Trade credit lending by non-financial firms.

Trade credit lending by non-financial firms.

“The importance to predict the probability of credit default”

Applied to Heineken.

August 2009

Anouk Blankhorst

MSc Thesis International Economics and Business Faculty of Economics and Business

University of Groningen The Netherlands

a.blankhorst@student.rug.nl

Supervisor:

M. Koetter

Faculty of Economics and Business University of Groningen

The Netherlands

m.koetter@rug.nl

2 ABSTRACT

The financial relationship between non-financial firms and customers established by trade credit is a well-known and important phenomenon. This trade credit can be seen as a binding relationship between non-financial firms (suppliers) and customers. In the literature the non-financial firms in the US serve as an example of lenders and users of trade credit. However, less can be found on trade credit applied in real, nor examples of non-financial firm being trade credit lenders in the Netherlands.

This is remarkable while the trade credit relation between the beer brewers (suppliers) and their customers, the owners of pubs, hotels, and restaurants, characterizes the Dutch beer market. In this paper, I explore the existing theory of non-financial firms being trade credit lenders and the credit risk they bear. This theory is applied to the largest beer brewer of the Netherlands and lender of trade credit Heineken NL. I investigate the risk Heineken, as a lender of trade credit, faces and if prediction of this risk is possible based on indicators specific to Heineken. For the prediction of the credit default rate, I use a hazard rate model. Heineken made an exclusive dataset of customer characteristics available for this research.

Keywords: Trade credit; probability of credit default; Heineken.

3 TABLE OF CONTENT

1.INTRODUCTION 4

2.LITERATURE REVIEW 8

2.1 Trade credit motives 8

2.2 Trade credit risk 11

2.3 Hypotheses and application 12

3.DATA & METHODOLOGY 14

3.1 Data 14

3.2 Description of the variables 15

3.3 Development of a binary depedent model 18

4.RESULTS 20

5.CONCLUSION 24

6.REFERENCES 26

7.APPENDIXES 28

A. Table 1 Variable description 28

B. Table 2 Summary of the independent variables 29

C. Graph 1 The ROC-curve 30

D. Table 4 The independent variable tested with the logit model 31

E. Graph 2a Sales growth versus BKR=0 32

Graph 2b Sales growth versus BKR=1 32

F. Graph 3a The default versus non-default customers 33

Graph 3b The default versus non-default customer (%) 33

G. Heineken NL, the beer market and the customers 34

4 1.INTRODUCTION

Due to the current economic situation, anno 2009, trade credit funding by non- financial firms has become more important than before. In the current economic recession, the possibility to receive credit from financial institutions is more difficult and almost made impossible. Hence, credit demand is shifting to the non-financial firm

, the suppliers, because of financial institutions not willingly to provide this (Cheng and Pike, 2003). For a customer who is in need of financial aid, the alternative of funding with trade credit by non-financial firms instead of funding provided by a financial institution can be the only way to survive in a recession. Therefore, the customer sees the supplier as an alternative source of providing financial aid. Biais and Gollier (1997) argue that suppliers with market power are in times of monetary contraction of great importance. They state, “when suppliers have market power, buyers have better access to credit. An empirical result of this is that, during episodes of tight monetary policy and high interest rates, industries with oligopolistic suppliers should be better able to accommodate the shock by relying on the use of trade credit” (Biais and Gollier, 1997:926). Furthermore, they emphasize that customers, often small firms, who are credit rationed by the financial institution turn to their supplier for credit (Biais and Gollier, 1997). In the current economic situation, anno 2009, the result is an increasing role of the suppliers as a lender of trade credit.

The Dutch horeca

segment is an example of a business area that relies heavily on trade credit and is struggling to survive in the current economic recession. The prohibited smoking law in the horeca is another reason why especially this horeca segment faces difficulties to survive. Since this law has been in force, since July 2008, the horeca market has decreased by more than 4.9 % in the last quarter of 2008 and 6.1% in the first quarter of 2009. This reduction in total sales is also expected for the rest of the year on a quarterly basis (www.cbs.nl).

1

A non-financial firm means that the firm has a core business not related to lending money. Here a non-financial firm refers to a supplier of goods, who also can/will facilitate trade credit to his customers. In this paper non-financial firms and suppliers are equivalent.

2

HORECA is the Dutch abbreviation for Hotel, Restaurant, and Café. Cafe means pub. In the rest of the paper the

abbreviation horeca is used to indicate the segment of hotels, restaurants, and pubs.

5 Because of the recession and the prohibited smoking law, the demand for trade credit in the horeca segment is high. Therefore, the associated trade credit risk and the probability of default are increasing

. Because the financial institutions are not willing to lend credit, the horeca owners are trying to get this credit from their main supplier, namely the beer brewer

. This indicates the role of Heineken. As more Heineken customers are demanding trade credit, Heineken faces the option to lend or not to lend this trade credit. Although not commonly known, in the Dutch beer market the way of binding customers to their suppliers is mostly done by using trade credit. In general, the brewers use this trade credit as a binding contract to their customer. The trade credit contract guarantees the brewer the selling of his products during the period of trade credit receiving. Heineken, as market leader with more than forty percent market share in the Netherlands

, is due to its market dominance

not allowed to offer the same kind of other binding contracts their competitors can. This is the quintessential difference for Heineken compared to other brewers. Heineken has therefore only the trade credit tool to bind customers to sell Heineken products for several years. However, the reason for Heineken to lend trade credit is not only a financial debate. Biais and Gollier (1997) motivate that indeed other reasons exist for a supplier to grant trade credit. Sentiment, long lasting relationships with customers, and keeping certain locations in their portfolio are other drivers for Heineken to be considered in lending trade credit or not.

The above discussion shows why Heineken would lend trade credit during times financial institutions are not eager to do that. However, by lending trade credit the beer brewer also has to deal with the associated financial risk in the form of trade credit default

. Financial institutions use hazard rate models based on financial data for the prediction of the credit default in their decision to lend trade credit. Dealing with the trade credit risk and the default rate is for non-financial firms, like Heineken, often more

3

The probability of default is increasing; this is based on the total numbers of default compared to the numbers of previous years of the horeca customer of Heineken NL. See also graph 2a and 2b in appendix F.

4

Horeca owners prefer to receive credit from a financial institution as the disadvantage of a contract with brewers outweighs the advantages of credit receiving from their brewer. Due to the privacy policy of Heineken, these advantages/disadvantages cannot be discussed here.

5

See Heinekeninternational.nl/annual report.

6

See ‘mededingings’ law for exact explanation of contracts and market share.

7

The term “trade credit default” is here defined as a customer whose trade credit contract will be dilapidated based

on disobedience of the contract.

6 difficult to interpret than it is for financial institutions. Heineken has often no insight in the financial situation of customers due to incorrect or missing data. Important therefore is that the default rate of the trade credit can be estimated or monitored. (Here, monitoring can be possible as time invariant variables are used to predict the probability of default.) In case the probability of default can be estimated, it would be possible to take action upon this estimate. Examples of the action taken are the decision of Heineken to stop the trade credit or to offer other trade credit terms. In other words, from an economic perspective, Heineken is able to identify customers having a high probability of trade credit default based on these estimates of the default rate. Thereby Heineken indirectly minimizes the risk of default associated with the lending of trade credit for itself. To conclude this discussion results in the following research question:

Can the probability of credit default be predicted for trade credit lender Heineken?

An exclusive customer dataset of Heineken enables me to investigate this research question by estimating the risk and therefore the probability of default specific for customer of Heineken. By considering customer characteristics in this beer market, the advantages are as follows. First, the focus in earlier research relies mainly on financial institutions lending trade credit or on non-financial firms in general. By considering only a specific niche, can thus provide insights into what determinants for trade credit specified to this niche can be. This research on trade credit in this beer market is therefore new.

Next, other research on trade credit focuses mainly on the financial determinants of the trade credit default. This research concentrates on the characteristics of customers only

. It can indicate that the commonly used financial variables are not the only explanatory variables to rely on in determining the probability of trade credit default.

Finally, this research can show if the risk of default in this niche can be estimated, with alternative determinants. Therefore, this research can indirectly indicate if trade credit in this niche is either a useful tool or not, for a non-financial firm in the role of a trade credit lender.

8

The research only concentrates on customer characteristics and therefore bears the risk of omitted variables.

7 In general the main disadvantages of this unique dataset are: the dataset is only applicable and representative for Heineken, no financial variables are investigated, and the influence of time is neglected. Conclusions drawn from this research can only be applicable to the Dutch beer market.

In this paper, I will explore the importance of trade credit and its default and apply this to Heineken. With the use of the existing literature, I will examine the unexplored trade credit system in the Dutch beer-brewing market. Furthermore, a hazard rate model is developed based on the dataset of Heineken. This model can help Heineken to identify risky customers by signalling customers who have a high probability of trade credit default.

The results show that only two determinants, sales growth and BKR testing

, of the six tested determinants are significant and therefore useful for Heineken in the estimate of the probability of trade credit default of their customers. Consequently, based on these results a non-financial firm cannot rely exclusively on alternative (non- financial) determinants in order to predict the probability of credit default.

This paper continues in the following way: in chapter 2, the literature concerning trade credit and the risk of trade credit will be discussed. Afterwards hypothesises and an application to Heineken is made. In the methodology, chapter 3, the dataset is explained and a hazard rate model, I named ‘early warning system’, will be developed.

Chapter 4 explains the results of this early warning system and its predictions. Chapter 5 will present the conclusions.

9

BKR stands for bureau credit registration. BKR is a company that collects data, like debt or payment irregularities,

of customers who received any kind of credit in the past by any kind of firm. For detailed description see www.bkr.nl

8 2. LITERATURE REVIEW

Trade credit is part of the supplied cash lent to the small firm. Financial institutions identify these small firms as a highly risky group (Cheng and Pike, 2003).

Trade credit can differ from a business loan with respect to the lenders of the trade credit. In case of trade credit the lender can be a financial institution as well as a non- financial institution. With a business loan the lender is in general a financial institution.

Furthermore, the terms of the business loan contract differ from that of a trade credit contract. A business loan is an amount of cash received by a customer who can spend it the way he/she likes to. Trade credit on the other hand, can be cash but also be a reduction of the invoice. Where the credit is used for, is often decided in accordance with the supplier. Often the credit is used directly or indirectly for the supplier’s product, making it attractive for a supplier to lend trade credit.

Reasons for lending and extending trade credit clearly depends on who the lender of the trade credit is. The motive varies, depending on either a financial institution or the supplier being the trade credit lender. The difference between a financial institution and a supplier as a trade credit lender lies in the advantage the lender receives by providing trade credit. Here, only trade credit lending by non- financial firms (suppliers) is highlighted. For a supplier to lend trade credit to its customers the following motives are discussed in this review: the price discrimination motive, the marketing motive and the financial motive.

2.1 Trade credit motives

“Trade credit can be viewed as part of the firm’s pricing policy”, Cheng and

Pike (2003:423). This quote reflects the price discrimination issue. Price discrimination

is one of the advantages a supplier has over financial institutions in offering trade credit

(Peterson and Rajan, 1997). Price discrimination can therefore be a reason why

suppliers are more keen than financial institutions to lend trade credit. In case the

market of suppliers and customers makes it impossible to discriminate through prices,

trade credit can be a useful tool to enable a supplier to discriminate although not by

prices but by trade credit terms. The suppliers are in general not bound to any kind of

obligations or rules in setting the terms for the trade credit. The trade credit terms are

9 negotiated on an individual basis which makes it possible for a supplier to charge different customers different terms for the trade credit. The suppliers can use the interest as a device to identify customers with a higher risk of default. The higher the interest rate the customer is willing to pay, the riskier the trade credit lent to the customer is. So, the customers who are demanding more trade credit or relaxation of the repayment terms in combination with a high interest rate, should be seen as a warning signal (Cheng and Pike, 2003). Consistent with the paper of Cheng and Pike (2003), this is especially the case in oligopolistic markets in which suppliers are characterized by great market power.

Another motive why a supplier wants to lend trade credit is called the marketing motive. This motive holds that a supplier can tie their customers to him and thereby guarantee that his products will be bought. It is based on the establishment of a long- term relationship between supplier (lender of trade credit) and customer. Wilner (2000) also highlights this relationship in his paper. He argues that a lender of trade credit who is willing to maintain a relationship with the customer is more flexible in lending trade credit during times the customer faces financial distress. Petersen and Rajan (1997) share the same opinion. They state that suppliers are eager to provide trade credit, especially to growing firms, when this trade credit is expected to capture the customer.

Summers and Wilson (1997:440) also refer to this, “allowing both customers and suppliers to establish a reputation which may be benefit in future negotiations and to build a working relationship”.

In a more recent paper of Summers and Wilson (2003) they emphasize again that building a customer relationship by means of trade credit is helpful for both the customer and the lender of trade credit in times of high or low sales. They highlight that an important motive for a supplier to lend trade credit is that “it can be used as a many faceted marketing/ relationship management tool and/ or as a means of signalling information to the market or to specific buyers about the firms, its products and its future prospects/ commitment” Summers and Wilson (2003: 439). Cheng and Pike (2003) call this the investment motive and indicate that creating shareholder value is based on the customers who are therefore willing to establish a long-lasting relationship.

The customer views a long-term relationship by means of trade credit as positive. The

10 marketing motive also holds the other way around. Buyers selecting suppliers who offer the best trade credit terms (Cheng and Pike, 2003).

In an imperfect capital market, suppliers are better able to finance customers at lower costs than financial institutions can, this is called the financial motive (Smiths, 1987). Suppliers who are better in receiving trade credit from banks are taking the role of intermediates by putting this trade credit forward to their customers. Cheng and Pike (2003) emphasize that this financial motive indeed holds. They mention that due to the available market information it is more difficult for a financial institution to use and interpret this information than it is for a supplier who has market knowledge and a supplier-buyer relationship. Peterson and Rajan (1997: 662) share the same argument by emphasizing that suppliers might be better at “evaluating and controlling the credit risk of their buyers”. This is possible because suppliers are in the same markets as their customers and as a result have important market knowledge. Hence, the cost of granting trade credit might be higher for a financial institution.

So based on the motives discussed above, a supplier is eager to offer trade credit and set competitive trade credit terms in order to attract customers. All the reasons given for these motives are in favour of suppliers instead of financial firms being trade credit lenders. However, these motives are based on soft information sources like relationships between supplier and customer, payment behaviour of the customer or being owner of multiple establishments (Biais and Gollier (1997) and Wilner (2000)).

However, financial institutions are able to consider hard information sources, like

financial customer information, that the suppliers do not have. This hard information

gives the financial institutions an advantage over the suppliers in lending trade credit.

11 2.2 Trade credit risk

The risk of trade credit default is normally reflected in the interest rate charged for the trade credit. In general, this relationship between the interest rate charged and the probability of default is consistent with the following: the higher the probability of default is expected, the higher the interest rate charged, and the more likely renegotiation of the trade credit will occur (Wilner, 2000).

Being a lender of trade credit is attractive in case the default rate of the trade credit can be estimated and the lender is able to price the risk of the trade credit properly. Financial institutions use the interest rate as a tool to indicate the risk of the customer (Stiglitz and Weiss, 1981). Financial institutions estimate this interest rate based on hard financial data. Examples of this ‘hard data’ used to determine the interest rate are variables like the customers’ own access to capital, sales, profits, debt-equity ratio, cash flow, and the characteristics of the customers firm, used in the models of Peterson and Rajan (1997) and Banerjee et al. (2004).

However, in case the lender of trade credit is a non-financial institution, a supplier, the estimate of the default rate is often more difficult to make. As mentioned in the introduction of this paper, the reason for this is that suppliers simply do not have the access to correct data or permission to check their customers’ financial data.

Furthermore, non-financial firms often have too many customers demanding trade credit who cannot easy be categorized into interest groups. In general, especially the smaller non-financial firms find it too time consuming and too expensive to charge every customer an individual interest rate. As a result, a reliable link between high interest rates charged and the risk of trade credit is often not possible to estimate for the supplier. Suppliers have to rely on other ‘softer’ sources to indicate the risk of trade credit. Ideally, these soft sources should be used as a complement to the available hard information.

Clearly, a supplier uses other information then financial institutions to estimate the risk. This information can be different for every supplier and depends on what the supplier values as important knowledge on identifying the credit risk (Biais and Gollier, 1997). A supplier can accept the credit risk if he is able to rely on, for him, suitable alternative information on which an estimate of the probability of default can be made.

A supplier can use the collateral often better and more efficient than financial institution

12 can. This is a reason why a supplier can be more flexible in accepting credit risk (Biais and Gollier, 1997). Suppliers are better able to use the collateral, like location or inventory, as the suppliers operate in the same market as the customer and have specialized market knowledge.

2.3 Hypotheses and application

Overall, financial and non-financial lenders of trade credit use the interest rate to reflect the risk of trade credit and therefore the probability of default. The literature discussion above points out that the probability of default of trade credit can be based on the interest rate or on financial data. Unfortunately, both cannot be applied to Heineken as the required data is not available and as a consequence Heineken does not set an interest rate which reflects the trade credit risk. The interest rate charged by Heineken is conform to the money market interest rate and is the same for all their trade credit customers.

Interestingly, Heineken is mostly concerned in the gaining of the location of the customer and on establishing a relationship with the customer. Therefore, the

‘marketing motive’ is the main reason why Heineken is a lender of trade credit. Based on the marketing motive and gaining of the location, Heineken lends trade credit to a customer based on a “commercial deal”. This makes it difficult to eliminate perilous customers. Being unable to eliminate these perilous customers results in a higher probability of default. This and the current economic situation, make the prediction of credit risk extremely important to Heineken. As explained above, the credit risk prediction should be based on other determinants than the interest rate or financial data.

Heineken evaluates the risk of trade credit based on the characteristics of the receivers

of the trade credit, the customers. For Heineken, the characteristics of the customers are

the only source of customer information. Heineken collects and evaluates this

information about the characteristics of the customer before lending trade credit. The

investigated characteristics are time independent and customer specific.

13 In general, the characteristics of the customer can be split into two groups; the payment characteristics and the ‘financial’ characteristics. The payment characteristics are based on variables that influence the payment behaviour of the customer. The financial characteristics contain variables regarding the financial contribution of the customer to the establishment at the time of receiving trade credit.

This has led to the following hypotheses:

Hypothesis 1: The probability of trade credit default increases when the payment characteristics of the customer of trade credit are negative.

Hypothesis 2: The probability of trade credit default increases when the financial

characteristics of the customer of the trade credit are negative.

14 3. DATA & METHODOLOGY

3.1 Data

The collected data to set up this empirical research is provided by Heineken. The sample includes only customers of Heineken and provides customer specific information. The investigated group of customer received trade credit of Heineken during the years 2004 until 2009. The sample size consist of n=455 customers that have Heineken as their lender of trade credit. 30.1 percent of the sample represent the default rate and reflect 137 customers of the sample. 318 customers are still receiving the trade credit and represent 69.9 percent of the total sample size. The time span (2004 until 2009) is divided into quarters of a year and contains therefore t=18 quarters of observation periods. The observations include only customers who received their trade credit during this period

. The first observation is of the last quarter of 2004. In this last quarter of 2004, Heineken established a new way of lending trade credit by providing this through a financial institution owned by Heineken. The third quarter of 2008 has been chosen as the last quarter customers in the time span could receive trade credit as otherwise not enough data would be available about the variables investigated. The final quarter of observation is the first quarter of 2009 therefore customers of trade credit were observed for at least six months.

The customers are randomly selected from the total trade credit customer database of Heineken. After the selection, the dataset is modified by deliberately excluding customers of multiple establishments who were selected in the dataset with more than one establishment. This deselecting is done to assure that the characteristics of a single customer are not taken into account twice. The 455 customers can be owners of multiple establishments with multiple trade credit contracts. Here only one establishment and one trade credit contract per customers is taken into account.

.

This dataset is used to estimate the probability of default for a customer i, with starting date of receiving trade credit t and in case of default, an end date of t*. Five out of the six independent variables do not vary over time. Therefore, the data is a cross-

10

Note that the customers do not have the same start and/ or end date of receiving the trade credit.

11

By deliberately excluding customers, the dataset does not fully represent the trade credit customer database of

Heineken. The dataset of this research therefore represents 90 percent of the real trade credit customer database of

Heineken.

15 sectional data set. The investigated group of customers did not receive the trade credit at the same time for the same period. Additionally, the customers are different in size (measured by total sales), types of establishments (e.g. Pubs, restaurants, etc) and located only in the Netherlands. This is done to ensure that the focus is on the customer characteristics and not on other external characteristics that cannot be estimated by Heineken.

3.2 Description of the variables

The dependent variable in the model is a binary variable and indicated by y, it can have the value 1 or 0. 1 represents a customer still receiving trade credit while the value 0 indicates that a customer defaulted on the trade credit. The reason for this trade credit default is not investigated. In general, we can assume that trade credit default emerge if a customer is not able to meet the contract terms and as a result of this, Heineken no longer lends trade credit. Table 1 provides an overview of the variables, their definition, expected signs, and their source.

[Insert Table 1 here]

The following six independent variables are specific indicators for the characteristics of the entrepreneur of the firm. All these variables are exogenous

and five variables are categorical dummies. As mentioned in section 2.3, two groups of customer characteristics are tested, namely the payment characteristics and the financial characteristics. All independent variables split up into to the payment characteristics or to the financial characteristics of the customer. The independent variables refinancing, BKR-testing, payment behaviour and entrepreneurial experience belong to the payment characteristics and relate to the customer and his/her paying ability. The financial characteristics relate to the customer and the establishment, it include the remaining independent variables sales growth and financing with own money. Table 2 shows the summary statistics of the independent variables used in this research.

[Insert Table 2 here]

12

The test results confirm that the independent variables are the exogenous.

16 The first independent dummy variable is called refinancing, DUMREF.

Refinancing occurs if a customer who receives trade credit faces difficulties with the trade credit contract and applies for trade credit extension. If refinancing is indicated with a 0, this means that the customer did not receive a refinancing of trade credit in the past. The value 1 corresponds to the opposite. The trade credit lender, Heineken, has been refinancing the customer by extension of the trade credit. The importance of including the measure refinancing is clear; the reasons why a customer demands trade credit extension are in general not positive and occur due to financial problems or difficulties with the repayment terms. In general, if the refinancing is requested the risk of default for the current trade credit is expected to increase. Based on this the expected sign of the variable is positive showing that a refinancing increases the probability of default.

BKR testing (DUMBKR). This dummy variable can have the value 0 or 1. The value 0 indicates that the outcome of the BKR testing is positive and 1 means that the BKR test has a negative outcome. A negative outcome shows that the customer is registered with a history of payment irregularities. BKR testing shows if the customer has a history of bad paying behaviour. BKR testing is a tool for the lender of trade credit. It is used to decrease or limit the risk of credit and should help in preventing a debt burden not able to bear by the customer. BKR testing works therefore as an indicator by showing if the supplier faces a high or low risk by granting credit. This risk is based on the payment behaviour and age of the customer. BKR is registered for all the credit the customer received in the past. (www.bkr.nl) The expected sign of this variable is positive, a customer with a negative BKR testing, (DUMBKR equal to 1) increases the probability of default.

If the customer adds own equity in the financing of the exploitation, the

probability of default is expected to decrease. In the literature, this is discussed as the

agency theory. The agency theory states that in case the principal hires an agent the

interest between the two of them can be different. To converge to a common interest in

favour of the principal the agent could for example be compensated in the form of

equity or be rewarded based on company results (Jensen and Murphy, 1990). Here

Heineken as a supplier of trade credit fulfils the role of the principal. The interest of

Heineken with respect to the outstanding customer trade credit is a low probability of

17 trade credit default. The customer as an agent might be less concerned with trade credit default if he/she does not have own money at stake. When own money has to be paid when default occurs, the expectation is that a customer is eager to work harder to prevent this trade credit default. To measure if this holds, ‘financing with own money’

is taken into account, DUMOWN. Based on the agency theory the sign of this dummy variable is positive, meaning that if own equity input is less than ten percent of the total investment, the probability of default is higher. The dummy variable can have the value 0 representing a customer who invested more than ten percent own money of the total funding of the firm. Value 1 shows the opposite.

The fourth dummy variable is payment behaviour, DUMPAY, here defined as payments that are collected within the terms agreed upon in the contract. Payment behaviour is indirectly influencing the probability of default of the trade credit (Wilner, 1997). Payment behaviour can be seen as a signal of the customer’s ability to pay the supplier as agreed upon in the contract. The payment behaviour can be unknown if the customer is new to Heineken, value 1 represents this. If payments behaviour is known, it is reflected in the value 0. Payment behaviour is known for existing customer who receives a trade credit extension. For new customers, the payment behaviour is unknown. The distinction between regular and irregular payment

behaviour could not be made as there was not enough data available for this.

Entrepreneurial experience, DUMENTEXP, is the fifth exogenous dummy variable. Entrepreneurial experience is classified as good if the customer has more than two years

of entrepreneurial experience, indicated by the value 0. The value 1 shows that the customer has less than two years of entrepreneurial experience. Entrepreneurial experience is taken into account as having more experience in running an own business will decrease the probability of default. Experience helps to survive the starting up period and dealing with the associated, often unforeseen, problems that are associated with running a new business. Hadjimanolis (2000) confirms this by indicating that indeed the managerial/owner skills are of great influence especially in small firms.

These managerial/owner skills are based on the experience of the entrepreneur and are

13

Irregular payment behaviour is payment behaviour not according to/as agreed upon in the trade credit contract.

14

The number of years, 2, of entrepreneurial experience is defined as good based upon the opinions of controllers and

account managers of Heineken.

18 present due to practice and experience over time. Entrepreneurial experience here does not per se have to be in the same sector.

Growth in sales (€), SALESGROWTH, is the last independent variable. Lower sales over an extended period can result in lower revenue. This lower revenue leads to a decreasing ability to meet for example the contract terms. In turn, difficulties in meeting the contract will result in an increase of the probability of default. Therefore, negative sales growth increases the probability of default. The sign of this variable is expected to be negative showing that a negative growth in sales will increase the probability of default. The sales growth is calculated per quarter and the values represent the actual growth numbers of the sales. To find out if the sales growth is increasing or decreasing the quarter of year x will be compared with the same quarter of the previous year. By calculating it per quarter, the four seasons are taken into account. This allows me to control for seasonal influences, as some of the customers are highly dependent on seasonal sales.

3.3. Development of the binary dependent model

A hazard rate model is often used for probability estimates of trade credit and its risk (see Madan and Unal, 2000). The hazard model should enable a prediction of default possible for the total group of customers who receive trade credit from Heineken. The common way to estimate this probability of trade credit default is done by using a linear or logistic regression (Thomas et al., 2005). The type of regression applied in this research will be discussed in the next section.

To estimate the probability of default a so-called early warning system (EWS) will be developed for Heineken. This model is based on the basic probability density function model. As mentioned earlier, the dependent variable, Y, is binary and can have the value 0 or 1. The value Y= 1 means that default occurred, Y= 0 shows that the customer still receives trade credit. The probability outcome of the EWS has to lie between the values 0-1 with 1 indicating that the probability of default is hundred percent. An outcome of 0 corresponds with a probability of zero percent. All percentages of probabilities between zero and hundred can be an outcome of the EWS.

A simple linear regression cannot be applied to estimate the probability of the

dependent variable Y. The outcome of a linear regression can have any value between -

19

∞ and ∞. This is not possible as the outcome of the EWS has to be a value between zero and hundred percent. An alternative for running a regression with an outcome of Y corresponding to 1 or 0 is a logistic regression in a log linear model. “Here the log of the probability odds by a linear combination of the characteristics variables” (Thomas, et al., 2005: 1009). However, this model is also not applicable as the model assumes a linear outcome of the right hand side of the function (Thomas, et al., 2005).

Therefore, a logit model for binary choice is preferred to estimate the probability of trade credit default. This preference is based on the advantages a logit model has compared to a logistic regression. The first advantage of the logit model relies in the dependent variable, which is categorical. The other advantage of the logit model is that it allows multiple categorical and continuous independent variables to predict the binary dependent variable. As explained above all the independent dummy variables used, are categorical. However, “the odds between any pair of alternatives are independent of irrelevant alternatives” (Hill, Griffiths, and Lim, 2008:429). This is an important assumption of the logit model and must not be violated. Additionally, observation with outliers ought to be noticed and be excluded as these outliers could bias the results. The following logit model for binary choice (1) is applied:

(1)

Inserting the independent variables into this logit model for binary choice, results in the following equation (2). This equation is used to determine the probability of trade credit default for a customer of Heineken

€

1

1 + e

(

) (2)  

Where:

Y, is the dependent variable and can be: .

X, represents the customer whose characteristics are investigated.

20 4. RESULTS

The cross-sectional dataset have to be tested with precaution. Heteroskedasticity should be investigated as heteroskedasticity is often encountered when cross-sectional data is used (Hill et al. 2008:200). A Wald-test is performed to detect this heteroskedasticity (Hill et al. 2008). The result showed that the dataset does not suffer from heteroskedasticity. In order to perform estimations with the logit model the independent variables are also tested on outliers. To detect outliers in the observations, which can bias the results, a correlation plot is made for the independent variable sales growth

. This plot showed no outliers in the data hence no corrections of the dataset have to be made. Additionally a check is performed to ensure that the independent variables are exogenous. The above test results lead to the conclusion that the assumptions of the logit model are not violated.

The outcome of the logit model for a 95% confidence interval is presented in table 3. The same logit test is done for a 90% and 99% confidence interval. In general, the interpretations of the results are the same, hence only the 95% is displayed here. As the number of observations is high, n=418, z-values instead of t-values are displayed.

The (Mc Fadden’s) pseudo R2, with a value of 0.264 can be interpreted as the goodness of fit of the logit model. This value indicates that the goodness of fit of this model is relatively high. To confirm this test result, a roc-curve is plotted. The outcome of the area under the curve has a value of 0.853. This outcome shows that the test is accurate, concluding that indeed the model fits well.

[Insert Graph 1 here]

The LRchi2 (6) of 133.86, with the associated p-value, is the result of the join test that the coefficients are zero (Adkins and Hill. 2008). This test shows the significance of the overall model. The results is conform the outcome of the Wald test.

15

Only this variable can have outliers in the observation, as the other independent variables are dummies

21 TABLE 3

The results of the logit model.

Number of observations = 418

LR chi2(6) = 133.86

Prob> chi2 = 0.0000

Pseudo R2 = 0.2642

Independent variables

β Odds ratio P>│z│

Salesgrowth -0.024 0.976 0.000

DUMBKR 0.587 1.797 0.050

DUMREF 0.362 1.435 0.236

DUMOWN -0.444 0.641 0.111

DUMPAY -0.023 0.977 0.939

DUMENTEXP -0.357 0.700 0.236

_cons -1.457 - 0.000

Note: DUMFIN dropped because of collinearity.

In general, the significance of the tested variables is shown by the result of the p-value. The first independent variable, sales growth, is significant. The sign of the coefficient is as expected negative, it has a β-value of -0.024. This means that if a firm faces an increase in the growth of its sales, the probability default declines with β (= - 0.024) multiplied with the growth in sales. The interpretation also holds vice versa, a firm facing a negative growth in sales will have a higher probability of trade credit default. This default rate increases with β (= -0.024) multiplied by the negative growth in sales, which results in a positive value and an increase in the probability of credit default.

The second independent variable is the dummy variable BKR. This variable is

also significant and has a positive coefficient of 0.587. This positive sign is indeed as

expected, meaning that a positive BKR, dummy value of 0, does not contribute to a

higher probability of trade credit default. A negative outcome of BKR is represented by

the value 1. This implies that this negative outcome increase the probability of trade

credit default by 1 times the value of the coefficient (0.587). This result is also

confirmed by the odds ratio. The outcome of the odds ratio with a value of ≥1,

22 shows that the odds of trade credit default is higher for the independent variable DUMBKR with value 1. This means that a customer with a negative BKR outcome has 1.797 times a higher odd of trade credit default compared to a customer with a positive BKR testing. This outcome is according to the expected sign of this dummy BKR- testing. A negative value of the dummy variable corresponds to higher probability of trade credit default.

The other independent variables tested: refinancing (DUMREF), own money input (DUMOWN), payment behaviour (DUMPAY), entrepreneurial experience (DUMENTEXP), are insignificant. The results of the odds ratio for the other binary independent variables, DUMOWN, DUMPAY, DUMENTEXP are not as expected and indeed there p-value shows no significance.

To see if the independent variables show significance in a certain combination of independent variables, every independent variable is tested separately. In the first test, besides sales growth and BKR-testing, DUMENTEXP is also significant. Any combination of two of these significant independent variables is tested against the dependent variable. The results show no significance for the variables in any kind of combination.

[Insert Table 4 here]

To find out which significant independent variable, BKR-testing or sales growth, has the strongest influence on the probability of default two graphs are plotted.

Both with the independent variable sales growth on the x-as and the predicted probability of default on y-as are plotted. The first graph is plotted with a BKR=0 and the second with BKR=1. Comparing both graphs, the density in graph 2b (with BKR equal to 1) is higher in the area with a high probability of default (upper left corner).

This reflects that more customers with a BKR equal to 1 have a higher probability of default compared to customers with the same sales growth but with a BKR equal to 0.

Therefore, I conclude that the variable DUMBKR has the strongest influence in the estimate of the probability of trade credit default.

[Insert Graph 2a & Graph 2b here]

23 Next, the development over the years 2005-2008 of the customers receiving trade credit versus the customers no longer receiving trade credit is investigated. A graph for the years 2005- 2008 is plotted. The graph shows an increase of trade credit customers for the years 2005-2007. The increase is high for the year 2005-2006. This can be the result of Heineken’s target, to increase and ensure their market share in the Dutch beer market. For the customer with trade credit default the observations for the year 2005 are relative low. This can be related to the starting date of the observations, the last quarter of 2004. Customers receiving trade credit from that moment have had less time to default by the year 2005. For the year 2008, in combination with an increase in the number of customer with trade credit default a decline in the number of customers receiving trade credit is observed. This shows that indeed the economic crisis and the new smoking law do have impact on the number of customers defaulted.

Because of the economic crisis and smoking law, Heineken is also more careful in lending trade credit, which can be seen in the decline of the customers receiving trade credit. In general, based on this decline and the negative economic prospects for the year 2009, customers will have a higher probability of default and will face difficulties to receive or remain to receive trade credit from their supplier.

[Insert Graph 3a & Graph 3b here]

Based on the results for the variables it is impossible to make a general conclusion. Therefore, I cannot accept hypothesis 1 that the probability of trade credit default increases when the payment characteristics of the customer of trade credit are negative. Alternatively, I conclude that the payment characteristic “BKR testing” does influence the probability of trade credit default. If the characteristic is negative, BKR is equal to 1, the probability of trade credit default increases.

Furthermore, the second hypothesis that the probability of trade credit default

increases when the financial characteristics of the customer of the trade credit are

negative, can also not be accepted as a general conclusion for the financial

characteristics of the customer. Instead, here I conclude that if ‘the growth in sales is

negative, the probability of trade credit default increases’.

24 5. CONCLUSION

This research aimed to find alternative (non-financial) variables that could be used in determining the probability of trade credit default for non-financial firms. The results of the logit binary dependent model show that out of the six investigated independent variables, only sales growth and DUMBKR are significant. These two alternative variables can be seen as useful determinants of trade credit default. However, an accurate prediction of the probability of default of a customer cannot be based on two explanatory determinants. The other four possible determinants of trade credit default turn out to be insignificant. Consequently, they cannot be used in the possibility of predicting the trade credit default for a non-financial firm. Other additional variables should be included will these insignificant variables be used in further research. The result for the variable payment behaviour is not confirming to the finding that payment behaviour does indirectly influence the probability of trade credit default (Wilner, 1997). A possible reason for this might be the difference in the definition of payment behaviour used.

The results of this research partly can be explained by the limitations of the dataset. Although it was the goal of the paper to find out if alternative variables could be used to estimate an accurate probability of default, a limitation of the dataset is that it does not contain financial data. This might have influenced the results. Additionally, to include five dummies out of the six independent variables might have biased the results.

Suggestions for further research would be to combine financial variables (hard data) with alternative variables (soft data) and limit the number of dummy variables.

Although Heineken serves more than forty percent of the beer market and is the largest

beer brewer of the Dutch market, it is difficult to draw a general conclusion for all the

trade credit lenders in this beer market. Another drawback of this research is that the

influences of time spans, like the average period of receiving trade credit before default

occurs, are not investigated. In the current economic crisis, it is important to monitor the

average period a customer receives trade credit before trade credit default occurs and

find out if this period is decreasing or increasing. If the average period is decreasing, the

probability of default is also likely to increase.

25 The general conclusion drawn for this research is that trade credit default cannot be interpreted for the beer market based only on these alternative variables. The probability of default predicted by only the two significant determinants would not represent an accurate estimate of the customers’ risk. Therefore, the risk for the lender of trade credit with this estimation of the probability of default would be extreme high.

Based on the investigated determinants it would not be in the interest of the trade credit lender in this beer market to lend trade credit. Consequently, I state that for a non- financial firm like Heineken predicting the trade credit default exclusively on alternative determinants cannot be done.

To conclude, although we now from this research that the determinants BKR testing and sales growth are significant, other, financial or non-financial, determinants might improve the accuracy of such an early warning system. The preference remains for a system that enables non-financial firms, especially those in the beer market, to predict the probability of trade credit default based not only on financial determinants.

26 6. REFERENCES

Adkins, L.C., and Hill, R.C. 2008. Qualitative and limited dependent variable models.

Using stata for principles of econometrics. Hoboken: John Wiley Sons, Inc.

Banerjee, S.,and Dasgupta, S., and Kim, Y. 2004. Buyer-Supplier relationship and trade credit. Working paper. Department of Finance, Hong Kong University of science and Technology

Biais, B., and Gollier, C. 1997. Trade credit and credit rationing. The Review of Financial Studies, 10(4): 903-937

Brady, H.E, and Johnston, R. 2008. The rolling cross-section and causal attribution.

Electoral Studies, 21 (2): 283-295

Cheng, N.S., and Pike, R. 2003. The trade credit decision: evidence of UK firms.

Managerial and decision economics, 4(6/7): 419-438

Dayton, K.N. 1984. Corporate governance: the other side of the coin. Harvard Business review, 62, 34-37

Jensen, M.C., and Murphy, K. J. 1990a. CEO incentives - It’s not how much you pay, but how. Journal of Applied Corporate Finance, 3649.

Hadjimanolis, A. 2000. A resource-based view of innovativeness in small firms.

Technology analysis and strategic management 12(2): 263-281

Hill, R.C, Griffiths, W.E, Lim, G.C. 2008. Heteroskedasticity. Principles of Economics :200. Hoboken: John Wiley Sons, Inc.

Lee, A.L, Niemeier, D. 1996. Advantages and disadvantages: longitudinal vs. repeated

cross section surveys. Project Battelle 94 (16), FHWA, HPM-40

27 Madan, D., and Unal, H. 2000. A two-factor hazard rate model for pricing risky debt and the term structure of credit spreads. Journal of financial and qualitative analysis, 35(1): 43-65

Peterson, M.A.,and Rajan, R. G. 1997. Trade credit: theories and evidence. The Review of Financial Studies, 10 (3): 661-691

Smith, J.K. 1987. Trade credit and informational asymmetry. The Journal of Finance, 42(4): 863-872

Stiglitz, J.E., and Weiss, A. 1981. Credit rationing in markets with imperfect information. The American economic review, 71(3): 393-410

Summers, B., and Wilson, N. 2003. Trade credit and customer relationship. Managerial and Decision Economics, 24(6/7): 439-455

Thomas, L.C., and Oliver, R.W., and Hand, D.J. 2005. A survey of the issues in consumer credit modelling research. The journal of the operational research society, 56(9): 1006-1015

Wilner, B.S. 2000. The exploitation of relationships in financial distress: the case of trade credit. The Journal of Finance, 55(1): 153-178

http://www.bkr.nl 29-04-2009

http://www.cbs.nl/nl-NL/menu/themas/handelhoreca/publicaties/artikelen/archief/ 2009/

2009-042-pb.htm 29-05-2009

28 APPENDIX A

Table 1 Variable description

Variable Abbreviation Description Expected

sign

Source

Trade credit position of the

customer

Y 0= customer i, no longer receives trade credit (default occurred) 1= customer i, is receiving trade credit

Inapplicable Heineken NL

Refinancing, dummy

DUMREF 0= no refinancing in the past

1= customer received a refinancing in the past

Positive Heineken NL

BKR testing, dummy

DUMBKR 0= BKR testing outcome is positive (good) 1= BKR testing outcome is negative (bad)

Positive Heineken NL

More than 10%

own money input, dummy

DUMOWN 0= own money input more than 10% of total investment 1= own money input less than 10% of total investment

Positive Heineken NL

Payment behaviour

DUMPAY 0= payment behaviour of customer is known 1= payment behaviour of customer is unknown

Positive Heineken NL

Entrepreneurial experience,

dummy

DUMENTEXP 0= entrepreneurial experience is more than 2 years 1= entrepreneurial experience is less than 2 years

Positive Heineken NL

Growth of sales SALESGROWTH Total sales of the customers business is measured during the trade credit period as growth of the sales during this period.

Sales growth is measured on a quarterly basis

Negative Heineken NL

29 APPENDIX B

Table 2 Summary of the independent variables

Variable Obs.= 0 Obs.=1 Total Obs.

DUMREF 338

117

455

DUMBKR 351

104

455

DUMOWN 233

222

455

DUMPAY 307

148

455

DUMENTEXP 129

326

455

SALESGROWTH

Mean -7.42

Std.Dev.

89.47 455

Note: the percentage in brackets is the percentage of the number of observation for a variable with observations=0

and observations=1 to the total number of observations. For example DUMREF 74.29% = 338/455

30 APPENDIX C

Graph 1 The ROC curve.

31 APPENDIX D

Table 4 The independent variables tested with the binary dependent logit model.

This table shows the results for the test of the single independent variables against the dependent variable. The independent variables that where significant in this first test, dumbkr, dumref, and sales growth, are subjected to a second test. In this second test, I grouped the significant variables in pairs to see what happens to their significance. Sales growth is the only variable that remained significant in both test. The results in red are the results from the second test.

Independent Variable:

DUMBKR DUMPAY DUMREF DUMOWN DUMENT-

EXP

Sales growth .605*

(.235)

.320 (.216)

.261 (.231)

-.323 (.212)

-.484*

(.235)

-.026*

(.003) .420

(.282)

-.026*

(.003) Dependent

Variable: Y

-.256 (.275)

-.026*

(.003)

Note: standard errors in brackets. *sign at 95% CI

32 APPENDIX E

Graph 2a Sales growth versus BKR=0

Y-as represents the probability of default

x-as represents the actual values of the growth of sales. -200 means that the growth in sales on average for a customer is negative (decreasing sales growth of -200 average)

Graph 2b Sales growth versus BKR=1

Y-as represents the probability of default

x-as represents the actual values of the growth of sales. -200 means that the growth in

sales on average for a customer is negative (decreasing sales growth of -200 average)

33 APPENDIX F

Graph 3a The default vs non-default customer

Graph 3b The default vs non-default customer (%)

34 APPENDIX G

Heineken NL, the beer market and the customers.

Heineken NL:

Heineken NL owns the beer brands Heineken, Amstel, and Brand. The trade credit applications can be for a customer of one of these three brands. Heineken NL is part of the original Dutch conglomerate Heineken Corporate. The headquarters of this conglomerate is situated in Amsterdam. Heineken NL is positioned in Zoeterwoude.

Heineken produces beer in Zoeterwoude, Den Bosch and Wijlre. In Zoeterwoude and Den Bosch, the brands Heineken, Amstel and other varieties are produced. In Wijlre, the beer brand Brand is produced.

Heineken NL is divided into 9 regional units in the Netherlands.

Heineken supplies, due to these three brands, more than 40% of the Dutch beer market. Therefore, Heineken is not allowed to bind a customer to herself by means of a long-term contract. This rule is set by the ‘mededinging’ law.

The Dutch beer market:

Competitors of Heineken are other beer brewers. In the Netherlands, the largest competitors are Grolsch, Bavaria and Inbev (all three also lend trade credit to their customers). They compete with Heineken in the beer market but also in the market of the non-alcoholic beverages.

The customers of Heineken:

Heineken is positioned with all their products in both the retail market and the horeca

market. In this paper only the horeca customers are taken into account. Heineken is

their supplier of beer. Some of these horeca customers also receive trade credit from

Heineken. Only these customers are investigated here.

30