DEFAULT RISK DRIVERS IN SHIPPING BANK LOANS

MANOLIS G. KAVUSSANOSa,* and DIMITRIS A. TSOUKNIDISb, a

a
Department of Accounting and Finance, School of Business Administration, Athens University of
Economics and Business, Athens, TK 10434, Greece.
b
Department of Commerce, Finance and Shipping, Faculty of Management and Economics, Cyprus
University of Technology, Limassol, 3036, Cyprus.

July 2016

ABSTRACT

This paper proposes a credit scoring model for the empirical assessment of default risk

drivers of shipping bank loans. A unique dataset, consisting of the credit portfolio of a

ship-lending bank is used to estimate a logit model with two-way clustered adjusted

standard errors, ensuring robust inferences. Industry specific variables, captured through

current and expected conditions in the extremely volatile global shipping freight markets,

the risk appetite of borrowers–the shipowners – expressed through the chartering policy

they follow – and a pricing variable, are shown for the first time to be the important

factors explaining default probabilities of bank loans.

Keywords: Default risk, bank loans, credit scoring models, shipping.

JEL Classification: G21, G33, C25.
* Corresponding author. Address: Department of Accounting and Finance, Athens
University of Economics and Business (AUEB), 76 Patission St, 10434, Athens, Greece.
Tel: +30 210 8203167, Fax: +30 210 8203196, Email: mkavus@aueb.gr (M.G.
Kavussanos).
 1. INTRODUCTION
The international ocean-going shipping industry is characterized capital intensive as
investment to even a single vessel can often require funds in excess of $100mln,
depending on type and size (Stopford, 2009). Such large amounts of capital are not easy
to raise. Bank loans to finance shipbuilding’s or second-hand ship acquisitions are
historically the most popular source of capital in the shipping industry (see Albertijn et
al., 2011 ). They are based on relationship banking and are providing companies with the
required capital faster in comparison to other sources of finance. The latter is important in
a market where speed of decision making can make the difference when striving to
achieve the best deal. Moreover, it leaves the ownership structure of the company
unchanged, which is not the case for sources of finance, such as Initial Public Offerings
(IPO’s). In addition, bank loans do not require the shipping company to disclose its
business information to the general public as with IPO’s and bond issues - see for
instance Kavussanos and Tsouknidis (2014).

Typically, shipping bank loans are assessed by credit analysis departments of banks
specializing in transportation and ship finance or by shipping departments of commercial
and investment banks lending to the industry. As observed in figure 1, the total value of
the loans advanced globally to shipping firms fluctuates, in line with the cyclical shipping
markets. It remains substantial, reaching a high of $115 billion in 2007 at the peak of the
market, while falling later in 2012 to $46 billion. Before the subprime financial crisis of
2007-2009, 75% of the external funding in the shipping industry took the form of bank
loans. However, this has changed substantially after the subprime crisis period, and the
onset of the shipping market crisis after the middle of 2008. As a consequence, mainly
because of liquidity issues, the substantial restructuring of the banking sector and the very
low earnings of shipping freight markets, banks have either decided to withdraw or
significantly reduce the financing of the sector. It came to a point when, in 2009 for
instance, funding through bank loans only accounted for 32% of the total ship finance
market - see Albertijn et al. (2011). During the post 2009 period, banks’ main operations
shifted from originating new loans to restructuring the repayments of defaulted or

1
technically defaulted1 shipping bank loans - see Gong et al (2013). This was a
consequence of the excess uncertainty in the global economy and the absence of liquidity
in banking - see Santos (2011). However, for the reasons outlined earlier, bank finance is
likely to remain the most important source of capital for the financing of the shipping
industry.

As experienced in financial markets on a massive scale, starting from 2007, default risk
can lead banks and bank-dependent firms to bankruptcy - see for instance Chava and
Purnanandam (2011). Therefore, identifying the factors which drive default risk in bank
loans is extremely important for the survival of banks and other financial market
participants, the stability of the financial system and ultimately the global economy.

Traditionally, loans account for the largest part of total assets in the balance sheets of
commercial banks.2 Thus, lending activity exposes financial institutions to substantial
default risk, as potential default events in individual bank loans lead to losses in the value
of the loan portfolio. In this case, at the micro level, the bank’s profitability deteriorates,
while at the macro level significant losses can affect the stability of the financial system.
Specifically, if a default occurs, typically there is negotiation between the two parties
where the bank may agree to provide a refinancing of the loan by extending its maturity
and/or charge a higher spread. In this case, the borrower is downgraded in the internal
credit rating scale of the bank and the whole amount of the loan is added to non-
performing loans (NPLs). Thus, it is extremely useful for banks to mitigate the credit risk
originating from NPL’s by identifying from the outset the correct risk level entailed in a
bank loan application and as a consequence the potential defaults. Even if a bank does not
eventually lose the whole amount of a defaulted loan, the process of refinancing and
amending the terms of the loan (e.g. extending its maturity, charge higher spreads,

1
A technical default of a bank loan occurs when the obligor of a loan agreement cannot comply with at
least one of the covenants (terms) in the agreement for a pre-determined period of time. For example, due
to adverse movements in vessel prices, the asset coverage ratio (ACR) of a loan - being the ratio of the
ship’s value over the amount of the loan granted – may fall below the agreed threshold, making the obligor
face a technical default situation. In this case, the lending bank may require additional collateral from the
obligor in order for the ACR ratio to be maintained above the agreed threshold.
2
According to end-of-2014 Federal Deposit Insurance Corporation (FDIC) annual report, the balance sheet
item “net loans and leases” accounts for 52.63% of total assets for all commercial banks in the US.

2
transfer the missed payment to the remaining payments) is a major issue since it creates
substantial operational cost and significantly alters the bank’s asset value and strategy.
For this reason it is a matter of paramount importance to develop a model that can detect
such cases from the outset.

This paper proposes a systematic framework for assessing default risk entailed in bank
shipping loans by examining for the first time a complete commercial credit loan
portfolio from the international ocean-going shipping industry. Apart from the
importance of the shipping industry for the global economy, being the leading mode of
transportation worldwide and carrying over 90% of the global trade in volume terms –
see for instance World Trade Organization (2014), there are a number of special
characteristics of the sector that set it apart from other industries in the economy. This is
mainly due to the ocean-going cargo carrying shipping companies: being affected by
international events see Amiti and Weinstein (2011); being structured as one-ship - one-
company offshore legal entities; involving high levels of capital investment, where
individual ships can cost up to 250 million dollars; operating in markets in which
conditions of perfect or near-perfect competition prevail; and facing substantial
investment and operating business risks - for a comprehensive overview of business risks
in shipping see Kavussanos (2010) and Kavussanos and Visvikis (2006).3 As a
consequence, the cash-flow generating ability of ships is uncertain, mainly due to the
very high volatility4 in freight rates, which carries through to substantial variation in asset
(ship) prices. These characteristics directly affect the ability of the companies to generate
sufficient cash-flows for covering both the operating costs of the vessels and servicing
debt repayments due, while at the same time retaining sufficiently high collateral value of
the asset to guarantee the loan. Thus, the complexity of the default risk assessment of

3
The risky shipping business environment has traditionally led market participants and especially investors
to hedge the documented high levels of risk by holding diversified portfolios of shipping assets, by
employing vessels in longer term time charter contracts and by using Forward Freight Agreements (FFA’s)
– for more formal analysis of this see Kavussanos (2003 and 2010).
4
Figure 2 depicts daily standard deviations with 1-year rolling window for the Baltic Dry Index (BDI)
which is considered a barometer for the shipping market, the MSCI world stock index and the Goldman
Sachs Commodities index. As observed, from 1997 to 2008 the BDI index exhibits similar variation with
the stock and commodity indices. However, its standard deviation is consistently higher in the post-
subprime financial crisis period.

3
shipping bank loan agreements is substantially higher when compared to the general
corporate finance loans and deserves special attention. Moreover, it should be noted that
an important feature is “relationship” based bank lending, whereby the “name” of the
ship-owner requesting the loan is considered of substantial importance in the decision
making process.

In response to the above issues, credit rating agencies have also introduced specialized
rating methodologies for shipping companies, aiming to capture the unique
characteristics of the industry, which are distinct from the standard corporate framework
- see for e.g. Moody’s Investor Services (2005). As stated in Moody’s report, “shipping
companies share several relevant business characteristics and common credit
considerations”. The report also recognizes the unique characteristics of the industry,
being the pronounced cyclicality, the capital intensity, the fragmentation of the industry
and the highly volatile financials.

The empirical findings of this study reveal that the important factors in predicting the
probability of default in shipping bank loans are those that measure current and expected
conditions in shipping freight markets, the risk appetite of the obligor as indicated by
his/her chartering policy of the vessel and the pricing of the loan as measured by the
arrangement fee percentage variable that reflects the bank’s assessment of the risk of the
loan. These results indicate that the set of factors driving loan defaults in the shipping
industry are different in comparison to those in other industries (e.g. mortgage loans),
where financial characteristics and/or other characteristics (e.g. ownership) might be
more important in assessing bank loan default risks.

Kavussanos and Tsouknidis (2011) estimate for the first time in the literature a credit
scoring model based on secondary shipping loan data. This paper, being a more advanced
version of the aforementioned preliminary study, contributes to the literature in the
following ways: First, it sets the theoretical framework to investigate for the first time
shipping bank loan agreements within the 6C’s of credit theory – see for instance Smith
(1964). Second, it is the first to investigate empirically through a panel data logit model

4
the probability of default of such loans with two-way clustered adjusted standard errors -
see Petersen (2009) and Thompson (2011). In this way, the “correct” standard errors are
reported, which are adjusted for loan and time effects, along with correlation
(autocorrelation and cross-correlation) and heterogeneity effects. This correction has been
shown to be necessary, to enable reliable inferences in panel data logit models. Failing to
adopt it, leads to biased standard errors and doubtful statistical inferences; see for
instance Kavussanos and Tsouknidis (2014) for an application in the shipping bond
market. Third, a unique dataset of 128 shipping bank loans is utilized to empirically
investigate the theoretical relationship between default risk and a number of potential
explanatory variables in the credit scoring model. Fourth, a set of unique variables have
been utilized for the first time to explain probabilities of default in shipping bank loans,
which include current and expected conditions in shipping freight markets, the chartering
policy of the shipping company applying for the loan and proxies for the ‘relationship
banking’ effect apparently prevalent in shipping bank loans decisions. The results of the
study are of interest to: the credit risk departments of banks financing shipping projects,
ship-owners, individual and institutional investors, regulatory authorities and academics.

The rest of this paper is organized as follows: Section 2 reviews the related literature;
section 3 outlines the methodology; section 4 describes the dataset; section 5 presents the
results; section 6 discusses the main findings, whereas section 7 concludes the paper.

2. LITERATURE REVIEW
2.1 Basel agreements framework
The importance of default risk assessment is recognized through the Basel I (1988), Basel
II (2004) and Basel III (2010)5 regulatory agreements. The Basel framework introduces
common rules regarding capital reserves requirements, which are linked to the degree of

5
In general Basel III is focused on strengthening bank capital requirements by enhancing bank liquidity
and limiting bank leverage (for details see Basel III, 2010). Financial institutions are required to comply
with the proposed changes from 1st April 2013 until 31st March 2019. A full list of member countries of the
Basel Committee on Banking Supervision (BCBS) can be found in:
http://www.bis.org/bcbs/membership.htm.

5
default risk for each loan. These rules are incorporated into national laws in the great
majority of developed and developing countries around the world. Specifically, banks are
required to use internal or external credit rating models to classify borrowers according to
their default risk probabilities. Thus, Basel II provides banks with two options regarding
the methodology followed to determine their capital reserves against default risk: the
standardized approach (SB), where risk weights are defined in accordance to external
ratings, and the internal ratings based (IRB) approach, which allows financial institutions
to estimate capital reserves using their own internally developed risk models. That is, the
agreement specifies that banks should hold at least 9% of their risk-weighted assets as
reserve capital, which is higher than 8% that it was before 2011. Thus, different classes of
assets are weighted according to their perceived industry risk. For example, shipping
loans have a risk weighting of 100%, whereas house mortgage loans have a risk
weighting of 50%. Then, loan applications, say, within shipping, are classified according
to the credit rating of the obligor applying. In this way, capital requirements are
determined based on the classification of loans, through the credit rating process rather
than being constant irrespective of the credit type, as they were under previous versions
of the capital accord. Basel III agreement retains the non-flat risk weightings originally
introduced in Basel II.

As pointed out by Blum (2008), banks have the incentive to underestimate the probability
of default (PD) for a borrower in order to keep a lower capital reserve. For this reason,
regulators pay more and more attention to testing the accuracy of the internally developed
credit risk models by using several validating procedures. Efforts to understand the
impact of Basel II risk weighting include Ruthenberg and Landskroner (2008) who, by
using Israeli economic data of a leading bank for the period 1998 to 2006, show that high
quality corporate and retail customers usually enjoy lower interest rates for loans drawn
by relatively big banks, which, most probably, will adopt the more sophisticated IRB
approach. Mainly due to the late 2000’s subprime crisis, the Basel III (2010) agreement
poses further controls on Tier I and Tier II pillars introduced originally in Basel I (1998)
and retained in Basel II (2004) agreement. Specifically, Basel III (2010) agreement
requires banks to hold 4.5% of common equity (2% in Basel II) and 6% of Tier I capital

6
of risk-weighted-assets (RWA) (4% in Basel II). Also, the agreement poses a leverage
ratio in excess of 3% and liquidity requirements based on the computation of “Liquidity
Coverage Ratio” (LCR). The leverage ratio is calculated by dividing Tier I capital by the
bank's average total consolidated assets and the "Liquidity Coverage Ratio" requires a
bank to hold sufficient high-quality liquid assets to cover its total net cash outflows over
30 days. Tier I (core capital) consists primarily of shareholders' equity. It is not inclusive
of any "goodwill", defined as an intangible asset of the firm which represents the
company’s brand name, solid customer base, good customer relations, good employee
relations and any patents or proprietary technology. Tier ΙΙ (supplementary capital)
comprises long-term subordinated loans and fixed assets reserves, and is limited to 50%
of Tier I capital.

2.2 Default risk drivers of bank loans

Previous studies on identifying the credit risk drivers of firms’ default risk and corporate
bank loans, through the estimation of credit scoring models, date back to Altman’s (1968)
z-score methodology.6 In the general corporate finance literature, a system of credit risk
assessment, namely the “5C’s of credit”, standing for Capacity, Capital, Collateral,
Conditions, and Character, has emerged from the 1960’s (for details see Smith, 1964),
and is widely used for the systematic classification of the default risk drivers in loan
facilities (see for instance Chen et al., 2009). Later, a 6th “C” is added to account for the
“Company” effect.

Past studies on the issue include: Bhimani et al. (2014), who examine an extensive
dataset of corporate bank loans in Portugal, over the period 1997-2003. The authors

6
Altman’s (1968) z-score utilizes financial data to compute the z-score of each company. Five financial
ratios are required to estimate the z-score, namely: Working Capital (WC), Retained Earnings (RE),
Earnings Before Interest and Taxes (EBIT), Sales (S), Total Assets (TA), Market Value of Equity (ME) and
Total Liabilities (TL). The following weightings for the calculation of the z-score were estimated by
Altman, utilizing a sample for the US market. Thus, the z-score for a company is given as:
z = 1.2 x (WC/TA) + 1.4 x (RE/TA) + 3.3 x (EBIT/TA) + 0.6 x (ME/TL) + 0.999 (S/TA).
The ratio (WC/TA) captures the short-term liquidity of a firm, the RE/TA and EBIT/TA measure the
historical and the current profitability respectively, the ME/TL is a measure of leverage and finally the
S/TA indicates the market competitiveness of a company. The model is constructed so that, the higher the
z-score the less the default risk of a company. A z-score above 3 is an indication that default is very
unlikely to happen, a z-score below 1.8 indicates that default is very likely, while values between 1.8 and 3
are a “gray” area.

7
highlight the importance of the liability status and show that it can be used as a signal to
reduce information asymmetries between borrowers and lenders in credit markets.
Limited liability leads to higher probability of default because the owners are liable only
up to the amount of the capital contributed. A negative relationship is also shown
between probabilities of default and earnings, liquidity and asset coverage.

Next, Bonfim (2009) examines an extensive dataset with detailed financial information,
including historical records of bank loans’ defaults, for more than 30,000 Portuguese
firms covering the period 1996 to 2002. By modeling default probabilities with discrete
choice models the author finds that default probabilities of bank loans are influenced by
several firm-specific characteristics, such as their financial structure, profitability and
liquidity. Chang et al. (2014) provide evidence on the importance of ‘relationship
banking’ in the assessment of probabilities of default, analyzing a dataset of corporate
loans from China for the period 2004-2006. The authors highlight that firm-specific hard
information, such as firms’ financial ratios, are significantly related to the probability of
default. In addition, non-financial attributes of obligors, such as their attitude towards
their propensity to default on mortgages are also highlighted as important in explaining
the observed defaults by Guiso et al. (2013). Jiménez and Saurina (2004) examine a
dataset from the Spanish market for the period 1988 to 2000. They show that:
collateralized loans have a higher probability of default (PD); loans granted by savings
banks are riskier and that a close bank-borrower relationship increases the willingness of
banks to take more risks.

2.3 Shipping bank loans

The evaluation of a new shipping loan application is not a simple process, since several
pieces of information about the potential borrower have to be combined in order to assess
in the best possible way the probability of default entailed in the loan agreement. Thus,
financial institutions advancing loans face the challenge of developing a model for the
assessment of default risk which combines the information available to the bank and to
classify loans according to their default probability. When assessing default risk of

8
shipping bank loan agreements, banks involved in shipping finance face some
challenging issues. These include: often, the lack of financial transparency of shipping
companies – being mostly private family business with non-audited accounts; the
inadequate evidence of their clients’ personal wealth and other activities and the need to
find “reliable” collateral to secure the loan. Overcoming these problems, leads financial
institutions involved in ship-lending projects to rely heavily on “relationship banking”,
where the “name”, the track record and the personal wealth of the shipowner are very
important criteria in assessing the loan application.

The special features of shipping companies, compared to other sectors of the economy
financed, call for careful handling of the vessel-backed loans advanced by banks. A
vessel-backed loan is one where the ship itself is used for security. Typically, shipping
bank loans lie within the “object-finance” class of loans, where the cash-flows generated
by the asset (ship) are used as collateral for the repayment of its own loan due. In this
way, the bank gains ownership of the vessel if the obligor cannot honor its contractual
obligations and follow the repayment schedule of the loan due. Thus, a bank can provide
a company with the required funds, which would otherwise not be granted due to the
difficulties involved in assessing its creditworthiness.

Several types of covenants are used in practice in order to secure repayment. The
covenants which are mostly used in shipping bank loans include: a first mortgage on the
ship, which dictates that the lender (mortgagee) has priority in a claims sequence and that
she can take possession of the asset if the borrower (mortgagor) fails to fulfill his pre-
agreed financial obligations; the assignment of income, where the borrower assigns to the
lender the portion of the revenue from the operation of the vessel required to pay interest
and capital repayments; a corporate or personal guarantee, where the lender demands
guarantees from the owning company or the owner himself and security maintenance
clauses, which protect the bank in case the remaining market value of the asset is less
than the outstanding loan. Close examination of these loans is needed to reveal whether
the aforementioned characteristics, set apart the factors that the credit risk departments of

9
banks should analyze to assess the default risk involved in shipping loan agreements and
predict defaults.

Previous empirical work in the area include: Dimitras et al. (2002), who employ the
utilities additives discriminants (UTADIS)7 method to group into homogeneous classes a
sample of 17 shipping bank loans, drawn over the period 1999-2001. The authors
investigate a set of credit criteria, the weights of which are pre-determined by a financial
consulting firm and then use the UTADIS method to specify a utility function and the
cut-off utility level for granting a loan. In another study, Gavalas and Syriopoulos (2014)
draw inferences from primary data, which are collected from a bank survey
questionnaire. They utilize a small sample of 16 managers from different commercial
banks in Greece, which are active in lending in the shipping sector. Bank managers were
asked to identify the factors which they considered important when assessing shipping
bank loan applications. Gavalas and Syriopoulos (2014) also utilize the UTADIS method
to determine a utility function and the cut-off utility thresholds in a manner similar to
Dimitras et al. (2002). It is found that a credit rating migration probability, debt-to-equity
ratio and the asset coverage ratio are significant in explaining defaults in the sample
utilized. However, specifying the relevant factors when assessing default risk of shipping
bank loans through survey’s, entails subjectivism, as each manager may evaluate
differently the weighting and the composition of the relevant factors when assessing a
loan agreement.

The UTADIS method applied in both the above studies, apart from other issues discussed
in the next section of the paper, is less easily implemented when compared to credit
scoring (logit) models.8 In contrast to previous efforts, the current study lets the data
speak for themselves and identifies the default risk drivers of shipping bank loans

7
The UTADIS method belongs to the wider family of the UTA multicriteria methods - for details see
Devaud et al. (1980). Given a pre-specified grouping of criteria, in this case credit loan criteria, the
UTADIS method seeks to provide an additive utility function and the corresponding utility thresholds that
leads to the minimum error when evaluating default risk.
8
This is due to the fact that the UTADIS method requires linear programming and constrained optimization
skills which are computationally and technically more demanding in comparison to the estimation of
logistic regression models applied in this paper.

10
through a logit regression model, fitted to an extensive sample of 128 bank loans drawn
over the period 1997-2011.

Mitroussi et al. (2016) adopt the logit credit scoring model methodology, introduced first
in the analysis of shipping loan data in Kavussanos and Tsouknidis (2011), to examine
the performance risk drivers of a limited sample of 30 shipping bank loans, for the
acquisition of dry-bulk vessels only, over the period 2005-2009. Albeit attempting to
justify the choice of focusing in the dry bulk sector only, it is well known in the literature
(see Kavussanos 2010, 2003, 1997, 1996; and Tsouknidis, 2016) that the shipping
industry is highly segmented. Thus, focusing only on dry-bulk shipping loans might not
be sufficient to enable inferences to generalize over the whole shipping sector. Credit
analysis departments of banks evaluating shipping loan applications are typically
interested in the spectrum of shipping subsectors, particularly as they wish to construct
diversified portfolios of shipping loans. Moreover, the aforementioned study covers a
period of time which is shorter than the typical shipping cycle of 6-7 years (see Stopford,
2009), and thus results may not generalize to all parts of the cycle. As observed later,
through the results of the current study, the current and expected state of the shipping
markets is indeed significant in explaining shipping loan defaults. In addition, they do not
use two-way cluster adjusted standard errors (see Pettersen, 2009) to draw inferences, an
omission that has been shown in Kavussanos and Tsouknidis (2014) also to lead to
possible incorrect inferences regarding the significant variables to explain shipping
bonds’ spreads. Finally, the descriptive work of Grammenos (2010) outlines the 6 Cs of
credit as applied to shipping bank loans. Kavussanos and Tsouknidis (2011) and its
extension in the current paper are actually the first ones to implement these empirically
for shipping loan portfolios.

3. METHODOLOGY
Since Altman's (1968) influential paper, several studies have adopted Multivariate
Discriminant Analysis (MDA) or Linear Discriminant Analysis (LDA) techniques in
order to develop z-scores to assess the default risk of companies. However, it is well

11
known in the literature (see for e.g. Karels and Prakash, 1987) that MDA suffers from
restrictive assumptions. These include that the discriminating variables are distributed as
jointly multivariate normal and that the covariance matrices for failed and non-failed
companies are equal. Typically, these assumptions are violated, thus limiting the
appropriateness of the method. In contrast, a logistic regression model is free from such
restrictive assumptions. In fact, as documented in the literature9, logit models overcome
the problems of MDA and form superior methods of developing a credit scoring model
for different markets. Moreover, in the credit risk literature standard discrete choice
models such as logit or probit models are widely used (see for instance Campbell et al.,
2008), since they are free from the restrictive assumptions of other methods. Based on the
results of previous studies on the issue and the objective of this study to identify the
default risk drivers of shipping bank loans, this paper utilizes the standard logistic
regression model approach for estimation purposes.

A critical issue in a study on default risk drivers is the definition of default. Three
“default” definitions are used in the extant literature: First, a loan is characterized as
“doubtful” if “full payment appears to be questionable on the basis of the available
information”. Second, a loan is characterized as in “distress” if a payment, i.e. interest
and/or principal, has been missed for more than 90 days. Third, a loan is characterized as
in “default” if a formal restructuring process or bankruptcy procedure has started. In
Basel II (2004) and Basel III (2010), a loan is considered to be in default if a payment is
missed for more than 90 days. This definition is adopted also in Chen and Deng (2013)
and Gerardi et al. (2013), amongst others.

9
In an early study, Laitinen and Kankaanpaa (1999) compare six alternative methods based on MDA and
logit models, which have been applied to build credit scoring models, and show that the logit models and
the MDA, despite its restrictive assumptions, are the most prominent and accurate methods. However, in a
related study, Laitinen and Laitinen (2000) adopt the logit instead of the MDA model, in order to construct
a default prediction model free from the problems associated with the normality of variables assumption. In
another study, Barniv et al. (2002) support the view that the logistic regression is the most popular
technique in the relevant literature when the objective is to construct a default prediction model. Charitou et
al. (2004) perform a comparison of neural networks and logit models and conclude that both are accurate
methods for evaluating the probability of default.

12
Typically, several factors can affect the borrowers’ PD in bank loan agreements. A
scoring model can be utilized to combine the different bits of information a bank may
have at its disposal and reveal the relevant ones to assess the PD of a borrower. In this
paper, the development of a credit scoring model is empirically examined through the
identification of the relevant default risk drivers for shipping bank loans. The model in its
general form is shown below:

Scoreit = b0 + b1x1it + b2x2it + … + bKxKit + uit = xit + uit (1)

with ( ) ( ) ( ) ( )

where i=1, 2,…, n identifies the loan; t=1, 2,…,T denotes the time period; Scoreit
represents the credit scoring for loan i at time t, taking continuous values in the range
[0,1] – see equation (2) below for the exact definition of this dependent variable; xkit
(k=1,…,K) denotes the matrix of the K independent variables used to explain default risk
in shipping bank loans; b (bi; i=1,…,K) stands for the vector of the corresponding
parameters; and uit is the random error term, which follows a white noise error process
with a distribution that has mean zero and variance σu2 and stands for the within-loans
errors. A consequence of the above is that uit is orthogonal with the regressors in the
model; that is, E(uitxkit) = 0.

A logit model is used to estimate equation (1) by utilizing historical records of a cross
section of shipping bank loans. The aim is to forecast the probabilities of default (PD’s)
of loans and to discriminate between defaulted and non-defaulted loans. In practice, this
is achieved by representing default probabilities as a logistic function, which when
applied to (1) leads to:

( )
( ) ( ) = (2)
( ) ( )

13
where, the discrete indicator variable yit takes the value one (1) if the loan defaulted in
year t and zero (0) otherwise. The estimated coefficients, represented by the b vector, are
obtained by maximizing the following log likelihood function:

∑ ( ) ( ) ( ) (3)

The dataset utilized follows a panel data structure, where each obligor may have more
than one loans with the bank at any point in time. A logit model with two-way clustered
adjusted standard errors is estimated, which corrects for the presence of autocorrelation
and heteroscedasticity in the residuals (see Petersen, 2009 and Thompson, 2011). This
“Petersen”-correction is deemed important in revealing the “true” model in panel data
estimation frameworks (see Kavussanos and Tsouknidis, 2014), as it enables estimation
of unbiased standard errors, and as a consequence reliable inferences on the significance
of the estimated coefficients. A potential problem with the estimation and testing of Eq.
(1) is the presence of endogeneity. This could derive either from simultaneity (reverse
causality) or from omitted variables bias. A simple, albeit imperfect remedy, is to include
a time-constant entity fixed effect. The joint significance of the time dummies
coefficients is tested through an F-test.

In order to test for a varying impact across loans, borrowers’ or time, loan-fixed effects,
firm-fixed effects and time-fixed effects as well as the joint significance of all of them are
introduced and tested for in the above model through Wald tests. Thus, loan, firm and
time fixed effects are included as explanatory variables in the estimated models through
the introduction of loan, firm and time dummies, respectively. This allows for potential
unobserved heterogeneity, i.e. when variables related to loan, firm or time characteristics
are not included in the estimated models, this may introduce heterogeneity in the
available sample and alter the results (omitted variables issue). In all cases, Wald tests do
not reject the null hypothesis that the coefficients of these dummy variables are jointly
equal to zero.10 To that effect, all estimations are performed using pooled OLS logit

10
The results are available from the authors on request.

14
regressions with two-way clustered adjusted standard errors, without the inclusion of
fixed-effects dummy variables.

Following the 6C’s of credit theoretical framework, for the identification of default risk
drivers of shipping bank loans, all explanatory variables xkit in eq. (1) are classified and
listed under each C of credit. Thus, all xkit explanatory variables entering equation (1),
along with their precise definitions and their a-priori expected signs, are presented next
and summarized in table 1.

Financial-specific variables are included to capture the effect on the probability of

default of the capital (financial) structure, the liquidity and the profitability of the
company. They include: (1) Leverage, defined as Total Liabilities (TL) / Total Assets
(TA). This is indicative of the loan gearing of the company; that is, of the mix of debt and
equity used to finance the company. The expected sign is positive since the higher the
financial leverage, the higher is the expected probability of default of the loan. (2) The
debt coverage ratio, used as an indication of the ability of the company to cover the short
and long term debt burden through its operating profits. For comparison and robustness,
four different definitions of this ratio are used (not simultaneously): Current Liabilities
(CL) / Earnings Before Interest Taxes Depreciation and Amortization (EBITDA); Long
Term Debt (LTD) / EBITDA; TL / EBITDA; and Interest Expenses / EBITDA. The
higher the Debt Coverage Ratio the higher is the expected probability of default for a
shipping bank loan. (3) The liquidity ratio, used as an indication of the level of liquidity
maintained in comparison to the debt principal and interest payments due. Two different
definitions are used: the current ratio, defined as Current Assets (CA) / Current
Liabilities (CL); and the cash reserves ratio, measured as Cash and Cash Equivalents /
TA. Liquidity ratios provide an indication of the ability of the company to pay all of its
debts as they come due, say over the next year or so. As such, a negative sign is expected
of its estimated coefficient, since higher liquidity increases the chances that the loan
repayment is followed as scheduled and no default occurs during its life time. (4) The
profit margin, measured as EBITDA / Revenue, indicates the profitability of the shipping

15
firm. It shows the combined effect of leverage, liquidity, asset management and debt on
the operating profit of the company. A negative sign of its coefficient is expected since
the higher the profit margin, the lower is the expected probability of default. All
financial-specific variables are drawn from the audited financial statements that the
shipping companies provided to the bank as part of the loan evaluation process.

Firm characteristics’-specific variables include: (1) The age of the firm, measured as
the time period in years since the shipping company was established. The expected sign
is negative, since younger firms are expected to have less experience with the shipping
industry and as a consequence higher probabilities of default. (2) The years of
cooperation with the bank (yocb), measures the number of years since the first loan
agreement is made between the bank and the shipping company. A negative sign is
expected since a longer relationship between the bank and the shipping company would
be a consequence of a good track record by the later, leading to lower PD’s. (3) The time-
charter (TC) policy, captured through a dummy variable, measures whether the shipping
company operates the majority of its vessels (over 50%) on time-charter contracts (taking
the value 1) or on trip voyage charters (taking the value 0), when the loan is granted.
There was no case where a shipping company had exactly or almost 50% of its fleet in
time-charter and the rest 50% in trip voyage agreements. A negative sign is expected
since, ships employed under time-charters have more stable and predictable cash flows,
thereby making it less likely to default on their loan obligations.

Loan-specific variables include: (1) The Arrangement fee / Amount of loan, is a ratio
representing one of the main sources of profit for the bank, expressed as a percentage of
the loan granted. A positive sign of its coefficient is more likely, since typically the bank
will charge higher arrangement fee as a percentage of the amount of the loan granted to
customers with higher risks, as higher PD are expected in such cases. (2) The variable
internal bank rating, refers to the internal risk rating assigned by the bank, based on the
overall assessment of each credit loan application (1=Completely Safe, 7=Acceptable

16
risk, >7 Non-acceptable risk). However, the ‘internal bank rating’ variable exhibits high
pairwise correlation coefficient (0.6104) with the variable ‘arrangement fee / amount of
loan’. For this reason the variables ‘internal bank rating’ and ‘arrangement fee / amount
of loan’ are not used simultaneously during estimation of the models later in the paper.
Based on Schwarz's (1978) Bayesian information criterion (SBIC) and McFadden’s
pseudo-R2 diagnostics, only the ‘arrangement fee / amount of loan’ variable is used
during estimations, as it gives better results. (3) Margin refers to the (margin) spread over
LIBOR charged by the bank on the nominal amount of the loan granted. Higher margins
typically reflect higher credit risks and therefore higher PD for the loan. Thus, a positive
coefficient is expected. (4) The life to maturity, measured as the number of years from the
initiation of the loan until its maturity, has no certain a-priori expected sign of its
coefficient. Arguments about how this variable can affect the probability of default can
be made in either direction. For instance, when a loan repayment schedule is followed
normally for the first several years of a loan’s life, it may be argued that this reduces the
probability of default for the remaining years until its final maturity. This is also the case
in the popular equal payment amortizing loans where the values of the interest payments
are reduced as one get closer to the end of the tenor. However, in the risky shipping
sector, due to the high volatility in freight rates and revenues, surviving the first years of
a loan’s life may not be a guarantee that the remaining payments of interest and principal
will take place. Thus, both a positive and a negative value of the estimated coefficient can
be justified. (5) The asset coverage ratio (ACR) contractual is defined as the ratio of the
market value of the ship over the amount of the loan at the initiation of the loan
agreement. Its threshold is defined according to the bank’s risk appetite, potential credit
policy regulations and the state of the capital and freight markets. A negative sign is
expected since, the higher the ACR ratio, the higher is the security for the bank and the
lower is the probability of default. (6) The asset coverage ratio (ACR) actual is the same
as the ACR contractual but this time the market value of the ship, and as a consequence
the ACR ratio, changes during the life of the loan, according to the demand and supply
forces in the vessel market. The expected sign is negative as, higher values of the ratio
indicate lower risk of the loan defaulting.

17
Industry-specific and macro variables include: (1) The spreads of 1yr TC – Spot and
3yr TC – Spot measure the difference between 1-year time charter rates and the spot
voyage rates and of the 3-year time-charter rates and the corresponding spot voyage rates,
for each of the individual subsectors of cargo carrying ships. A broad classification of
cargo-carrying shipping sub-sectors includes: dry bulk carriers, tankers, containerships
and gas carriers. The spreads11 are used to capture expectations in freight rate markets. As
shown in Kavussanos and Alizadeh (2002) these spreads are negative on average,
indicating the relative safety of TC contracts in comparison to their corresponding spot
ones. This is because, amongst other reasons, TC contracts “lock” the revenues of the
ship for a specific period of time and in this way protect against the uncertainty of
seeking to hire the vessel again in unknown market rate(s) at the end of the voyage
contract. It is shown further that the spread is time-varying and depends on the volatility
of the freight markets. Adland and Cullinane (2005) support this view. Apart from being
generally negative, to reflect the safety of “locking” cash flows for a period of time, the
spread can also become positive when the freight market is strong, in which case short-
term freight contracts are preferred over longer-term ones. When a shipowner enters a
period charter agreement the exposure to volatile spot freight rates is eliminated, but his
inability to terminate or sell the agreement creates liquidity risk. However, if the spot
market moves against the charterer he may be tempted to default or renegotiate the terms
of the agreement. Thus, the a-priori expected signs of the estimated coefficients for the
spread variables cannot be determined with certainty. This is because, a small negative or
even positive value of the spread corresponds to the time-charter rate being close or even
above the spot rate, which generally indicates an optimistic expectation for the long-term
spot market in the future but can also indicate a pessimistic expectation for the short-term
spot market if the prevailing freight rate level is high. In the former scenario, shipping
bank loans are expected to exhibit lower PD’s, whereas in the latter higher PD’s. (2) The
orderbook / total fleet, measures for each subsector the percentage of vessels (in dwt) on
order or under construction in the market over the total fleet capacity. This ratio is
computed by using orderbook figures for each specific shipping sub-sector according to

11
1yr and 3yr spreads are computed separately for dry bulk and for tankers. However, due to lack of data,
this was not possible for other shipping sub-sectors examined in this paper, such as diversified,
containership and gas.

18
the vessel profile of each loan in the sample12. Orderbook forms a forward-looking
indicator for the vessel market. Here the expected sign is positive since, a higher
orderbook compared to the total fleet of the sector would reflect lowering freight rates
and thus income in the future, thereby increasing the PD’s of shipping bank loans. (3)
The inactive tonnage / total fleet variable measures the total inactive tonnage at a global
scale for all cargo-carrying shipping sectors13 over the total fleet. This ratio is computed
by using the total dwt of the vessels removed (removals) for each specific shipping sub-
sector according to the vessel profile of each loan in the sample. Here, the expected sign
is positive since higher inactive tonnage compared to the total world fleet indicates lower
freight market rates and thus lower income for the shipping companies, leading to higher
PD’s for shipping bank loans. (4) The Clarksea index is a freight rate index published by
Clarkson’s Research Limited. It is constructed to be representative of freight rates in all
cargo carrying sectors of the global shipping industry. ClarkSea indices are published by
Clarkson’s Research on a weekly basis as indicators of earnings for all the main
commercial vessel types involved in ocean cargo transportation of various commodities.
The sectors covered in the ClarkSea Index are Oil tankers (VLCC: 200,000 – 399,999
dwt; Suezmax: 120,000 – 199,999 dwt; Aframax: 75,000 – 119,999 dwt and clean
product carriers), Dry bulk carriers (Capesize: 150,000+ dwt; Panamax: 60,000 – 70,000;
Handymax: 50,000 – 60,000 dwt and Handysize: 15,000 – 35,000 dwt), Gas carriers
(Very Large Gas Carriers: 180,000 – 320,000 dwt) and fully cellular containerships.
Separate Clarksea indices are constructed for each of these subsectors of shipping. The
overall ClarkSea index is constructed as a weighted average of freight rates in all
shipping subsectors, with the weights used reflecting the number of vessels in each fleet
sector. The indices are constructed from rates directly collected from Clarksons’ brokers
on a daily and weekly basis. The expected sign of the Clarksea freight index used is
negative since higher values of the index indicate higher freight income and as a

12
The ratio is also computed with the general - not sector-specific - orderbook values with no significant
changes in the results.
13
Inactive tonnage, as defined by Clarkson’s Shipping Intelligence Network (SIN) database, can be due to:
(i) damaged and out of service vessels, (ii) idle vessels and awaiting orders, (iii) idle for more than six
weeks, (iv) laid-up vessels, (v) vessels in long term storage (for more than 60 days), (vi) vessels being
repaired. The ratio is also computed with the general - not sector-specific - removals values with no
significant changes in the results.

19
consequence improved cash flows for shipping companies, thereby reducing the PD’s for
shipping bank loans. (5) The MSCI world stock index, is representative of global stock
market prices. It is a stock market index maintained by MSCI Inc., formerly Morgan
Stanley Capital International and is used as a benchmark for a global stock portfolio. The
index includes 1,500 stocks from 23 developed markets in the world. Higher levels of the
index indicate in turn: a stronger global economy, improved international and therefore
seaborne trade, higher freight rates and thus lower PD’s for bank loans. A negative sign
of the coefficient of the variable is thus expected.

4. DATASET
The dataset used for the analysis in this paper is unique. It is the most extensive portfolio
of shipping loans examined empirically. The entire shipping loan portfolio of a ship-
lending bank is collected, loan by loan, from the credit assessment files of the bank. The
data comprises 484 loan-year observations, and includes 128 loans advanced to 63
shipping companies14 over a 14 year period, from 1997 to 2011. In order to ensure that
we avoid any sample selection bias in the dataset of companies analyzed, the mean values
of the characteristics of shipping firms examined here are compared to those of the
shipping industry. It is found that profitability, debt, interest cover ratio, etc., are very
close to the ones reported in such studies as Grammenos et al. (2008) and Kavussanos
and Tsouknidis (2014), indicating that the companies used in the current paper are typical
of shipping companies. The time period over which the loan decisions are made includes
all phases of the general business and shipping business cycles. This is deemed important
for results to generalize irrespective of the phase of the cycle that the decisions are taken
but also to capture the possibility that the phase of the shipping and general business
cycle affect decision making by the bank. As shown later in the paper, shipping industry
cycle characteristics indeed influence decision making.

14
Out of the 63 companies, 13 (20.63%) are listed. The inclusion of a listed vs. non-listed dummy variable
does not alter significantly the results of this paper and the estimated dummy coefficient is not statistically
significant at any reasonable significance level.

20
Out of the 128 loans, 102 (79.68%) are syndicated, where a group of banks cooperate to
provide funds to a single borrower, and 26 (20.31%) are bilateral loans where the bank
which provided the dataset acted as the sole agent. For all syndicated loans, the bank that
provided the data was the arranging bank and assessed the bank loan applications. No
data were available for the number of banks involved in a syndicated loan or their
percentage contributions to the loan. The great majority of the obligors (>95%) have
provided the bank with audited financial statements, comprising historical information on
company balance sheets, income and cash-flow statements. The frequency of the data is
annual for the great majority of the obligors (>95%). If quarterly data are available, only
end of year values are used. In several cases, data for the explanatory variables are not
available for the whole period of the sample. In such cases, they are treated as missing
variables. The final number of unique loan-year observations included in the sample is
474. Out of the 128 loans, 7 have missed a payment for more than 90 days and therefore
are classified as in default15. The record of historical payments forms our dummy
dependent variable, taking the value 1 when the loan is due for more than 90 days and 0
otherwise. The sample does not include loan applications that were rejected at the outset,
only the loans advanced. This is because the bank does not keep any records of rejected
loans.

Regarding the vessel profile of loans included in the sample, 192 (or 40.50%) loan-year
observations are tanker loan-years, 127 (or 26.79%) are assigned to the diversified
category, 88 (or 18.56%) refer to the dry bulk subsector, 44 (or 9.28%) come from the
containership subsector and 23 (or 4.85%) are from the gas subsector. Each shipping firm
is classified according to the type of vessels included in its fleet at the time the loan
application is assessed by the bank. If no shipping sub-segment (i.e. dry bulk, tanker,
containers and gas) accounts for more than 50% of the total fleet then the obligor is
classified as diversified. The number of loans per vessel profile are the following: 46
loans (or 35.93%) are in the tanker segment, 34 loans (or 26.56%) belong to the

15
Robustness tests using different number of days to classify a loan in default, for example 30 to 60 days or
60 to 90 days, were not possible to be performed, since the bank does not record in a systematic way
missed payments for periods less than 90 days. Also, records on any restructuring or renegotiation of
problematic loans were not available in the bank’s database.

21
diversified category, 33 loans (or 25.78%) are in the bulker segment, 9 loans (or 7.03%)
are in the containerships category while 6 loans (or 4.68%) belong to the gas market
segment.

Table 2 presents descriptive statistics of the variables used in the estimation of the
econometric model. All variables are winsorized16 at 1% level in order to reduce the
effect of possible outliers. Jarque and Bera (1980) test statistics strongly reject the null
hypothesis of normality at the 5% significance level for the majority of the variables
examined. The null hypothesis of normality is not rejected for in some sectors of the
variables internal bank rating, arrangement fees, ACR Contractual, ACR Actual and life
to maturity of loan. Both excess skewness and excess kurtosis are observed in several
cases.

As observed in Panel A of table 2, the average amount of loans advanced is over $30
million, ranging from $0.65 million for working capital loans to $310 million for
syndicated loans - typically to finance the acquisition of newbuildings or second-hand
vessels. Loans advanced in the containership and tanker segments are on average larger
when compared to the dry bulk or gas segments. The same pattern holds for the tenor of
the loans, where loans financing containership and tanker vessels are on average longer.
The tenor of the loans is relatively short on average and equal to 5.14 years, with a range
of 4 months for short-term working capital loans up to 15 years for long-term tanker
loans. Arrangement fees are higher in the containership segment, followed by bulkers,
tankers and gas carriers. Typically, arrangement fees are collected in a single installment
at the initiation of the loan agreement and range from 15 to 175 thousand dollars. Gas
loans carry higher margins on average, followed by bulker, containership and tanker
segments. Margins charged range from 1% to 4% over LIBOR, and reflect the credit
assessor’s risk assessment of each loan agreement. The average internal rating by the
bank for the loans advanced is 3.63 on a scale 1 to 7, where 1 corresponds to ‘excellent’
(very safe) and 7 to ‘acceptable’ level of risk. It is worth noting that some loans are on

16
Winsorizing sets all the data points less than the 1st percentile of each variable equal to the 1st percentile
and all the data points exceeding the 99th percentile equal to the 99th percentile, thereby excluding extreme
observations from the sample.

22
the acceptable category. Internal bank ratings assigned to containership and tanker
segments’ loans are on average larger in comparison to bulker and gas segments. As for
the rest of the loan-specific variables, the ACR contractual ratios range between 125%
and 185%, with an average value of 146%. That is, the bank has provided loans which
range from 54% to 80% of the value of the vessel, while on average they provided bank
finance to the level of 69% of the value of the asset. The bank requires on average higher
ACR contractual ratios for bulkers and containerships in comparison to tankers and gas
carriers. However, with vessel values changing during the lifetime of the loan, the ACR
actual ratio has ranged from 125% to 275%, with an average value of 182%. That is, the
loan to vessel value ratio has ranged during the life of the loan from 36% to 80%, with an
average value of 55%. ’ACR actual’ ratios are larger for bulkers (214%) when compared
to the rest of the segments, where for containerships the ratio stands at 166%, for gas
carriers at 163% while for tankers it is 162%. These ratios are considered important, as
reduction of ‘ACR actual’ below 100% poses substantial risk for the bank providing the
loan. For the loans examined in this paper ‘ACR actual’ has not fallen below 125% in
any case.

Panel B of table 2 presents descriptive statistics for all the Financial and Firm
characteristics’ variables. Given the capital intensive nature of the shipping industry,
firms in the sample exhibit moderate leverage ratios, computed as total liabilities over
total assets, with an average value of 0.53, and range of values from 0.12 to 0.94. Large
debt ratios are involved when the debt of the obligor is expressed as a percentage of the
profits (EBITDA); for example, the total liabilities over EBITDA ratio exhibits an
average value of 5.03. The interest expenses over EBITDA ratio has an average value of
0.21, indicating that shipping companies in the sample can cover on average almost 5
times their interest expenses through their earnings. When turning to liquidity ratios,
firms in the sample exhibit an average current ratio of 1.40, indicating that current assets,
where maturity is less than a year, account for 1.4 times the value of current liabilities.
The cash reserves ratio is on average low and equal to 0.09; shipping companies usually
avoid keeping cash reserves, as no significant returns can be earned on them, and invest

23
instead their funds on shipping or other projects. Last, the profit margin ratio is relatively
high on average and equals 0.34.

Regarding the statistics of other firm-specific characteristics, the following is noted. As

can be observed, the average age of the companies in the sample which raised capital
through shipping loans, equals 19.10 years, with a range of ages from 7 to 38 years.
These values indicate that most shipping firms included in the sample are experienced
participants in the shipping market. Most obligors have on average a short time of
cooperation with the bank; the average is 3 years, with values ranging between 0 and 9
years. Regarding the chartering policy, as indicated by the average value of 0.77 of the
relevant dummy variable, the majority of the firms financed operate their ships under
time charter contracts rather than in voyage trips ones. It is well known that a steady cash
flow, such as that emanating from a time charter agreement, is seen positively by banks
as a security towards the repayment of the loan.

Panel C of table 2 reports descriptive statistics for the shipping industry and macro
variables utilized as explanatory variables in the model. As expected, the average values
of the spreads between 3 year time charter and spot freight rates for both dry bulk and
tankers are negative. The spreads between 1-year time charter and spot freight rates
remain negative for bulkers but are positive in the tanker sector. As shown in Kavussanos
and Alizadeh (2002), owners in the market are prepared to accept a discount for the
relative safety of time charter contracts in comparison to operating vessels in a series of
spot voyage agreements. However, it is also shown that these discounts are time varying
and depend on the volatility of the market and expectations. Further, as Adland and
Cullinane (2005) propose, spreads are generally negative but can be positive when
focusing on short-term freight rates instead of long-term ones and the state of the freight
market is strong. This can explain the positive sign in the average (1yr TC – Spot) spread
for tankers. Next, it is observed that for the sample period of the loans, the average
orderbook is over 10% of the total fleet, ranging from 2.25% to 22.25%. The percentage
orderbook is higher on average for bulker and tanker segments, standing at around 8%-
9% of the corresponding fleets, but lower for containerships and gas segments, standing

24
at around 4% and 0.13% of the fleet, respectively. High proportions of the orderbook are
typically expected to lead to pressures on the freight income of shipping companies due
to the possibility of leading to excess supply of ships in the market. The inactive tonnage
is considered to be an important indicator of the current conditions in shipping freight
markets, with high lay-up levels being indicators of bad market conditions, leading to low
freight rates and incomes for the shipping companies repaying their loans. The average
inactive tonnage over the sample period is 0.33% of the total fleet, ranging in values from
0.02% to 1.20%. As observed, the inactive tonnage is higher for the gas and tankers
segments in comparison to bulkers and containerships. The Clarksea indices, which
represent the state of the shipping freight markets are other important indicators of the
ability of obligors to earn income and therefore repay their loans. For the period
examined, tankers are on average generating more freight rate income (31,226 $/day),
followed by dry bulkers (18,099 $/day), then containerships (16,079 $/day) and finally
gas carriers (15,140 $/day). Bulkers and tankers exhibit larger ranges of values, standing
at around 38 and 34 thousand dollars, respectively, whereas the containership and gas
segments exhibit much smaller ranges, of around 20 and 12 thousand dollars,
respectively. In all cases, average freight rates stand higher than operating expenses,
thereby earning operating profits for shipowners.

Table 3 presents the pair-wise linear correlations among all explanatory variables
included in equation (1). As observed, only a few sets of variables exhibit high linear
correlation coefficients between themselves. For example, two pairs (debt coverage ratio
3 with debt coverage ratio 2 and with debt coverage ratio 1) of the three different
definitions used for the debt coverage ratio exhibit correlation coefficients over 50%.
Liquidity ratio variables do not exhibit high pairwise linear correlations. Finally, the
following pairs of variables exhibit linear correlations in excess of 60%: ACR
Contractual and ACR Actual (77.10%); and (1yr TC – Spot) and (3yr TC – Spot)
(73.11%). In order to avoid multicollinearity issues which may result in biased estimates,
care is taken not to include simultaneously during estimation sets of variables which
exhibit high linear correlations. Further, as discussed later in the paper, variance inflation
factors (VIF’s) are computed and reported per estimated specification, in order to further

25
test the potential existence of multicollinearity among the explanatory variables of the
estimated models.

Another potential issue with the model described in eq. (1) is the possible existence of
endogeneity and simultaneous determination of the variable ‘Arrangement fee / Amount
of loan’ that appears on the right hand side of equation (1) and the dependent variable, in
the form of reverse causality or simultaneity. This may arise, as the level of the
‘independent’ variable is set by the officers of the bank at a level which reflects the
probability of default of the loan, following careful evaluation by them of all the
information that relates to the loan. One could argue then that the level of this variable is
determined simultaneously with that of the dependent variable of the equation. In order to
test for that, a Hausman (1978) test is due. However, since an instrumental variables (IV)
logit model cannot be estimated, as the IV regressions require linearity in the relationship
under examination, the relevant Hausman test statistic for endogeneity is estimated
through an IV probit model.

5. RESULTS
The estimated coefficients and the respective t-statistics for different specifications of the
logit regression model described in equation (1) are reported in table 4. In order to assess
the explanatory power of different categories of potential shipping loan default risk
drivers, models M1 to M4 include variables classified as follows: Financial (M1), firm
characteristics’ (M2), loan (M3) and industry and macro (M4) specific variables. These
models are estimated in order to assess the explanatory power of each of the above
groups of variables on shipping bank loan defaults, and not to draw inferences about the
significance of individual variables in each group. This role is assumed by Model M5; it
is the most parsimonious model and includes only statistically significant variables,
which may come from any of models M1-M4. The final estimated model M5 has been
specified by taking into account the maximization of McFadden’s pseudo-R2 coefficient,
the minimization of Schwarz's (1978) Bayesian information criterion (SBIC), as well as
the statistical significance of the estimated coefficients.

26
In all models, two-way cluster adjusted standard errors are estimated in order to report
unbiased standard errors, robust to heteroscedasticity and autocorrelation effects, as
shown in Petersen (2009). This adjustment leads to unbiased estimation of standard
errors, thus leading to correct inferences regarding the significance of the explanatory
variables, and is carried out during estimation of the logit models in this paper. As shown
in Kavussanos and Tsouknidis (2014) this correction is also important for the shipping
industry. Previous studies in the literature, such as that of Mitroussi et al. (2016), do not
use this adjustment.

When adding/removing variables from the model, multicollinearity issues may arise,
which would lead to biased estimated coefficients and standard errors. In order to avoid
such issues, variables with bilateral linear correlations in excess of 60%17 are not used
simultaneously in the estimated models. Apart from this rule of thumb, judgment guides
us not to use simultaneously variables which potentially contain similar economic
information. Such potential pairs/sets of variables are the four debt coverage ratios and
the two liquidity ratios, the pairs of ‘ACR Contractual’ with ‘ACR Actual’ and the ‘1yr
TC rates minus spot rates’ with ‘3yr TC rates minus spot rates’. As a further test for the
presence of multicollinearity, the Variance Inflation Factors (VIF) are estimated for each
variable that enters the M1-M5 specifications , as a further diagnostic test for the
presence of collinearity in the estimated models.18 Only the mean (over all coefficients)

17
The threshold of 60% is adopted also by several papers in the finance literature. For example, when
examining bonds’ liquidity premia, Dick-Nielsen et al. (2012) adopt the 60% threshold to avoid the issue of
multicollinearity, when estimating linear panel data regressions.
18
VIF coefficients for each explanatory variable are computed through the following formula: ,
where Ri2 is the coefficient of determination of an auxilliary regression in which the dependent variable is
the independent variable under scrutiny for multicollinearity in the original equation, while the independent
variables in this auxiliary regression are the rest of the independent variables of the original model (for
details see Gujarati and Porter 2008). As a rule of thumb, variables with VIF values greater than 5 indicate
high collinearity, i.e. that the variable could be considered as a linear combination of other independent
variables. Only the mean VIF values are reported in table 4 for economy of space; however, all individual
coefficient values are well below 5, indicating the absence of collinearity between the explanatory variables
in the estimated models. The analytical results are available from the authors on request.

27
VIF values are reported in table 4. These values are far lower than 5 in all cases,
verifying that multicollinearity is not detected in any of the estimated models.

As observed in table 4, in the M1 specification, only the financial variables are included
as potential determinants of the observed defaults. They capture the Company and
Capital specific information of the companies applying for loans. McFadden’s pseudo-R2
is 14.97%. Two financial ratios appear to be statistically significant in explaining
defaults, and they have the correct a-priori expected signs. They are: the ‘debt coverage
ratio’, defined as ‘TL / EBITDA’ and the ‘liquidity ratio’, defined as ‘cash reserves over
total assets’. The M2 specification includes loan applicant-related variables only,
capturing the Character, Capacity and Company characteristics of the loan applicants.
The pseudo-R2 is 8.99%. The time-chartering policy dummy is the only statistically
significant variable. It has the a-priori negative expected sign, indicating that the adoption
of a time-charter hiring policy results in lower PD’s for the loans, as it leads to more
stable cash flows for the obligor. Next, model M3 considers the loan-related set of
variables, which capture the Collateral assigned to the loan, the Capacity of the company,
the Conditions in the market as well as the rest of the Cs of Credit. The model’s R2 is
higher than model M2 and stands at 20.33%. Only the ‘ACR contractual ratio’ is
statistically significant in this case. The M4 specification, considers shipping industry and
macro-related variables, capturing the Conditions in the shipping and world markets. The
pseudo-R2 is 34.76%, and is the highest out of the four models. The ‘1yr TC minus Spot’,
the ‘orderbook as a percentage of the total fleet’, the ‘inactive tonnage as a percentage of
the total fleet’ and the ‘MSCI’ variables are found statistically significant. In all
specifications, except for M2, the likelihood ratio (LR) test strongly rejects the null
hypotheses that all estimated coefficients are jointly equal to zero at the 5% significance
level.

Having evaluated the explanatory power of each category of default risk drivers as well
as the potentially important variables from each group, specification M5 reports the

28
estimation results from the most parsimonious model of equation (1). This is achieved by
following the step-wise method used also in Barniv et al. (2002) and Charitou et al.
(2004). We first assess the individual significance of each variable used in isolation, by
estimating a univariate logit regression. Next, we modify the model by adding or
removing variables according to their statistical significance and the likelihood ratio test,
when at the same time we seek to: maximize the McFadden pseudo-R2; minimize the
SBIC criterion and avoid using simultaneously variables which exhibit high pair-wise
linear correlations (>60%) or are carrying the same economic information. This process
ends when no further variables can be added or removed from the model based on the
rules described above.

Model M5 is well specified and has a reasonable goodness of fit: The two-way adjusted
clustering of the standard errors ensures that there are no issues of autocorrelation or
heteroskedasticity in the panel data regression error terms; the VIF test statistic shows
absence of multicollinearity; the likelihood ratio (LR) test statistic strongly rejects the
null hypothesis that all estimated coefficients are simultaneously equal to zero;
McFadden’s pseudo-R2 equals 33.53%. This is similar to the 35% value reported in
Altman and Rijken (2004), who estimate a logit model utilizing a dataset of US listed
corporates. Of course, this depends on the size and nature of the dataset used. As a further
goodness of fit test, the Hosmer and Lemeshow (1989) test statistic has a value of 3.21
with p-value of 0.9203, where lower values of the statistic and higher p-values indicate a
good fit of the model; Last, as discussed earlier, in order to test for the possible
endogeneity of the variable ‘Arrangement fee / Amount of loan’, an instrumental variable
(IV) probit model is estimated, where the aforementioned variable is instrumented by the
‘internal credit rating’ variable19. The theoretical rationale behind the choice of this
instrument is that the internal credit rating is associated with the default indicator (the
dependent variable) through the ‘Arrangement fee / Amount of loan’ variable. The
Hausman (1978) test statistic for endogeneity yields a p-value of 0.1649, thus not
rejecting the null hypothesis of exogeneity.

19
The full results for the IV probit model are available from the authors on request.

29
The variables found significant in the final M5 specification all have the a-priori expected
sign, and are the following: The bank loan pricing variable, ‘Arrangement fee / Amount
of loan’; the shipping freight market variable, ‘1yr TC minus the Spot freight rate’; the
freight market condition variable, ‘Inactive tonnage / Total fleet’; and the company’s ship
operation policy variable, the ‘Time-charter policy’ dummy. They are presented next,
starting with the variable which has the highest impact, where the relative importance of
the variables is shown in the last column of table 4. It reports changes in the odds of
observing a shipping bank loan default, given a standard deviation increase in the
respective explanatory variable, ceteris paribus. This is a way to assess the relative
importance of the variables in this most parsimonious (M5) model. The reason these
values are reported is that the originally estimated coefficients of the most parsimonious
specification, model M5, show the change in the log of odds, but do not reflect their
relative importance. This is because they depend on the scaling of each explanatory
variable; that is, variables may have different units of measurement between them.

The ‘Arrangement fee as a percentage of the amount of the loan’ is set by the bank after
careful consideration of all the information related to the project; that is, after considering
all the 6 Cs of credit, Company, Capital, Capacity, Character, Collateral and Conditions
in the market, and setting it at a level which reflects the riskiness of the project financed.
‘Relationship banking’ plays a great role in this assessment, as the credit officers of the
bank are well aware of the personal characteristics of obligors through their current and
past experience with them as well as the information that exists for them in the ‘market’.
Thus, as verified by the positive sign of the estimated coefficient, loans with higher
probability of default are classified by the bank as riskier, and as a consequence the bank
requires a higher return as compensation for the higher risk taken. The change in odds of
observing a default for a standard deviation increase in this explanatory variable is
3.3646.

30
The ‘1year Time charter minus the spot freight rate’ spread appears with a positive sign
and is the second most significant variable in explaining probabilities of default. It
captures the expected conditions in freight markets and implies that, other things being
equal, a larger value of the spread between the TC and the spot freight rate signals lower
prospects for the freight market of the shipping industry, as a consequence expectations
of lower freight income for shipping companies, leading to higher PD’s for shipping
loans. The change in odds of observing a default is 2.7220 for a standard deviation
increase in this spread.

The next most significant variable is related to the current and expected conditions of the
freight market, and is the ‘inactive tonnage as a percentage of the total fleet’. Inactive
tonnage includes the laid up fleet, the vessels idle for more than six-weeks and the vessels
used for storage for periods longer than 60 days. Thus, it is a variable which captures the
current and to some extend the future short to medium term prospects of shipping freight
markets. As expected, it appears with a positive sign, which indicates that when more of
the fleet is inactive in the market, it is because current and prospective short term freight
rates and therefore the income of shipping companies is significantly low, thereby
increasing the probability of default of ship bank loans. The relative importance of this
variable is 2.4124, as indicated in the last column of the table.

Finally, the ‘time-charter policy’ dummy variable, relates to the Company / Character of
the shipowner – loan applicant - as it reflects their risk aversion towards the freight
market and other business risks involved in their operating policy of the ships. It also
reflects their expectations towards the shipping market conditions, and has a negative
sign, as expected. Essentially, adopting a time-charter operating policy for ships rather
than one of operating them in spot markets leads to more stable and predictable cash
flows and reduced overall risks – see for instance Kavussanos and Alizadeh (2002) and
Adland and Cullinane (2005) for this. This increases the probability of meeting the
repayment schedule of the loan due and in this way decreases the probability of default.

31
As a consequence, banks consider favorably shipping loan applications which are backed
up with a time charter contract, thereby providing security of cash flows over the whole
or part of the tenor of the loan. The relative importance of the time-charter policy variable
is captured in the standardized coefficient 0.3935.

In contrast to results from classical corporate sectors, none of the financial variables
measuring the Company and Capacity characteristics are found directly significant in the
shipping industry, once all the above variables are taken into account. This is believed to
be the case because ‘relationship banking’ and the conditions of this highly cyclical
industry are so important in shipping. Thus, the very important overall relationship
between the bank and the shipowner is captured through the ‘Arrangement fee / Amount
of the loan’ variable, while the current and expected conditions in the sector and the risk
appetite of the owner are the significant factors that determine whether a shipping loan
can be served or not during its tenor. The financial standing of the company can be so
easily overturned from the aforementioned factors, that it turns out not to be significant in
the shipping industry. Moreover, financial statements of companies are ‘old’, reflecting
end-of-year financial information, which is valued much less (more accurately,
insignificant as shown by the statistical insignificance of the estimated coefficients) in the
rapidly changing shipping environment. This explains why in the final model where all
groups of variables are jointly examined, that variables which capture the fast changing
conditions in the highly volatile shipping industry – such as, the current and expected
conditions in freight markets and the time charter policy of the obligor, are more
important in judging probabilities of default, when compared to the more static / past
information reflected in company financial statements. This explains the difference in the
results in the current paper in relation to those found in similar studies in evaluating bank
loans to obligors outside the shipping industry.

5.1 Estimation and analysis of PD’s

32
Having revealed the factors which are significant in explaining defaults in shipping bank
loans, the ability of the model to discriminate between non-defaulted and defaulted loans
is examined next. Broadly speaking, the aim of credit scoring models in practice is
twofold: first, to provide an accurate ex-ante probability of default associated with a
specific loan agreement; and second to assign a credit score which correctly classifies
each potential loan, according to the associated default risk, into the ‘defaulting’ or ‘non-
defaulting’ category.

The success rate of the estimated PD’s is of high importance for a credit scoring model.
In other words, an ideal credit scoring model would assign as high a PD as possible to a
subsequently defaulted loan and as low a PD as possible to an eventually non-defaulted
loan.20 In order to investigate this for the dataset of this study, table 5 reports the default
probabilities that specification M5 would assign to each one of the seven eventually
defaulted loans at the time the loan was assessed. Moreover, the descriptive statistics for
the ex-post PD’s, assigned by the M5 specification, are presented in table 6 for the non-
defaulted and defaulted loans. As observed, the first moment of the ex-post PD’s for the
eventually-defaulted loans is significantly higher than the eventually non-defaulted ones,
indicating the ability of the model to assign on average higher PD’s on eventually
defaulted loans.

However, apart from the single value of the PD assigned in a bank loan, the goodness-of-
fit tests are informative of the overall ability of the model to discriminate between
eventually defaulted and non-defaulted bank loans. In that direction, the accuracy ratio or
equivalently, the area under the Receiver Operating Characteristic (ROC) curve, forms
the most widely used goodness of fit tests in the literature. As Stein (2005) reports, a
ROC curve analysis involves a non-parametric test similar to the Kolmogorov-Smirnov

20
Ideally, one would like to estimate ex-ante or out-of-sample probabilities of default. However, since there
are only 7 defaults in the sample utilized, this does not leave any margin to perform such forecasts.
Therefore, ex-post (in-sample) forecasts are performed for the purposes of this paper, acknowledging that
out-of-sample default probabilities may differ.

33
test, which examines the equality of continuous, one-dimensional probability
distributions. The ROC curve plots the percentage of true positive defaults versus the
false positive defaults, i.e. the proportion of true defaults classified as “defaults” by the
model against the proportion of false defaults classified as “defaults” by the model. In
this way, it helps visualize the Type I errors (an eventually defaulted bank loan classified
as “non-default”) and Type II errors (an eventually non defaulted bank loan classified as
in “default”). Therefore, it depicts the level of discrimination of the model between the
eventually defaulted and eventually non-defaulted possible alternatives. Thus, if a model
assigns a high PD to a loan which has low actual PD, then a client may be lost for the
bank (Type-I error). If a model assigns a low PD to a loan which has high actual PD, then
the bank may face a loan default (Type-II error) and potential losses. The ROC ratio
equals the area covered by the ROC curve over the area covered from a model which
discriminates perfectly; a ROC curve ratio equal to 1 indicates a model with “perfect”
discriminating ability.

Thus, in order to investigate formally the discriminatory power of the selected model, the
Receiver Operating Characteristic (ROC) curve21 for this dataset is depicted in figure 3.
As can be observed, the area under the ROC curve is high and equal to 92.93%,
indicating the high discriminatory power of the proposed model, where a model with
“perfect” discriminatory power between eventually non-defaulted and eventually
defaulted bank loans would have an area under the ROC curve equal to 100%. The y-
axis, named “Sensitivity”, is also known as true default rate. It measures the percentage
of true defaults, out of all the defaults indicated by the model. In the present setting, this
would be equal to the forecasted defaulted loans out of the total defaulted ones. The x-
axis, named “1 minus specificity”, is also known as the false positive rate and represents
the fraction of false defaults out of the defaults indicated by the model. In the present
setting this would be equal to the falsely forecasted defaulted loans out of the total
defaulted ones.

21
The step-style ROC curve is due to the limited number of defaults in the sample, where a sample with
more defaults would exhibit a smoother ROC curve.

34
Another way to measure not only the discrimination power of the model but both its
discrimination and calibration, i.e. the accuracy of the model, is the Brier (1950) Score
(BS). It measures prediction accuracy and is calculated as: ∑ ( ) ,

where mi is the estimated probability of default for a specific bank loan and ni is a binary
indicator for the actual default or not (1 if default, 0 if no default). Thus, it examines the
squared forecast errors to probability forecasts, defined as the differences between the
forecasted (by the model) PD for a specific loan and the true PD of the loan. The Brier
score takes values between 0 and 1, with lower values indicating better prediction
performance for the estimated model, since the squared forecast errors are lower. The BS
for the preferred M5 model, as observed in table 4, is 0.0337. This is quite close to 0,
indicating the high discriminatory power and high ability of the model to predict defaults
in shipping bank loans. Medema et al. (2009) report a Brier score of 0.015 for their
estimated model using loan data from a commercial bank in the Netherlands over the
period 2000 to 2003.

The fact that in some cases the models estimated do not succeed in assigning high PD’s
in all eventually defaulted loans is not necessarily an indication of an inadequate scoring
model for at least two reasons: First, a borrower can default even if their default
probability is very low. Second, even if a model is adequate in predicting defaults on the
whole, as suggested by a high value of the McFadden pseudo-R2, it may fail at predicting
some individual default probabilities.

6. DISCUSSION

Empirical results in this study indicate that some of the 6-Cs of credit are more important
than others in evaluating credit risk of bank shipping loans. Specifically, the current and
expected Conditions of the market as captured by the freight rate spread and the
proportion of the inactive fleet, the Character of the shipowner in terms of their risk

35
appetite as expressed through the chartering policy they follow22, and the rest of the Cs of
credit as assessed by the credit analysis department of the bank and captured indirectly
through the pricing of the loan, are the significant factors that seem to drive loan defaults
in this risky industry. The finding that the conditions of the shipping market would be so
important in capturing default events is expected, given the characteristics and the rapidly
changing freight market conditions of this particularly risky global business. The
financial statements of the company, even though requested by the bank when assessing
shipping loans, seem to take second role as they are deemed more static and that they
reveal ‘old’ information relative to current and expected events in the shipping industry.
Moreover, the analysis in this paper indicates that default risk assessment in bank
shipping loans can take place by relying on quantitative rather than qualitative variables.

These results are compared next to other findings in the literature, starting with those in
the corporate finance literature investigating sectoral differences. Thus: Bhimani et al.
(2014) examine an extensive corporate bank loans dataset from Portugal over the period
1997-2003. They find that company specific characteristics, such as the limited versus
full ownership and the age of the firm have an effect on probabilities of default, with
firms’ financial ratios being the second most important category of variables to that
effect. Specifically, they report a negative relationship between probabilities of default
and age, cash reserves ratio, leverage and asset coverage ratio. Moreover, they document
high joint significance for the coefficients of industry dummies and find the dummy
coefficient for the transportation sector, which includes the shipping industry
(transportation by sea), to be the highest.

Agarwal et al. (2015) investigate a comprehensive dataset of all mortgage transactions in

the US real estate market over the period 2005 to 2008 and provide evidence supporting
the existence of industry-effects when assessing default risk of bank loans. Moreover,

22
The choice of a time charter policy versus spot operation of the vessel financed essentially ensures
predictable cash flows. An alternative way of creating such predictable cash flows is through the use of
Forward Freight Agreements (FFAs) – see Kavussanos and Visvikis (2006) for details.

36
they show that borrowers with large initial payments i.e. larger committed Capital are in
general less likely to default as opposed to those with low initial payments. Also, a high
loan-to-value (LTV) ratio seems to be associated with higher default probabilities.
Equivalents to the LTV ratio, the ‘ACR Contractual’ and the ‘ACR actual ratio’ of
shipping loans, are used in the current study, in order to capture this effect. However they
are found insignificant in explaining shipping bank loan defaults. Thus, it seems that the
capital coverage of the bank loan from the financed asset’s value is not a universal
indicator for assessing default risk in bank loan agreements.

Chang et al. (2014) also reveal industry-effects when assessing default risk of bank loans.
They utilize a dataset of corporate loans from the Chinese market over the period 2004 to
2006 and document a significant association between relationship banking and
probabilities of default for corporate bank loans. The authors confirm the existence of
“industry” effects and “bank-specialty” effects, where the accurate assessment of bank
loans requires a model which is tailored-made to the industry applied to. In their study,
“bank-specialty” is defined as the information revealed to the company over time and/or
through repeated interactions with the borrowing firm. It is measured as the residual
component of the internal credit rating using an all-industries sample of bank loans for
China. However, the effect of firms’ financial ratios is still present, as leverage and cash
reserves’ ratios remain significant. Last, the authors measure relationship banking
through the years of cooperation of each borrower with the bank. The current paper also
investigates whether the years of cooperation plays a role in explaining default
probabilities of bank shipping loans. It appears to be insignificant in the estimated
models. Relationship banking is captured in other ways in the current study, such as
through the arrangement fee charged by the bank. .

Overall, the aforementioned findings from the literature support that

“Capital”/“Company”/“Capacity” factors are more important than the rest of the 6 C’s of
Credit when assessing default risk in corporate bank loans. At the same time, evidence in

37
previous studies supports the presence of differences between industries. In the current,
shipping sector-specific, study, the current and expected Conditions in the shipping
markets (‘1yr TC minus spot spread’ and ‘inactive tonnage over total fleet’), the
Character / Company (TC dummy for hiring policy) factor and the ‘Arrangement fee /
Amount of the loan’ variable, capturing the rest of the 6 Cs’ effect, are revealed to be the
significant factors in explaining shipping loan defaults. In that respect, it seems that the
Conditions of the market and the Character of the shipowner on how he tackles the very
volatile, capital intensive, risky international shipping sector, can explain assumed risk
levels in shipping bank loans, with the importance of the rest of the Cs of credit
following.

Differences between previous studies and the current study are notable. This paper is the
first to utilize a complete shipping credit loan portfolio and a logit model with two-way
clustered adjusted standard errors. The latter provides a necessary condition for correct
inferences on the significant factors affecting probabilities of default in shipping bank
loans. Other studies, such as Bhimani et al. (2014), Chang et al. (2014) and Bonfim
(2009), do not apply the two-way clustering of standard errors when estimating their logit
/ probit models. This is also true in the study by Mitroussi et al (2016). In addition, in the
latter study, the sample of loans is relatively small, concentrates only on loans to the dry
bulk sector while these loans were drawn over a period which does not cover the entire
shipping business cycle. As a consequence, the results may not generalize over entire
shipping industry bank loan portfolios, over different market conditions, while inferences
on relevant factors determining probabilities of default may not lead to correct
conclusions.

Moreover, in the current study the proposed model can be practically applied to assess
default probabilities of bank loans in shipping, and in that respect differs from the
UTADIS method used in Dimitras et al. (2002) and in Gavalas and Syriopoulos (2014).
In addition, the questionnaire methodology applied in latter study contains a certain

38
degree of subjectivity, while the pre-determined ‘subjective’ weights obtained through
financial consulting firms, as in the former study has similar issues. As a consequence,
the Conditions are not playing any role in driving default risks in shipping loans but
rather report that mainly firm’s characteristics and loan characteristics are relevant,
captured through the capital structure represented by the debt to equity ratio and the asset
coverage ratio of the shipping company.

The results of the current paper have important implications for the maritime industry.
Specifically, shipping companies seeking to raise capital through bank loan agreements
should take into account the default risk drivers of shipping bank loans revealed in this
paper. This is because the margin charged by the bank on the loan agreement reflects its
assessment regarding the default risk entailed in the loan agreement, which directly
affects the cost of capital for the shipping company. In addition, it reveals the factors that
shipping companies should consider as important when deciding to apply for a bank loan,
as they reveal their ability to serve their loan to the bank. Indirectly, it reveals the factors
that the obligors should be concerned about in order to make their investment project
succeed. More specifically, shipping companies applying for a credit facility should take
into account the current and expected conditions in the shipping market as indicated by
the proportion over the total of the inactive fleet, and the TC-spot spread, respectively.
Moreover, the chartering policy adopted by a shipping firm is shown to be an important
default risk driver of shipping bank loans. Thus, shipping firms should consider very
carefully the risk exposure of the income generated by the financed project – the ship,
which can be moderated through the use of time-charter agreements or perhaps through
freight derivative contracts, and thus create stable – predictable cash flows.

Evaluating risks, pricing accordingly and channeling funds in the most promising
investment projects is a core function of financial institutions, and the results of this study
add to existing knowledge on how to assess these in shipping projects more accurately. In
this way, the findings of this paper have important implications regarding primarily the

39
financial institutions involved in financing shipping projects. As already mentioned,
shipping companies applying for bank loans should take into account the results of this
study in order to be aware of the default risk drivers that financial institutions seem to
assess as important in loan applications. Thus, the results of the paper enable better
decision making both by shipping companies and banks advancing loans to shipping
business.

7. CONCLUSIONS
This paper is the first to investigate the default risk drivers of shipping bank loans using a
credit scoring model with two-way cluster-adjusted standard errors. The later ensures
correct inferences of estimated models. It draws measurable explanatory factors from all
C’s of credit theory, and through careful modelling reveals the important factors that
drive defaults of these loans. The dataset analyzed is unique in that it covers the entire
bank lending portfolio of a ship lending bank, over a period of time that covers entire
shipping business cycles, spans loans across all subsectors of the cargo carrying shipping
industry while the obligors’ characteristics are typical of shipping company business.
These attributes are important in that they allow results to generalize and be applicable to
bank shipping loans.

The unique characteristics of the shipping industry set the study apart, in comparison to
the general finance literature. The results justify this. Specifically, in the general finance
literature the important factors which seem to drive default probabilities of bank loans are
financial ratios and other characteristics (e.g. ownership) of the companies / individuals
applying for loans, while results differ between industries. These are also considered by
the banks in shipping loan applications but they are not the important factors driving
defaults. In shipping, the assets to be financed are of very high value, with cash flows
generated being greatly affected by the very volatile freight and other international
markets these companies are exposed to. Thus, the conditions are changing rapidly and
the ‘old’ information carried in end-of-year financial statements are not driving decisions.
40
As a consequence, important factors in predicting the probability of default, uncovered in
this paper, seem to be those that measure current and expected future conditions in these
markets, the risk appetite of the obligor as captured from their choice of the chartering
policy of the vessel, and of the other risk factors and the relationship banking effect
involved in the rest of the 6 C’s of credit as assessed by the credit analysis department of
the bank and captured in the pricing of the loan through the ‘arrangement fee’ variable.
These results could well generalize to risky international industries operating under
conditions of perfect competition; factors mediating the volatile cash flow risks are
considered to be the most important, while more traditional type of information such as
that carried in financial ratios seem to come second in line, without being irrelevant.

Acknowledgements: This paper has benefited from comments of seminar participants at:
Research seminars series of Hong Kong Polytechnic University, Hong Kong, 2015;
ICMA Centre, University of Reading, Reading, UK, 2014; International Association of
Maritime Economists (IAME) Conference, Santiago De Chile, Chile, 2011; International
Atlantic Economic Society (IAES) Conference, Athens, Greece, 2011 and the
2nd Hellenic Conference in Applied Economics, Volos, Greece, 2011. The authors would
like to thank a ship-lending bank for providing access to the data; the Heads of bank's
Departments of Shipping and Risk management for useful discussions and two
postgraduate students at Athens University of Economics and Business, Maria
Velimvasaki and Eleanna Kotsaleni, for their help with data collection. Any remaining
errors or omissions are the authors’ responsibility.

REFERENCES

Adland, R., Cullinane, K., 2005. A Time-Varying Risk Premium in the Term Structure of
Bulk Shipping Freight Rates. Journal of Transport Economics and Policy 39,
191–208.
Agarwal, S., Green, R.K., Rosenblatt, E., Yao, V., 2015. Collateral pledge, sunk-cost
fallacy and mortgage default. Journal of Financial Intermediation 24, 636–652.
doi:10.1016/j.jfi.2014.10.001
Agarwal, V., Taffler, R., 2008. Comparing the performance of market-based and
accounting-based bankruptcy prediction models. Journal of Banking & Finance
32, 1541–1551. doi:10.1016/j.jbankfin.2007.07.014
Albertijn, S., Bessler, W., Drobetz, W., 2011. Financing shipping companies and
shipping operations: a risk-management perspective. Journal of Applied
Corporate Finance 23, 70–82. doi:10.1111/j.1745-6622.2011.00353.x

41
Altman, E.I., 1968. Financial Ratios, Discriminant Analysis and the Prediction of
Corporate Bankruptcy. The Journal of Finance 23, 589–609. doi:10.1111/j.1540-
6261.1968.tb00843.x
Altman, E.I., Rijken, H.A., 2004. How rating agencies achieve rating stability. Journal of
Banking & Finance 28, 2679–2714. doi:10.1016/j.jbankfin.2004.06.006
Amiti, M., Weinstein, D.E., 2011. Exports and Financial Shocks. The Quarterly Journal
of Economics 126, 1841–1877. doi:10.1093/qje/qjr033
Barniv, R., Agarwal, A., Leach, R., 2002. Predicting Bankruptcy Resolution. Journal of
Business Finance & Accounting 29, 497–520. doi:10.1111/1468-5957.00440
Basel I, C. on B.S., 1988. International convergence of capital measurement and capital
standards. BIS report.
Basel II, C. on B.S., 2004. International convergence of capital measurement and capital
standards: A revised framework. BIS report.
Basel III, 2010. Group of Governors and Heads of Supervision announces higher global
minimum capital standards. BIS report.
Bhimani, A., Gulamhussen, M.A., Lopes, S. da R., 2014. Owner liability and financial
reporting information as predictors of firm default in bank loans. Rev Account
Stud 19, 769–804. doi:10.1007/s11142-013-9269-0
Blum, J.M., 2008. Why Basel II may need a leverage ratio restriction. Journal of Banking
& Finance 32, 1699–1707. doi:10.1016/j.jbankfin.2007.12.003
Bonfim, D., 2009. Credit risk drivers: Evaluating the contribution of firm level
information and of macroeconomic dynamics. Journal of Banking & Finance 33,
281–299. doi:10.1016/j.jbankfin.2008.08.006
Brier, G.W., 1950. Verification of forecasting expressed in terms of probability. Monthly
Weather Review 78, 1–3. doi:10.1175/1520-
0493(1950)078<0001:VOFEIT>2.0.CO;2
Campbell, J.Y., Hilscher, J., Szilagyi, J., 2008. In Search of Distress Risk. The Journal of
Finance 63, 2899–2939. doi:10.1111/j.1540-6261.2008.01416.x
Chang, C., Liao, G., Yu, X., Ni, Z., 2014. Information from Relationship Lending:
Evidence from Loan Defaults in China. Journal of Money, Credit and Banking 46,
1225–1257. doi:10.1111/jmcb.12139
Charitou, A., Neophytou, E., Charalambous, C., 2004. Predicting corporate failure:
empirical evidence for the UK. European Accounting Review 13, 465–497.
doi:10.1080/0963818042000216811
Chava, S., Purnanandam, A., 2011. The effect of banking crisis on bank-dependent
borrowers. Journal of Financial Economics 99, 116–135.
doi:10.1016/j.jfineco.2010.08.006
Chen, J., Deng, Y., 2013. Commercial Mortgage Workout Strategy and Conditional
Default Probability: Evidence from Special Serviced CMBS Loans. J Real Estate
Finan Econ 46, 609–632. doi:10.1007/s11146-012-9374-z
Chen, W., Ma, C., Ma, L., 2009. Mining the customer credit using hybrid support vector
machine technique. Expert Systems with Applications 36, 7611–7616.
doi:10.1016/j.eswa.2008.09.054
Devaud, J.M., Groussaud, G., Jacquet-Lagreze, 1980. UTADIS: Une méthode de
construction de fonctions d’utilité additives rendant compte de jugements
globaux. UTADIS: [A method for the construction of additive utility functions

42
 based on global judgments], in: Proceedings of the European Working Group on
MCDA. Bochum, Germany.
Dick-Nielsen, J., Feldhütter, P., Lando, D., 2012. Corporate bond liquidity before and
after the onset of the subprime crisis. Journal of Financial Economics 103, 471–
492. doi:10.1016/j.jfineco.2011.10.009
Dimitras, A.I., Petropoulos, T., Constantinidou, I., 2002. Multi-criteria evaluation of loan
applications in shipping. J. Multi-Crit. Decis. Anal. 11, 237–246.
doi:10.1002/mcda.332
Gavalas, D., Syriopoulos, T., 2014. An integrated credit rating and loan quality model:
application to bank shipping finance. Maritime Policy & Management 0, 1–22.
doi:10.1080/03088839.2014.904948
Gerardi, K., Lambie-Hanson, L., Willen, P.S., 2013. Do borrower rights improve
borrower outcomes? Evidence from the foreclosure process. Journal of Urban
Economics 73, 1–17. doi:10.1016/j.jue.2012.06.003
Gong, S.X., Ye, H.-Q., Zeng, Y.Y., 2013. Impacts of the recent financial crisis on ship
financing in Hong Kong: a research note. Maritime Policy & Management 40, 1–
9. doi:10.1080/03088839.2012.745202
Grammenos, C.T., 2010. Revisiting credit risk, analysis and policy in bank shipping
finance, in: The Handbook of Maritime Economics and Business. Informa, pp.
777–810.
Grammenos, C.T., Nomikos, N.K., Papapostolou, N.C., 2008. Estimating the probability
of default for shipping high yield bond issues. Transportation Research Part E:
Logistics and Transportation Review 44, 1123–1138.
doi:10.1016/j.tre.2007.10.005
Guiso, L., Sapienza, P., Zingales, L., 2013. The Determinants of Attitudes toward
Strategic Default on Mortgages. The Journal of Finance 68, 1473–1515.
doi:10.1111/jofi.12044
Gujarati, D., Porter, D., 2008. Basic Econometrics. McGraw-Hill/Irwin, New York.
Hosmer, D.W.J., Lemeshow, S., 1989. Applied Logistic Regression. John Wiley & Sons,
New York.
Jarque, C.M., Bera, A.K., 1980. Efficient tests for normality, homoscedasticity and serial
independence of regression residuals. Economics Letters 6, 255–259.
Jiménez, G., Saurina, J., 2004. Collateral, type of lender and relationship banking as
determinants of credit risk. Journal of Banking & Finance 28, 2191–2212.
doi:10.1016/j.jbankfin.2003.09.002
Karels, G.V., Prakash, A.J., 1987. Multivariate Normality and Forecasting of Business
Bankruptcy. Journal of Business Finance & Accounting 14, 573–593.
doi:10.1111/j.1468-5957.1987.tb00113.x
Kavussanos, M., 2010. Business risk measurement and management in the cargo carrying
sector of the shipping industry - an update, in: The Handbook of Maritime
Economics and Business. Lloyd’s List Publications, London.
Kavussanos, M., Alizadeh, A., 2002. The Expectations Hypothesis of the Term Structure
and Risk Premia in Dry Bulk Shipping Freight Markets; An Egarch-M Approach.
Journal of Transport Economics and Policy 36, 267–304.
Kavussanos, M.G., 2003. Time varying risks among segments of the tanker freight
markets. Maritime Economics and Logistics 5, 227–250.

43
Kavussanos, M.G., 1997. The dynamics of time-varying volatilities in different size
second-hand ship prices of the dry-cargo sector. Applied Economics 29, 433–443.
Kavussanos, M.G., 1996. Price risk modelling of different size vessels in the tanker
industry using Autoregressive Conditional Heteroskedasticity (ARCH) models.
The Logistics and Transportation Review 32, 161–176.
Kavussanos, M.G., Tsouknidis, D.A., 2014. The determinants of credit spreads changes
in global shipping bonds. Transportation Research Part E: Logistics and
Transportation Review 70, 55–75. doi:10.1016/j.tre.2014.06.001
Kavussanos, M.G., Tsouknidis, D.A., 2011. Default Risk Drivers in Shipping Bank
Loans, in: Proceedings of the 15th International Association of Maritime
Economists Conference.
Kavussanos, M.G., Visvikis, I., 2006. Derivatives and risk management in shipping.
Witherby’s, London.
Laitinen, E.K., Laitinen, T., 2000. Bankruptcy prediction: Application of the Taylor’s
expansion in logistic regression. International Review of Financial Analysis 9,
327–349. doi:10.1016/S1057-5219(00)00039-9
Laitinen, T., Kankaanpaa, M., 1999. Comparative analysis of failure prediction methods:
the Finnish case. European Accounting Review 8, 67–92.
doi:10.1080/096381899336159
Medema, L., Koning, R.H., Lensink, R., 2009. A practical approach to validating a PD
model. Journal of Banking & Finance 33, 701–708.
doi:10.1016/j.jbankfin.2008.11.007
Mitroussi, K., Abouarghoub, W., Haider, J.J., Pettit, S.J., Tigka, N., 2016. Performance
drivers of shipping loans: An empirical investigation. International Journal of
Production Economics, Operations Management and Production Research Using
Secondary Data 171, Part 3, 438–452. doi:10.1016/j.ijpe.2015.09.041
Moody’s Investor Service, 2005. Global Shipping Industry.
Petersen, M.A., 2009. Estimating Standard Errors in Finance Panel Data Sets: Comparing
Approaches. Rev. Financ. Stud. 22, 435–480. doi:10.1093/rfs/hhn053
Ruthenberg, D., Landskroner, Y., 2008. Loan pricing under Basel II in an imperfectly
competitive banking market. Journal of Banking & Finance 32, 2725–2733.
doi:10.1016/j.jbankfin.2008.07.009
Santos, J.A.C., 2011. Bank Corporate Loan Pricing Following the Subprime Crisis. Rev.
Financ. Stud. 24, 1916–1943. doi:10.1093/rfs/hhq115
Schwarz, G., 1978. Estimating the dimension of a model. The Annals of Statistics 6, 461–
464. doi:10.1214/aos/1176344136
Smith, P.F., 1964. Measuring Risk on Consumer Instalment Credit. Management Science
11, 327–340.
Stein, R.M., 2005. The relationship between default prediction and lending profits:
Integrating ROC analysis and loan pricing. Journal of Banking & Finance 29,
1213–1236. doi:10.1016/j.jbankfin.2004.04.008
Stopford, M., 2009. Maritime Economics, 3rd ed. Routledge, London and New York.
Thompson, S.B., 2011. Simple formulas for standard errors that cluster by both firm and
time. Journal of Financial Economics 99, 1–10. doi:10.1016/j.jfineco.2010.08.016

44
Tsouknidis, D.A., 2016. Dynamic volatility spillovers across shipping freight markets.
Transportation Research Part E: Logistics and Transportation Review 91, 90–111.
doi:10.1016/j.tre.2016.04.001
World Trade Organization, 2014. Maritime transport services (Background note by the
secretariat).

45
 Table 1: Default risk drivers used as explanatory variables of shipping bank loans defaults in equation (1)

Panel A: Financial, firm and loan related variables

Group of
Explanatory 6C’s Mapping Description Expected Sign
Variables

Dependent Default Indicator: Takes value 0 if the loan is still active and follows the repayment schedule and value 1 if a missed
Variable payment for a loan is recorded for a period longer than 90 days.

Capital / Company Leverage = Total Liabilities (TL) / Total Assets (TA): Leverage measures the total debt burden of the company
+
compared with its total assets.

Capital / Company Debt Coverage Ratio 1 = Current Liabilities (CL) / EBITDA: The debt coverage ratio measures the current liabilities
+
of the company in comparison with its current EBITDA.

Capital / Company Debt Coverage Ratio 2 = Long Term Debt (LTD) / EBITDA: This debt coverage ratio measures the long term +
liabilities as a percentage of EBITDA.

Capital / Company Debt Coverage Ratio 3 = TL / EBITDA: This ratio measures the total liabilities of the company in comparison to its +
EBITDA.

Financial- Capital / Company Debt Coverage Ratio 4 = Interest Expenses / EBITDA: This ratio measures the interest expenses of the company as a +
specific percentage of its EBITDA.
variables

-
Liquidity Ratio 1 ≡ Current Ratio = Current Assets (CA) / Current Liabilities (CL): The current ratio measures how
Capital / Company many times the current assets of the company cover its current liabilities.

-
Liquidity Ratio 2 ≡ Cash Reserves Ratio = Cash and Cash Equivalents / Total Assets (TA): This ratio measures the
Capital / Company proportion of the total assets of a company which is kept as cash and cash equivalents.

Capital / Company Profit Margin Ratio = EBITDA / Revenue: The profit margin measures the percentage of profit for each unit of
revenue. -

Character / Capacity / Age: The time period the shipping company exists. -
Company

Firm Character / Capacity / Years of cooperation with the bank (yocb): The number of years of cooperation between the bank and the shipping -
characteristics’- Company company since the first loan agreement made between them.
specific
variables

Character / Capacity / TC Policy: A dummy variable taking the value of 1 for time chartered vessels and 0 for voyage chartered ships. -
Company

All 6 C’s of credit Arrangement Fee / Amount of Loan: The arrangement fee is the main source of profit for the bank along with the
+
margin (spread) charged. These fees are computed as a percentage of the whole amount of the loan.

All 6 C’s of credit Life to Final Maturity (tenor of the loan): The number of years from the initiation of the agreement until the final +/-
maturity of the loan and the balloon payment.

All 6 C’s of credit Margin: The spread over LIBOR charged. +

All 6 C’s of credit Internal Bank Rating: The variable internal bank rating, refers to the internal risk rating assigned by the bank, based +
Loan-specific on the overall assessment of each credit loan application (1=Completely Safe, 7=Acceptable risk, >7 Non-acceptable
variables risk).

Collateral Asset Coverage Ratio (ACR) Contractual: The market value of the ship over the amount of loan. This ratio is defined -
by the bank as a threshold which should be maintained at lower levels than the ACR actual.

Collateral / Conditions Asset Coverage Ratio (ACR) Actual: The market value of the ship as estimated by ship-brokers or internal sources of -
the bank over the amount of the loan. This ratio changes over time since the price of the vessel may fluctuate and the
debt outstanding is also been reducing during the repayment schedule of the loan.

46
 Panel B: Industry-specific and macro variables

6C’s Mapping
Group of Explanatory
Description Expected
Variables
Sign

Conditions 1yr TC – Spot for Bulkers (Tankers): Difference between the 1yr TC Bulker
(Tanker) rates and the Average Spot Bulker (Tanker) rates for the subsegments of
Capesize, Panamax, Supramax and Handy in the case of Dry Bulk, and over all
Tanker routes in the case of Tankers. Both the TC and spot rates are in $/day.
Baltic Exchange freight rate data are used for the subsectors as follows: The
average of the 4TC routes for the Capesize, the 4TC routes for Panamax, the 6TC
routes for the Supramax and the 6TC routes for the Handymax. For Tankers the
Spot rate is the average of all the spot tanker routes. According to the SIN SIW
+/-
report the spot time series for tankers and bulkers are average earnings for each
ship type of the voyages earnings for selected routes. The constituent routes of
these average earnings figures are listed in Annex 4 (b) of the Clarkson’s SIN
SIW report.

Conditions 3yr TC – Spot Bulkers (Tankers): These are defined in the same way as the 1yr +/-
TC – Spot Bulkers (Tankers) case, where the 3yr TC’s are used in this case.

Conditions Orderbook / Total Fleet: Defined as the world orderbook expressed in

deadweight (DWT) over the total fleet measured in mln. dwt. Also the orderbook
+
for each broad subsector (Bulker, Tanker, Containership, Gas) of the sample is
used with no change in the results.

Conditions Inactive Tonnage / Total Fleet: This is the total removals of tonnage for the world
fleet due to several reasons including lay-up over the total fleet measured in mln.
dwt. This inactive tonnage can be due to (i) damaged and out of service vessels,
Industry-specific and
(ii) idle and awaiting orders, (iii) idle for more than six weeks, (iv) laid-up, (v) in +
macro variables
long term storage (>60 days) or (vi) being repaired. We also use the removals for
each broad subsector (Bulker, Tanker, Containership, Gas) of the sample with no
changes in the results.

Conditions ClarkSea index: An index constructed as a weighted average of earnings in all

shipping subsectors, with the weights used reflecting the number of vessels in
each fleet sector. We also use the ClarkSea index for each broad subsector -
(Bulker, Tanker, Containership, Gas) of the sample with no significant changes in
the results.

Conditions MSCI world stock index: An index constructed to be representative of the stock
-
market sentiment at a global level.

Notes: This table lists all the variables examined in this paper as possible default risk drivers of the defaults of shipping bank
loans. These factors can be classified into five broad categories as financial-specific, firm-specific, bank-specific, macro and
industry-specific and bank-judgment specific variables. They correspond to the 6 C’s of credit analysis discussed in the text.
The column “Expected Sign” refers to the a priori theoretical sign that a risk factor is expected to carry after the estimation of
the logit regressions.

47
Table 2: Descriptive Statistics of the winsorized variables in Equation (1) - Sample period 1997-2011, Annual data
Panel A: Descriptive statistics of key loan-specific variables overall and by vessel profile: Dry Bulk, tankers, containerships and gas carriers
JB stat
Mean Median Minimum Maximum Standard Deviation Skewness Kurtosis
[p-value]
Amount of Loan (mln. $) – Over all sectors 30.41 13.00 0.65 310.00 49.25 3.83 20.59 22.13
[0.0000]
Amount of Loan (mln. $) – Bulkers 24.89 9.00 0.65 110.00 33.07 1.81 4.92 27.36
[0.0000]
Amount of Loan (mln. $) – Tankers 41.36 17.00 0.65 310.00 66.68 3.16 12.79 26.45
[0.0000]
Amount of Loan (mln. $) – Containerships 46.30 31.00 6.00 148.00 47.05 1.15 2.85 7.98
[0.0185]
Amount of Loan (mln. $) – Gas 19.21 11.00 3.00 90.90 23.76 2.45 7.93 19.86
[0.0000]
Life to maturity of loan (in years) – Aggregate over all sectors 5.14 3.00 0.30 15.00 3.86 0.71 2.19 56.97
[0.0000]
Life to maturity of loan (in years) - Bulkers 4.14 4.00 0.50 12.00 2.66 0.99 3.51 11.73
[0.0028]
Life to maturity of loan (in years) - Tankers 6.14 5.00 0.30 15.00 4.52 0.16 1.46 21.26
[0.0000]
Life to maturity of loan (in years) - Containerships 5.72 5.00 3.00 12.00 3.20 1.04 2.69 7.01
[0.0301]
Life to maturity of loan (in years) – Gas 5.08 4.00 3.00 12.00 3.10 1.04 2.70 5.00
[0.0822]
Arrangement Fees (‘000 $) – Aggregate over all sectors 76.03 64.75 15.00 175.00 51.43 0.67 2.31 10.28
[0.0057]
Arrangement Fees (‘000 $) – Bulkers 85.02 60.00 15.00 175.00 52.66 0.68 2.26 5.89
[0.0525]
Arrangement Fees (‘000 $) – Tankers 73.78 64.75 15.00 175.00 51.04 0.68 2.36 13.58
[0.0011]
Arrangement Fees (‘000 $) – Containerships 105.66 112.80 20.00 175.00 43.04 -0.09 2.71 2.14
[0.2145]
Arrangement Fees (‘000 $) – Gas 39.05 37.50 15.00 68.17 17.43 0.24 1.96 1.54
[0.4632]
Margin (%) – Aggregate over all sectors 1.91 2.00 1.00 4.00 0.92 1.18 4.74 31.59
[0.0000]
Margin – Bulkers (%) 2.11 2.00 1.00 4.00 1.04 1.90 7.65 34.85
[0.0000]
Margin – Tankers (%) 1.79 2.00 1.00 4.00 0.83 1.00 3.64 22.60
[0.0000]
Margin – Containerships (%) 1.90 1.50 1.00 4.00 1.09 0.82 2.28 6.06
[0.0483]
Margin – Gas (%) 2.21 1.00 1.00 4.00 1.34 0.28 1.28 16.52
[0.0003]
Internal Bank Rating – Aggregate over all sectors 3.63 4.00 1.00 7.00 1.32 0.17 3.20 3.18
[0.2040]
Internal Bank Rating – Bulkers 3.98 4.00 1.00 6.00 1.09 -0.34 3.06 2.23
[0.3286]
Internal Bank Rating – Tankers 3.20 3.00 1.00 7.00 1.17 -0.16 3.87 5.29
[0.0711]
Internal Bank Rating – Containerships 4.52 4.00 1.00 7.00 2.02 0.01 1.73 7.92
[0.0190]
Internal Bank Rating - Gas 3.00 3.00 2.00 4.00 1.05 0.01 1.11 8.15
[0.0112]
ACR Contractual (%) – Aggregate over all sectors 146 145 125 185 20 0.68 2.18 51.18
[0.0000]
ACR Contractual (%) – Bulkers 153 150 125 185 23 0.31 2.25 5.17
[0.0753]
ACR Contractual (%) – Tankers 136 130 125 170 19 1.88 6.29 51.36
[0.0000]
ACR Contractual (%) – Containerships 140 130 125 175 18 0.77 2.26 4.83
[0.0891]
ACR Contractual (%) – Gas 136 130 125 155 15 0.25 1.17 24.03
[0.0000]
ACR Actual (%) – Aggregate over all sectors 182 168 125 275 48 0.66 2.22 44.55
[0.0000]
ACR Actual (%) – Bulkers 214 194 133 275 65 0.68 2.25 9.13
[0.0104]
ACR Actual (%) – Tankers 162 151 125 275 47 2.52 9.55 73.47
[0.0000]
ACR Actual (%) – Containerships 166 151 125 250 43 0.98 2.83 5.74
[0.0567]
ACR Actual (%) – Gas 163 134 125 235 45 0.73 1.86 5.35
[0.0690]
Panel B: Financial and company related variables
JB stat
Mean Median Minimum Maximum Standard Deviation Skewness Kurtosis
[p-value]
Financial figures
Leverage = TL / TA 0.53 0.54 0.12 0.94 0.21 -0.08 2.90 45.49
[0.0000]
Debt Coverage Ratio 1 ≡ Current debt = CL / EBITDA 1.08 0.58 0.06 6.27 1.54 2.47 8.20 28.64
[0.0000]
Debt Coverage Ratio 2 ≡ Long-term debt = LTD / EBITDA 2.29 1.89 0.02 7.21 2.02 1.00 3.22 27.34
[0.0000]
Debt Coverage Ratio 3 ≡ Total liabilities = TL / EBITDA 5.03 3.55 0.17 17.04 4.53 1.45 4.20 24.97
[0.0000]
Debt Coverage Ratio 4 = Interest Expenses / EBITDA 0.21 0.13 0.02 0.93 0.23 1.93 5.99 25.11
[0.0000]
Liquidity Ratio 1 ≡ Current ratio = CA / CL 1.40 1.13 0.21 5.01 1.11 1.89 6.59 20.56
[0.0000]
Liquidity Ratio 2 ≡ Cash reserves = Cash and Cash Equivalents / TA 0.09 0.05 0.00 0.39 0.11 1.87 5.76 29.01
[0.0000]
Profit margin ratio = EBITDA / Revenue 0.34 0.34 0.02 0.75 0.23 0.11 1.68 20.11
[0.0000]
Borrowing company characteristics’ variables
Age of Obligor (years) 19.10 17.00 7.00 38.00 9.01 0.75 2.49 46.35
[0.0000]
Years of Cooperation with the bank (Years) 3.00 3.00 0.00 9.00 2.05 0.74 2.87 44.48
[0.0000]
Time-charter policy (TC=1, Voyage=0) 0.77 1.00 0.00 1.00 0.42 -1.46 3.13 164.64
[0.0000]

49
 Panel C: Shipping industry and Macro variables
JB stat
Mean Median Minimum Maximum Standard Deviation Skewness Kurtosis
[p-value]
14.04
1yr TC – Spot, Bulkers ($/Day) -2,033 -1,844 -4,942 4,816 2,430.28 0.62 3.04
[0.0000]
9.01
3yr TC – Spot, Bulkers ($/Day) -6,725 -8.929 -16,099 1,072 5,546.27 -0.03 1.64
[0.0111]
7.28
1yr TC – Spot, Tankers ($/Day) 3,011 3,364 -4,115 17,712 4,759.95 0.32 3.34
[0.0263]
4.11
3yr TC – Spot, Tankers ($/Day) -3,303 -4,078 -13,492 16,241 6,659.52 0.28 3.07
[0.1280]
34.95
Orderbook (mln. dwt) / Total Fleet (mln. dwt) (%) – Over all sectors 10.106 8.365 2.251 22.256 6.075 0.76 2.63
[0.0000]
29.34
Orderbook (mln. dwt) / Total Fleet (mln. dwt) – Bulkers (%) 8.226 8.106 3.136 22.256 5.194 1.81 5.62
[0.0000]
30.94
Orderbook (mln. dwt) / Total Fleet (mln. dwt) – Tankers (%) 8.765 7.949 3.187 16.068 2.781 1.18 4.31
[0.0000]
9.80
Orderbook (mln. dwt) / Total Fleet (mln. dwt) – Containerships (%) 3.786 4.108 1.149 7.693 1.966 0.27 1.73
[0.0074]
2.88
Orderbook (mln. dwt) / Total Fleet (mln. dwt) – Gas (%) 0.134 0.122 0.022 0.299 0.078 0.33 2.08
[0.23666]
45.21
Inactive Tonnage (mln. dwt) / Total fleet (mln. dwt) (%) – Over all sectors 0.339 0.225 0.021 1.204 0.355 1.51 4.21
[0.0000]
42.80
Inactive Tonnage (mln. dwt) / Total fleet (mln. dwt) – Bulkers (%) 0.045 0.047 0.021 0.088 0.023 0.37 1.65
[0.0000]
68.87
Inactive Tonnage (mln. dwt) / Total fleet (mln. dwt) – Tankers (%) 0.381 0.331 0.125 1.204 0.235 2.08 7.51
[0.0000]
23.25
Inactive Tonnage (mln. dwt) / Total fleet (mln. dwt) – Containerships (%) 0.165 0.064 0.021 1.104 0.234 2.02 7.07
[0.0000]
10.03
Inactive Tonnage (mln. dwt) / Total fleet (mln. dwt) – Gas (%) 0.797 0.904 0.021 1.201 0.447 -0.46 1.62
[0.0066]
55.03
ClarkSea ($/day) – Over all sectors 24,080 23,836 6,211 44,258 10,492.03 0.23 2.07
[0.0000]
41.00
ClarkSea Average Bulker Earnings ($/day) 18,099 15,457 6,200 44,267 10,358.78 0.87 3.22
[0.0000]
42.56
ClarkSea Average Tanker Earnings ($/day) 31,226 29,047 11,613 44,860 10,295.17 -0.36 2.01
[0.0000]
42.97
ClarkSea Average Containerships Earnings ($/day) 16,079 14,368 5,072 24,981 5,788.04 0.14 1.78
[0.0000]
45.04
ClarkSea Average Gas Carrier Earnings ($/day) 15,140 14,153 11,825 24,317 3,269.12 0.94 3.09
[0.0000]
12.06
MSCI world (index points) 1,172 1,169 792 1,589 218.11 0.03 2.41
[0.0024]
Notes: See table 1 for definitions of variables. Min and max are the minimum and maximum values of the sample data, respectively. Skewness and kurtosis are the estimated centralized third and
fourth moments. J-B is the Jarque Bera (1980) test for normality; the JB statistic is χ2(2) distributed. Numbers in square brackets [.] indicate p-values.

50
Table 3: Correlation matrix
Years of Inactive
Orderbook
Debt Debt Debt Debt Age cooperating Time- Arrangement Life to 1yr 3yr Tonnage MSCI
Margin / Total
Coverage Coverage Coverage Coverage Liquidity Liquidity Profit of the with the charter fee / Amount maturity ACR ACR TC - TC - / Total ClarkSea world
Fleet
Default Leverage Ratio 1 Ratio 2 Ratio 3 Ratio 4 Ratio 1 Ratio 2 Margin firm bank policy of loan Contractual Actual Spot Spot Fleet index index

Default 1.000

Leverage -0.132 1.000

Debt
Coverage
Ratio 1 0.078 0.264 1.000
Debt
Coverage
Ratio 2 0.359 0.069 0.125 1.000
Debt
Coverage
Ratio 3 0.302 0.275 0.657 0.551 1.000
Debt
Coverage
Ratio 4 -0.036 0.015 0.123 -0.134 0.157 1.000
Liquidity
Ratio 1 0.079 -0.275 -0.131 0.012 -0.124 0.041 1.000
Liquidity
Ratio 2 -0.092 -0.343 -0.168 -0.050 -0.234 0.039 0.547 1.000
Profit
Margin 0.067 -0.235 -0.366 0.016 -0.363 -0.086 0.097 0.087 1.000
Age of the
firm -0.019 -0.142 -0.190 0.109 -0.165 0.005 0.430 0.526 0.141 1.000
Years of
cooperating
with the
bank 0.014 0.093 0.183 -0.032 0.143 -0.093 -0.052 -0.154 -0.321 0.041 1.000
Time-charter
policy -0.205 -0.038 0.221 0.103 0.149 0.068 -0.062 0.033 0.060 -0.372 -0.258 1.000
Arrangement
fee / Amount
of loan 0.211 -0.113 -0.068 -0.013 0.023 0.076 -0.022 -0.088 0.084 -0.213 -0.128 -0.050 1.000

Margin
0.079 -0.018 0.120 0.135 0.124 -0.006 0.146 -0.205 0.146 0.104 0.117 -0.022 0.251 1.000
Life to
maturity -0.182 0.027 -0.205 0.075 -0.105 0.025 0.079 0.219 0.098 0.269 -0.047 -0.005 -0.427 -0.317 1.000
ACR
Contractual 0.246 -0.008 0.432 -0.023 0.361 0.021 -0.074 -0.236 -0.074 -0.452 0.120 0.163 0.319 0.342 -0.497 1.000

ACR Actual -0.006 0.141 0.489 -0.065 0.288 -0.055 -0.029 -0.136 -0.070 -0.307 0.164 0.256 0.151 0.323 -0.376 0.771 1.000
1yr TC –
Spot 0.194 -0.146 -0.030 0.145 -0.022 -0.084 -0.049 -0.048 -0.207 0.143 0.157 -0.194 -0.125 -0.012 0.113 -0.180 -0.236 1.000
3yr TC -
Spot 0.139 -0.015 0.243 0.157 0.178 -0.022 -0.132 -0.120 -0.372 0.032 0.171 -0.056 -0.160 -0.124 0.069 -0.077 -0.096 0.731 1.000
Orderbook / -
Total Fleet 0.551 -0.133 -0.186 0.427 0.217 -0.051 0.152 -0.017 0.105 0.021 -0.060 -0.204 0.304 0.316 -0.087 0.184 -0.069 0.234 0.069 1.000
Inactive
Tonnage /
Total Fleet 0.408 -0.238 -0.302 0.029 -0.182 -0.010 0.130 0.038 -0.015 0.078 0.123 -0.298 -0.007 -0.081 0.111 -0.127 -0.352 0.345 0.048 0.521 1.000
ClarkSea -
index 0.041 -0.204 -0.489 -0.007 -0.302 0.019 0.170 0.139 0.205 0.212 -0.060 -0.246 0.023 0.052 0.126 -0.252 -0.351 0.019 0.487 0.464 0.472 1.000
MSCI world - -
index 0.271 -0.097 -0.036 0.036 0.060 -0.070 0.129 -0.029 0.215 -0.070 0.046 -0.004 0.266 0.215 -0.228 0.260 0.168 0.032 0.234 0.321 0.056 0.324 1.000
Notes: This table presents the pairwise linear correlations among the explanatory variables used in eq. (1). Linear correlations above 0.6 are shown in bold.

51
Table 4: Logistic regression estimates of the shipping bank loans on default risk drivers – Dependent variable: default dummy
M1 M2 M3 M4 M5 Change in odds, M5

Financial-specific variables
Constant -2.2552* -1.9517*** -5.5428*** -18.2600*** -5.8693*** -
(-1.6924) (-5.2923) (-30.2341) (-2.8810) (-3.9838)
Leverage = TL/TA -3.3569 - - - - -
(-1.0478)
Debt Coverage Ratio 3 = 0.0649** - - - - -
TL/EBITDA
(2.1254)
Liquidity Ratio 2 = Cash -20.4106** - - - - -
Reserves/Total Assets (-2.1047)
Profit Margin Ratio = 0.0335 - - - - -
EBITDA/Revenue (0.0745)
Firm characteristics’–specific variables
Age of the firm - -0.0681 - - - -
(-1.1984)
Years of cooperation with the bank, - 0.0330 - - - -
yocb (0.1852)
Time-charter policy - -1.8874*** - - -2.3590* 0.3935
(-3.0486) (-1.7607)
Loan-specific variables
Arrangement fee / Amount of loan - - 0.2561 - 0.3405** 3.3646
(1.3212) (2.0850)
Margin - - 10.5811 - - -
(0.1012)
Life to maturity - - -0.3543 - - -
(-1.3724)
ACR Contractual - - 1.0953* - - -
(1.7421)
ACR Actual - - - - - -

Industry-specific and macro variables

1yr TC - Spot - - - 0.0001* 0.0002*** 2.7220
(1.6818) (3.2665)
3yr TC - Spot - - - - - -
***
Orderbook / Total Fleet - - - 21.3951 - -
(3.2284)
Inactive Tonnage / Total Fleet - - - 1.2271** 0.8595*
2.4124
(2.4671) (1.6980)
ClarkSea index - - - 0.0000 - -
(0.7107)
MSCI world index - - - 0.0063*** - -
(2.6713)
Observations (loan-years) 461 470 307 271 208 -
McFadden pseudo-R2 14.97% 8.99% 20.33% 34.76% 33.53% -
H-L 5.94 7.07 8.24 6.41 3.21 -
[p-value] [0.6542] [0.5296] [0.4102] [0.6019] [0.9203] -
LR stat 25.38 7.12 12.02 16.76 20.53 -
[p-value] [0.0000] [0.0681] [0.0172] [0.0050] [0.0004] -
SBIC information criterion 85.10 90.58 75.71 76.01 67.39 -
Loan clusters 123 128 78 83 56 -
Time clusters 14 14 12 14 11 -
Mean VIF 1.03 1.01 1.28 1.22 1.01
Brier Score - - - - 0.0337 -
Notes: This table presents the results of the estimated logit regressions between the defaults of shipping bank loans and different specifications of the econometric model
described in eq. (1). t-statistics are reported in parentheses below the estimated coefficients. Statistical significance of the estimated coefficients is denoted with *, ** and ***
for 10%, 5% and 1% significance levels, respectively. All standard errors reported are two-way cluster-adjusted, as described in Petersen (2009). Columns M1-M4 report
estimates for specifications which test the explanatory power of individual groups of default risk drivers (i.e. finance-specific, firm-specific, bank-specific, industry-specific
and macro factors). Column M5 reports the most parsimonious model and includes all variables which are statistically significant in M1-M4, while minimizing SBIC criterion
and maximizing McFadden pseudo-R2. The latter is an indication of the goodness of fit of the estimated model. The column change in odds shows the relative importance of
the estimated coefficients for the M5 model and represent the change in the odds (the probability of observing a default over the probability of non-observing a default) for
one standard deviation increase in the value of the corresponding regressor, ceteris paribus. The “change in odds” is computed as: (
), where beta stands for the estimated coefficient of the ith explanatory variable presented in column M5 and SD stands for the standard deviation of that explanatory
variable. The SBIC information criterion assesses the explanatory power of each estimated model with smaller values indicating higher explanatory power of the regressors.
The likelihood ratio (LR) test, following the chi-square distribution with d.f. equal to the number of explanatory variables, suggests that the null hypothesis that all estimated
coefficients of each model are jointly zero is rejected at the 5% confidence level, apart from specification M2. The H-L statistic is the Hosmer and Lemeshow (1989) test
statistic, where lower values of the test statistic and higher p-values indicate a good fit. . Mean Variance Inflation Factors (VIF’s) are presented as diagnostic tests for the
presence of multicollinearity in the estimation of M1-M5 specifications. For each of the specifications M1-M5, a separate VIF estimate is obtained for each variable included.
However, only the mean (over all coefficients) VIF value for each model is reported here, due to lack of space. As a rule of thumb, VIF values below 5 indicate absence of
multicollinearity. Thus, there seems to be no multicollinearity issues in the estimated models overall and for each individual coefficient – analytical coefficient VIF values are
available from the authors on request. Standardized coefficients are computed as: , where, and are respectively the standardized and the estimated
coefficients, and are respectively the standard deviations of the ith explanatory variable and of the shipping bond spread. Last, the Brier (1950) score is estimated to
assess the forecast and discrimination power of the proposed model. Brier score is measured as the mean squared difference between the predicted probabilities assigned to the
default or non-default outcomes for a specific loan and the actual outcome. Numbers in square brackets [.] indicate p-values.
 Table 5: Model M5 estimated probabilities of default (PD’s) for each of the defaulted
shipping loans granted

Default year Amount of Loan PD from model M5

2005 9,000,000 29.46%
2006 1,500,000 20.31%
2006 2,000,000 40.93%
2006 9,000,000 32.94%
2007 6,000,000 1.41%
2007 6,500,000 14.86%
2008 6,300,000 6.11%
Notes: This table presents the PD’s of the eventually defaulted bank loans at the time of the initial credit risk assessment
from the bank.

Table 6: Descriptive statistics for ex-post probabilities of default (PD) for non-defaulted and
defaulted loans estimated by model M5

Model M5
Non-defaulted Defaulted
Observations 201 7
Mean 2.75% 20.86%
Median 0.37% 20.31%
Min 0.01% 1.41%
Max 77.17% 40.93%
Standard deviation 6.71% 14.46%
Skewness 6.64 -0.01
Kurtosis 18.94 1.69
Notes: This table presents descriptive statistics for the ex-post probabilities of default (PD) for defaulted and non-
defaulted loans produced by model M5; the most parsimonious model.
 Figure 1: Total volume and number of deals for global shipping bank loans

Time

Notes: This figure presents the total volume and the number of new deals for shipping bank loans worldwide for the
period 2002-2012. Volume is in $ million. Source: Marine Money (based on data by Dealogic).

Figure 2: Daily Standard Deviation with 1 year rolling window MSCI World Stock Index vs. BDI vs. Goldman
Sachs Commodities index

1.80%
1.60%
1.40%
1.20%
1.00%
0.80%
0.60%
0.40%
0.20%
0.00%

Time

MSCI World Stock index BDI index Goldman Sachs Commodities index

Notes: This figure presents the standard deviation with 1 year rolling window for MSCI World Stock Index vs. BDI vs.
Goldman Sachs Commodities index. Source of data: Bloomberg.

54
 Figure 3: Receiver Operating Characteristic (ROC) Graph

Notes: This figure depicts the Receiver Operating Characteristic (ROC) curve of the proposed model (M5) of
default risk assessment for shipping bank loans. The ROC curve is a plot which indicates the performance of a
binary classifier as its discrimination threshold varies. As observed, the area under the ROC curve equals 0.9293,
where a model with “perfect” discriminatory power between eventually defaulted and eventually non-defaulted bank
loans would have an area under ROC curve equal to one. The y-axis named sensitivity is also known as true positive
rate, which is measured as the true positives out of the total actual positives. The x-axis named 1 minus specificity is
also known as the false positive rate and represents the fraction of false positives out of the total actual negatives.
The step-style ROC curve is due to the limited number of defaults in the sample.