikolaos Balafas & Alexandros Kostakis (2017) Financial constraints and
asset pricing: comprehensive evidence from London Stock Exchange, The European Journal of
Finance, 23:1, 80-110, DOI: 10.1080/1351847X.2015.1115773

To link to this article: http://dx.doi.org/10.1080/1351847X.2015.1115773

Published online: 31 Dec 2015.

Submit your article to this journal

Article views: 82

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=rejf20

Download by: [Bahauddin Zakariya University] Date: 27 November 2016, At: 20:00
The European Journal of Finance, 2017
Vol. 23, No. 1, 80110, http://dx.doi.org/10.1080/1351847X.2015.1115773

Financial constraints and asset pricing: comprehensive evidence from London


Stock Exchange

Nikolaos Balafasa and Alexandros Kostakisb


a University of Liverpool Management School, Chatham Building, Liverpool L69 7ZH, UK; b The Manchester Business

School, University of Manchester, Booth Street, Manchester M15 6PB, UK

(Received 3 February 2013; final version received 2 October 2015)

This study provides comprehensive evidence on the pricing of financial constraints (FC) risk on London
Stock Exchange during the period 19882013. Utilizing a large number of proxies for FC, we find that
investors are not compensated with higher premia for holding shares of financially constrained firms. To
the contrary, in most of the cases, the most constrained firms significantly underperform, both statistically
and economically, the least constrained ones. Focussing on the WhitedWu index to construct a zero-cost
FC factor that goes long the most constrained firms and sells short the least constrained ones, we find that
this factor carries a significantly negative premium and it is priced in the cross-section over and above the
commonly used risk factors.

Keywords: asset pricing; financial constraints; London Stock Exchange; risk-adjusted performance

JEL classification: G12; G15

1. Introduction
The recent global financial crisis has highlighted the importance of capital markets well-
functioning and the availability of credit to firms. The classical result of Modigliani and Miller
(1958), that the firms capital structure decisions are irrelevant for its value under perfect and
frictionless markets, may still be theoretically appealing but it crucially fails in the real world
of imperfect capital markets. Agency costs, asymmetric information and market incomplete-
ness may cause a prolonged disruption in credit markets, to the extent that external finance may
become impossible to acquire (see the seminal studies of Jaffee and Russell 1976; Jensen and
Meckling 1976; Stiglitz and Weiss 1981; Myers and Maljuf 1984). There is a well-established
strand of the corporate finance literature that recognizes the existence of frictions in firms finan-
cial decision-making and examines their impact on corporate investment policy. The frictions
that prevent the firm from funding all desired investments are termed as financial constraints
(FC) (Lamont, Polk, and Saa-Requejo 2001, 529).
The seminal studies of Fazzari, Hubbard, and Petersen (1988), Kashyap, Lamont, and Stein
(1994) and Hubbard (1998) have demonstrated that firms FC do affect their financing, invest-
ment decisions and hence their profits. Similar is the evidence provided for UK firms in a series
of recent studies by Guariglia and Mateut (2006), Carpenter and Guariglia (2008) and Guariglia
(2008). Accepting that external finance (e.g. new equity or debt issue) is not a perfect substitute

Corresponding author. Email: alexandros.kostakis@mbs.ac.uk

2015 Taylor & Francis


The European Journal of Finance 81

for internal finance (e.g. retained profits), in the sense that it may be costlier and scarcer, obliges
us to take into account firms capital structure as well as their financial status. Interestingly,
theories of capital market imperfections have important implications not only for firms invest-
ment decisions but also for their asset prices (see, e.g. the credit cycle model of Kiyotaki and
Moore 1997). In line with this literature, our study examines the asset pricing implications of
FC risk. The conventional asset pricing hypothesis is that financially constrained firms should
yield a higher rate of return relative to the unconstrained firms as compensation to risk-averse
investors. This is due to the fact that FC are typically binding during high marginal utility (bad
economic) states, causing highly constrained firms to perform poorly in these states.
The concept of FC is rather elusive, since a firm may become constrained either due to the
size, liquidity and structure of the assets and liabilities of its balance sheet or due to the level
and the variability of its cash flows. To capture this aspect of risk and to provide comprehensive
evidence, we utilize a plethora of proxies defined by these stock and flow variables as well as their
combinations. These proxies refer to firms total assets, debt-to-book equity and debt-to-market
value ratios, cash holdings-to-assets ratio, interest coverage ratio, debt capacity as measured by
tangible assets as well as two composite indices of FC: the KZ index according to Kaplan and
Zingales (1997) and the WW index according to Whited and Wu (2006). Each of these two
indices combines linearly the previously mentioned as well as other accounting and market firm
characteristics into a single measure of the degree of firms FC. The higher the value of the index,
the more constrained the firm is.
The issue we examine has important implications for firms optimal capital structure (see
Myers 2001 for an overview of the topic and Beattie, Goodacre, and Thomson 2006 for a recent
survey of UK firms). A crucial feature of all these theories is the relationship between leverage,
or more generally external finance, and the required rate of return on the securities issued by
the corporation. For example, the trade-off theory predicts an optimal target level of debt-to-
equity ratio, assuming that any increase beyond this level would actually increase the cost of
both equity and debt, offsetting the benefits from substituting costly equity for cheaper debt (see
Myers 1984). In other words, it implicitly assumes that the risk of FC will be priced by the
market. Our study essentially tests this crucial assumption too.
A series of previous studies have tested whether FC constitute a priced source of risk yielding
puzzling results. Lamont, Polk, and Saa-Requejo (2001) employed the composite KZ index as a
measure of FC and found that the most constrained firms actually yielded lower returns relative
to the least constrained ones. Along the same lines, Gomes, Yaron, and Zhang (2006), relying
on an investment-based model, showed that financing frictions play only a negligible role for
the pricing of cross-sectional premia. More recently, Campello and Chen (2010) confirmed that
a FC factor constructed on the basis of the KZ index does not yield a significant premium. They
only found some evidence of FC risk being priced on the basis of an ad hoc composite factor
that combines various proxies. Moreover, they showed that the returns of such a FC factor are
correlated with macroeconomic movements, indicating that financially constrained firms have
higher systematic risk relative to the unconstrained ones.
Whited and Wu (2006), introducing their novel WW index as a proxy for FC, found that
the most constrained firms yielded higher returns relative to the least constrained ones, but
the premium of the corresponding factor was not significant. On a related issue, Hahn and
Lee (2009) provided evidence that debt capacity, proxied by asset tangibility, is priced only
within the subgroup of financially constrained firms, classified as such on the basis of their asset
size, payout ratio and bond ratings. To the contrary, tangibility is not priced in the subgroup of
unconstrained firms. Therefore, they argue that it is the interaction of firms debt capacity with
82 N. Balafas and A. Kostakis

high FC that is priced, but they do not examine whether FC risk per se is priced. Finally, Liv-
dan, Sapriza, and Zhang (2009) found a positive relationship between the degree of firms FC
and their expected stock returns using simulated data from an extended investment-based asset
pricing model, highlighting the importance of collateral constraints that restrict firms flexibility
to smooth dividend streams.1
All of the previously mentioned studies focus on the US market. With respect to international
markets, Chan et al. (2010), using discriminant analysis on the basis of changes in firms dividend
payout, found that constrained firms yielded lower returns than their unconstrained counterparts
in the Australian stock market and that a FC factor is systematically priced, carrying a negative
premium. Moreover, examining the same issue in the Chinese stock market, Ling and Chen
(2012) also found that a FC factor is priced beyond the commonly used risk factors but it carries
a positive premium.
Our study contributes to the literature in three main ways. Firstly, to the best of our knowledge,
our study is the first to provide comprehensive evidence on the pricing of FC risk on London
Stock Exchange (LSE).2 Secondly, we provide comprehensive evidence using a large number of
FC proxies to capture this elusive concept, while previous studies have only examined subsets
of these proxies for other markets. Thirdly, we examine the pricing of FC using a sample period
that includes the recent global financial crisis. The global financial crisis period provides an
ideal setup to examine the pricing of FC risk, since we expect that the impact of FC would be
exacerbated during the crisis period as credit supply is reduced, the cost of borrowing is increased
and firms cash flows are reduced.
Utilizing a carefully constructed dataset of shares listed on LSE during the period 19882013,
we sort firms on the basis of a series of alternative FC measures into portfolios. The post-ranking
returns of these portfolios are calculated to examine the potentially differential behaviour of
the most constrained firms versus the least constrained ones. Standard asset pricing tests are
conducted to examine whether risk factors such as market, size, value and momentum can help
explain the cross-section of these portfolios returns. Our key finding is that the most constrained
firms do not yield higher average returns relative to the least constrained ones. To the contrary, it
is typically the most constrained firms that puzzlingly yield lower risk-adjusted returns relative
to the least constrained ones.
A series of further checks confirm the robustness of the relative underperformance of the most
constrained firms. This pattern is highly statistically and economically significant particularly
when the WW index is used as a proxy. Motivated by the latter finding, we construct a zero-
cost FC factor that goes long the most constrained firms and sells short the least constrained
ones, using independently double-sorted portfolios on firms WW values and size. This factor
is found to carry a significantly negative premium, equal to 13.76% p.a. (t-value = 4.34),
while FamaMacBeth tests show that this factor captures a systematic source of risk over and
above the commonly used factors and that it is significantly priced in the cross-section of stock
returns. Moreover, this factor yielded extremely negative returns during the 20072009 global
financial crisis period. Even though the underperformance of the most constrained firms relative
to the least constrained ones was aggravated during the financial crisis, it was also economically
and statistically significant even before the crisis.
The rest of this study is structured as follows. Section 2 provides the details of the utilized
proxies of FC and discusses various data issues. Section 3 presents the characteristics of the
decile portfolios and their risk-adjusted performance and presents a number of robustness checks.
In Section 4, we construct a FC factor and examine its performance as well as its ability to price
the cross-section of WW-sorted portfolios returns, while Section 5 concludes.
The European Journal of Finance 83

2. FC proxies and data issues


Our initial dataset consists of all common shares listed on LSE that are available in Datastream
(Thomson Reuters) from 1988 to 2013. This is the earliest starting period for which all account-
ing data that are necessary to calculate the utilized FC proxies are available in Worldscope for a
sufficient number of UK firms. Following common practice in the literature (see, e.g Florackis,
Gregoriou, and Kostakis 2011), we include both listed and de-listed firms, so that our dataset is
free of any potential survivorship bias. We then exclude unit trusts, investment trusts and Amer-
ican Depositary Receipts. We impose several screening criteria to our initial sample, excluding
firms for which the necessary accounting and market data are not available for a year. Moreover,
financial and insurance companies are excluded, since their capital structure is fundamentally dif-
ferent, and hence they are not comparable with the rest firms. We end up with a final sample of
2743 firms for the entire period. Monthly total returns inclusive of dividends are calculated using
the RI datatype. Particular attention is also paid to a firms delisting reason. Utilizing the London
Share Price Database (LSPD) and following Soares and Stark (2009), we set the return in the
delisting month equal to 100% when a shares death code is assigned by LSPD as 7, 14, 16, 20
or 21. The market portfolio returns are proxied by the FTSE All Share Index returns and the risk-
free rate by the UK interbank rate. For the benchmark asset pricing tests, we use the size, value
and momentum factors constructed for the UK by Gregory, Tharyan, and Christidis (2013).3
We use a series of proxies to measure the degree of FC for each firm in our sample. These prox-
ies utilize information embedded into the assets and the liabilities side of the firms balance sheet
as well as its cash flows. In particular, the following measures are used: firms size proxied by
the book value of its total assets, its ratio of tangible-to-total assets as a measure of debt capacity,
its total debt-to-common equity and total debt-to-market value ratios, its cash holdings-to-total
assets ratio, its interest coverage ratio, the composite KZ index proposed by Kaplan and Zingales
(1997) and the composite WW index proposed by Whited and Wu (2006).4 Table 1 presents the
list of these measures along with their definitions and the corresponding Worldscope/Datastream
codes used to calculate them.5
It should be also noted that there is great variation in the fiscal year end for firms whose shares
are listed on LSE. In other words, contrary to the common practice in US studies, December
accounting values by no means correspond to fiscal year end values for all UK firms. As a result,
if we calculate FC proxies using December accounting values for the entire subsequent year,
we run the risk of relying on far outdated data for the firms that have fiscal year end in the
subsequent months, because December values would correspond to the previous fiscal year in
this case. Soares and Stark (2009, 326327) discuss analytically this issue. To avoid this problem
and to use the most up-to-date information, we update these proxies on a monthly basis. Finally,
following Soares and Stark (2009) to ensure that accounting data are publicly available, and
hence they could be used by an investor in real time to construct portfolios, we lag them by 6
months.6 For example, if the firms fiscal year end is in April, the accounting variables can be
firstly used to calculate the FC proxies and construct portfolios in October.
Having calculated these FC measures for each firm i and each month t, in our benchmark
analysis we construct decile portfolios by sorting the available shares on the basis of each of
these alternative measures. Portfolio 1 (P1) contains the shares of the least constrained firms,
while Portfolio 10 (P10) contains the shares of the most constrained firms in each case. Next, for
each portfolio we calculate post-ranking (i.e. next month, t + 1) returns in excess of the risk-free
rate. Portfolios are rebalanced on a monthly basis. For robustness, we have also repeated the
analysis with annual rebalancing using December accounting values. Results are qualitatively
similar to the ones reported and they are readily available upon request.
84 N. Balafas and A. Kostakis

Table 1. Definitions of FC proxies.

FC measure Definition Data items used

(1) Total assets Book value of total assetst Worldscope item:


WC02999
Tangible assetst
(2) Tangible-to-total assets ratio Worldscope item:
Total assetst
WC02501
Total debtt
(3) Total debt-to-common equity Worldscope item:
ratio Book value of common equityt
WC08231
Total debtt
(4) Total debt-to-market value Worldscope item:
ratio Market valuet
WC03255 and MV
Cash holdingst
(5) Cash holdings-to-total assets Worldscope item:
ratio Total assetst
WC02001 and
WC02999
EBITt
(6) Interest coverage ratio Worldscope item:
Total interest expense ratiot
WC08291
Cash flowt
KZt = 1.002
Prop, plant and equipt1
+0.283 Tobin sQt
Total debtt
+3.139
(7) KZ index Total capitalt Worldscope item:
Dividends paidt WC01250, WC01151,
39.368 WC02501, WC02999,
Prop, plant and equipt1
WC03501, WC03451,
Cash holdingst
1.315 MV, WC03255,
Prop, plant and equipt1 WC03998, WC04551
and WC02001
Cash Flowt
WWt = 0.091
Total Assetst
0.062 Dividend dummyt
Longterm Debtt
(8) WW index +0.021 Worldscope item:
Total Assetst
WC01250, WC01151,
0.044 ln(Total Assetst ) WC02999, WC04551,
+0.102 Industry Sales Growtht WC03251, WC01001
0.035 FirmSales Growtht and FTAG3 for industry
classification

3. FC-sorted portfolios
3.1 Portfolio characteristics
We begin the discussion of our empirical results by presenting the characteristics of the decile
portfolios constructed on the basis of each of the eight alternative proxies of firms FC.
Tables 25 contain the average equally- and value-weighted post-ranking excess portfolio returns
during the full sample period July 1988December 2013, the average value of the financial con-
straint proxy across firms in each portfolio and their market value. Moreover, we report the full
sample capital asset pricing model (CAPM) beta estimate for each portfolio. In each case, P1
always stands for the portfolio of the least constrained firms and P10 for the portfolio of the
most constrained firms according to each proxy. The pre-last column in each table reports the
The European Journal of Finance 85

difference between P10 and P1. The last column contains the value of the t-test under the null
hypothesis that the characteristic of portfolio P10 is equal to the characteristic of portfolio P1.
The first sorting criterion we examine is the book value of firms total assets and the
corresponding results are reported in Panel A of Table 2. Interestingly, the post-ranking value-
weighted excess returns of the portfolios containing the most constrained firms are much lower
than the corresponding returns of the portfolios with the least constrained ones. The annualized
spread between the extreme decile portfolios (P10 P1) is significantly negative and equal to
12.78%. Panel B of Table 2 contains similar information for portfolios constructed on the basis
of the firms ratio of tangible-to-total assets. Focussing on value-weighted returns, the most con-
strained firms tend to underperform the least constrained ones. The spread between the extreme
decile portfolios is equal to 4.29% p.a.
In Table 3, we utilize leverage ratios as proxies of FC. Panel A uses the total debt-to-common
equity ratio, while Panel B employs the total debt-to-market value ratio.7 We assign firms with
zero leverage ratios to portfolio P0, separating them from portfolios P1 to P10 that contain firms
with strictly positive leverage ratios.8 There is a large cross-sectional variation in firms degree
of leverage. In general, the firms with the lowest leverage ratios typically have very low market
values. However, there is no monotonic size pattern as we move towards portfolios of firms with
high leverage ratios. Regarding their returns, there is no particular relationship between average
premia and leverage ratios. Contrary to conventional wisdom, highly leveraged firms, though
appearing as more risky, do not yield higher premia relative to low-leverage firms.
Table 4 reports the corresponding results, using as sorting criterion cash holdings-to-total
assets ratios in Panel A and interest coverage ratios in Panel B. Panel A shows that there is
no particular relationship between average size, CAPM beta or excess returns and firms cash
holdings. Though there is a huge variation across firms cash holdings ratios, these differences
are related neither to the riskiness of these firms shares nor the premia that they yield. In other
words, the shares of cash-poor firms that appear as highly constrained did not yield higher premia
to reward investors who would be averse to withholding them. Similar are the results in Panel B.9
The most interesting finding is that P10, containing the firms with the lowest interest coverage
ratios, significantly underperformed P1 that contains the firms with the highest interest coverage
ratios. In particular, the strategy P10 P1 that goes long the most constrained firms and sells
short the least constrained firms would yield a significantly negative return of 14.28% p.a.
Arguably, the previously utilized proxies reveal only some aspects of the FC that a firm may
face. To overcome this potential limitation, we also utilize two composite indices that have been
suggested in prior literature in an attempt to simultaneously capture all relevant dimensions
of FC: the KZ index and the WW index. Table 5 presents the characteristics of the portfolios
constructed according to each of these two indices. Utilizing the KZ index in Panel A, there
seems to be a hump-shaped relationship between KZ index values and firms capitalizations,
with larger firms classified in the middle deciles. Moreover, though there is no monotonic rela-
tionship between average excess returns and KZ index values, the portfolio containing the most
constrained firms, P10, significantly underperforms portfolio P1 with the least constrained firms.
The spread return P10 P1 is significantly negative, equal to 10.39% p.a.
Even more interesting are the findings reported in Panel B of Table 5 utilizing the WW index.
We find a stark size pattern as we move from the portfolios of the least constrained firms, which
are typically very large capitalization firms, to the portfolio of the most constrained and typically
small capitalization firms. Most importantly for our study, we report an almost monotonic decline
in average excess returns as we move from P1 to P10. The most constrained firms severely
underperformed the least constrained ones. In particular, a spread strategy P10 P1 would yield
86
N. Balafas and A. Kostakis
Table 2. Characteristics of decile stock portfolios constructed on the basis of total assets and tangible-to-total assets ratio.

P1 P2 P3 P4 P5 P6 P7 P8 P9 P10 P10 P1 t-test


Least Constrained Most Constrained

Panel A: Total assets


EW returns (% p.a.) 7.05 6.88 3.86 4.87 4.33 2.50 2.86 0.92 4.85 6.64 0.42 0.09
VW returns (% p.a.) 4.99 4.93 2.70 3.73 4.68 4.74 0.16 2.32 6.44 7.79 12.78 2.71
Total assets (m) 8328.55 648.28 227.99 112.07 62.72 38.00 23.40 14.09 7.61 2.82 8325.73 26.90
MV (m) 6443.34 665.62 236.04 135.44 70.77 43.88 28.46 19.29 12.26 7.63 6435.71 31.78
CAPM beta 0.91 1.10 1.01 0.97 0.88 0.90 0.92 0.86 0.73 0.74
Panel B: Tangible/total assets ratio (%)
EW returns (% p.a.) 3.78 4.65 5.03 4.65 2.95 4.84 3.19 4.00 2.53 7.04 3.26 0.93
VW returns (% p.a.) 6.63 6.77 2.76 4.32 6.75 6.43 5.09 3.74 3.81 2.35 4.29 1.14
Tangible/total assets (%) 80.65 57.49 43.03 33.86 26.40 20.55 15.92 11.82 7.82 2.76 77.89 301.56
MV (m) 1155.21 1829.23 713.31 968.38 1068.31 720.51 430.91 338.65 251.77 197.81 957.40 18.57
CAPM beta 0.89 0.95 1.04 0.83 0.90 0.78 0.91 1.16 1.14 1.08

Note: This table reports the characteristics of portfolios constructed on the basis of total assets (Panel A) and tangible-to-total assets ratio (Panel B). All stocks listed on the LSE
from 1988 to 2013 are sorted at month t and they are assigned to 10 portfolios. P1 is the decile portfolio containing the stocks of the least constrained firms (highest values of
total assets in Panel A and highest values of tangible-to-total assets ratio in Panel B) and P10 is the decile portfolio containing the stocks of the most constrained firms (lowest
values of total assets in Panel A and lowest values of tangible-to-total assets ratio in Panel B). The returns of these portfolios in excess of the risk-free rate at month t + 1 are
calculated (i.e. post-ranking returns). P10 P1 stands for the spread return between P10 and P1. Portfolios are rebalanced on a monthly basis. EW returns correspond to the
annualized average monthly excess returns of equal weighted portfolios. VW returns refer to the annualized average monthly excess returns of value-weighted portfolios. MV
is the average market value of the stocks in each portfolio (in m). The last column reports values for t-tests referring to the null hypothesis of no difference in means between
portfolios P10 and P1 characteristics.
Table 3. Characteristics of decile stock portfolios constructed on the basis of Debt-to-Common Equity and Debt-to-Market Value ratios.

P0 P1 P2 P3 P4 P5 P6 P7 P8 P9 P10 P10 P1 t-test


Zero Least Constrained Most Constrained

Panel A: Total debt/common equity ratio (%)


EW returns (% p.a.) 9.65 6.68 7.70 6.12 4.41 7.55 4.25 4.38 2.94 0.68 1.46 8.14 2.90
VW returns (% p.a.) 8.72 6.68 7.01 7.42 4.03 6.37 6.89 4.91 4.32 3.48 5.86 0.81 0.23
Debt/equity (%) 0.00 1.23 6.29 13.96 22.87 32.91 44.54 58.74 78.95 117.01 789.61 788.38 11.01
MV (m) 110.99 183.63 307.92 752.97 784.72 1421.26 1276.29 1080.88 1113.93 1245.09 1251.80 1068.17 18.40
CAPM beta 1.08 1.04 0.91 0.94 1.08 0.99 0.99 1.01 0.93 0.82 0.89
Panel B: Total debt/market value ratio (%)
EW returns (% p.a.) 9.61 3.75 4.15 3.78 3.92 3.11 3.35 3.00 1.79 2.87 7.36 3.61 0.94
VW returns (% p.a.) 8.63 4.76 2.63 7.32 4.88 3.93 5.95 8.22 5.44 7.57 5.07 0.31 0.05
Debt/ MV (%) 0.00 0.57 3.14 7.32 12.60 19.29 27.48 38.43 55.22 88.91 699.65 699.08 13.02
MV (m) 110.46 269.26 469.01 806.28 1543.57 1657.81 1422.83 1330.69 777.45 523.89 161.05 108.21 11.58

The European Journal of Finance


CAPM beta 1.08 0.96 0.79 0.84 0.91 0.91 1.01 0.94 1.01 0.97 1.42

Note: This table reports the characteristics of portfolios constructed on the basis of Debt-to-Common Equity (Panel A) and Debt-to-Market Value ratio (Panel B). All stocks
listed on the LSE from 1988 to 2013 are utilized. P0 contains the firms with zero leverage ratios in each month. The rest firms are sorted at month t and they are assigned to
10 portfolios. P1 is the decile portfolio containing the stocks of the least constrained firms (lowest values of Debt-to-Common Equity ratio in Panel A and lowest values of
Debt-to-Market Value ratio in Panel B) and P10 is the decile portfolio containing the stocks of the most constrained firms (highest values of Debt-to-Common Equity ratio
in Panel A and highest values of Debt-to-Market Value ratio in Panel B). The returns of these portfolios in excess of the risk-free rate at month t + 1 are calculated. P10 P1
stands for the spread return between P10 and P1. Portfolios are rebalanced on a monthly basis. EW returns and VW returns correspond to the annualized average monthly excess
returns of equally weighted and value-weighted portfolios, respectively. MV is the average market value of the stocks in each portfolio (in m). The last column reports values
for t-tests referring to the null hypothesis of no difference in means between portfolios P10 and P1 characteristics.

87
88
N. Balafas and A. Kostakis
Table 4. Characteristics of decile stock portfolios constructed on the basis of cash holdings-to-total asset and interest coverage ratios.

P1 P2 P3 P4 P5 P6 P7 P8 P9 P10 P10 P1 t-test


Least Constrained Most Constrained

Panel A: Cash holdings/total assets ratio (%)


EW returns (% p.a.) 8.19 12.41 7.31 5.86 4.53 3.35 3.42 1.25 0.29 1.92 10.11 2.42
VW returns (% p.a.) 0.44 8.75 4.42 4.86 4.65 5.68 4.94 4.66 3.64 0.09 0.54 0.11
Cash/ T. Assets (%) 58.50 30.28 19.73 13.69 9.71 6.76 4.45 2.59 1.15 0.19 58.31 79.99
MV (m) 159.16 349.00 671.43 924.72 909.61 1199.94 1284.42 1039.96 857.51 228.17 69.01 3.45
CAPM beta 1.18 1.10 0.91 0.85 0.90 0.95 0.95 0.98 0.89 0.93
Panel B: Interest coverage ratio
EW returns (% p.a.) 14.35 14.32 11.48 13.07 10.05 9.52 8.63 8.23 2.25 2.94 17.29 7.59
VW returns (% p.a.) 9.12 9.31 6.63 9.81 4.66 7.93 5.92 4.41 0.27 5.17 14.28 4.06
Interest Coverage 85.07 46.65 20.18 12.44 8.57 6.31 4.78 3.52 2.38 1.07 84.00 98.13
MV (m) 358.59 1032.26 1712.63 1377.37 1452.69 1542.88 1271.88 1122.34 687.74 455.22 96.64 3.74
CAPM beta 0.98 0.84 0.92 0.94 0.93 0.85 0.90 0.94 0.90 1.09

Note: This table reports the characteristics of portfolios constructed on the basis of cash holdings-to-total assets ratio (Panel A) and interest coverage ratio (Panel B). All stocks
listed on the LSE from 1988 to 2013 are sorted at month t and they are assigned to 10 portfolios. P1 is the decile portfolio containing the stocks of the least constrained firms
(highest values of cash holdings-to-total assets ratio in Panel A and highest values of interest coverage ratio in Panel B) and P10 is the decile portfolio containing the stocks of
the most constrained firms (lowest values of cash holdings-to-total assets ratio in Panel A and lowest values of interest coverage ratio in Panel B). The returns of these portfolios
in excess of the risk-free rate at month t + 1 are calculated (i.e. post-ranking returns). P10 P1 stands for the spread return between P10 and P1. Portfolios are rebalanced on
a monthly basis. EW returns correspond to the annualized average monthly excess returns of equal weighted portfolios. VW returns refer to the annualized average monthly
excess returns of value-weighted portfolios. MV is the average market value of the stocks in each portfolio (in m). The last column reports values for t-tests referring to the
null hypothesis of no difference in means between portfolios P10 and P1 characteristics.
Table 5. Characteristics of decile stock portfolios constructed on the basis of composite KZ and WW indices.

P1 P2 P3 P4 P5 P6 P7 P8 P9 P10 P10 P1 t-test


Least Constrained Most Constrained

Panel A: KZ index
EW returns (% p.a.) 6.78 7.16 5.33 7.34 4.65 7.96 5.60 5.18 0.62 8.56 15.34 5.62
VW returns (% p.a.) 1.97 5.20 6.44 3.32 5.23 4.82 5.45 3.62 4.53 8.42 10.39 2.50
KZ index 25.22 18.90 10.36 6.13 3.69 2.08 0.89 0.03 1.07 8.23 33.45 38.32
MV (m) 279.24 731.37 1430.38 855.43 1053.42 1250.98 1050.48 778.58 439.53 148.92 130.32 16.02
CAPM beta 1.04 0.86 0.78 0.88 0.98 1.03 1.06 0.96 1.03 1.12
Panel B: WW index
EW returns (% p.a.) 5.93 7.02 6.10 6.03 5.92 4.42 2.46 1.43 0.42 2.00 3.93 0.90
VW returns (% p.a.) 5.13 4.07 5.86 4.34 1.65 2.06 0.97 2.92 8.99 15.02 20.15 3.93
WW index 0.54 0.35 0.31 0.28 0.25 0.22 0.19 0.16 0.12 0.02 0.52 25.09
MV (m) 6577.33 768.41 279.04 148.13 84.00 52.96 32.71 21.88 17.30 12.69 6564.65 30.71

The European Journal of Finance


CAPM beta 0.91 1.13 0.98 0.92 0.91 1.00 0.95 0.96 0.85 0.94

Note: This table reports the characteristics of portfolios constructed on the basis of KaplanZingales (Panel A) and WhitedWu (Panel B) composite indices of FC. All stocks
listed on the LSE from 1988 to 2013 are sorted at month t and they are assigned to 10 portfolios. P1 is the decile portfolio containing the stocks of the least constrained firms
(lowest values of KZ index in Panel A and lowest values of WW index in Panel B) and P10 is the decile portfolio containing the stocks of the most constrained firms (highest
values of KZ index in Panel A and highest values of WW index in Panel B). The returns of these portfolios in excess of the risk-free rate at month t + 1 are calculated (i.e.
post-ranking returns). P10 P1 stands for the spread return between P10 and P1. Portfolios are rebalanced on a monthly basis. EW returns correspond to the annualized average
monthly excess returns of equal weighted portfolios. VW returns refer to the annualized average monthly excess returns of value-weighted portfolios. MV is the average market
value of the stocks in each portfolio (in m). The last column reports values for t-tests referring to the null hypothesis of no difference in means between portfolios P10 and P1
characteristics.

89
90 N. Balafas and A. Kostakis

a significantly negative premium of 20.15% p.a. during the examined period. As a result,
using the WW index as a sorting criterion, we report a severe underperformance of the portfolios
containing the most constrained firms. In sum, investors on LSE not only were not rewarded
with higher premia for holding shares of highly constrained firms, but also they have actually
underperformed the market as well as the portfolios of the least constrained firms.

3.2 Risk-adjusted performance


In this subsection, we analyze the performance of FC-sorted portfolios adjusting for their
exposure to a series of commonly used risk factors. In particular, we estimate the abnormal
performance of portfolios P1P10 using three popular asset pricing models. Firstly, we estimate
Jensen alpha from the CAPM regression:
f f
Ri,t Rt = i + i,MKT (Rm,t Rt ) + i,t , (1)
f f
where Ri,t is the return of portfolio i in month t, Rt is the risk-free rate for month t, (Rm,t Rt ) is
the excess market portfolio return in month t and i,MKT is the market beta of portfolio i. Secondly,
we compute FamaFrench alpha, that is, the intercept of the FamaFrench (1993) model:
f f
Ri,t Rt = i + i,MKT (Rm,t Rt ) + i,SMB SM Bt + i,HML HM Lt + i,t , (2)
where SM Bt and HM Lt stand for the size and value factors, respectively, while i,SMB and
i,HML denote the corresponding factor loadings of portfolio i. Thirdly, we estimate Carhart alpha,
that is, the intercept of the four-factor Carhart (1997) model:
f f
Ri,t Rt = i + i,MKT (Rm,t Rt )t + i,SMB SM Bt + i,HML HM Lt + i,MOM MOMt + i,t , (3)
where MOMt stands for the momentum factor and i,MOM denotes the corresponding factor
loading of portfolio i.
We report alphas estimated via GMM with NeweyWest standard errors corrected for het-
eroscedasticity and autocorrelation. Performing a system estimation enables us to test the joint
significance of the 10 portfolios alphas, using a Wald test under the null hypothesis that all
alphas are equal to zero. This Wald test has an asymptotic chi-squared distribution, its func-
tional form is proportional to the finite-sample test of Gibbons, Ross, and Shanken (1989) and
it can help us evaluate the magnitude of each models pricing errors (see Goyal 2012, 8 for a
detailed discussion). We further test and report in each case the significance of the risk-adjusted
performance of the return differential (spread) between the extreme portfolio deciles (P10 P1).
Table 6 reports the estimated alphas using total assets (Panel A) and the ratio of tangible-
to-total assets (Panel B) as sorting criteria. With respect to total assets, we observe that the
portfolios containing the least constrained firms have higher alphas relative to the portfolios
containing the most constrained firms. This holds true for all three asset pricing models. Most
interestingly, the spread P10 P1 that goes long the portfolio of the most constrained firms and
sells short the portfolio of the least constrained ones yields significantly negative alphas. Finally,
the Wald test strongly rejects the null hypothesis of all 10 alphas being equal to zero for all asset
pricing models. Similar is the evidence when the ratio of tangible-to-total assets is used as a
sorting criterion in Panel B. Again, P10 underperforms P1 and the corresponding spread strategy
(P10 P1) yields a negative, though insignificant, abnormal performance. The main difference
with Panel A is that here the null hypothesis of all 10 alphas being equal to zero is not rejected
because the alphas of the middle portfolios are very close to zero and insignificant.
Table 6. Alphas of value-weighted decile stock portfolios constructed on the basis of total assets and tangible-to-total assets ratio.

P1 P2 P3 P4 P5 P6 P7 P8 P9 P10 P10 P1 Chi-sq.


Least Constrained Most Constrained

Panel A: Total assets


CAPM alpha (% 0.69 0.24 2.05 0.83 0.51 0.48 4.18 6.39 9.88 11.30
11.98 31.79
p.a.) (1.09) ( 0.12) ( 0.74) ( 0.26) (0.15) (0.12) ( 0.95) ( 1.35) ( 2.01)** ( 1.90)*
( 1.96)** (0.00)
[0.01]
FamaFrench 0.78 0.58 2.36 1.15 0.04 0.19 4.41 6.81 10.34 11.43 12.21 35.67
alpha (% p.a.) (1.38) ( 0.47) ( 1.62) ( 0.73) (0.02) (0.07) ( 1.66)* ( 2.22)** ( 3.25)*** ( 2.56)** ( 2.77)*** (0.00)
[0.46]
Carhart alpha 0.35 1.09 2.56 1.97 0.75 0.41 5.53 7.96 11.11 12.70 13.06 45.88
(% p.a.) (0.61) ( 0.84) ( 1.60) ( 1.34) ( 0.40) (0.16) ( 2.21)** ( 2.86)*** ( 3.46)*** ( 2.94)*** ( 3.11)*** (0.00)
[0.46]
Panel B: Tangible/total assets ratio (%)
CAPM alpha (% 2.46 2.31 2.16 0.38 2.51 2.75 0.81 1.72 1.57 2.77 5.23 9.67
p.a.) (1.04) (1.39) ( 1.22) (0.23) (1.32) (1.53) (0.36) ( 0.65) ( 0.44) ( 0.88) ( 1.29) (0.47)
[0.02]
FamaFrench 2.19 2.31 2.49 0.59 2.42 2.72 0.60 1.85 1.55 3.13 5.32 10.72
alpha (% p.a.) (1.41) ( 1.59) ( 0.85) ( 0.63) ( 1.13) ( 1.59)

The European Journal of Finance


(1.06) (0.35) (1.26) (1.52) (0.27) (0.38)
[0.31]
Carhart alpha 1.37 0.55 1.35 1.65 2.85 2.23 1.02 0.96 0.58 2.64 4.02 5.14
(% p.a.) (0.56) (0.33) ( 0.79) (0.93) (1.42) (1.12) (0.43) ( 0.48) ( 0.26) ( 1.09) ( 1.25) (0.88)
[0.32]

Note: This table reports the abnormal performance of 10 value-weighted stock portfolios constructed on the basis of total assets (Panel A) and tangible-to-total assets ratio
(Panel B). P1 is the decile portfolio containing the stocks of the least constrained firms (highest values of total assets in Panel A and highest values of tangible-to-total assets
ratio in Panel B) and P10 is the decile portfolio containing the stocks of the most constrained firms (lowest values of total assets in Panel A and lowest values of tangible-to-
total assets ratio in Panel B). P10 P1 stands for the zero-cost strategy that goes long the portfolio of the most constrained firms (P10) and sells short the portfolio with the
least constrained firms (P1). CAPM, FamaFrench and Carhart alphas are the annualized alpha estimates derived from the corresponding asset pricing models. t-statistics are
reported in parentheses. R2 for the spread P10 P1 is reported in square brackets for each model. The last column reports the chi-square ( 2 ) statistic of the Wald test for the
null hypothesis that the 10 portfolios alphas are jointly equal to zero in each case; p-values are reported below the statistic.
*Statistical significance at the 10% level.
**Statistical significance at the 5% level.
***Statistical significance at the 1% level.

91
92 N. Balafas and A. Kostakis

The same asset pricing tests are repeated using portfolios that have been constructed on the
basis of leverage ratios. Panel A of Table 7 reports the results for the portfolios sorted accord-
ing to total debt-to-common equity ratio. In this case, the estimated alphas are quite low and
none is statistically different from zero. All three asset pricing models explain well the per-
formance of these decile portfolios and the Wald tests reported in the last column confirm this
argument. These results show that investors were by no means rewarded for holding highly lever-
aged firms. As expected, very similar are the results reported in Panel B of Table 7 using the total
debt-to-market value ratio. Alphas across decile portfolios are typically statistically insignificant,
confirming the ability of these models to explain the time-series performance of leverage-sorted
portfolios returns. Moreover, we do not find any significant differential performance between
the most and the least constrained firms portfolios (P10 and P1, respectively). This evidence is
in partial contrast with the negative relationship between leverage and future returns documented
for the US market by Penman, Richardson, and Tuna (2007), but this can be explained by the
argument of Caskey, Hughes, and Liu (2012) claiming that it is the deviation from an optimal
leverage target that affects future returns, not the leverage ratio itself.
Table 8 reports the risk-adjusted performance of portfolios sorted on the basis of firms cash
holdings-to-total assets ratios (Panel A) and interest coverage ratios (Panel B). With respect to
cash holdings-sorted portfolios in Panel A, apart from portfolio P10, the rest yielded statistically
insignificant alphas. Overall, the employed asset pricing models are successful in explaining
the performance of these portfolios returns. The performance of the spread strategy P10 P1
is not found to be significantly different from zero. A potential explanation for this finding can
be provided by the argument of Simutin (2010), who claims that it is excess cash relative to a
target ratio that matters for asset pricing, rather than the level of cash holdings. Quite different
are the results reported in Panel B of Table 8 for the portfolios using firms interest coverage
ratios. In this case, there is a clear pattern with portfolios of the most constrained firms (P9 and
P10) yielding significantly negative alphas and portfolios of the least constrained firms (P1
P4) yielding significantly positive alphas in most of the cases. In particular, the spread P10 P1
yields a highly economically and statistically significant negative alpha for all three asset pricing
models used. In other words, the most financially constrained firms according to their interest
coverage ratio severely underperformed their least constrained counterparts and this massive
relative underperformance cannot be explained by differential exposure to market, size, value or
momentum factors.
Having examined the risk-adjusted performance of portfolios constructed according to unilat-
eral proxies of FC, we report in Table 9 the corresponding results for portfolios sorted on the
basis of the composite KZ index and WW index. With respect to KZ-sorted portfolios, results
reported in Panel A show that portfolio P10 containing the most constrained firms performed very
poorly both in statistical and in economic terms, yielding significantly negative alphas. Most
interestingly, the spread strategy P10 P1 yielded significantly negative Jensen, FamaFrench
and Carhart alphas ( 10.78% p.a., 11.06% p.a., 12.91% p.a., respectively). It should be
also noted that even though there is no particular pattern in the sign and the magnitude of alphas
across the KZ-sorted portfolios, Wald tests show that none of these three asset pricing models
can sufficiently explain the joint time-series performance of all 10 portfolios.
Panel B of Table 9 reports the corresponding results using the WW index. This is a very
interesting case because we observe an almost monotonic deterioration in the risk-adjusted per-
formance as we move from the least to the most constrained firms portfolios (from P1 to P10).
Actually, portfolios P8, P9 and P10 exhibit highly negative alphas. Along these lines, the spread
P10 P1 exhibits an annualized Jensen alpha of 20.30% (t-value = 3.30), FamaFrench
Table 7. Alphas of value-weighted decile stock portfolios constructed on the basis of Debt-to-Common Equity and Debt-to-Market Value ratios.

P0 P1 P2 P3 P4 P5 P6 P7 P8 P9 P10 P10 P1 Chi-sq.


Zero Least Constrained Most Constrained

PANEL A: Total Debt/ Common Equity Ratio (%)


CAPM alpha (% p.a.) 3.62 1.77 2.71 2.99 1.05 1.70 2.22 0.14 0.09 0.39 1.68 0.09 10.51
(0.77) (0.53) (1.10) (1.22) ( 0.49) (0.84) (1.17) (0.08) ( 0.05) ( 0.21) (0.89) ( 0.02) (0.40)
[0.02]
FamaFrench alpha (% p.a.) 3.81 1.89 2.65 2.87 1.10 1.57 2.17 0.03 0.24 0.31 1.84 0.05 9.59
(1.19) (0.76) (1.12) (1.13) ( 0.51) (0.79) (1.15) ( 0.02) ( 0.14) ( 0.17) (1.00) ( 0.02) (0.48)
[0.21]
Carhart alpha (% p.a.) 1.93 1.11 2.12 3.16 0.53 0.07 1.89 0.75 0.23 0.17 1.31 0.20 6.28
(0.67) (0.49) (0.82) (1.21) (0.30) ( 0.03) (0.96) ( 0.41) (0.13) (0.08) (0.65) ( 0.06) (0.79)
[0.21]
Panel B: Total Debt/Market Value ratio (%)
CAPM alpha (% p.a.) 3.52 0.23 1.09 3.37 0.59 0.38 1.19 3.80 0.66 3.01 1.62 1.85 9.91
(0.75) (0.07) ( 0.42) (1.66)* (0.39) ( 0.20) (0.77) (1.75)* (0.23) (1.15) ( 0.33) ( 0.30) (0.45)
[0.05]
FamaFrench alpha (% p.a.) 3.72 0.63 0.96 3.62 0.63 0.29 1.11 3.58 0.20 2.49 3.39 4.02 10.14
(1.16) (0.29) ( 0.39) (1.96)** (0.42) ( 0.16) (0.72) (1.73)** (0.08) (1.03) ( 0.97) ( 0.98) (0.43)

The European Journal of Finance


[0.37]
Carhart alpha (% p.a.) 1.84 0.69 1.61 2.04 0.39 1.65 0.60 3.76 1.75 5.34 3.73 4.41 10.29
(0.64) ( 0.31) ( 0.63) (1.08) (0.22) ( 0.81) (0.38) (1.68)* (0.68) (2.19)** (0.88) (0.99) (0.42)
[0.47]

Note: This table reports the abnormal performance of value-weighted portfolios constructed on the basis of Debt-to-Common Equity (Panel A) and Debt-to-Market Value ratios
(Panel B). P0 contains the firms with zero leverage ratios in each month. All the rest firms are assigned to 10 portfolios. P1 is the decile portfolio containing the least constrained
firms (lowest positive values of Debt-to-Common Equity ratio in Panel A and lowest positive values of Debt-to-Market Value ratio in Panel B) and P10 is the decile portfolio
containing the most constrained firms (highest values of Debt-to-Common Equity ratio in Panel A and highest values of Debt-to-Market Value ratio in Panel B). P10 P1 stands
for the zero-cost strategy that goes long the portfolio of the most constrained firms (P10) and sells short the portfolio with the least constrained firms (P1). CAPM, FamaFrench
and Carhart alphas are the annualized alpha estimates derived from the corresponding asset pricing models. t-statistics are reported in parentheses. R2 for the spread P10 P1 is
reported in square brackets for each model. The last column reports the chi-square statistic of the Wald test for the null hypothesis that the 10 portfolios alphas are jointly equal
to zero; p-values are reported below the statistic.
*Statistical significance at the 10% level.
**Statistical significance at the 5% level.
***Statistical significance at the 1% level.

93
94
Table 8. Alphas of value-weighted decile stock portfolios constructed on the basis of cash holdings-to-total assets and interest coverage ratio.

N. Balafas and A. Kostakis


P1 P2 P3 P4 P5 P6 P7 P8 P9 P10 P10 P1 Chi-sq.
Least Constrained Most Constrained

Panel A: Cash holdings/total assets ratio (%)


CAPM alpha (% 5.99 3.55 0.15 0.85 0.40 1.20 0.44 0.05 0.56 4.31 1.68 12.08
p.a.) ( 1.08) (1.29) (0.09) (0.56) (0.21) (0.79) (0.27) (0.03) ( 0.21) ( 1.62) (0.26) (0.28)
[0.02]
FamaFrench alpha 5.75 3.60 0.16 1.04 0.34 1.37 0.39 0.16 0.93 4.85 0.90 18.30
(% p.a.) ( 1.39) (1.58) (0.09) (0.70) (0.18) (0.93) (0.27) ( 0.10) ( 0.37) ( 2.14)** (0.16) (0.05)
[0.17]
Carhart alpha (% 5.96 2.85 0.95 0.99 0.34 0.47 0.03 0.39 1.55 5.17 0.79 13.42
p.a.) ( 1.56) (1.36) (0.53) (0.60) (0.16) (0.35) ( 0.02) ( 0.21) ( 0.56) ( 2.36)** (0.16) (0.20)
[0.17]
Panel B: Interest coverage ratio
CAPM alpha (% 4.22 5.12 2.05 5.13 0.00 3.69 1.42 0.27 4.23 10.60 14.82 42.70
p.a.) (1.63) (2.07)** (1.07) (3.01)*** (0.00) (2.27)** (0.55) ( 0.13) ( 1.87)* ( 3.46)*** ( 3.68)*** (0.00)
[0.01]
FamaFrench alpha 4.07 5.51 2.19 5.28 0.16 3.72 1.29 0.49 4.67 10.82 14.89 47.21
(% p.a.) (1.87)* (2.35)** (1.15) (3.12)*** (0.09) (2.26)** (0.52) ( 0.25) ( 2.29)** ( 3.61)*** ( 3.81)*** (0.00)
[0.02]
Carhart alpha (% 4.82 2.24 1.34 3.83 1.65 3.25 1.72 0.93 4.30 11.15 15.96 33.20
p.a.) (2.28)** (0.90) (0.64) (2.02)** ( 1.01) (1.94)* (0.65) (0.50) ( 1.79)* ( 3.60)*** ( 3.94)*** (0.00)
[0.03]

Note: This table reports the abnormal performance of 10 value-weighted stock portfolios constructed on the basis of cash holdings-to-total assets (Panel A) and interest coverage
ratios (Panel B). P1 is the decile portfolio containing the stocks of the least constrained firms (highest values of cash holdings-to-total assets ratio in Panel A and highest values
of interest coverage ratio in Panel B) and P10 is the decile portfolio containing the stocks of the most constrained firms (lowest values of cash holdings-to-total assets ratio in
Panel A and lowest values of interest coverage ratio in Panel B). P10 P1 stands for the zero-cost strategy that goes long the portfolio of the most constrained firms (P10) and
sells short the portfolio with the least constrained firms (P1). CAPM, FamaFrench and Carhart alphas are the annualized alpha estimates derived from the corresponding asset
pricing models. t-statistics are reported in parentheses. R2 for the spread P10 P1 is reported in square brackets for each model. The last column reports the chi-square ( 2 )
statistic of the Wald test for the null hypothesis that the 10 portfolios alphas are jointly equal to zero in each case; p-values are reported below the statistic.
*Statistical significance at the 10% level.
**Statistical significance at the 5% level.
***Statistical significance at the 1% level.
Table 9. Alphas of value-weighted decile stock portfolios constructed on the basis of composite KaplanZingales and WhitedWu indices.

P1 P2 P3 P4 P5 P6 P7 P8 P9 P10 P10 P1 Chi-sq.


Least Constrained Most Constrained

Panel A: KZ index
CAPM alpha (% 2.96 1.12 2.76 0.86 0.58 0.07 0.42 0.93 0.34 13.74 10.78 13.17
p.a.) ( 1.11) (0.55) (1.55) ( 0.40) (0.30) ( 0.04) (0.23) ( 0.34) ( 0.16) ( 2.66)*** ( 2.21)** (0.21)
[0.00]
FamaFrench 3.13 1.38 3.24 0.85 0.28 0.16 0.31 1.13 0.50 14.19 11.06 30.26
alpha (% p.a.) ( 1.32) (0.81) (2.26)** ( 0.39) (0.15) ( 0.09) (0.17) ( 0.44) ( 0.24) ( 3.40)*** ( 2.42)** (0.00)
[0.06]
Carhart alpha 2.36 1.78 3.13 0.98 0.91 0.55 0.25 2.28 0.08 15.27 12.91 21.25
(% p.a.) ( 1.13) (0.96) (2.03)** ( 0.45) (0.48) ( 0.28) ( 0.12) ( 0.84) ( 0.03) ( 3.47)*** ( 2.67)*** (0.02)
[0.07]
Panel b: WW index
CAPM alpha (% 0.82 1.30 1.20 0.03 2.68 2.68 5.49 7.47 13.03 19.49 20.30 33.66
p.a.) (1.19) ( 0.64) (0.56) ( 0.01) ( 0.78) ( 0.63) ( 1.23) ( 1.58) ( 2.47)*** ( 3.35)*** ( 3.30)*** (0.00)
[0.00]
FamaFrench 0.91 1.62 0.83 0.46 2.96 2.83 5.87 8.00 13.42 19.91 20.82 40.31
alpha (% p.a.) (1.58) ( 1.32) ( 0.29) ( 1.41) ( 1.11) ( 2.05)**( 2.47)** ( 3.40)*** ( 4.73)*** ( 4.84)***

The European Journal of Finance


(0.59) (0.00)
[0.50]
Carhart alpha 0.40 0.99 0.64 1.95 3.23 3.20 5.31 7.78 13.65 20.10 20.51 31.37
(% p.a.) (0.68) ( 0.79) (0.45) ( 1.11) ( 1.32) ( 1.30) ( 1.86)* ( 2.32)** ( 3.48)*** ( 4.61)*** ( 4.69)*** (0.00)
[0.50]

Note: This table reports the abnormal performance of 10 value-weighted stock portfolios constructed on the basis of KaplanZingales (Panel A) and WhitedWu (Panel B)
indices of FC. P1 is the decile portfolio containing the stocks of the least constrained firms (lowest values of KZ index in Panel A and lowest values of WW index in Panel
B) and P10 is the decile portfolio containing the stocks of the most constrained firms (highest values of KZ index in Panel A and highest values of WW index in Panel B).
P10 P1 stands for the zero-cost strategy that goes long the portfolio of the most constrained firms (P10) and sells short the portfolio with the least constrained firms (P1).
CAPM, FamaFrench and Carhart alphas are the annualized alpha estimates derived from the corresponding asset pricing models. t-statistics are reported in parentheses. R2
for the spread P10 P1 is reported in square brackets for each model. The last column reports the chi-square ( 2 ) statistic of the Wald test for the null hypothesis that the 10
portfolios alphas are jointly equal to zero in each case; p-values are reported below the statistic.
*Statistical significance at the 10% level.
**Statistical significance at the 5% level.
***Statistical significance at the 1% level.

95
96 N. Balafas and A. Kostakis

alpha of 20.82% (t-value = 4.84) and Carhart alpha of 20.51% (t-value = 4.69). As
expected from the reported alphas estimates, Wald tests confirm that the null hypothesis of
jointly zero alphas across the WW-sorted portfolios is strongly rejected for all three models.
The previous asset pricing results, referring to the most sophisticated proxies of FC suggested
in the literature, confirm the preliminary evidence presented in Table 5. Taken together, these
results strongly support the argument that investors in highly constrained UK firms not only did
not earn a premium for holding these shares and being exposed to this risk, but contrary to the
conventional asset pricing hypothesis, they experienced a significant underperformance relative
to the other listed firms and the market index. The most pronounced return differentials are
reported when the WW index is used as a sorting criterion. The following subsection performs
a series of checks confirming the robustness of this evidence. In Section 4, we construct a FC
factor on the basis of the WW index and examine whether this factor is priced.

3.3 Robustness checks on KZ- and WW-sorted portfolios


3.3.1 Industry effects
A potential issue arising in our benchmark results is that firms from particular industries may
be clustered into the same decile portfolios because they exhibit similar balance sheet or cash
flow features. For example, firms belonging to an industry may be altogether exhibiting a higher
degree of leverage or having a lower level of cash holdings, if this is standard practice for firms
in this particular industry. In this case, some firms could be erroneously characterized as finan-
cially constrained relative to the entire universe of UK firms, even though they are actually
unconstrained relative to their industry competitors. To address this concern, in this section we
standardize the values of each FC measure that we previously utilized relative to the industry
that each firm belongs to.10 To this end, we use firms Level-3 FTSE Industrial Classification
(Datastream datatype: FTAG3). Using the industry-adjusted values of FC proxies, we repeat the
procedure of sorting firms, assigning them to the corresponding decile portfolios and calculating
their post-ranking monthly returns. To save space, we focus on the risk-adjusted performance
of decile portfolios constructed on the basis of the KZ and WW industry-adjusted composite
indices.11
The estimated alphas during the sample period 19882013 are reported in Table 10. Panel
A contains the alphas of the 10 industry-adjusted KZ-sorted portfolios. These are qualitatively
similar to the benchmark results reported for KZ-sorted portfolios in Panel A of Table 9. Along
the same lines, the spread P10 P1 still yields economically and statistically significant neg-
ative alphas of the same order as in the benchmark results. Panel B of Table 10 contains the
corresponding alphas for portfolios sorted on the basis of industry-adjusted WW index values.
Comparing these alphas with the benchmark estimates reported in Panel B of Table 9, they are
remarkably robust to the industry-adjustment we perform. The almost monotonic deterioration
of performance as we move from P1 to P10 also remains intact. Moreover, the spread P10 P1 is
still characterized by highly significant underperformance. Hence, we confirm our main finding
that the most constrained firms not only failed to yield a premium as compensation to investors
for holding their shares, but they rather severely underperformed. This result cannot be attributed
to an industry effect.

3.3.2 Higher co-moments risk factors


The limited ability of the commonly used models to price the cross-section of KZ- and WW-
sorted portfolios motivates us to investigate whether other risk factors may be more successful.
Table 10. Alphas of value-weighted decile portfolios constructed on the basis of industry-adjusted KZ and WW indices.

P1 P2 P3 P4 P5 P6 P7 P8 P9 P10 P10P1 Chi-sq.


Least Constrained Most Constrained

Panel A: KZ index
CAPM alpha (% 1.36 4.51 0.80 3.17 0.11 1.46 1.66 1.91 5.06 10.23 8.87 23.32
p.a.) ( 0.63) (2.66)*** ( 0.45) (1.99)** ( 0.06) ( 0.51) (0.89) (0.98) ( 1.51) ( 2.32)** ( 1.85)* (0.01)
FamaFrench 1.37 4.71 0.56 3.11 0.23 1.57 1.49 1.59 5.62 10.81 9.45 34.64
alpha (% p.a.) ( 0.68) (2.92)*** ( 0.33) ( 1.93)* ( 0.13) ( 0.56) (0.82) (0.86) ( 1.97)** ( 2.97)*** ( 2.12)** (0.00)
Carhart alpha 1.63 4.85 0.62 2.54 0.47 2.18 0.72 1.94 5.20 10.96 9.33 27.55
(% p.a.) ( 0.74) (2.87)*** ( 0.33) (1.36) ( 0.24) ( 0.67) (0.35) (0.98) ( 1.69)* ( 2.82)*** ( 1.91)* (0.00)
Panel B: WW index
CAPM alpha (% 0.34 2.05 0.45 2.34 0.03 0.68 2.17 9.03 10.86 17.41 17.75 46.75
p.a.) (0.46) (1.25) (0.27) (0.90) (0.01) ( 0.23) ( 0.52) ( 2.35)** ( 2.41)** ( 2.67)*** ( 2.64)*** (0.00)
FamaFrench 0.44 1.92 0.27 2.27 0.26 1.18 2.66 9.48 11.37 17.77 18.21 38.03
alpha (% p.a.) (0.64) (1.39) (0.21) (1.26) ( 0.14) ( 0.53) ( 1.02) ( 3.32)*** ( 3.33)*** ( 3.90)*** ( 3.99)*** (0.00)
0.54 2.14 1.45 2.65 10.12 11.44 20.63 20.67

The European Journal of Finance


Carhart alpha 0.04 1.58 0.91 44.83
(% p.a.) (0.06) (1.07) ( 0.41) (0.50) ( 1.12) ( 0.59) ( 0.90) ( 3.33)*** ( 3.20) ( 4.65)*** ( 4.69)*** (0.00)

Note: This table reports the abnormal performance of 10 value-weighted stock portfolios constructed on the basis of industry-adjusted KZ (Panel A) and WW (Panel B) indices
of FC. P1 is the decile portfolio containing the stocks of the least constrained firms adjusting for industry characteristics (lowest values of industry-adjusted KZ index in
Panel A and WW index in Panel B) and P10 is the decile portfolio containing the stocks of the most constrained firms adjusting for industry characteristics (highest values of
industry-adjusted KZ index in Panel A and WW index in Panel B). P10 P1 stands for the zero-cost strategy that goes long the portfolio of the most constrained firms (P10)
and sells short the portfolio with the least constrained firms (P1). CAPM, FamaFrench and Carhart alphas are the annualized alpha estimates derived from the corresponding
asset pricing models. t-statistics are reported in parentheses. The last column reports the chi-square ( 2 ) statistic of the Wald test for the null hypothesis that the 10 portfolios
alphas are jointly equal to zero in each case; p-values are reported below the statistic.
*Statistical significance at the 10% level.
**Statistical significance at the 5% level.
***Statistical significance at the 1% level.

97
98 N. Balafas and A. Kostakis

To this end, we employ the higher co-moments asset pricing model in the spirit of Harvey and
Siddique (2000), Hung, Shackleton, and Xu (2004) and Kostakis, Muhammad, and Siganos
(2012). The motivation to employ this model is that the shares of financially constrained firms
may exhibit tail risk, since these firms are more likely to perform very poorly during bad eco-
nomic times. These features are undesirable and a typical investor would require a premium to
hold such shares (see Dittmar 2002; Dichtl and Drobetz 2011). Hence, we test whether coskew-
ness or cokurtosis risk factors can explain away the KZ- and WW-sorted portfolios alphas. To
this end, we augment the CAPM with the coskewness (S S + ) and cokurtosis (K + K )
factors, leading to the following higher co-moments model:12

Rp,t Rt = p + MKT (Rm,t Rt ) + S S+ (S S + )t + K + K (K + K )t + p,t .


f f
(4)

We estimate this model for each portfolio for the period up to December 2008 due to factor
data availability and report the estimated coefficients in Table 11.
With respect to KZ-sorted portfolios (Panel A), we get results very similar to the benchmark
ones reported in Panel A of Table 9. The portfolio containing the most constrained firms (P10)
yields a highly significant negative alpha (15.47% p.a.) even after adjusting for coskewness
and cokurtosis risks. Similar to the benchmark results, the portfolio containing the least con-
strained firms (P1) also yields a significantly negative alpha, while the portfolios in the middle
of the classification yield insignificant alphas. The spread P10 P1 still yields a highly negative
abnormal performance ( 8.31% p.a.), though this is not statistically significant. The estimates
for market, coskewness and cokurtosis factor loadings do not reveal any particular pattern. In
conclusion, even though the higher co-moments model does a better job in explaining overall the
performance of all ten KZ-sorted portfolios and the null hypothesis of jointly zero alphas cannot
be rejected, the abnormal performance that we previously documented in the extreme deciles
cannot be explained by their exposure to coskewness and cokurtosis risk factors.
Even more interesting are the estimates of this model for the WW-sorted portfolios in Panel
B of Table 11. Overall, the null hypothesis of jointly zero alphas across the portfolios can be
rejected at any level of statistical significance, showing the inability of coskewness and cokur-
tosis risk factors to explain the abnormal performance of WW-sorted portfolios. As with the
previously used asset pricing models (see Panel B of Table 9), we still observe an interesting
negative gradient as we move from P1 to P10 and the spread P10 P1 yields a remarkable alpha
of 22.32% p.a., reflecting the highly negative abnormal performance of the portfolio contain-
ing the most constrained firms (P10). Moreover, we find no particular pattern with respect to
market and coskewness risk exposure. However, the portfolios of the highly constrained firms
(P8P10) tend to be less exposed to cokurtosis risk, but this differential is by no means sufficient
to explain the differential performance relative to portfolios of the least constrained firms (P1
P3). These findings lead to the conclusion that the negative performance of the most constrained
firms, as classified according to either KZ or WW index, is a genuine finding that cannot be
explained by exposure to higher co-moments risk factors.

3.3.3 Quintile portfolios


A potential concern regarding the robustness of the puzzling negative abnormal performance of
the most constrained firms could be that this significant return differential appears only across
extreme decile portfolios, and hence it is concentrated only on a small fraction of stocks uni-
verse. To address this concern, we also construct quintile portfolios that naturally contain a much
larger number of stocks and we calculate their post-ranking value-weighted monthly excess
Table 11. Higher co-moments alphas and factor loadings of decile portfolios constructed on the basis of KaplanZingales and WhitedWu indices.

P1 P2 P3 P4 P5 P6 P7 P8 P9 P10 P10 P1 Chi-sq.


Least Constrained Most Constrained

Panel A: KZ index
Higher co-moments 7.16 0.03 2.95 0.04 2.08 0.81 1.65 1.24 0.94 15.47 8.31 14.38
alpha (% p.a.) ( 2.33)*** ( 0.01) (1.26) (0.02) (0.79) ( 0.44) (0.73) ( 0.42) ( 0.36) ( 2.34)*** ( 1.32) (0.16)
Market loading 1.06 0.93 0.82 0.89 0.94 0.98 1.00 0.86 1.09 1.15 0.09
(S S + ) loading 0.38 0.09 0.50 0.11 0.38 2.04 0.77 0.23 0.36 0.31 0.07
(K + K ) loading 0.50 0.27 0.66 0.88 0.14 1.45 1.00 1.02 0.34 0.24 0.74
Panel B: WW index
Higher co-moments 1.02 3.17 0.03 0.30 2.98 6.18 6.73 7.44 10.67 21.29 22.32 27.02
alpha (% p.a.) (1.19) ( 1.28) ( 0.01) ( 0.08) ( 0.65) ( 1.21) ( 1.22) ( 1.25) ( 1.60) ( 2.87)*** ( 2.81)*** (0.00)
Market loading 0.88 1.17 0.98 0.94 0.94 1.17 1.03 1.04 0.94 0.96 0.08
(S S + ) loading 0.11 0.53 0.77 0.64 0.89 1.32 1.17 1.00 0.71 0.65 0.54
(K + K ) loading 0.28 0.35 0.86 0.45 1.23 0.83 0.86 1.37 1.14 1.03 0.75

The European Journal of Finance


Note: This table reports the abnormal performance and the higher co-moments factor loadings of 10 value-weighted stock portfolios constructed on the basis of KZ (Panel A)
and WW (Panel B) indices of FC. P1 is the decile portfolio containing the stocks of the least constrained firms (lowest values of KZ index in Panel A and WW index in Panel
B) and P10 is the decile portfolio containing the stocks of the most constrained firms (highest values of KZ index in Panel A and WW index in Panel B). P10 P1 stands for
the zero-cost strategy that goes long the portfolio of the most constrained firms (P10) and sells short the portfolio with the least constrained firms (P1). Higher co-moment alpha
stands for the annualized alpha estimate derived from the higher co-moments asset pricing model in (4). Market loading, (S S + ) loading and (K + K ) loading stand for
the point estimates of each portfolio returns exposure to market, coskewness and cokurtosis factor returns, respectively. The sample period is from 1988 to 2008. t-statistics for
the alphas are reported in parentheses. The last column reports the chi-square ( 2 ) statistic of the Wald test for the null hypothesis that the 10 portfolios alphas are jointly equal
to zero in each case; p-values are reported below the statistic.
*Statistical significance at the 10% level.
**Statistical significance at the 5% level.
***Statistical significance at the 1% level.

99
100 N. Balafas and A. Kostakis

Table 12. Characteristics and alphas of value-weighted quintile portfolios constructed on the basis of
WhitedWu index values.
t-test/
Q1 Q2 Q3 Q4 Q5 Q5 Q1 Chi-sq.
Least Constrained Most Constrained

FC proxy: WW index
WW index 0.59 0.29 0.24 0.18 0.06 0.53 28.76***
MV (m) 3675.11 213.80 68.60 27.36 14.99 3660.12 30.96***
VW excess returns (% 4.96 5.52 2.06 1.65 10.90 15.86 3.70***
p.a.)
CAPM alpha (% p.a.) 0.55 0.95 2.44 6.19 15.06 15.61 22.06
(0.99) (0.41) ( 0.67) ( 1.39) ( 2.88)*** ( 2.90)*** (0.00)
FamaFrench alpha (% 0.62 0.56 2.66 6.62 15.46 16.08 16.63
p.a.) (1.13) (0.45) ( 1.34) ( 2.38)** ( 4.20)*** ( 4.28)*** (0.01)
Carhart alpha (% p.a.) 0.15 0.06 3.04 6.10 15.91 16.06 19.16
(0.27) ( 0.05) ( 1.43) ( 2.21)** ( 4.32)*** ( 4.34)*** (0.00)

Note: This table reports the characteristics and abnormal performance of quintile portfolios constructed on the basis of
firms WW index of FC. All stocks listed on the LSE from 1988 to 2013 are sorted at month t and they are assigned to five
portfolios. Q1 is the quintile portfolio containing the stocks of the least constrained firms (lowest values of WW index)
and Q5 is the quintile portfolio containing the stocks of the most constrained firms (highest values of WW index). The
returns of these portfolios in excess of the risk-free rate at month t + 1 are calculated (i.e. post-ranking returns). Q5 Q1
stands for the zero-cost strategy that goes long the quintile portfolio of the most constrained firms (Q5) and sells short the
quintile portfolio with the least constrained firms (Q1). MV is the average market value of the stocks in each portfolio
(in m). VW returns refer to the annualized average monthly excess returns of value-weighted portfolios. For the first
three rows, the last column reports values of t-tests for the null hypothesis of no difference in means between portfolios
Q5 and Q1 characteristics. CAPM, FamaFrench and Carhart alphas are the annualized alpha estimates derived from the
corresponding asset pricing models. t-statistics for the alphas are reported in parentheses. The last column next to alphas
reports the chi-square ( 2 ) statistic of the Wald test for the null hypothesis that the five portfolios alphas are jointly
equal to zero in each case; p-values are reported below the statistic.
*Statistical significance at the 10% level.
**Statistical significance at the 5% level.
***Statistical significance at the 1% level.

returns. Apart from reporting the characteristics and premia of quintile portfolios constructed
on the basis of WW index in Table 12, we also estimate their Jensen, FamaFrench and Carhart
alphas from time-series regressions (1), (2) and (3), respectively.
The most striking result is that the spread between the quintile portfolio containing the most
constrained firms (Q5) and the quintile portfolio containing the least constrained ones (Q1)
is highly negative ( 15.86% p.a.) and strongly significant (t-value = 3.70). This negative
spread mainly derives from the severe underperformance of Q5. Adjusting these returns for their
risk factor exposures, this finding remains intact. Q5 significantly underperformed Q1 and their
spread yielded highly significant negative alphas. Finally, we find a clear monotonic deteriora-
tion of raw and risk-adjusted performance as we move from the quintile portfolio with the least
constrained firms (Q1) to the one with the most constrained firms (Q5).

3.3.4 Longer investment horizon returns


The asset pricing tests conducted in this study are based on 1-month post-ranking returns. Even
though this is the standard approach in the literature, a potential concern could be that the abnor-
mally negative performance of the most constrained firms is short-lived and is reversed if one
considers longer investment horizons. To address this concern we calculate longer buy-and-hold
The European Journal of Finance 101

portfolio returns. In particular, sorting stocks into decile portfolios according to their KZ or WW
index values in a given month t, we alternatively calculate 3-, 6- and 12-month cumulative post-
ranking portfolio returns.13 Unreported results, which are readily available upon request, show
that the abnormally negative performance of the most constrained relative to the least constrained
firms on the basis of their WW index values remarkably persists as we increase the investment
horizon to 3 or 6 months, and hence this pattern cannot be characterized as a short-lived phe-
nomenon. As expected, the magnitude of this relative underperformance is somewhat reduced
when we consider 12-month buy-and-hold returns, but its economic and statistical significance
remains intact even when we risk-adjust these returns for the commonly used risk factors. With
respect to KZ-sorted portfolios, results show that the relative underperformance of the most con-
strained firms reported in Panel A of Table 5 and Table 9 using 1-month returns remains intact
when we consider 3-, 6- or 12-month buy-and-hold returns.

4. FC factor
4.1 Construction and factor returns
The previous section discussed the characteristics and performance of portfolios sorted according
to a series of FC proxies. It also reported that the extreme deciles or quintiles yielded significantly
different risk-adjusted returns, indicating that FC risk may not be captured by the commonly used
asset pricing models. Though prevalent across most of the used proxies, this finding was most
pronounced when we sorted firms into portfolios on the basis of the WW index, and hence we
focus on this measure for the rest of this study. This evidence motivates us to examine whether
there is a priced systematic source of risk related to FC. To this end, we construct a zero-cost
FC factor and examine its raw and risk-adjusted returns, while in the following subsection we
examine whether this factor is priced in the cross-section of stock returns.
To construct the zero-cost FC factor, we follow standard practice in the asset pricing literature
as well as in Lamont, Polk, and Saa-Requejo (2001) and Chan et al. (2010), and we use 6 (2 3)
independently double-sorted portfolios on the basis of firms size and WW index values, ensuring
that there is a sufficient number of firms in each portfolio.14 In particular, a firm is categorized
as Small (S) if its market value is below the median market value of all firms in a given month
and Big (B) if it is above the median. A firm is categorized as Least Constrained (LC) if its
WW value is below the 30th percentile of the firms WW values distribution in the same month,
Neutral (N) if its value is between the 30th and the 70th percentiles and Most Constrained (MC)
if its value is above the 70th percentile of the distribution. The FC factor return (i.e. FC = MC
LC) is defined as the difference between the average post-ranking return of the MC portfolios
across the two size categories, i.e. MC = (MCS + MCB)/2, and the average post-ranking return
of the LC portfolios across the two size categories, that is, LC = (LCS + LCB)/2.15 The average
returns, CAPM, FamaFrench and Carhart alphas of the six double-sorted portfolios are reported
in Table 13, while the corresponding FC factor returns and alphas are reported in Table 14.
With respect to the double-sorted portfolio returns and alphas reported in Table 13, the most
interesting observation is that the severe underperformance of the most constrained firms relative
to the least constrained ones is remarkably robust across both size categories. The Most-Least
Constrained spread yields an average return of 14.16% p.a. among Small firms and 13.36%
p.a. among Big firms. These negative spreads remain intact and highly significant even after
adjusting for their risk factor exposures. For example, the Carhart alpha of the Most-Least
Constrained spread is 13.86% p.a. (t-value = 4.45) for Small firms and 14.09% p.a.
(t-value = 3.04) for Big firms.
102 N. Balafas and A. Kostakis

Table 13. Performance of value-weighted double-sorted portfolios constructed on the basis of Size and WW
index values.

Least Constrained Neutral Most Constrained Most-Least t-test

Panel A: VW excess returns (% p.a.)


Small 7.56 3.45 6.60 14.16 4.35***
Big 4.92 1.30 8.44 13.36 2.67***
Panel B: CAPM alphas (% p.a.)
Small 2.73 0.04 10.27 12.99
(0.57) ( 0.01) ( 2.33)** ( 3.23)***
[0.05]
Big 0.52 3.45 13.64 14.16
(0.95) ( 0.86) ( 2.33)** ( 2.38)**
[0.01]
Panel C: FamaFrench alphas (% p.a.)
Small 1.39 0.88 10.85 12.24
(0.40) ( 0.48) ( 3.70)*** ( 3.66)***
[0.20]
Big 0.56 3.57 14.04 14.60
(1.01) ( 1.61) ( 3.14)*** ( 3.20)***
[0.39]
Panel D: Carhart alphas (% p.a.)
Small 2.80 0.06 11.07 13.86
(0.81) ( 0.03) ( 3.63)*** ( 4.45)***
[0.21]
Big 0.05 3.86 14.04 14.09
(0.10) ( 1.72)* ( 3.05)*** ( 3.04)***
[0.39]

Note: This table reports the performance of double-sorted Size and WhitedWu (WW) portfolios. All stocks listed on the
LSE from 1988 to 2013 are independently sorted at month t according to their Size and their WW index values. A firm is
categorized as Small if its market value is below the median market value of all firms in this month and Big if it is above
the median. A firm is categorized as Least Constrained if its WW value is below the 30th percentile of the firms WW
values distribution in month t, Neutral if its value is between the 30th and the 70th percentiles and Most Constrained if
its value is above the 70th percentile of the distribution. The interaction of these two classifications yields six portfolios.
The returns of these portfolios in excess of the risk-free rate at month t + 1 are calculated. Most-Least stands for the
zero-cost strategy that goes long the portfolio of the Most Constrained firms and sells short the portfolio with the Least
Constrained firms for each Size category. VW returns refer to the annualized average monthly excess returns of value-
weighted portfolios and the t-test refer to the null hypothesis of zero average returns for the Most-Least spread strategy.
CAPM, FamaFrench and Carhart alphas are the annualized alpha estimates derived from the corresponding asset pricing
models. t-statistics for the alphas are reported in parentheses. R2 for each model are reported in square brackets.
*Statistical significance at the 10% level.
**Statistical significance at the 5% level.
***Statistical significance at the 1% level.

With respect to the zero-cost FC factor returns reported in Table 14, we find that it
yielded a highly negative premium ( 13.76% p.a.), which was also strongly significant (t-
value = 4.34). A natural question is whether this negative premium that the FC factor yielded
can be attributed to its market, size, value or momentum factor exposure.16 To address this issue,
we estimate the corresponding asset pricing models. The results reported in Table 14 indicate
that the negative premium of the FC factor cannot be explained by its exposure to any of these
factors. To the contrary, this negative premium remains intact in the form of a highly significant
alpha for each of the considered models. Moreover, the explanatory power of these models is
The European Journal of Finance 103

Table 14. FC factor return and alphas.

Factor return CAPM FamaFrench Carhart

Average return (% p.a.) 13.76


( 4.34)***
Alpha (% p.a.) 13.57 13.42 13.98
( 3.25)*** ( 4.22)*** ( 4.53)***
Market loading 0.04 0.02 0.01
( 0.49) ( 0.34) ( 0.21)
SMB loading 0.55 0.56
(4.82)*** (4.88)***
HML loading 0.46 0.43
( 3.51)*** ( 3.23)***
MOM loading 0.05
(0.59)
R2 0.01 0.31 0.31

Note: This table reports the average annualized return of the FC factor as well as its CAPM, FamaFrench and Carhart
alphas and factor loadings estimated from the corresponding asset pricing models during the period August 1988
December 2013. The last row reports the R2 of each model. t-statistics are reported in parentheses. ***, ** and * indicate
that the corresponding coefficient is statistically significant at 1%, 5% and 10% level, respectively. To construct the
FC factor, firms are classified into six portfolios based upon independent sorts of their Size and WW index values. A
firm is categorized as Small (S) if its market value is below the median market value of all firms in this month and
Big (B) if it is above the median. A firm is categorized as Least Constrained (LC) if its WW value is below the 30th
percentile of the firms WW values distribution in month t, Neutral (N) if its value is between the 30th and the 70th
percentiles and Most Constrained (MC) if its value is above the 70th percentile of the distribution. The FC factor return
(i.e. FC = MC LC) is defined as the difference between the average return of the MC portfolios across the two size
categories, i.e. MC = (MCS + MCB)/2, and the average return of the LC portfolios across the two size categories, i.e.
LC = (LCS + LCB)/2.

very low. This is especially true for the CAPM, whose R2 is as low as 1%. Even though the
returns of the FC factor are correlated with SMB and HML factor returns, this correlation neither
subsumes nor explains, even partly, its highly negative premium. Taken together, these results
provide robust evidence that the constructed FC factor carries a significantly negative premium
that is genuinely priced over and above the commonly used risk factors.

4.2 Performance during the financial crisis period


FC attracted the attention of policymakers and investors during the global financial crisis because
it was during this period that agency costs were exacerbated and a prolonged disruption in credit
markets was observed, rendering firms access to external finance very costly or even impossible.
Therefore, it is interesting to examine the performance of the constructed FC factor during this
period. In particular, we plot in Figure 1 the cumulative returns of the FC factor from August
2007 to December 2009. The choice of the starting point for the crisis period is motivated by
the analysis of Brunnermeier (2009) and it coincides with the bank run of Northern Rock, which
brought the UK financial system to a near collapse.
As Figure 1 shows, the FC factor performed extremely poorly during the crisis period, exhibit-
ing a cumulative return of approximately 90%. In other words, the most constrained firms
shares severely underperformed relative to the least constrained firms shares. The extremely
negative performance of this factor was persistent throughout the crisis period, yielding an aver-
age return of 3.1% per month. Interestingly, the worst performance of the FC factor was
104 N. Balafas and A. Kostakis

Figure 1. This figure shows the cumulative monthly returns of the zero-cost FC factor during the recent
financial crisis period, from August 2007 until December 2009. To construct the FC factor, firms are
classified into six portfolios based upon independent sorts of their Size and WW index values. A firm
is categorized as Small (S) if its market value is below the median market value of all firms in this month
and Big (B) if it is above the median. A firm is categorized as Least Constrained (LC) if its WW value
is below the 30th percentile of the firms WW values distribution in month t, Neutral (N) if its value is
between the 30th and the 70th percentiles and Most Constrained (MC) if its value is above the 70th per-
centile of the distribution. The FC factor return (i.e. FC = MC LC) is defined as the difference between
the average return of the MC portfolios across the two size categories, that is, MC = (MCS + MCB)/2,
and the average return of the LC portfolios across the two size categories, that is, LC = (LCS + LCB)/2.

observed in the aftermath of the Lehman Brothers collapse, particularly in December 2008 and
January 2009.17
Given the extremely poor performance of the FC factor during the crisis period, it is legitimate
to ask whether this performance drives its negative premium that we documented for the full
sample period. To this end, we compute the premium of the FC factor excluding the global
financial crisis period and beyond, that is, for the period August 1988August 2007. For this
period, we find that the zero-cost FC factor yields an average return of 11.20% p.a., which is
also strongly significant (t-value = 2.94). This negative premium remains intact in the form of
a significant alpha even when we control for the exposure of the FC factor returns to the market,
size, value and momentum factors. In particular, the Carhart alpha of the FC factor returns for
this period is 8.86% p.a. (t-value = 2.52).
Moreover, Figure 2 shows the cumulative performance of an investment in the FC factor
together with the cumulative performance of an investment in the market portfolio, for the
period August 1988August 2007. Confirming the previous results on the significantly negative
premium that the FC factor carried even before the crisis period, Figure 2 shows that the most
The European Journal of Finance 105

Figure 2. This figure shows the cumulative performance of an investment in: (i) the FC (solid line) factor
and (ii) the market portfolio (dashed line) from August 1988 to August 2007, that is, before the onset of
the financial crisis period. To construct the FC factor, firms are classified into six portfolios based upon
independent sorts of their Size and WW index values. A firm is categorized as Small (S) if its market value
is below the median market value of all firms in this month and Big (B) if it is above the median. A firm
is categorized as Least Constrained (LC) if its WW value is below the 30th percentile of the firms WW
values distribution in month t, Neutral (N) if its value is between the 30th and the 70th percentiles and
Most Constrained (MC) if its value is above the 70th percentile of the distribution. The FC factor return
(i.e. FC = MC LC) is defined as the difference between the average return of the MC portfolios across
the two size categories, that is, MC = (MCS + MCB)/2, and the average return of the LC portfolios across
the two size categories, that is, LC = (LCS + LCB)/2. The market portfolio return is proxied by the return
on the FTSE All Share Index in excess of the risk-free rate.

constrained firms considerably underperformed relative to the least constrained ones throughout
the pre-crisis period.
Therefore, we can conclude that the puzzling, highly negative premium that the FC factor
yielded during the examined period is not exclusively driven by the financial crisis, but it was
both statistically and economically significant even before the crisis. This negative premium
arises from the fact that, contrary to the predictions of standard asset pricing theory, the most
constrained firms did not outperform the least constrained ones during good economic times
so as to offset their losses during bad economic times and to reward risk-averse investors for
withholding these stocks.

4.3 Cross-sectional asset pricing tests


Having established that the constructed FC factor yields a highly significant negative premium
in the time-series dimension, it is also important to examine whether this factor is priced in the
106 N. Balafas and A. Kostakis

cross-section of portfolio returns. To this end, we employ the standard two-stage FamaMacBeth
methodology, using as test assets the 10 WW-sorted value-weighted portfolios. The first stage of
the FamaMacBeth methodology involves the estimation of betas from time-series regressions
of excess portfolio returns on a set of risk factors. To examine whether the FC factor is priced
beyond the market, size, value and momentum factors, we augment the CAPM, FamaFrench
and Carhart models, respectively, by adding this factor. In particular, the most general asset
pricing model we estimate is:
f f
Ri,t Rt = i + i,MKT (Rm,t Rt )t + i,SMB SM Bt + i,HML HM Lt + i,MOM MOMt
+ i,FC FCt + i,t . (5)
The second stage involves the estimation of monthly cross-sectional regressions of the 10
excess portfolio returns on the corresponding betas estimated in the first stage. Corresponding to

Table 15. Cross-sectional asset pricing tests.

0 MKT SMB HML MOM FC R2

CAPM 0.0267 0.0275 0.10


( 3.31)*** (3.64)***
[ 2.78]*** [3.10]***
CAPM + FC 0.0004 0.0067 0.0132 0.39
(0.08) (1.07) ( 4.01)***
[0.07] [1.02] [ 3.94]***
FamaFrench 0.0136 0.0210 0.0067 0.0042 0.44
( 2.28)** (3.07)*** ( 2.09)** (0.64)
[ 2.00]** [2.73]*** [ 1.93]* [0.57]
FamaFrench + FC 0.0105 0.0066 0.0028 0.0035 0.0190 0.57
(1.82)* ( 1.01) (0.91) (0.52) ( 4.97)***
[1.60] [ 0.90] [0.84] [0.47] [ 4.63]***
Carhart 0.0289 0.0356 0.0045 0.0074 0.0337 0.52
( 3.52)*** (4.01)*** ( 1.44) (1.11) (3.21)***
[ 2.02]** [2.36]** [ 0.98] [0.66] [1.88]*
Carhart + FC 0.0063 0.0024 0.0028 0.0041 0.0054 0.0181 0.64
(0.68) ( 0.24) (0.90) (0.62) (0.47) ( 4.42)***
[0.60] [ 0.21] [0.84] [0.55] [0.41] [ 4.09]***

Note: This table reports the mean of the monthly risk premium coefficients i estimated from the second stage of the
FamaMacBeth methodology. This methodology involves the monthly cross-sectional regressions of excess returns for
10 value-weighted stock portfolios, sorted on the basis of the WW index of FC, on the corresponding full sample risk
factor loadings estimated in the first stage of the procedure. These risk factor loadings have been respectively estimated
using the CAPM, FamaFrench and Carhart models as well as their augmented versions adding the FC factor FC. To
construct the FC factor, firms are classified into six portfolios based upon independent sorts of their Size and WW index
values. A firm is categorized as Small (S) if its market value is below the median market value of all firms in this month
and Big (B) if it is above the median. A firm is categorized as Least Constrained (LC) if its WW value is below the
30th percentile of the firms WW values distribution in month t, Neutral (N) if its value is between the 30th and the
70th percentiles and Most Constrained (MC) if its value is above the 70th percentile of the distribution. The FC factor
return (i.e. FC = MC LC) is defined as the difference between the average return of the MC portfolios across the two
size categories, i.e. MC = (MCS + MCB)/2, and the average return of the LC portfolios across the two size categories,
i.e. LC = (LCS + LCB)/2. t-statistics are reported in parentheses. We also report in brackets t-statistics using Shanken
(1992)-corrected standard errors to mitigate the errors-in-variables bias. The last column reports the average R2 of the
second-stage regression models.
*The corresponding coefficient is statistically significant at the 10% level.
**The corresponding coefficient is statistically significant at the 5% level.
***The corresponding coefficient is statistically significant at the 1% level.
The European Journal of Finance 107

model (5), the most general cross-sectional regression model is:


Rp,t Rt = 0 + MKT MKT + SMB SMB + HML HML + MOM MOM + FC FC + wp,t ,
f
(6)
where lambdas represent the risk premium coefficients associated with each beta.18 For robust-
ness purposes, apart from standard FamaMacBeth t-statistics for the lambda coefficients, we
also follow Shanken (1992) to calculate t-statistics using standard errors that have been corrected
for the error-in-variables bias arising due to the fact that the factor loadings used as regressors in
model (6) are pre-estimated rather than being the true, unobservable ones.
Results from the FamaMacBeth regressions are reported in Table 15. For all asset pricing
models considered, when we add the constructed FC factor, its negative premium is found to be
highly economically and statistically significant. For example, a FC lambda estimate of 0.0181
in the augmented Carhart model indicates that a unit factor loading of a portfolio on FC factor
corresponds to a negative excess return of 1.81% per month or 21.72% p.a. The magnitude
of this negative price of risk for the FC factor in the cross-section of WW-sorted portfolios
is in line with its time-series performance presented in Table 14. Interestingly, adding the FC
factor subsumes the significance of the rest factors in explaining the cross-section of WW-sorted
portfolios returns. Moreover, the average R2 of the second-stage regressions is considerably
increased when the FC factor is added relative to the benchmark models.
In addition to FamaMacBeth regressions, we follow Chan et al. (2010) and Ling and Chen
(2012) in estimating time-series regressions of the 10 WW-sorted and the 6 size and WW double-
sorted portfolios returns using model (5). Unreported results show that the estimated coefficients
on FC factor are statistically significant and carry the anticipated signs. This finding together with
the FamaMacBeth results indicate that the WW-sorted portfolios returns systematically covary
and that the constructed FC factor captures this systematic source of risk, being priced in the
cross-section over and above the commonly used risk factors.

5. Conclusions
The present study contributes to the literature by providing for the first time comprehensive
evidence on the pricing of FC risk on LSE using a rich dataset of non-financial firms from 1988
to 2013. To this end, we examine a large number of proxies suggested in prior studies to capture
this elusive concept, including the KaplanZingales and the WhitedWu index. The examined
period in our study contains the recent global financial crisis, which provides an ideal set-up,
since we expect that the impact of FC would be exacerbated during the crisis period as credit
supply is reduced, the cost of borrowing is increased and firms cash flows are reduced.
The main finding of our study is that investors in highly constrained firms were not rewarded
for being exposed to this aspect of risk. To the contrary, the most constrained firms under-
performed the least constrained ones in most of the cases we have examined. The reported
underperformance was even more striking using the sophisticated WW index that jointly cap-
tures various aspects of FC. Focussing on this proxy to construct a zero-cost FC factor that goes
long the most constrained firms and sells short the least constrained ones, we find that this factor
carried a highly economically and statistically significant negative premium and it was priced in
the cross-section of stock returns beyond the commonly used risk factors.
Our findings have a number of highly important implications for asset pricing and investment
management. Consistent with the findings of Lamont, Polk, and Saa-Requejo (2001) and Whited
and Wu (2006) for the US market, we challenge the conventional wisdom that financially con-
strained firms should in equilibrium command a higher risk premium to compensate investors
108 N. Balafas and A. Kostakis

for withholding them. To the contrary, we find that investors on LSE would have been better off
forming portfolios that do not include these shares and, to the extent that this would be feasible,
they would gain considerable profits by selling them short.
Our results also have important implications for firms optimal capital structure decisions.
Since investors on LSE did not penalize financially constrained firms by requiring higher premia
for holding their shares, one could argue that managers were acting rationally in leveraging up
their firms. The offsetting mechanism of the trade-off theory broke down, the cost of capital did
not increase in line with the debt-to-equity capital mixture and hence managers rationally chose
to utilize the cheaper source of external capital. Empirical tests of capital structure theories in the
UK market should take our findings into account.

Disclosure statement
No potential conflict of interest was reported by the authors.

Notes
1. Our study is also related to the literature on financial distress and asset pricing, even though the concept of financial
distress, which is primarily related to the probability of a firms bankruptcy, is not identical to the concept of financial
constraints, which mainly measures the strength of a firms restrictions to grow at its desirable pace (see Whited and
Wu 2006, 534 for a discussion on their subtle difference).
2. Arguably, financial constraints may be even more severe for unlisted companies, which are predominantly small and
do not have access to credit (Guariglia 2008). However, the issue we examine relates to the pricing of this aspect
of risk in the stock market, and hence we can only examine listed companies, whose shares are tradable and market
prices are observable.
3. Factors are available at: http://xfi.exeter.ac.uk/researchandpublications/portfoliosandfactors/index.php.
4. To calculate firms KZ and WW index values, we use the coefficients suggested in the original studies by Kaplan and
Zingales (1997) and Whited and Wu (2006), respectively. We acknowledge that for a UK study one would ideally re-
estimate these structural models for UK firms and utilize the corresponding coefficients to calculate indices values.
However, we do not have a long enough sample to estimate these coefficients, to use them to construct portfolios
and to evaluate their out-of-sample (i.e. post-ranking) performance. Therefore, we opt for the original US-estimated
coefficients in order to have a long enough period for an asset pricing study and we expect that these should also
be applicable for UK firms given the similarities in the functioning of these two countries financial systems and in
their legal frameworks.
5. We have also considered firms payout ratios but there is not enough cross-sectional variation in this measure to
render it an informative sorting criterion. Similarly, bond/commercial paper ratings exist only for a small number
of UK-listed firms. Therefore, unlike US studies, credit ratings are not a useful sorting criterion for the UK market.
Finally, we also utilized total debt-to-total assets and total debt-to-tangible assets ratios that yielded similar results
to the presented leverage measures, and hence they are omitted for the sake of brevity.
6. For accounting periods beginning before 2007, firms listed on LSE were allowed by the Financial Services Author-
ity (FSA) to make their financial reports public up to 6 months after the end of their fiscal year. For details, see
http://fsahandbook.info/FSA/html/handbook/DTR/4/1.
7. It should be noted that we have excluded firms with negative total debt-to-common equity values that derive from
negative common equity values.
8. We do so to avoid classifying firms with zero leverage ratios to the portfolio of the least constrained firms, P1.
Firms may have zero leverage either by choice (in which case they are unconstrained) or because they are unable
or untrustworthy to borrow (in which case they are highly constrained). The number of firms with zero leverage
increases dramatically after 2000.
9. It should be noted that in this table we have excluded firms with negative interest coverage ratios values deriving
from negative earnings before interest and taxes. As a robustness check, we have alternatively assigned the firms
with negative such values to the portfolio of the most constrained firms (P10); the spread return P10P1 was even
more negative and highly significant.
10. In particular, for each month we deduct from the calculated measure the corresponding mean industry value and
divide by the corresponding standard deviation of industry values.
The European Journal of Finance 109

11. The corresponding tables of descriptive statistics and alphas for the rest industry-adjusted financial constraints
proxies are readily available upon request.
12. The coskewness factor (S S + ) has been constructed as a zero-cost spread between the portfolio containing the
20% of the shares exhibiting the most negative coskewness with respect to the market portfolio (S ) and the portfolio
containing the 20% of the shares exhibiting the most positive coskewness (S + ). Similarly, the cokurtosis factor
(K + K ) has been constructed as a zero-cost spread between the portfolio containing the 20% of the shares
exhibiting the most positive cokurtosis with respect to the market portfolio (K + ) and the portfolio containing the
20% of the shares exhibiting the most negative cokurtosis (K ). For more details, see Kostakis, Muhammad, and
Siganos (2012) as well as Kostakis (2009) for the importance of coskewness risk for mutual fund performance
evaluation.
13. We would like to thank an anonymous referee for suggesting this robustness check.
14. We would like to thank an anonymous referee for this suggestion.
15. We have also experimented with alternative percentiles for firms portfolio classification. Results, which are readily
available upon request, are very similar to the ones presented here, and hence the reported results are very robust to
the choice of the cut-off percentiles.
16. We would like to thank an anonymous referee for suggesting this test.
17. We have also examined whether the constructed FC factor returns are related to a series of macroeconomic variables.
The results, which are readily available upon request, point to a significant relationship between the FC factor
returns and certain monetary and credit variables. In particular, we find that the change in BoEs policy rate has a
significant contemporaneous effect on FC factor returns, while M2 growth and the credit spread affect these returns
with one or two lags, respectively. Moreover, we find that the change in the effective exchange rate has significant
contemporaneous and lagged effects, while we also find significant lagged effects from industrial production growth.
To the contrary, we do not find significant effects for inflation rate, real rate and the term spread.
18. This is the standard way to test whether a set of factors comprising an asset pricing model are priced and whether
they can explain the cross-section of a set of portfolio returns. An alternative approach that is commonly used in the
literature is to regress individual stock returns on firm characteristics. We do not pursue this approach here, as we
do not aim at testing whether particular financial constraints proxies are priced in the cross-section of stock returns.
Instead, we want to examine whether the constructed financial constraints factor is priced in the cross-section of
portfolio returns and whether it adds explanatory power to the commonly used CAPM, FamaFrench and Carhart
models.

References
Beattie, V., A. Goodacre, and S. J. Thomson. 2006. Corporate Financing Decisions: UK Survey Evidence. Journal of
Business Finance and Accounting 33: 14021434.
Brunnermeier, M. 2009. Deciphering the Liquidity and Credit Crunch 200708. Journal of Economic Perspectives 23:
77100.
Campello, M., and L. Chen. 2010. Are Financial Constraints Priced? Evidence from Firm Fundamentals and Stock
Returns. Journal of Money, Credit and Banking 42: 11851198.
Carhart, M. 1997. On Persistence in Mutual Fund Performance. Journal of Finance 52: 5782.
Carpenter, R. E., and A. Guariglia. 2008. Cash Flow, Investment, and Investment Opportunities: New Tests using UK
Panel Data. Journal of Banking and Finance 32: 18941906.
Caskey, J., J. Hughes, and J. Liu. 2012. Leverage, Excess Leverage, and Future Returns. Review of Accounting Studies
17: 443471.
Chan, H., X. Chang, R. Faff, and G. Wong. 2010. Financial Constraints and Stock Returns-Evidence from Australia.
Pacific-Basin Finance Journal 18: 306318.
Dichtl, H., and W. Drobetz. 2011. Portfolio Insurance and Prospect Theory Investors: Popularity and Optimal Design
of Capital Protected Financial Products. Journal of Banking and Finance 35: 16831697.
Dittmar, R. F. 2002. Nonlinear Pricing Kernels, Kurtosis Preference, and Evidence from the Cross Section of Equity
Returns. Journal of Finance 57: 369403.
Fama, E. F., and K. R. French. 1993. Common Risk Factors in the Returns on Stocks and Bonds. Journal of Financial
Economics 33: 356.
Fazzari, S. R., G. Hubbard, and B. Petersen. 1988. Financing Constraints and Corporate Investment. Brooking Papers
in Economic Activity 1988: 141195.
110 N. Balafas and A. Kostakis

Florackis, C., A. Gregoriou, and A. Kostakis. 2011. Trading Frequency and Asset Pricing on London Stock Exchange:
Evidence from a New Price Impact Ratio. Journal of Banking and Finance 35: 33353350.
Gibbons, M. R., S. A. Ross, and J. Shanken. 1989. A Test of the Efficiency of a Given Portfolio. Econometrica 57:
11211152.
Gomes, J. F., A. Yaron, and L. Zhang. 2006. Asset Pricing Implications of Firms Financing Constraints. Review of
Financial Studies 19: 13211356.
Goyal, A. 2012. Empirical Cross-Sectional Asset Pricing: A Survey. Financial Markets and Portfolio Management
26: 338.
Gregory, A., R. Tharyan, and A. Christidis. 2013. Constructing and Testing Alternative Versions of the Fama-French
and Carhart Models in the UK. Journal of Business Finance and Accounting 40: 172214.
Guariglia, A., and S. Mateut. 2006. Credit Channel, Trade Credit Channel, and Inventory Investment: Evidence from a
Panel of UK Firms. Journal of Banking and Finance 30: 28352856.
Guariglia, A. 2008. Internal Financial Constraints, External Financial Constraints and Investment Choice: Evidence
from a Panel of UK Firms. Journal of Banking and Finance 32: 17951809.
Hahn, J., and H. Lee. 2009. Financial Constraints, Debt Capacity, and the Cross-Section of Stock Returns. Journal of
Finance 64: 891921.
Harvey, C., and A. Siddique. 2000. Conditional Skewness in Asset Pricing Tests. Journal of Finance 55: 12631295.
Hubbard, G. 1998. Capital Market Imperfections and Investment. Journal of Economic Literature 35: 193225.
Hung, D. C., M. Shackleton, and X. Xu. 2004. CAPM, Higher Co-Moment and Factor Models of UK Stock Return.
Journal of Business Finance and Accounting 31: 87112.
Jaffee, D. M., and T. Russell. 1976. Imperfect Information, Uncertainty, and Credit Rationing. Quarterly Journal of
Economics 90: 651666.
Jensen, M. C., and W. H. Meckling. 1976. Theory of the Firm: Managerial Behaviour, Agency Costs, and Ownership
Structure. Journal of Financial Economics 3: 305360.
Kaplan, S. N., and L. Zingales. 1997. Do Financing Constraints Explain Why Investment Is Correlated with Cash
Flow? Quarterly Journal of Economics 112: 169215.
Kashyap, A. K., O. A. Lamont, and J. C. Stein. 1994. Credit Conditions and the Cyclical Behavior of Inventories.
Quarterly Journal of Economics 109: 565592.
Kiyotaki, N., and J. Moore. 1997. Credit Cycles. Journal of Political Economy 105: 211248.
Kostakis, A. 2009. Performance Measures and Incentives: Loading Negative Coskewness to Outperform the CAPM.
European Journal of Finance 15: 463486.
Kostakis, A., K. Muhammad, and A. Siganos. 2012. Higher Co-Moments and Asset Pricing on London Stock
Exchange. Journal of Banking and Finance 36: 913922.
Lamont, O., C. Polk, and J. Saa-Requejo. 2001. Financial Constraint and Stock Returns. Review of Financial Studies
14: 529554.
Ling, Z., and S. Chen. 2012. Financial Constraints and Stock Returns: Evidence from Stock Market in China. 2012
International Conference on Management Science & Engineering: 14221428.
Livdan, D., H. Sapriza, and L. Zhang. 2009. Financially Constrained Stock Returns. Journal of Finance 64: 18271862.
Modigliani, F., and H. M. Miller. 1958. The Cost of Capital, Corporate Finance, and the Theory of Investment.
American Economic Review 48: 261297.
Myers, S. C. 1984. The Capital Structure Puzzle. Journal of Finance 39: 575592.
Myers, S. C. 2001. Capital Structure. Journal of Economic Perspectives 15: 81102.
Myers, S. C., and N. S. Maljuf. 1984. Corporate Financing and Investment Decisions When Firms Have Information
That Investors Do Not Have. Journal of Financial Economics 13: 187221.
Penman, S., S. Richardson, and I. Tuna. 2007. The Book-to-price Effect in Stock Returns: Accounting for Leverage.
Journal of Accounting Research 45: 427467.
Shanken, J. 1992. On the Estimation of Beta Pricing Models. Review of Financial Studies 5: 133.
Simutin, M. 2010. Excess Cash and Stock Returns. Financial Management, 39: 11971222.
Soares, N., and A. W. Stark. 2009. The Accruals Anomaly-Can Implementable Portfolio Strategies Be Developed That
Are Profitable Net of Transaction Costs in the UK? Accounting and Business Research 39: 321345.
Stiglitz, J. E., and A. Weiss. 1981. Credit Rationing and Markets with Imperfect Information. American Economic
Review 71: 393411.
Whited, T. M., and G. Wu. 2006. Financial Constraints Risk. Review of Financial Studies 19: 531559.

You might also like