Columbia University

Graduate School of Business

Managerial Economics (B6006)

Classes 1-5

Managerial decisions, cost-benefit analysis and economic costs

I. Introduction

Individuals make numerous decisions everyday. These decisions range from small, minor ones,
to decisions that have major effects on their lives. Managers frequently make business related
decisions. Typical examples of such decisions include the levels of production and employment,
various investment decisions, choices concerning the quality and price of products and services,
the locations of businesses, offices, plants, and so on.
We also take various decisions in our personal lives concerning housing, health, children,
marriage, divorce, vacations, schooling (MBA?), etc.

These decisions vary dramatically in their nature. Our objective, however, is to develop a
common methodology to address any decision making, regardless of its nature.

The basic approach of economists to any decision-making is the use of Cost-Benefit Analysis.
Although it sounds trivial, it is worth pointing out that we are concerned ONLY with decisions
that involve both costs and benefits. Decisions that involve only benefits are not really decisions
because we should always do something if it only has benefits, and never take any action that has
only costs and no benefits. When a decision involves both costs and benefit, one must weigh one
against the other, and favor a decision only when its benefit exceeds its cost. This rule should
apply to any decision, regardless of its nature.

While this approach sounds extremely simple and logical, applying it is not always easy. In fact,
a substantial amount of time in this course will be devoted to this simple decision making rule. It
is also our belief that the failure to follow such a simple rule is at the heart of many mistakes
done by managers and decision makers in a variety of settings.

We will, therefore, start with a contrived, but useful example.

II. A Simple Example.

Consider a small company that produces a standard, generic, product; for example, plastic lunch
boxes. Further assume that our firm produces only one type of boxes and, therefore, choices
regarding quality, design, etc. are irrelevant.

The company has several long time customers whom it supplies with its product on a regular
basis, and it does not expect new business proposals with the exception of the following:
1
A buyer from one of the largest buyers of lunch boxes, 4M Inc., calls our small company one day
with the following proposal:
“Hello, this is Mr. McNerny of 4M. I would like to buy as many of your lunch boxes as you can
produce. I am willing to pay $9 per box. Are you interested? If yes, how many lunch boxes are
you willing to supply within 90 days?”

There are two essential features to this business proposal:

1. The price is given and independent of the production level.

2. There is only one dimension to our decision problem, namely, the number of lunch boxes
to deliver to 4M.

Thus, our company has to take two sequential decisions:

a. Should we produce or not?
b. If yes, how many units should we produce and supply to 4M?

What is the company’s goal?

Profit maximization
Quite obviously, it is impossible to take a decision if we do not spell out first precisely what the
goal of our company is. At this stage, we can agree that we are going to accept the 4M proposal
if we make money out of it. However, there is too much ambiguity in such a statement.

For instance, we do not want to face a situation described by the following pretty common
sentence: “Yes, I can make some money in this deal, but it’s not worth it going into this deal.”

What does “some money” mean? What does “worth it” mean?

We need a precise indicator of profitability for this business proposal. We require this indicator
to satisfy two properties:

1. If the business proposal that we are looking at is our only business opportunity, a positive
number generated by the indicator implies that we should accept the proposal. A negative
number indicates that we should reject it.
2. If we are analyzing alternative, competing, business proposals, and several of them
generate positive values for the indicator, we should take the proposal attached to the
highest value indicator.

We call this indicator economic profits. Economic profits are equal to:

Revenues – Economic costs.

In our case the price is given and equals to $9 per box. Thus, the profits generated by this deal
will equal:

(number of boxes sold)*9 – (economic cost of producing these boxes).

If we decide to accept 4M’s proposal, it is obvious that we will want to supply the amount of
2
boxes that makes our economic profits as large as possible.
Thus, we can agree, at this stage, that the goal of our company (at least as far as 4M proposal is
concerned) is to maximize economic profits.

The revenue side of our story is trivial. Revenues are just 9*q, where q denotes the number of
lunch boxes sold to 4M. Thus our real, challenging problem is to figure out the economic cost of
production.

Since calculating the cost of production is not a trivial task, we will leave our lunch boxes
example for a while to discuss some of the fundamentals of cost analysis.

III. Understanding costs.

There are two problems related to the correct evaluation of the costs related to a business
decision, e.g., level of production. The first is to understand and single out the resources that are
needed to accomplish the task. The second is to impute the right price (cost) to the use of each
resource. In the three examples that follow, the used resources are unambiguously indicated.
However, it is not immediate what the right cost to attach to them is.

Example 1: Using your grandma’s apartment (Equity).

You are running a desktop publishing business, using, as an office space, a two bedroom
apartment in the Upper West area that you inherited from your grandmother. At the end of the
year you conduct a profitability analysis of your business. What cost should be allocated to the
apartment?

Example 2: Borrowing a sales person from one department to the other (transfer
pricing).
The manager of women’s wear at Macy’s is asking to “borrow” for one month a sales person
from men’s wear. When calculating the cost of operating the women’s wear department for that
month, what cost should be allocated to the use of this sales person?

Example 3: Apple Computers (a true story)

In August 1988 Apple ordered millions of 1 megabit DRAM chips at $38 per chip.
Before Apple could use up its inventory, the price fell to $23 per chip (in January 1989).
How should Apple calculate the cost of its new Macintosh?

In the above example we have identified resources that have been used by the firm (or
department), but do not always appear in financial statements. This, however, does not mean that
they should not be taken into consideration in a variety of managerial decisions.
We can easily show that ignoring these resources, or assigning them an incorrect cost, can result
in a wrong managerial decision.

3
Analysis:

1. Decision Related Costs

Our objective is to improve decision making. When we analyze costs it is crucial that we have a
specific decision in mind. Therefore, for each of the above examples, and in general, whenever
asked to calculate costs, as a first step, identify the decision that is at hand.

2. What resources should be included and at what price?

Our approach can be labeled “decision based cost analysis”. We are interested in costs ONLY if
they could affect a decision. Therefore, as a first rule, we consider ONLY resources/costs that are
affected by our decision.

Students familiar with financial statements or have any knowledge in accounting will
immediately realize that some “costs” do not appear in the firm’s income statement and,
therefore, are not taken into account when calculating (accounting) profits. When it comes to
managerial decision making, however, ignoring such costs could result in wrong decisions.
We call these costs “Phantom Costs”, or opportunity costs.

The general rule is very simple:

All resources that participate in an activity (associated with a decision) should be taken into
account!

The next question is: what price tag (cost) should be attached to any of these resources?

Here we come to the heart of what is considered a cost by economists and the meaning of the
term “economic costs”.

A resource has a cost ONLY if it has an alternative use. The cost of using such a resource is,
therefore, the loss associated with not using it in its next best use, thus the term “opportunity”.

It should be pointed out that standard direct monetary costs are also “opportunity costs”. The lost
opportunity when we spend a dollar on materials is anything else that could be purchased with
this dollar. Therefore, the “opportunity cost” of spending a dollar on anything is simply one
dollar.

3. What costs should NOT be included?

Given our rule of what costs should be included, the answer to this question is obvious: Any cost
that is NOT going to be affected by our decision should be IGNORED.

Sunk Costs
Sunk costs are costs that have already been spent or committed to in a way that they cannot be
recovered. By definition, therefore, such costs are not affected by any decision and should
always be ignored.
Sunk costs are of a particular interest because there are numerous examples in business as well as
in our personal lives showing that human beings have hard time ignoring sunk costs. Therefore,
ignoring sunk costs, as logical it may sounds, may require an extraordinary effort and discipline
in real life.
4
IV. Back to the “Lunch Boxes” example.
Now, that we have a better sense of how to allocate costs when calculating the profits associated
with any decision or project, we can go back to our earlier example.

We want to understand what are the “relevant” resources needed to produce lunch boxes for 4M.
The difficulty is to make the word “relevant” precise.

Some cost items are obvious. Consider the following cost information concerning the production
of lunch boxes:

1. Each box requires $1 of materials, $.50 of labor, and $.50 of energy and packaging.
2. In addition, the production of the boxes requires space, tools and machines. Suppose, for
example, that our firm uses a 10,000sqf space that it rents for $10,000 per month. It also
rents all its tools and machines for another $10,000 per month. Producing the boxes for
4M does not require any additional space, nor will it require getting additional tools or
machines. Given that the firm does not operate in full capacity, these resources are
available for the production of the boxes.

While it is obvious how to treat the first group of resources, the items listed under (2) are not that
trivial to analyze.
Space, machine, and tools, are resources used to produce the boxes. As such, they should be
considered in our cost analysis. On the other hand, had we decided NOT to supply the boxes to
4M, we would still have to pay the $20,000 per month. So the question is, should we include any
of these resources in calculating the cost of producing the boxes?

The answer is NO, and the reason is simple. These are costs that are NOT AFFECTED by our
decision of whether to produce the boxes or not, and as a result the opportunity cost of using
them is zero.

To make this point clear, let’s illustrate why any other approach could result in making the
wrong decision and, therefore, lower profits for our firm.

Let’s say that it will take about a month to produce and deliver the boxes to 4M, and during this
period about 10% of space, tools, and machines will be allocated to this project, which in terms
of costs is about $2000.

Now, let’s assume that, excluding the cost of space, tools, and machine, the profit from this
project is going to be $1500.

Now, if we do not include these costs, our decision will be to undertake this project, because it is
going to generate net profits of $1,500. If, however, we decide to include these costs by
allocating the relative use of space, tools, and machines to the project, we have a project that has
a negative profit, or a loss of $500, and should, therefore, be rejected.
What is the result of this decision?
At the end of the month we still pay all the expenses related to space, tools, and machines, but by
rejecting the project we let $1,500 go. On the other hand, if we accepted the project, we would
have been $1,500 richer by the end of the month.
5
Repeating the principles that we have already discussed, whenever facing a business proposal,
we can look at all costs related to the proposal in terms of avoidable and unavoidable costs.
“Unavoidable” costs are those cost that we cannot avoid, whether we undertake or reject the
proposal, and such, should be ignored when calculating the economic profits of the project.
“Avoidable” costs are costs that can be avoided by NOT taking the project. These costs
SHOULD be included when calculating the project profitability.

Fixed, Variable, Avoidable, Unavoidable. It starts getting confusing.

In our specific example space is a fixed cost. It does not matter how many boxes we produce for
4M, we do not need additional space. “Materials”, on the other hand, is a variable expense. It
varies with the level of production. We need $1 worth of materials per box. Thus, if we produce
100 boxes we need $100 worth of materials, 200 boxes $200 of materials, and so on. If you look
carefully at the example, all unavoidable expenses are fixed.

So, is it always the case that fixed costs are unavoidable?

The answer to this question is largely semantic. However, at this stage, it will avoid confusion to
give a negative answer. The TV listing case will clarify once and for all this issue.

If you want to think about this point a little bit before we discuss the TV Listing case, let’s get
back to the example and let’s change the story a little bit. Assume that in order to produce the
boxes for 4M we need to rent an additional cutting machine for a fee of $1000 per month. We
need to rent this machine for one month regardless of the number of boxes we will produce. This
additional cost of $1000 does not change with the level of production and could be considered in
some sense, a fixed cost. However, quite obviously it is an avoidable cost and should, therefore,
be considered in the computation of the profits.

Last, as another illustration of this issue: here is a little problem that you might want to solve at
home.

Consider a proposal to start a new business. The start-up costs of this business are €60M. Let's
assume that these costs are annualized at 20% per year, which gives us a cost of €12M/year.
In addition, the annual fixed costs of operating this business are also €12M/year.
Net yearly income from sales (but without any of the above costs included) is €16M.

Questions:
1. If you already run this business, will you consider going out of business?
2. If you were asked to start this business, would you?

Notice that we didn’t answer yet the second part of our initial question: if we decide to produce
the lunch boxes, how many units should we produce given the price of $9 per unit?

We will come to this problem in our “profit maximizing” class.

6
A Warning!!!
This remark is meant to clarify a possible confusion that may arise by reading almost any
standard Microeconomic textbook. There, all avoidable costs are variable, while all fixed costs
are unavoidable. As we have already argued, this is a simplification, which is not true in a lot of
practical situations.

7
 Cost Analysis: Unit Costs, Marginal Costs

I. Introduction

We analyze the cost of production of an individual firm. We focus on avoidable costs only.
Three different examples will be used to illustrate all the relevant cost concepts. Before turning
to the examples we discuss and define several concepts of cost.

I) Costs of Production

We analyze the costs associated with the production of output of a particular firm in a particular
market. Consider, for example, the costs faced by the plastic box producer. Remember that we
are only focusing on avoidable costs, i.e., on the costs directly related to the task of producing
boxes for 4M.

There are two types of avoidable costs: fixed and variable. Fixed costs are those that do no vary
with the number of units produced. For instance, the machine used in the 4M contract is
independent of the number of boxes produced. Variable costs are resources that vary with the
level of output. For example, the resine film used to produce boxes for 4M is proportional to the
number of boxes sold.
Thus, the following equality holds true:

Total (avoidable) costs = fixed (avoidable) costs + variable (avoidable) costs

As a formula:

C(q) = FC + VC(q),

where q denotes the units of output produced, FC fixed cost, VC(q) variable cost associated with
the production of q units of output, and C(q) represents the total cost associated with the
production of q units of output.

Average Cost and Marginal Cost

It is convenient to report cost information on a per-unit basis. Average total costs are simply total
costs divided by the level of output. Since total costs are the sum of fixed and variable costs, we
get:

Average Cost=(Total avoidable Cost)/Output

= Average Fixed Cost + Average Variable Cost;

Using formulas:

AC(q) = AFC(q) + AVC(q),

AVC(q) = VC(q)/q,
And AFC(q) = FC/q,

8
where AC(q) is the total (avoidable) average cost, AFC(q) is the average fixed cost, and AVC(q)
is the average variable cost.

The marginal cost, evaluated at a level of output equal to q, is the cost generated by the
production of an additional unit of output (MC(q)). For example, assume that we are currently
producing 10,000 boxes. The marginal cost evaluated at this level of production (10,000) is the
cost of increasing production from 10,000 to 10,001 boxes.

In a lot of circumstances (problem sets), the marginal cost is identified as the derivative of the
total cost function with respect to output. In this case (see example 3) we set

MC(q)=dC(q)/dq.

Where dC(q)/dq denotes the first derivative of the total cost function C(q) with respect to output.

II. Examples

Example 1: Simplest conceivable cost structure (selling apples in Union Square).

Consider an entrepreneur who lives in Upstate New York and sells apples on the farmers market
on Saturdays, in Union Square. She buys the apples for $10 per case. In addition, she has the
following costs: To ship the apples to Union Square she needs to rent a truck for $250 for the
day. The truck has a capacity of 500 cases. She also has to hire one worker to ship the apples and
sell them. The labor costs are $150 for the day.

All these costs are avoidable. At any given weekend she has the option of not going to Union
Squarel that Saturday and to avoid all costs. However, the only costs that vary with the number
of apples sold are the $10 per case. The remaining costs of $400 are fixed.

Since it costs $10 to “produce” one case of apples, it costs 10*q to produce q units of output. In
terms of equations:

VC(q) = 10q and AVC(q) = 10, for each q < 500

Since fixed costs are $400, it follows that:

AFC(q) = 400/q, for each q < 500.

Hence:

AC(q) = (400/q) + 10, for each q < 500.

What is the marginal cost in this business? In other words, what is the additional cost associated
with increasing sales by one unit (case of apples), given, of course, that we are below 500, the
capacity constraint? In this specific example the answer is very simple: $10, regardless of the
level of sales.

To sum up, if you are producing q units, where q < 500, then in order to add an extra unit you
incur an additional cost of $10. Hence,
9
MC(q) = AVC(q) = 10.

Let’s plot all the above information in a graph:

$ per unit

AC

$10.80
$10
MC

Cases of apples
500
Observe that AFC(q) is a decreasing curve and that as q approaches 0, the AFC(q) (and,
therefore, the AC(q)) becomes very, very large (400/q goes to infinity). Moreover, when the firm
is producing at capacity, i.e., q = 500, the AC(500) = $10.80.

Example 2:
For this example, consider problem set #2 (Piggy Bank). There, the bank could use 3 different
sources of funds in order to “produce” loans. However, the cost associated with each source was
different and, moreover, the first two sources of funds, checking and savings deposits, were
limited in quantity. The example below will have a similar cost structure:

Consider a firm that can produce at most 200 units of output per day, i.e., capacity = 200. In
order to do that, it must incur in a fixed cost of $100 per day; i.e., FC = $100. The firm has to
pay $1 per unit in order to produce output up to a capacity of 100. For each unit produced above
100, the per-unit cost is $4, up to a capacity of 200.

Suppose that the firm is producing q units, q < 100. Since it pays $1 per unit in variable costs, the
variable cost of producing q units is:

VC(q) = $1*q, for q < 100,

and the total avoidable costs are

C(q) = 1*q+100, for q < 100

10
Suppose now that the firm is producing q units, q>100. Then, the firm pays $1 per unit in
variable cost for the first 100 units and $4 per unit for the remaining (q-100) units. Hence, the
variable cost of producing Q units is

VC(q) = 1*100 + 4*(q-100) = 4q-300 , for 100 < q < 200,

and the total avoidable costs are

C(q) = FC+VC(q) = 100 + 4q – 300 = 4q – 200, for 100 < q < 200.

Summarizing:

C(q) = 100 +q, if q < 100

C(q) = 4q – 200, if 100 < q < 200

Therefore, the average cost is:

AC(q) = 1 + (100/q), if q < 100

AC(q) = 4 – 200/q, if 100< q < 200,
.
and the average variable cost is:

AVC(q) = VC(q)/q = 1, if q < 100,

AVC(q) = 4- (300/q), if 100 < q < 200.

Finally, suppose that the firm is producing q units of output, q<100. In order to produce an extra
unit, the firm must pay $1of variable costs. Hence,

MC(q) = 1, for each q < 100.

However, if the firm is producing q, 100 < q < 200, the extra unit requires $4 of variable inputs.
Hence:

MC(q) = $4, for each 100< q<200.

Observe that when q < 100,
MC(q) = 1 < AC(Q) =1 + 100/q

Therefore, the AC(q) is declining for 0< q <100. However, when q > 100,

MC(Q) = 4 > AC(q) = 4 – (200/q)

and the AC(q) is increasing for q>100. In fact, the AC reaches its minimum exactly at q=100
where it is equal to $2.

Finally, observe that AVC(q) = 4 – (300/q) and MC(q) = 4, for q>100. This shows that it might
be very well true (depending on the technologies) that marginal costs and variable costs are
different.

See the graph below.

11
 $ /Unit

AC

MC
4

AC

1
MC
Q
100 200

Example 3: Textbook technology.

This example is the more general one, commonly used in economic textbooks. The cost function,
represented by the equation below, is continuous and, as you will see, is characterized by
constantly increasing marginal cost.

The firm is facing the following cost function:

C(q) = 100 + q + q2

We are going to calculate:

1. the average cost curve,
2. the marginal cost curve, and
3. the minimum average cost and its corresponding output.
1) The average total cost is:
AC(q) = C(q)/q = (100/q) + 1 + q.

2) The marginal cost is the derivative of the total cost with respect to Q:
MC(q) = dC(q)/dq =1 + 2q

This specific MC function means that the cost of producing an additional unit will differ,
depending on the existing level of production. As the level of production is increasing, the cost
of producing an additional unit is increasing. You should contrast this cost structure with the
earlier examples where (at least over a range of levels of production) the marginal cost was
fixed.

3) The average cost reaches its minimum at output level q* where MC(q*) = AC(q*).
We will discuss the intuition of this equality later on.
12
Hence, q* solves the following equation:
100/q + 1 + q = 1 + 2q
or q2 = 100.
And, therefore, q* = 10.
An alternative way to find the point of minimum AC is by taking the derivative of the AC
function with respect to output, and computing the level of output that will equate the derivative
to zero.
The graph below, produced with Excel, shows the AC and MC functions.

60
MC
50

40
AC
30
`
20

10

0
1 6 11 16 21 26 31 36

The minimum average cost is AC(10) = $21, and the marginal cost at this level of output is also
$21.

Notice that for q<10, the MC(q) is less then the AC(q) and the average cost is declining. For
q>10, on the other hand, the AC(q) < MC(q) and the average cost is increasing.
What is the intuition behind this observation? Why is it always true that if MC(q) < AC(q), then
AC is decreasing, and if MC(q) > AC(q), then AC is increasing?
Suppose that 10 of us are in a bar. We ordered drinks for 15 dollars per person. This is our
average bill. At this point, Bob (the marginal drinker) shows up and joins us for a drink. Bob’s
spending plays the same role of the marginal cost. If Bob orders drinks for more than $15, our
average bill (Bob’s included) will go above $15, i.e., it will be increasing. However, if Bob
orders drinks for less than $15, the average bill will decrease below $15, i.e., it will be
decreasing.

13
 Production Decisions and the Individual Supply

I. Profit maximizing output decisions and the supply curve of the firm.
So far we have seen how to allocate economic costs as a function of the decision that we face.
We also examined several typical cost functions. At this point we go back to the earlier example,
deciding if and how many units of lunch boxes to produce and supply to 4M. As you recall, 4M
was willing to buy any quantity at a price of $9 per box.

More generally, we are looking at a firm that considers the possibility of supplying a product to a
market. At this specific market there is a given price, p, and the firm can supply any quantity of
output without affecting the market price. Hence the firm has two sequential decisions to make:
a) Whether to produce or not, and b) if yes, what is the profit maximizing level of output.

Given the cost structure of the firm we will be able to determine the minimum price at which the
firm will be willing to supply the product. Then, for every price above this minimum, there will
be a quantity that will maximize the firm’s economic profits (given that price, of course). The
relation between market price and output supplied by the firm is called supply curve. More
precisely:

The firm’s supply curve is a function/graph that associates to each conceivable price the profit
maximizing output.

In this section we look at this problem by analyzing the three cost structures of the previous
section.

It will be useful to make the following observations:

i) The firm decides to produce positive output if and only if this decision generates positive
economic profits. Otherwise, the optimal choice is not to engage in any production activity.
ii) By definition, the economic profits generated by the production of q units of output are equal
to the profits per unit times the number of units:

profits = Revenues - Costs= (p*q) - C(q) = (p - AC(q))*q = (profit/unit)*(# of units)

These two observations imply that the firm will decide to produce if and only if the market price
allows for positive profits per unit, i.e., only if p > AC(q), for some level of output q. However,
the price is above the average cost of production of some output level q if and only if p > min
AC. Hence, in order to decide whether to produce at all or not, we have to compare the market
price to the minimum average cost of production.
If p > minAC, then produce; if p < minAC, don’t produce.

(In the perverse case where p = min AC, the firm is indifferent between two choices : a) produce
at q*, the output corresponding to the minimum average cost, and realize 0 economic profits, b)
do not produce at all and, again, realize zero economic profits.)

It is very likely that the above statements seem very complicated. Don’t panic. Go carefully over
the examples below and then see whether, within the content of the examples, these general
statements are clear.
14
Example 1
Go back to the apples seller example. As you can see in the graph, if the market price is below
$10.80 per case (the minimum AC within the 0-500 domain), the firm does not sell at all. The
intuition is very simple: There is no way to recover the cost of the apples, labor, and truck. Costs
will be greater than revenues no matter what output level we choose. Therefore, it is better to
stay home and avoid Union Square on any given Saturday in which the market price is below
$10.80. However, if p > 10.80, going to Union Square to sell apples will be profitable. The
problem is to find the profit maximizing output for p > 10.80.

It is useful to start from an arbitrary level and to perform marginal adjustments. Hence, say that,
for p > 10.80, the firm is considering the production of q units of output, q < 500. Is this choice
optimal (i.e., profit maximizing)?

Consider the possibility of increasing the production by one unit. How is this choice going to
change economic profits? Or, more precisely, what are the changes in revenues and the changes
in costs generated by the production of this extra unit?

Since the firm is producing an extra unit, its revenues from sales will increase by $p (selling one
more unit at the market price, p, increases sales by p!). However, the production of an extra unit
increases costs by MC(q). In example 1, MC(q) = $10, for all q < 500. Hence the change in
profits generated by the production of an extra unit of output is

(change in revenues - change in costs) = p - 10 > 0.

It follows that if p > 10.80, every time q < 500 the firm has an incentive to increase production.
This implies that the firm should optimally produce at capacity.

Summarizing:
i) If p < $10.80, the profit maximizing choice is to not produce at all (or to stay out of the
market);
ii) If p > $10.80, the firm produces at capacity, the profit maximizing level of output.

Items (i) and (ii) entirely characterize the supply curve of the firm, as the graph below shows:

15
 Price/Case
S
AC

$ 10.80
$ 10

Cases of apples
500
Example 2.
In this case the min AC = $2. Therefore, the firm will not produce for any price below $2, while
it will for any p > 2. The problem is to find the profit maximizing output associated to any p > 2,
i.e., to find the supply curve of the firm.

We analyze the problem by looking at three different price ranges:

i) Consider any market price 2 < p < 4. What is optimal output at p? Remember that MC(q) = 1,
for q < 100, while MC(q) = 4, for 100 < q < 200. Should the firm produce less than 100 units?
No. Why? By increasing output by one unit, the firm is going to increase revenues by $p, while
costs are going to increase by MC(q) = $1 (since q <100). Therefore, the change in profits is
equal to p - MC(q) = p - 1 > 0. Therefore, it is never profit maximizing to produce below 100.

Should the firm produce above 100? No. Why? By reducing output by one unit, the firm is
going to lose $p of revenues, while it will save MC(q) = $4 of costs. This generates a change in
profits equal to -p + MC(q) = 4 - p > 0. Therefore, it is never profit maximizing to produce
above 100 when 2 < p < 4. It follows that the optimal output decision is q = 100 for all price p, 2
< p <4.

ii) Suppose now that p > 4. What is the optimal output choice? Should the firm produce at less
than capacity, i.e., q < 200? No. Why? By increasing output by one unit, the firm increases
revenues by $p, while costs increase by MC(q). Therefore, the change in profits is equal to p -
MC(q) . If q < 100, p - MC(q) = p - 1, while if 100 < q <200, p - MC(q) = p - 4. In both
circumstances, since p > 4, the change in profits is positive and, therefore, the firm should
increase production. It follows that q = 200 is the optimal production level.

iii) Suppose now that p = 4. By point i), the firm will never stop production below 100 units and
it is indifferent toward producing any amount between 100 and 200 units. Why? Each unit in
addition to the first 100 costs MC(q) = $4 and yields p = $4 of revenues. The firm is neither
16
making profits nor losses on any of these units and it is therefore indifferent about producing
them.

Points i), ii) and iii) suffice to entirely characterize the supply curve of the firm. The thick curve
in the picture below represents the supply curve of the firm.

$ /Unit

S
AC

AC

Q
100 200
Example 3:
Surprisingly, this is the simplest setting to characterize the supply curve of the firm. As already
said, no production activity will take place for prices below the min AC, i.e., p < 21. For p > 21,
(as discussed in class) the profit maximizing output satisfies the condition

p = MC(q),

i.e.:

(1) p = 1 + 2q.

Solving equation (1) in terms of q yields:

q = (p - 1)/2.

Summarizing, the supply curve of the firm is:

q = (p - 1)/2, for p > 21,

q = 0, p < 21.

17
 P
S AC

21

Q
0
0 10 20 30 40 50 60

Caveat: Remember that C(q) = FC + VC(q). The fixed cost is independent of the level of output,
while the variable cost depends on it. Moreover, if q = 0, the variable cost is zero. However,
since the fixed cost is independent of output, it is equal to $FC even when q = 0. Does this mean
that the firm has to pay the (avoidable) fixed costs even when it does not produce at all? The
answer is no. The reason is that there is a subtle distinction between producing 0 and not
producing at all. Think about the TV Listing magazine situation. If the firm enters the Cheyenne
market, it needs plates, local data bases and so on. These costs are independent of how many
copies of the magazines are produced weekly. If the firm produces just one copy per week it
needs to pay all the fixed costs. However, if the firm stays out of Cheyenne, none of these costs
are incurred. Hence, to be a bit pedantic, when we say that the supply of the firm is identically
equal to zero for p < min AC, we really mean that the firm stays out of the market, i.e., does not
engage in any production activity.

18
 Decision Trees
I. Introduction

So far we have tackled only a simple class of decision problems using cost-benefit analysis. In
simple production decisions, a manager has to choose only one variable, the output or quantity
produced, with complete knowledge of the surrounding production, cost and business
environment. Many real-life managerial decisions are multidimensional, in that there are many
variables that have to be set, and this often comes in situations that have a time dimension, with a
lot of uncertainty.

For example, a studio producer will have to select the type of show to broadcast in a given air
segment, the writers to hire for the script, the director, technicians and actors who will form the
cast in the show, often as functions of their proven or perceived ability, of the budget and
financial constraints her division faces, as well as the forecast audiences for the season, the
moves of competing studios, and so on.

As a person likely to be involved in complex decisions such as this one, decisions that are
frequently dependent on other decisions or outcomes, some of which are uncertain, you need a
powerful, yet simple decision-making tool to help you sort out such problems. Decision Trees
are just the kind of tool that will make your life much easier.

The main virtues of decision trees are:

™ Help handle inter-related decisions (over time).

™ Help work with uncertainty.
™ Translate a complicated story into a simple structure.
™ Show when costs are sunk.
™ Assess the value of information.

As you will see, solving a decision tree is simple and mechanical. In fact, there are a variety of
computer programs that will do it for you. The main task, therefore, is to construct the tree itself
so that it accurately represents the problem at hand.
When a group of people is involved in a complex decision-making project, they frequently fail to
realize that they don’t fully agree about the details of the problem. Describing the problem via a
decision tree can solve such a problem. It describes a complex problem in a crystal clear way.

II. Notations:

= Decision Node

= Event Mode

⊳ =Terminal Mode

Decision Node: A point in which the decision maker has to select one out of several possible
decisions. All possible decisions at this point are described as branches coming out of the
19
squared node. For example:

Risky Investment

How should I invest $1000 CD Paying 5%

Under the Mattress

Event Node: A point in which several possible outcomes/states/events can take place/happen.
All possible states of nature are described as branches coming out of the circled node. For
example:

Rainy
Weather
Sunny

Each of the branches describes a possible state of nature. In the above example it is the weather.
In this simple case there are only two possible states, either rainy or sunny.
Since each of these possible states is uncertain, we have to assign to each of them a probability.
If we believe that the chance of rain is 60% then we will have the following event node:

Rainy
Weather .6
Sunny
.4

Notice that the sum of the probabilities of all states of nature is 1 (100%). This means that all
states of nature are mutually exclusive and all possible states are covered.

Terminal Node: A point in which the decision process ends and payoffs (profits) are realized.
To each terminal node there is a number attached, the payoff realized when that terminal node is
reached. Since there are no decisions to take at terminal node, there are no branches emanating
from it. For example:

Risky Investment
1200
How should I invest $1000? CD paying 5%
1050
Under the mattress
1000

In order to illustrate how to construct and solve decision trees we use two simple examples using
the same business setting. At first there will not be any uncertainty involved and, therefore, no
event nodes. In the second example we will introduce uncertainty.
20
 III. Example:

1. A deterministic world

We are in charge of a small farming business that has to make decisions concerning the production of
corn. The units of output are tons and the production span is one year.
There are several decisions that you have to make concerning production, using the following
information:
Price: The market price of corn is $40 per ton.
Technology: We can use two types of seeds: “regular” (R) or “drought resistant” (DR) seeds. There are
two available plots of land.
Plot A allows for a maximum production of 2,000 tons independently of the type of seeds used. The cost
of producing corn using regular seeds (R) is $20 per ton while using DR seeds increases the cost
to $40 per ton. The yearly rent of plot A is $20,000.
Plot B allows for a maximum output of 4,000 tons regardless of the type of seeds used. Given
differences in soil across the two plots, producing corn in Plot B costs $35 per ton when using
regular seeds but only $30 per ton when using DR seeds. The yearly rent of plot B is $25,000.
Timing and decisions: We have to decide, in the following order:
1) whether to produce or not,
2) what plot to rent,
3) what seeds’ type to use, and
4) how much corn to plant (produce).

We construct the decision tree by moving forward. We start from the first (period) decision node
drawing two branches, P and NP, corresponding to the two alternative decisions of either
producing (P) or not producing (NP). We repeat the same operation for all subsequent periods.
Accordingly, we have the following decision
tree:

21
 2000 Tons

R q

0 Tons
A
2000 Tons

DR q

0 Tons
P
4000 Tons

R q

0 Tons
B
4000 Tons

DR q

0 Tons

NP
0

Dressing the Tree:

So far, in the above tree only the decision to NOT produce (NP) has a terminal node with a
payoff of $0. We now need to compute the realized profits at each of the possible terminal nodes.

We simplify the analysis by computing for each land-seed choice the profit maximizing output
and the associated value of the profit. We proceed as follows. We suppose that a choice of land
and of the type of seed to be used has been made. These choices are: (A,R) (meaning, land A and
seeds of the R type), (A,DR), (B,R) and (B,DR). For each possible land-seeds choice, we
compute the profit maximizing output.

A–R: Given this choice, the marginal cost (equal to the average variable cost) is $20, while the
price is $40. Hence, the profit maximizing choice is to produce at capacity, i.e., q = 2,000. The
profits associated with the A-R choice are:
Π(A,R) = 40*2000 - 20*2,000 - 20,000 = 20,000

A-DR. Since the marginal cost is equal to the price, (equal to $40), we are indifferent about
producing any quantity between 0 and capacity. Thus,
Π(A,DR) = (40*q) - (40*q) - 20,000 = -20,000

B-R. Since the marginal cost is equal to $35 and the price is $40, profit maximization is
achieved at capacity. i.e., q = 4,000. Thus:
Π(B,R) = (40-35)*4,000 - 25,000 = -5,000

B-DR. Since the marginal cost is equal to $30 and the price is $40, profit maximization is
achieved at capacity. Thus:
Π(B,DR) = (40-30)*4,000 - 25,000 = 15,000.
22
Note that when deciding whether or not to use land A and seeds R, the decision to produce has
already been made. Clearly, the rent of the land is irrelevant as far as this decision is concerned.
Thus, the rent of the land is an unavoidable, or sunk, cost at this point. This also appears from
the fact that we subtract $20,000 from revenues in both cases, A and R. We said previously that
sunk costs should not be included in the profits computations: why do we include them here?

The answer is that when writing terminal payoffs, we are computing profits from the viewpoint
of the beginning of the decision process, that is, from the root of the tree, before the decision on
whether or not to produce is made. At that point, land rent is not a sunk, rather a fixed cost and it
should be included in the profits computation.

The tree can now be redrawn as:

R
20,000
A
DR
-20,000
P
R
-5000
B
Producing Corn
DR
15,000
NP
0

IV. Solving the Tree:

We can no w solve the tree, i.e., select the sequence of profit maximizing choices. We solve the
tree by using “backward induction.” The name is scary, but the idea is natural.

Look at all decision nodes immediately preceding the terminal nodes (excluding the initial node).
There are 2 of them. The first is associated to the choice of land A and the second of land B. Call
them A and B. Start with A. Select the branch (decision) leading to the highest payoff. It is R.
Delete the other branch (decision) DR. By doing this we are indicating that if we had to choose
land A, then it would be optimal to use seeds R.
Move to B and repeat the same operation. We select DR and we delete R.

We have eliminated one layer of decisions (the choice of the seeds) and, hence, we have moved
one step backward. We meet the decision node associated to the choice of producing (P). There
are two branches (decisions) emanating from it corresponding to the choice of land (A and B)
Select the branch leading to the highest payoff: if we choose A then it is optimal to choose R and
we realize a payoff of 20K, while the choice B is followed by the choice DR leading to a payoff
of 15 K. Hence, we delete B.

Move one step backward, now we are back to the initial node. If we do not produce (NP) we
realize a profit of 0, if we do (P), we “walk” through the branches A, R and we get 20K. We
delete the NP branch. Thus we have found the optimal sequence of optimal decisions (P,A,R)
and the maximum profits (20K) that we can realize, i.e., we have solved the tree.

23
The picture below illustrates the procedure. The symbol // illustrates deletion of a branch
(elimination of a decision). There is only one path (sequence of decisions) that we can take from
the initial node to the end of the tree (without falling off the tree and getting killed): it is (P,A,R)

R
R : 20,000.00 20,000.00; P = 1.000
A
DR
A : 20,000.00 -20,000.00
P
R
B -5,000.00
Producing Corn
P : 20,000.00 DR
DR : 15,000.00 15,000.00
NP
0.00

Warning: This example is so simple that it seems absurd to even indicate a way of solving the
tree. Do not get tricked. Trees become horrendously (“exponentially”) complicated when you
increase the number of periods and branches.

2. Introducing Uncertainty

We now introduce uncertainty to the picture. However, in order to avoid unnecessary

complications, we also simplify a bit the story.

Uncertainty: Each season can either be rainy or dry. The probability that the season is rainy is .6
(60%), while the probability that it is dry is .4.

Timing: If we decide to produce, we first have to choose the land. Second, we plant the seeds
that we chose, third, God© gives us either a sunny or a rainy season, and last, the corn comes up
and we collect our profits.

Profitability: We use the following information:

Plot A: If we use seeds type R, profits are 20K if it rains and 5K if it does not rain. If we use DR,
profits are 0, if it rains, and 25K it does not rain.

Plot B: If we use seeds R, profits are 10K if it rains and 5K if it does not rain. If we use DR,
profits are 10K, if it rains, and 30K it does not rain.
Thus, this is the tree:

24
 Rainy
20,000
R .6
Dry
5,000
A .4
Rainy
0
DR .6
Dry
25,000
P .4
Rainy
10,000
R .6
Dry
5,000
B .4
Producing Corn
Rainy
10,000
DR .6
Dry
30,000
.4
NP
0

IV. Solving the Tree:

With uncertainty, we face a new problem: we cannot arbitrarily choose a terminal node. For
instance, it would be great if we could select the terminal node associated with the value of 30K
in profits. Unfortunately, to get there we need to produce, use land B and seeds R (these
decisions are under our control), but we need as well a dry season (and this is not under the
control of the decision maker). This problem generates an additional conceptual difficulty. Let’s
compare the two strategies A-R and B-DR. If it rains, the first strategy is superior since it
generates a profit of 20K compared to the profits of 10K generated by the second. However if it
does not rain, the second strategy is superior since it generates 30K compared to 5K generated by
the first. Thus, in order to take a decision we need a criterion to rank strategies that generate a
random flow of profits. In simple words, in our example, we need a method to “summarize” the
two profits generated by each strategy (one profit for rainy and one for dry year) into a single
number.

We use the expected monetary value of the profits (EΠ). In simple words, we take the
weighted average of the profits, using the probabilities as weights. Courses in finance will clarify
the idea behind this choice. At this stage, it suffices to make two observations:
a) A decision maker using EΠ is risk neutral. Intuitively, we can agree that risk is very
much related to the variability of the payoffs. By definition, an EΠ decision maker does
not care about how variable the payoffs are, only about their expected value.
b) There are justifications for using the EΠ criterion if the decision maker is a firm.
Individuals, on the other hand, frequently display risk aversion (think about insurance,
mutual funds and so on).

Once we agree to use expected values, we proceed as follows:

a) First, we collapse all branches emanating from an event node into a terminal node with a
unique number, obtained by using EΠ.
b) Second, we proceed as before by backward induction.
25
For instance, the expected profits generated by using land A and seeds R, EΠ (A,R), is
EΠ (A,R) = .6*(20K) + .4*(5K) = 14K.

The solved tree is depicted below:

14,000.00 Rainy
20,000.00
R 0.600
Dry
5,000.00
A 0.400

R : 14,000.00 10,000.00 Rainy 0.00

DR 0.600
Dry
B : 18,000.00 25,000.00
P 0.400
Rainy
8,000.00 10,000.00
R 0.600
Dry
5,000.00
B 0.400
Producing Corn
P : 18,000.00 18,000.00 Rainy
DR : 18,000.00 DR 0.600
10,000.00; P = 0.600
Dry
0.400
30,000.00; P = 0.400
NP
0.00

By walking backward we get that the (expected) profit maximizing strategy is produce, B-DR.
This strategy generates expected profits equal to 18K.

Value of information
Decision trees are ideal tools to understand the value of information. With a bit of latitude we
can view information (consulting, forecasting and so on) as a way of “reducing” the uncertainty
that a business is facing (and, ultimately, as the rationale justifying our salaries).
Here we just outline the main idea. Let’s go back to our example. A weather research center is
able to predict with certainty if next year will be rainy or not. We can get this information at
some cost before we take any decision. What is the value, to us, of that information?
Equivalently, what is the maximum amount of money that we are willing to pay to the
meteorological center in order to buy its weather forecast?

Before we proceed we need to clarify two typical sources of confusion.

1) Why should the information be worth at all? The answer is: flexibility. For instance, in
our example, if we know that next season will be wet, the choice A-R is profit
maximizing. However, the choice B-DR is the profit maximizing choice for dry seasons.
Thus, being informed (i.e., knowing the weather) allows us to make a weather dependent
choice: choose A-R if it will rain, choose B-DR if it will not rain. This is the source of
the value of the information.
2) What probability should I attach to the event that the weather report will indicate a rainy
season? It is the same probability that we attach to the likelihood of a rainy season, i.e.,
.6. Since we believe that 60% is the correct probability of rain, we also believe that this is
the likelihood that the center will predict a rainy season. Therefore, to attach a probability

26
 to a report indicating rain is logically equivalent (since the report is never wrong) to
attach a probability to a rainy season. Thus, these probabilities must be equal.

We know that if we do not buy information (weather forecast), the choice (land B-seeds DR) is
optimal and will generate 18K of expected profits. In other words, we have the same tree that we
had before considering the option of buying a forecast. Therefore, the tree that we now face is
the following:
R
20,000
A
DR
0
Rainy
.6 R
10,000
B
DR
10,000
Buy Forecast The Forecast
R
5,000
A
DR
25,000
Dry
Should we buy forecast?
.4 R
5,000
B
DR
30,000
Don't buy forecast
18,000

Let’s solve the tree by backward induction:

Delete the branches DR, R (or DR, it is the same), R, DR, from the last decision nodes. We have
deleted all sub-optimal choices.
Move one step backward. If it rains, we pick A because it leads (followed by R) to a payoff of
20K. Thus, delete B. If it is dry, delete A. We have agreed that if it rains we select (A,R) and we
get a payoff of 20K, if it is dry we select (B,DR) and we get a payoff of 30K. Thus, the expected
profit generated by these choices, i.e., the (expected) profit generated by the information
acquisition, EΠ(I) is equal to:

EΠ(I) = .6* Π(A,R) + .4Π(B,DR) = .6*(20k) + .4*(30K) = 24K

Now we are back to the initial node. If we buy the forecast we have expected profits of 24K, if
we do not buy the forecast our expected profits are 18K. Thus the optimal sequence of choices is:
Buy forecast, and proceed with (A,R) if the forecast is rainy, and (B,DR) if it is dry.

27
 R
20,000.00; P = 0.600
A
DR
0.00
Rainy R : 20,000.00
0.600 R
B 10,000.00
A : 20,000.00 DR
Buy Forecast The Forecast 10,000.00
(Indifferent) : 10,000.00
R
24,000.00 5,000.00
A
DR
DR : 25,000.00 25,000.00
Buy Forecast : 24,000.00 Dry
Should we buy forecast?
0.400 R
B 5,000.00
B : 30,000.00
DR
30,000.00; P = 0.400
Don't buy forecast DR : 30,000.00
18,000.00

Since without the forecast the expected profits were EΠ(NI) = 18K, while with the forecast
expected profits are EΠ(I) = 24K, the value of the information, VI, is equal to
VI = EΠ(I) - EΠ(NI) = 24K - 18K = 6K

The value of the information is also equal to the maximum amount of money we are willing to
pay in order to get the weather forecast.

Partially Correct Information: How would your answer change if the forecast were not
certain? For example, assume that the forecast is 90% accurate (do it at home).

Additional questions:
Can you think of how the willingness to pay for the forecast will change with its accuracy? (Plot
the value of X as a function of accuracy, ranging from 0-100%)

Solving decision trees: Backward induction

Here, is a general, simple procedure to solve decision trees, i.e., to find out what is the best
(event contingent) course of decisions and the profit that it generates.

1. Start from a node immediately preceding a terminal node. If it is the initial node, wait. If
it is an event node, attach to it the expected payoff of the terminal nodes that follow it. If
it is a decision node different than the initial node, select the decision branch leading to
the terminal node with the highest payoff. That is the best decision that you can take at
that node. Keep track of the decision and “kill” all the others.
2. Because of 1), you have moved one step backward, i.e., you have transformed the nodes
which where immediate predecessors of terminal nodes (but not an initial node) into
terminal nodes. Iterate 1).
3. Stop when all terminal nodes are immediate successors of the initial node. If the initial
node is an event node take expected value of the payoffs. If it is a decision node, select
the branch leading to the highest payoff.

28