Advantages of Trade Credit

The firm must balance the advantages of trade credit against the cost of forgoing a cash discount,
the opportunity cost associated with possible deterio lbration in credit reputation if it stretches its
payables, and the possible increase in selling price the seller imposes on the buyer. There are
several advantages of trade credit as a form of short term financing. Probably the major
advantage is its ready availability. The accounts payable of most firms represent a continuous
form of credit. There is no need to arrange financing formally; it is already there. If the firm is
now taking cash discounts, additional credit is readily available by not paying existing accounts
payable until the end of the net period. There is no need to negotiate With the supplier; the
decision is entirely up to the firm. In stretching accounts payable, the firm will find it necessary,
after a certain degree of postponement, to negotiate with the supplier.

In most other types of short term financing, it is necessary to negotiate formally with the lender
over the terms of the loan. The lender may impose restrictions on the firm and seek a secured
position. Restrictions are possible with trade credit, but they are not nearly as likely. With other
sources of short term financing, there may be a lead time between the time the need for funds is
recognized and the time the firm is able to borrow them. Trade credit is a more flexible means of
financing. The firm does not have to sign a note, pledge collateral, or adhere to a strict payment
schedule on the note. A supplier views an occasional delinquent payment with a far less critical
eye than does a banker or other lender.

The advantages of using trade credit must be weighted against the cost. As we have seen, the
cost may be very high when all factors are considered, Many firms utilize other sources of short
term financing in order to be able to take advantage of cash discounts. The savings in cost over
other forms of short term financing, however, must offset the flexibility and convenience of trade
credit. For certain firms there are no alternative sources of short term credit.

Advantages of Trade credit

· Reduced capital requirements, this means that if a new business setting up has trade credit,
they will obviously require less money in capital to start up the business. This is a major
advantage to someone who has very little money but has a good idea about starting a new
business.
· Trade credit with improve the cash flows and therefore provide smoother operation for the
business
· Businesses can buy now and pay later which means even if they don't have the money at first
they can purchase items, sell them as a business and then make the payments at the end of the
month when the products have been sold and a profit has been made.
· Businesses can look to grow without having to worry of needing to make immediate payments
which may set them back.
· With trade credit, the business can focus on other areas such as sales, marketing and
research rather than worrying about meeting targets just to have enough money to pay the bills.
Disadvantages of Trade credit
· If repayments are not made by certain deadlines, the business will receive a poor credit history
which will be a big blow to any business as they will not trusted in the future if they require any
loans, trade credit, credit cards or leasing.
· Only companies with a good credit history will get trade credit and these can often be hard to
build up, especially for new businesses.
Optimal trade credit limits.

Trade credit is granted by one business to another in connection with the sale of goods or services. A trade
credit limit is a policy instrument that applies either a maximum dollar receivables balance or a maximum
order amount to a specific buyer. A seller can assign such a limit to any or all buyers. When the receivables or
the order amount exceeds the credit limit, it signals the seller to take action. As a mechanism for policy
implementation, one advantage of a credit limit is its simplicity. Once the credit limit is set, all that is required
is a comparison of the limit with the current receivables balance or the amount of the order.

Trade credit limits are the most common tool in credit management. Over 85% of large firms use this tool,
and they typically assign credit limits to more than 80% of their customers (Beranek and Scherr, 1991, and
Besley and Osteryoung, 1985). While such a widely-used technique must have practical value, practitioners
and researchers have little understanding of how it can enhance shareholder wealth.

By far the most popular method of setting credit limits is the analyst's judgment (Besley and Osteryoung,
1985). This may or may not lead to wealth maximization, depending on the analyst's skills and biases.
Further, while practitioners almost always cite "risk control" as the primary motivation for using credit limits
(Beranek and Scherr, 1991, and Besley and Osteryoung, 1985), there is little agreement among them as to
what type of risk is controlled, although there appears to be a strong connection between credit limits policy
and credit investigation expenses (Beranek and Scherr, 1991).

In this paper, I present a methodology for setting and using trade credit limits to enhance shareholder
wealth. The following section uses a single-period framework to show how credit limits can be set when the
seller is risk-neutral and the probability of the buyer's default does not increase with the amount of credit
granted. Under these assumptions, credit limits can act as triggers for expenditures on credit investigation. I
call these "information credit limits." In Section II, I relax the assumptions of risk neutrality and invariant
default probability and develop procedures for setting both information credit limits and "risk credit limits."
Risk credit limits address changes in risk, default probability, and other factors that are affected by the
amount of credit granted. (My discussions with trade credit practitioners indicate that they generally use the
term "credit limit" to refer to any dollar value of a particular customer's orders or receivables that is used to
trigger action, regardless of the type of action required. Consequently, though information and risk credit
limits are quite different in purpose and application, I call them both "credit limits.") Section III explores the
effects of administration costs, seasonality, future orders, and credit terms on information and risk credit
limits and then examines the relationship between trade credit limits and the credit limits used in
conjunction with consumer credit-card debt. Section IV summarizes and concludes the paper.

I. Credit Limits Under Risk Neutrality and Invariate Default Probability

Consider the trade credit-granting decision under the following assumptions:

1. The seller is risk-neutral.

2. The probability of the buyer's default or delinquency and the timing of the payment if the buyer does not
default do not vary with the amount of credit granted.

3. The seller produces goods for sale at a constant cost per unit and sells these at a constant price regardless
of order size. The seller faces no constraints on production capacity.

4. The seller can accept or reject orders without any effect on other cash flows (including those from future
orders from the same buyer or orders from other buyers).

5. The seller can obtain information about the buyer by expending fixed costs associated with various levels
of credit investigation. Information can come from any source, including the customer, credit-reporting
agencies, financial databases, etc. This information is gathered in a sequential fashion, starting with the least
expensive source of information and proceeding to more expensive sources. (This depiction of credit
investigation as a sequential process follows Mehta, 1968, and Stowe, 1985.)

In this paper, I view credit decisions in the context of present value. This framework for evaluating those
decisions was originally proposed by Hill and Riener (1979), Kim and Atkins (1978), Lieber and Orgler (1975),
and Sartoris and Hill (1981 and 1983) and has since been used by many other authors. The applicability of
these techniques and concepts to real-world sellers depends on whether the organizational structure of
sellers' credit departments allows decisions to be made on a present-value basis. For various reasons, in
some organizations the credit department's sole task (subject to certain organizational constraints) is
minimizing credit-related costs. In such cases, a present-value approach that considers both costs and
revenues is inappropriate.

With regard to risk-neutrality, I initially assume that the selling firm is indifferent to the variability of
outcomes. If this is true, the expected net present value (NPV) of the cash flows from granting credit at the
risk-free rate captures wealth effects. The risk-free rate is the appropriate discount rate since, when the seller
is risk-neutral, the discount rate need only reflect the time value of money.

Note that this assumption ignores the effects of credit granting on the seller's systematic risk. Ignoring
systematic risk effects is justified if this type of risk is irrelevant to the credit-granting decision or if the
correlation between default probability and the return on the market is sufficiently low that the effects on
required return are unimportant. (For an exploration of the relationship between credit granting and
systematic risk, see Copeland and Khoury, 1980.)

I also assume that the firm faces no production capacity constraints. Capacity constraints present the same
problem as when the seller's total investment in accounts receivable is limited. When a seller faces capacity
constraints, a portfolio approach to the credit decision is required. The seller chooses the best portfolio of
credit risks among all potential buyers. (For discussion and development of a model for credit granting under
capacity constraints, see Besley and Osteryoung, 1984.)

Under the prior assumptions, because cash flows from future orders are not affected by the credit decision,
the trade credit-granting decision has a one-period (one-order) horizon. This is consistent with much other
research on credit-granting decisions; see, for example, Beranek and Taylor (1976), Scherr (1992), and Stowe
(1985). Note also that for a given set of credit information, the size of the order is irrelevant to the credit-
granting decision. To see this, assume that:

1. There are no taxes.

2. There are no recoveries in default.

3. There are no collection costs.

4. There are only two possible outcomes in credit granting: payment or default.

5. Productions costs are paid immediately.

6. No interest charges or penalties are imposed if the buyer pays beyond the due date.

Let X be the order size in dollars, V be the production cost per dollar sold, P be the probability of payment, t
be the expected time to payment in days, and r be the yearly risk-free rate. Then, using a 360-day year:

E(NPV) = -XV + XP/[(1 + r).sup.t/360] (1)

Applying the risk-neutral acceptance rule of E(NPV) [greater than] 0, then grant credit if:

0 [less than] -XV + XP/[(1 + r).sup.t/360] (2)

and dividing by X:

0 [less than] -V + P/[(1 + r).sup.t/360] (3)

Thus, the decision to grant credit is invariant to X as long as the remaining variables are not functions of X.
For a given set of credit information about the buyer, order size is irrelevant if cost and price per unit do not
vary in X. (Under these assumptions, credit information consists of estimates of P and t, which are the buyer-
related parameters in Equation 3.)

This does not, however, mean that the credit-granting decision can be made regardless of order size because
the effects of credit investigation expenses have not yet been addressed. A given set of information is
assumed, but the seller's estimates of P and t will vary with the amount of credit investigation undertaken.

Mehta (1968) shows that in the presence of the fixed costs of different types of credit information, decisions
must be made simultaneously on the depth of credit investigation to be undertaken and to whom credit will
be granted. Using a decision-tree approach to the problem, he develops rules for credit investigation that
result in wealth-maximizing decisions based on the tradeoff between information costs and the other
revenues and costs associated with credit granting. Stowe (1985) shows that Mehta's procedures can be
formulated and solved as an integer program. This approach has an added advantage in that resource
constraints can be incorporated in the algorithm.
These two approaches allow the analyst to formulate optimal credit investigation strategies for a given order
size. But order sizes from a buyer can, and often do, grow over time. From a policy standpoint, credit
managers are concerned not only with the optimal strategy for a particular order size but also with strategies
for other order sizes. It is here that credit limits can be employed advantageously. If the credit investigation
problem is formulated as a linear integer program, it can be solved to determine the particular order sizes for
which additional credit investigation is optimal.(1) I call these switch-points information credit limits.

Using information credit limits, credit investigation expenses are triggered by increases in order size. Prior
credit investigation provides estimates of a customer's parameters, and the integer program's solutions
reveal the next level at which further investigation should be performed. Though the mathematical
programming processes are complex, the resulting policy is simple to understand and implement.

For example, retain all the prior assumptions, and also assume:

1. There are fixed costs associated with the various types of credit investigation.

2. There are only two stages of credit investigation: checking previous payment experience and ordering a
credit report. The first costs $5 and the second, $50. Both costs are payable immediately.

3. There are three possible results of checking prior experience: good, poor, or none.

4. There are two possible credit reports on financial condition: strong and weak.

5. Costs of production are 80% of sales.

6. Order amounts are in increments of $100.

Given these assumptions, there are six types of customers (three types of past experience times two levels of
financial strength for each experience type). There are therefore six sets of customer characteristics, each
consisting of an estimated payment time and payment probability. Table 1 shows the assumed values of
these characteristics and population probabilities of each buyer type. In this table, [f.sub.ij] is the joint
probability of a credit applicant with past payment experience i and rating j; [t.sub.ij] is the expected time to
payment if payment is made; and [P.sub.ij] is the probability of payment. Let X be the order amount and r the
risk-free rate. Applying Stowe's mathematics results in the integer programming formulation for this credit
investigation and credit-granting problem illustrated in Table 2 for X = $500. (Notation in this table and
throughout the paper follows Stowe, 1985.)

A brief review of Stowe's formulation may aid those who are not familiar with this methodology. The costs
and revenues from credit granting, the probabilities of various outcomes, and the credit investigation costs
are captured in the objective function.

The rules relating credit investigation and expected NPV are captured in the constraint equations. The
objective function includes separate variables for the credit investigation costs and the probability-weighted
expected NPVs of granting credit in various information situations. The coefficient of IMMGNT is the
expected NPV of granting credit without any investigation. The coefficients of PEGGNT, PEPGNT, and PENGNT
are the expected NPVs of granting credit to buyers with good, poor, and no prior experience, weighted by the
probabilities of these payment experiences. Note that the sum of the coefficients of PEGGNT, PEPGNT, and
PENGNT is the coefficient of IMMGNT. Similarly, the coefficients of the variables for the full-investigation
conditions (GSGNT, GWGNT, NSGNT, NWGNT, PSGNT, and PWGNT) are the expected NPVs of granting credit
for these values of [P.sub.ij] and [t.sub.ij], weighted by the probability of the buyer type's occurrence. Note
that the coefficient of GSGNT plus the coefficient of GWGNT equals the coefficient of PEGGNT (and similarly
for PENGNT and PEPGNT).

The credit investigation expenses that are required to assess states of experience and rating are captured by
the INVPE, PEGCR, PENCR, and PEPCR variables in the objective function. The coefficient of INVPE is the cost
of checking credit experience. The coefficients of PEGCR, PEPCR, and PENCR are the costs of obtaining a
credit report, weighted by the probability that past experience will be good, poor, or none. Note that in the
objective function no variables represent the NPV of refusing credit. The credit investigation costs that are
lost if credit is denied after investigation are captured by the INVPE, PEGCR, PENCR, and PEPCR variables.

Table 1. Parameter Values for Example Problem(a)

This table presents parameter values for a two-stage credit

investigation problem. The two stages are: 1) checking past payment
experience and 2) obtaining a credit report. There are three
possible outcomes of checking past payment experience: good, poor,
or no prior experience. There are two possible credit reports:
strong or weak financial condition.

Panel A. Joint Probabilities of Past Payment Experience and

Rating ([f.sub.ij]) and Marginal Probabilities of Past Experience
([f.sub.i])

Joint Probs. Marginal Probs.

Credit Rating

Strong Weak

Past Experience:

Good 0.60 0.15 0.75

Poor 0.01 0.04 0.05
No Experience 0.14 0.06 0.20

Panel B. Joint Collection Periods in Days ([t.sub.ij])

Credit Rating

Strong Weak

Past Experience:

Good 30 45
Poor 60 120
No Experience 30 90
Panel C. Joint Payment Probabilities ([P.sub.ij])

Credit Rating

Strong Weak

Past Experience:

Good 1.00 0.99

Poor 1.00 0.70
No Experience 1.00 0.80

a The following conditions exist: yearly risk-free rate = 10%; order

amounts (X) are in increments of $100; production costs (payable
immediately) = 0.80X; cost of checking prior experience (payable
immediately) = $5; cost of a credit report (payable immediately) =
$50.

Solving the integer program for various values of X gives the information credit limits. Solutions to the
example problem appear in Table 3. For this example, an order should be approved without any credit
investigation as long as the amount is less than or equal to $1,500 (since INVPE = 0 for these values of x).
When an order for less than or equal to this amount is received, the information credit limit should be set at
$1,500, and no credit investigation of the customer is required until order size exceeds this figure.(2)

An order above $1,500 requires further credit investigation. If the order is greater than $1,500 but less than
or equal to $8,800, the optimal strategy (see Table 3) is to:

1. Check prior payment experience (since INVPE = 1 for these values of X).

2. If prior credit experience is poor, obtain a credit rating (since PEPCR = 1).

3. If prior experience is poor and the credit rating is weak, reduce the amount of credit to be granted to zero
(since PWGNT = 0). (This class of buyer always has a negative expected value of credit granting; the
coefficient of PWGNT is negative for all values of X. For orders greater than $1,500, the revenues and costs of
credit granting are large enough relative to the investigation cost to warrant a separate policy decision for
these buyers.)

4. If past experience is poor but the credit rating is strong, grant unlimited credit. (This class of buyer has a
positive expected NPV of credit granting; PSGNT is positive for all X. Under the assumptions of the problem,
additional investigation is not possible, so the best decision is to grant unlimited credit to all buyers of this
type.)

5. Grant unlimited credit to all applicants with good prior payment experience. (The net present value of
granting credit to both strong and weak buyers with good payment experience is always positive; the
coefficients of both GSGNT and GWGNT are positive for all values of X, so it is never necessary to order a
credit rating to distinguish between them.)
6. For buyers with no prior experience, grant credit without ordering a credit rating and set the information
credit limit for these buyers to $8,800 (since PENCR is zero for volumes below this). Strong buyers with this
payment record have positive expected NPVs (the coefficient of PSGNT is always positive) and weak buyers
have negative NPVs (the coefficient of PWGNT is always negative). However, the proportions of these buyers
and the revenues and costs from granting them credit make it unprofitable to distinguish between them until
the amount owed is above $8,800.)

For amounts owed below $8,900, decisions resulting in either infinite (open) or zero credit are made for all
buyers with good or poor payment records. Only buyers with no prior payment experience have finite,
nonzero information credit limits. If the order volume reaches $8,900 for those buyers on whom the seller
has no payment experience, the optimal strategy (see Table 3) is to:

1. Order a credit rating (since PENCR = 1).

2. Reduce the credit limit for weak buyers to zero (since the coefficient of NWGNT is negative for all values of
X).

3. Grant unlimited credit to strong buyers (since the coefficient of NSGNT is positive of all values of X).

II. Credit Limits with Risk Aversion or Changing Default Probability

Earlier, I assumed that the seller was risk-neutral with respect to credit granting, that goods were produced
and sold at a constant dollar value per unit, and that the probability of the buyer's default did not increase
with the amount of credit granted. Based on these assumptions, the amount of the order did not affect the
decision to grant credit except as affected by the optimal amount of credit investigation. However, if these
assumptions are relaxed, the credit-granting decision can depend on the amount of the order.

If Equation (1) is positive for small values of X but attains a unique optimum in X, additional order volume
from the buyer beyond the optimum is not desirable. Several circumstances can result in such an optimum,
including increasing costs per unit (v), decreasing price per unit (perhaps from demand curve effects),
decreasing probability of payment (P), increasing time to pay (t), or increases in the seller's risk aversion with
increases in receivable size.(3,4) In this section, I discuss the effects of variations in V, P, t, and r on credit
limits policy.(5)

Table 3. Integer Programming Solutions to Example Problem for

Various Values of Amount Owing (x)

This table presents values of variables from the integer programming

solutions to the example problem for various order sizes (x). These
values are found by calculating the coefficients of the variables
using the formulas in Table 2, Panel A, for various values of X,
then solving the resulting integer programming formulations.

Value of X

Less than $1,600 to

Variable $1,600 $8,800 Over $8,800
IMMGNT 1 0 0
INVPE 0 1 1
PEGGNT 0 1 1
PEGCR 0 0 0
PENGNT 0 1 0
PENCR 0 0 1
PEPGNT 0 0 0
PEPCR 0 1 1
GSGNT 0 0 0
GWGNT 0 0 0
NSGNT 0 0 1
NWGNT 0 0 0
PSGNT 0 1 1
PWGNT 0 0 0
IMMDNY 0 0 0
PEGDNY 0 0 0
PENDNY 0 0 0
PEPDNY 0 0 0

One effect of increases in V, t, or r or of decreases in P with X is a reduction in those coefficients of the

variables in the objective function that capture the probability-weighted net present value of credit granting.
As V, t, or r increase or P decreases in X, the values of these coefficients are reduced relative to when V, P, t,
or r are invariant in X. The reduction in these coefficients results in lower information credit limits. It
produces different patterns of optimal credit investigation, particularly for values of X large enough that the
coefficients of one or more variables change sign from positive to negative. This change indicates that the
expected NPV of credit granting is negative for values of X greater than that resulting in the sign change.

In addition to triggering credit investigation, when V, t, or r increase with X or P decreases in X, rules based on
order size can be used to limit the size of the order to its optimal level. While information credit limits
address credit investigation expenses, risk credit limits are used to optimize the expected NPV of the order,
given the relationship between X and E(NPV).

Note that in this model, optimum order sizes caused by increases in r and decreases in P with X represent risk
or the seller's response to it; optimums that occur because of increases in V or t are not really risk-based. I
use the term "risk credit limits" to refer to all such optimums because I suspect that in practice most
optimums in E(NPV) are caused by risk, and the term "Optimum E(NPV) Credit Limits" and other alternatives
seem cumbersome.

Risk credit limits are obtained by finding the maximum E(NPV) in X (the maximum value of Equation 3), given
the available information about the buyer. The risk credit limit can be greater or less than the information
credit limit for the same buyer type. By enforcing risk credit limits (for example, by limiting order size), the
seller can manage orders to maximize their net present values. Note that because setting risk credit limits
requires estimates of the parameters of Equation (3), which in turn depend on the amount of credit
investigation undertaken, risk credit limits and information credit limits are not independent.
As an example illustrating both types of limits, assume that for some classes of buyer, the probability of
payment decreases with the amount of credit granted according to the logit function and that the probability
of payment is described by:(6)

Probability of Payment = [P.sub.ij] - [P.sub.ij]/(1 +[e.sup.a+bX]) (4)

where [P.sub.ij] is now interpreted as the probability of payment for an infinitesimally small amount of credit.
(7) The rate of decrease in payment probability with amount of credit granted depends on the parameters a
and b, which can vary among the types of buyer. In addition to the prior assumptions of the numerical
example, assume that the probability of payment declines for all buyers with weak financial positions and for
buyers with poor prior payment experience but strong financial positions, according to the following
conditions:

Financial Prior Payment

Condition Experience [P.sub.ij] a b

Weak Good 0.99 5 -0.0001

Weak No Experience 0.80 5 -0.0002
Weak Poor 0.70 5 -0.0003
Strong Poor 1.00 5 -0.0001

Figure 1 gives plots of probability of payment versus amount of credit granted for the first three conditions.

Given these data, calculating risk credit limits and information credit limits is straightforward. For information
credit limits, the probability of payment for each type of buyer is calculated using Equation (4) for each level
of X. These probabilities are then used in calculating the coefficients of the credit-granting variables in the
objective function, and the integer program is iterated as before to obtain the switch-points. For risk credit
limits, the order sizes resulting [TABULAR DATA FOR TABLE 4 OMITTED! in the maximum NPVs are then
calculated, given the investigation undertaken (and therefore the available estimates of payment probability
and time to pay).(8)

A. Information Credit Limits

The switch-points in the solutions to the integer program for the revised example problem appear in Table 4.
There are several differences in these solutions relative to the prior results. Increases in the probability of
default reduce the coefficients of all variables except GSGNT and NSGNT. (By assumption, the probabilities of
default do not increase in X for buyers of these types.) This results in lower values for the first two switch-
points. Specifically, the order value above which it is necessary to order credit reports on buyers with poor
prior experience declines to $1,400 (from $1,500). The value above which it is necessary to order credit
reports on buyers with no prior payment experience declines to $5,000 (from $8,800).

In addition to lowering these two investigation strategy switch-points, two other information credit limits are
introduced. The first of these occurs at $34,800. For orders of this size, the coefficient of PSGNT, which is
positive for smaller values of X, becomes negative. The value of PWGNT is always negative, so the value of
PEPGNT becomes negative. Granting credit of this amount to any buyer with poor prior experience has a
negative expected value. Consequently, for orders of this size or larger, credit should be denied to all buyers
with poor prior experience without ordering a credit report. Distinguishing between weak and strong buyers
who have poor prior payment experience is not necessary.

The second new switch-point occurs at $35,400. For orders of this size, the coefficient of GWGNT, which is
positive for smaller values of X, turns negative, and it becomes advantageous to order a credit report on
buyers with good prior payment experience. Those with weak financials will be denied credit.

A summary of the rules for information credit limits with changing default probabilities for the example
problem is presented in Table 5. Perhaps of greatest interest in this table is the set of rules for order sizes
greater than $1,400 but less than $5,100. For these order sizes, different information credit limits are set for
buyers with different prior payment experiences.

B. Risk Credit Limits

If the order amount is less than $1,500, no investigation is optimal. The seller's estimate of the expected NPV
for any level of credit granted is the sum of the expected NPVs for each type of buyer at that level of credit
weighted by the probability of occurrence of that buyer type. When no credit investigation is undertaken, the
maximum expected NPV occurs at an X of about $26,000 [ILLUSTRATION FOR FIGURE 2 OMITTED]. This is the
risk credit limit for all buyers whose order size is less than $1,500.

Values for risk limits for other order sizes and investigation conditions are determined similarly and are
shown in Table 6. Each is based on the expected present values of granting various levels of credit, given the
remaining possible outcomes of the credit investigation process. For example, for order sizes between $1,500
and $5,000, if checking prior credit experience shows good prior payments, ordering a credit report is not
necessary to distinguish between the two remaining outcomes: buyers who have good payment records and
either strong or weak financials. The expected NPV of granting any level of credit is the probability-weighted
average of the expected NPVs for these two buyer types. This average reaches a maximum at $38,000, which
is the risk credit limit for these buyers.

In Table 6, as receivables size increases, if both limits are finite and nonzero, the lower of the two limits
determines the action to be taken. If the information limit is lower than the risk limit, further investigation is
triggered before the risk limit is reached, and the risk limit is not binding. If the risk limit is less than the
information limit, the buyer's order size is restricted to the risk limit until order sizes are large enough to
trigger additional credit investigation, in which case the risk and information credit limits are revised based
on this investigation.

III. Extensions of the Model

The scope of the previously presented credit-limits model is limited by the assumptions under which the
model is derived. This section discusses the effects of altering some of these assumptions.

[TABULAR DATA FOR TABLE 6 OMITTED]

A. Credit Limits and Administration Costs

The credit-granting model presented earlier was developed under the assumption that the only cash outflows
associated with this decision were the production costs per unit sold (V) and the costs of credit investigation.
However, the seller can also incur fixed costs in setting up and administering the relationship with the buyer
(for example, mailing marketing materials, price changes notices, and other such expenses). It is partly as a
reflection of these fixed administration costs that sellers offer quantity discounts and other volume-based
incentives for larger purchases. Starting with the model in Equation (1), let A be the fixed costs of
administering the account. Then:

E(NPV) = -A - XV + XP/[(1 + r).sup.t/360] (5)

and the credit-granting rule is:

0 [less than] - A - XV + XP/[(1 + r).sup.t/360] (6)

Notice that in this situation, the decision to grant credit is no longer invariant in X. Dividing Equation (6) by X
results in:

0 [less than] - A/X - V + P/[(1 + r.sup.)t/360] (7)

Since -A/X decreases in X (amortizing the costs of fixed administration costs over more units sold), higher
sales volumes increase per-unit profitability. Selling to some buyers will be unprofitable for lower sales
volumes but will be profitable for higher ones.

The effect on credit limits of incorporating administrative costs depends on the relationship between the
administrative and production costs of the order (V). There are two possible cases. One is that the
administrative cost term captures an additional cost not included in the original model. The second is that the
administrative cost represents a fixed component of V.

When the administrative cost represents an additional cost, the expected present value from granting credit
for any level of sales is reduced by the amount of the administrative cost. Any buyer is less likely to represent
an advantageous sale. Credit investigation at lower sales volumes is required to separate positive-NPV from
negative-NPV buyers. Consequently, information credit limits are lower.

As an illustration, consider the original numerical example to which we add an administrative cost of $25.
With this cost, the minimum order size necessary to check prior payment experience is reduced to $1,300
from $1,600, and the minimum order size necessary to require a credit report on buyers with no past
experience is reduced to $7,600 from $8,900. (Calculations are available from the author.) However, risk
credit limits are unaffected by such administrative costs because the deduction of a fixed amount from
expected NPV does not change the point at which the expected NPV reaches its maximum.

When the administrative cost captures a fixed component of order cost, the addition of A to the model
breaks the total cost of the order into fixed and variable components. When a portion of the total cost is
fixed, lower sales volumes have higher total costs than when all costs are variable, while higher sales volumes
have lower costs. For order sizes below breakeven, information credit limits are lower; for order sizes above
breakeven, higher.

In the original information credit limits example, V = 0.80 and A = 0. Let A = $200 and V = 0.76. This structure
produces the same total cost of $4,000 for an order size of $5,000; total costs are higher below this order size
and lower above it. For these parameters, the minimum order size necessary to require a check of prior
payment experience is reduced to $1,000 from $1,600, the result of this particular information credit limit
being below the breakeven point. However, for V = 0.76 and A = $200, there is no order size large enough to
warrant ordering a credit report on buyers with no past payment experience (as required at X = $8,900 for A
= 0 and V = 0.80). This is so because for A = $200 and V = 0.76, buyers with no prior payment experience have
positive expected NPVs at reasonably large order sizes whether their credit ratings are strong or weak.
Consequently, it is not necessary to distinguish between them.

When the total cost of the order is decomposed into fixed and variable costs, all risk credit limits increase.
These limits are not affected by the deduction of the fixed cost from expected NPV, but the reduction in
variable cost causes the maximum in expected NPV to climb.

B. Credit Limits and Seasonality

When there is seasonality in demand for the goods or services, there are times during the year when buyers
come to the seller with larger order sizes. Using the model presented earlier, this means that on average
more credit investigation is performed during these times, since the larger order sizes trigger additional levels
of credit investigation. Credit limits are set based on these higher levels of investigation. There are more
buyers for whom credit ratings must be ordered, more buyers who have no information credit limits (a
complete credit investigation having been performed), and more buyers who have either open or zero risk
credit limits.

Seasonality also presents an interesting opportunity to consider the effects of inventory costs on credit
policy. Emery (1987) applied peak-load pricing theory to this problem and found that there are some
circumstances in which the seller should modify credit policy to address seasonality or uncertainty in
demand. In slack periods, Emery argued, under some conditions, a firm should loosen credit policy. It should
incur additional credit-related costs but reduce the inventory-carrying costs of producing goods during slack
periods and storing them until peak periods.

In the model, all the costs of goods or services, including inventory carrying costs, are captured by V. When
the firm sells from inventory and produces goods to replenish this inventory, it incurs inventory-carrying
costs resulting from holding these goods until the next sale. During peak periods, the time between sales is
less than during slack periods. Therefore, during slack periods there is a relatively greater opportunity to
reduce inventory-carrying costs, resulting in a reduced V. Lower V leads to higher information and risk credit
limits during these periods; optimal credit policy is looser.

C. Credit Limits and Future Orders

The model developed here is for a single order. An important extension concerns the effect of other orders
on the strategies produced by the model. Consider two cases:

1. The buyer places an additional order before payment for the initial order has been made.

2. The expected NPV of the initial order is negative (thus credit is not granted), but the buyer may place
future orders, perhaps for larger dollar amounts.

The first case requires no important modifications to the model. As long as the initial order is not past due, in
which case payment for the initial order may be required before the additional order is approved, the
additional order offers the same conditions to the seller as the initial one. This case does not differ from the
one in which the initial order is for a larger amount. The combined amounts of the two orders determine
credit investigation and credit granting.

The second case is more complicated. Fewings (1992) shows that credit decisions can be reduced to the
single current order even if additional orders are placed by the same buyer in any amount, as long as either
1) the estimates of default risk, time to pay, etc., are not expected to change over time or 2) these estimates
are expected to deteriorate over time.

However, when estimates of default risk and payment time are expected to improve over time (creating a
situation in which the current order has a negative expected NPV but future orders have positive expected
NPVs), future orders are relevant to the initial credit decision. Unfortunately, attempts to model such credit
decisions are mathematically cumbersome, limited by their assumptions about the evolution of default
probabilities and payment times, and may not lead to a closed-form solution (see, for example, Bierman and
Hausman, 1970, and Dirickx and Wakeman, 1976). This case is best regarded as a limitation to this and all
other single-order credit-decision approaches.

D. Credit Limits and Credit Terms

A trade buyer's information and risk credit limits are not the only controls that sellers can impose. When the
buyer's account is past due, the seller can require that the past-due items be paid for before additional
purchases are made, even if credit limits are not violated. (This is called "trading orders.") This process
accelerates payments to the seller, but it requires that buyers be monitored not only on their credit limits but
also on the due dates of prior purchases.

One way to enforce terms is to reduce the buyer's risk credit limit such that violation means that the buyer is
past due. For example, suppose that terms are net 30 days, that the buyer is expected to purchase $10,000
per month, and that the buyer's risk credit limit is $15,000. If this limit is reduced to $10,000, its violation
means that the customer is also past due, since more than one month's purchases are outstanding.

While this approach reduces the factors that have to be monitored, it has at least one disadvantage:
Purchases beyond anticipated levels trigger action even if these purchases are caused by advantageous
events, such as shifts in market share or the growth of the buyer's own sales volume. When risk credit limits
monitor past-due status, the concept of the maximum amount of credit to be granted is masked. Action can
be triggered when no action is required.

E. Credit Limits and Consumer Credit-Card Debt

While this article primarily concerns trade credit, it is useful to contrast this with consumer credit-card debt.
Consumer credit-card debt and trade credit differ in a number of important ways. First, the dollar amounts of
individual purchases and the total dollar amounts owing are typically much smaller for consumer debt. This
means that much less credit investigation is advantageous. Also, in the U.S., the kinds of information that can
be collected are limited by consumer protection laws, such as the Fair Credit Reporting Act and the Equal
Credit Opportunity Act.

A second difference concerns the treatment of balances beyond the due date. For credit-card debt, these
balances accrue interest at profitable rates. Consequently, granting credit to a consumer who pays after the
due date is at least as profitable as granting credit to a customer who pays promptly. Therefore, unlike trade
credit, analysis of credit granting is concerned almost entirely with a single aspect of the customer, the
probability of default (rather than the probability of default and the time to pay).

These differences result in a considerably lower level of investigation for consumer credit than for trade
credit. Typically, a consumer credit application is evaluated and an initial credit limit assigned using an
inexpensive credit-scoring system.(9) Updating the credit limit is based on payment experience, the cheapest
source of credit information. This limit is a risk credit limit in that if the consumer violates it without
contacting the credit-card firm first, credit may not be approved. However, the limit is also informational in
that credit-card firms are generally willing to review it, based on payment experience, if the customer so
requests.

IV. Summary

Credit managers must make decisions in the management of accounts receivable that maximize shareholder
wealth. Practicing managers find that credit limits are useful tools in this process. Comparing the order
amount or the receivables balance to a credit limit triggers wealth-enhancing action.

This normative paper provides methods for determining credit limits to manage credit investigation expenses
and credit risk. It prescribes two sets of credit limits for these purposes, information credit limits and risk
credit limits. While both sets of limits are expressed in terms of order size, they perform very different
functions and are to be used by the seller in quite different ways. Information credit limits signal
management of the order sizes at which additional credit investigation expenditures are most advantageous.
Risk credit limits maximize the present values of sales in the presence of risk or cost factors that increase with
order size. Risk credit limits are imposed on buyers as maximum order sizes.

The procedure described here allows for revision of information about the buyer as payments are received
(in the spirit of Bierman and Hausman, 1970). For example, suppose that a new buyer places an order and
that this order is large enough to warrant a check of prior experience. This check will show "no experience"
and the remainder of the credit investigation process is based on buyers of this experience type. Assume that
credit is granted and that the receivable is paid or defaulted. If this buyer then places a second order and this
second order is large enough to warrant a check of prior experience, the payment experience for the first
order will be retrieved, and credit investigation will progress using rules for a different experience type.

Table 2. Formulation and Solution to Example Problem: Integer Programming Approach

This table presents the two-stage credit investigation problem described in Table 1 as an integer program.
Panel A defines the variables to the used in the program's formulation and shows how their coefficients are
calculated from the parameters in Table 1. Panel B presents the integer programming formulation for an
order size of $500. Panel C presents the program's solution for this order size.

Panel A. Variable Definitions and Coefficient Calculation(a)

IMMGNT = a variable equal to 1 when credit is granted without any investigation and zero otherwise.

[Mathematical Expression Omitted]

PEGGNT = a variable equal to 1 when prior payment experience is good and credit is granted and zero
otherwise.

[Mathematical Expression Omitted]

PEPGNT and PENGNT are similar for poor and no prior experiences.

GSGNT = a variable equal to 1 when credit is granted when prior experience is good and credit rating is strong
and zero otherwise.

Coef. of GSGNT = [f.sub.ij](-0.8X + [P.sub.ij]X/[(1 + r).sup.[t.sub.ij]/360])

GWGNT, NSGNT, NWGNT, PSGNT, and PWGNT are similar for other combinations of prior experience and
credit rating.

INVPE = a variable equal to 1 when past payment experience is investigated and zero otherwise.

Coef. of INVPE = cost of checking prior experience

PEGCR = a variable equal to I when a credit rating is ordered if prior experience is good and zero otherwise.

Coef. of PEGCR = (cost of rating)[f.sub.i]

PENCR and PEPCR are similar for other past experiences.

IMMDNY, PEGDNY, PENDNY, PEPDNY = variables equal to 1 when credit is denied for various information
conditions and zero otherwise.
Panel B. Integer Programming Formulation for X = $500

Objective Function:

Maximize: 82.362 IMMGNT + 70.997 PEGGNT + 12.881 PENGNT - 1.517 PEPGNT + 57.627 GSGNT + 13.371
GWGNT

+ 13.446 NSGNT - 0.565 NWGNT + 0.921 PSGNT - 2.438 PWGNT - 5.000 INVPE - 37.500 PEGCR

- 10.000 PENCR - 2.500 PEPCR

Constraint Equations:

IMMGNT + INVPE + IMMDNY = 1

INVPE - PEGGNT - PEGCR - PEGDNY = 0

INVPE - PENGNT - PENCR - PENDNY = 0

INVPE - PEPGNT - PEPCR - PEPDNY = 0

2 PEGCR - GSGNT - GWGNT [greater than or equal to] 0

2 PENCR - NSGNT - NWGNT [greater than or equal to] 0

2 PEPCR - PSGNT - PWGNT [greater than or equal to] 0

Panel C. Integer Programming Solution for X = $500

OBJECTIVE FUNCTION VALUE = 82.362

VARIABLE VALUES:

IMMGNT = 1.0

PEGGNT = 0.0

PENGNT = 0.0

PEPGNT = 0.0

GSGNT = 0.0

GWGNT = 0.0
NSGNT = 0.0

NWGNT = 0.0

PSGNT = 0.0

PWGNT = 0.0

INVPE = 0.0

PEGCR = 0.0

PENCR = 0.0

PEPCR = 0.0

IMMDNY = 0.0

PEGDNY = 0.0

PENDNY = 0.0

PEPDNY = 0.0

a See Stowe (1985, pp. 70-71).

Table 5. Rules for Setting Information Credit Limits for Example Problem When the Payment Probability
Decreases with the Amount Purchased

This table interprets the results in Table 4 as information credit limits.

X [less than] 1,500

1. Grant credit to all buyers without checking credit; set information credit limit to $1,400.

$1,500 [less than or equal to] X [less than] $5,100

1. Check prior payment experience.

2. If prior credit experience is poor, obtain a credit rating. If prior experience is poor and the credit rating is
weak, do not grant credit. If past experience is poor but the credit rating is strong, grant credit and set the
information credit limit to $34,800.

3. If prior experience is good, grant credit and set the information credit limit to $35,400.
4. For firms with no prior payment experience, grant credit without ordering a credit rating and set the
information credit limit to $5,000.

$5,100 [less than or equal to] X [less than] $34,900

1. Check prior payment experience.

2. If prior credit experience is poor, obtain a credit rating. If prior experience is poor and the credit rating is
weak, do not grant credit. If past experience is poor but the credit rating is strong, grant credit and set the
information credit limit to $34,800.

3. If prior experience is good, grant credit and set the credit limit to $35,400.

4. For firms with no prior experience, obtain a credit rating. If financial position is strong, grant unlimited
credit. If financial position is weak, do not grant credit.

$34,900 [less than or equal to] X [less than] $35,500

1. Check prior payment experience.

2. If prior credit experience is poor, do not grant credit.

3. If prior experience is good, grant credit and set the credit limit to $35,400.

4. For firms with no prior experience, obtain a credit rating. If financial position is strong, grant unlimited
credit. If financial position is weak, do not grant credit.

$35,500 [less than or equal to] X

1. Check prior payment experience.

2. If prior credit experience is poor, do not grant credit.

3. If prior experience is good or none, order a credit report. If financial position is strong, grant unlimited
credit. If financial position is weak, do not grant credit.

1 Stowe also shows how to express the linear program as a set of simultaneous equations that can be solved
algebraically or graphically for the switch-points. However, in problems of practical size, this procedure may
be cumbersome.

2 Note that since order amounts are assumed to be in increments of $100, the next feasible order size above
$1,500 is $1,600, and order sizes of $1,600 require a different pattern of credit investigation and credit
decisions. The information credit limit is set at $1,500 because this is the maximum order size that does not
trigger action; this is the way a credit limit is interpreted by practitioners. A violation of a credit limit (in this
case, order sizes of $1,600 or greater) requires the credit manager to do something (in this case, perform
additional credit investigation).

3 For further discussion of these and other circumstances that can result in an optimum of E(NPV) in X, see
Besley and Osteryoung (1984), Chua (1995), Copeland and Khoury (1980), and Scherr (1992).

4 The prior discussion assumes that the seller is risk-neutral, thus cash flows are discounted at the risk-free
rate. When the seller is not risk-neutral, the discount rate reflects the risk associated with the cash flows
being discounted. One risk that can be priced this way is the effect on the seller's survival of a buyer's default.
As the size of a particular receivable grows relative to other receivables, a larger portion of the seller's cash
inflows come from that particular buyer, and default by the buyer increasingly affects the seller's ability to
survive (see Scherr, 1992). To reflect this risk, the seller can increase the required return as receivables size
increases, which can produce an optimum in E(NPV).

5 While I consider the effects of assumptions regarding these variables on the procedure for setting credit
limits, other prior assumptions, when relaxed, can also affect this process. Discussion of the effects of taxes,
collection costs, and similar other factors on the setting of information credit limits is available from the
author.

6 In the numerical example that follows, I allow P to vary rather than V, r, t, or some other parameter of the
sale that may cause E(NPV) to take on an optimum in X. While the analysis is similar regardless of the
parameter involved, survey evidence (Beranek and Scherr, 1991) indicates that about 30% of practitioners
believe that the probability of payment decreases with the amount of credit granted for some or all buyers.
Therefore, I center on this parameter in my example.

7 The logit function is employed in this illustration because of its simplicity and its mathematical properties: It
makes the probability of payment a decreasing function in X and bounds this probability between [P.sub.ij]
and zero. In practice, the function relating X and [P.sub.ij] is determined by the nature of the relationship that
induces the payment probability to decrease with order size.

8 Unfortunately, while the logit function provides a simple relationship between payment probability and
credit granted, substituting Equation (4) in Equation (3), differentiating with respect to X, and setting the
result equal to zero does not lead the a function that is easily solved for X. (Contact the author for details.)
The risk credit limits in the following example were obtained numerically.

9 Use of credit-scoring systems is not limited to consumer credit. These systems are useful whenever the
number of orders is large and the amount ordered by each customer is small, so that the costs of credit
analysis itself are important. When applied to trade credit, these scoring systems are typically "expert
systems," intended to replicate the judgment of a credit "expert," rather than systems based on present
value concepts or on statistical analysis of defaults and nondefaults. See Srinivasan and Kim (1987 and 1988)
for discussion of these systems.

References
Beranek, W. and F.C. Scherr, 1991, "On the Significance of Trade Credit Limits," Financial Practice and
Education (Fall/Winter), 39-44.

Beranek, W. and W. Taylor, 1976, "Credit-Scoring Models and the Cut-Off Point - A Simplification," Decision
Sciences (July), 394-404.

Besley, S. and J.S. Osteryoung, 1984, "Accounts Receivable Management: The Development of a General
Credit-Granting Algorithm for Determining Credit Limits under Funds Constraints," presented at the 1984
Financial Management Association Annual Meeting, held in Atlanta, Georgia.

Besley, S. and J.S. Osteryoung, 1985, "Survey of Current Practices in Establishing Trade-Credit Limits,"
Financial Review (February), 70-81.

Bierman, H., Jr. and W.H. Hausman, 1970, "The Credit-Granting Decision," Management Science (April), B-
591-B-632.

Chua, J., 1995, "A Trade Credit Decision Model with Order Size Dependent Credit Risks," University of Calgary
Working Paper.

Copeland, T.E. and N.T. Khoury, 1980, "A Theory of Credit Extensions with Default Risk and Systematic Risk,"
Engineering Economist (Fall), 35-52.

Dirickx, Y.M.I. and L. Wakeman, 1976, "An Extension of the Bierman-Hausman Model for Credit-Granting,"
Management Science (July), 1299-1237.

Emery, G.W., 1987, "An Optimal Financial Response to Variable Demand," Journal of Financial and
Quantitative Analysis (June), 209-225.

Fewings, D.R., 1992, "Trade Credit as a Markovian Decision Process with an Infinite Planning Horizon,"
Quarterly Journal of Business and Eco