Exponential Utility Function

Tomado de Spice-Logic INC documentation para su software

sobre Decision Trees
https://www.spicelogic.com/docs/DecisionTreeAnalyzer/UtilityFunction/exponential-utility-
function-299

What is the Exponential Utility Function?

In economics and finance, exponential utility refers to a specific form of the Utility Function.
This utility function is mainly used for mapping real-world Monetary gain to perceived value.
Formally, the exponential utility is given by:

Where R is the Risk Tolerance. x is the real-world value and u(x) is the utility value or perceived
value (the value of an outcome in utils). "a" and "b" are essentially scaling parameters. Decision
Tree Software can calculate that parameter based on the Minimum and Maximum possible
values in the decision context, which is collected from the user.

If you represent real-world money gain in X-axis and your level of satisfaction in Y-axis (in terms
of 0 to 1, where 0 means no satisfaction and 1 means the highest satisfaction), then the
Exponential Utility function will look like this:
This utility function is concave, and so it can be used to model risk aversion. In such a utility
function, R, the Risk tolerance parameter determines how concave the utility function is, which in
turn reflects how risk-averse the decision-maker is. In the above chart, we used the Risk
Tolerance value (R) = 1000. Greater values of R (Risk Tolerance) make the exponential utility
function flatter, while smaller values make it more concave or more risk-averse. When your risk
tolerance is infinite, the above function becomes a straight line equation. Thus, if you are less
risk-averse, (you can tolerate more risk), you would assess a greater value for R to obtain a
flatter utility function. If you are less tolerant of risk, you would asses a smaller R and have a
more curved utility function.

The exponential utility function is mainly used to measure the utility of monetary gain where
there is a chance of losing money.

From where this utility function comes from

Well, obviously, this function was not derived. The human brain and behavior are more complex
than such a modeling function. It is just an idea that, as such function is concave, it can be used
to model the behavior of a risk-averse decision-maker. If you look into this utility function (

), you will notice that, as x increases, U(x) approaches 1, which

means the highest utility. The utility of zero in this equation, U(0), is equal to 0. That makes
sense as someone who does not get anything, his utility is 0, right?. Then notice that the utility
for any negative x (being in debt) is negative, which makes perfect sense too. So, we can
whimsically use this utility function, but of course, there is no scientific ground that proves that
human attitude perfectly follows this utility function.

How can the Risk Tolerance (R) be

determined?

Approximate method

The Risk Tolerance 'R' has a very intuitive interpretation that makes its assessment very easy.

Consider a decision situation where you can choose to play a lottery or not to play a lottery. If
you do not play the lottery, you won't gain or lose anything. But, if you choose to play the lottery,
then you can win X with a 50% probability and lose X/2 with a 50% probability. This situation can
be depicted by the following decision tree.

Now, you can ask yourself a question, if this X is 1000$, will you play the lottery or won't play the
lottery? (Remember the chance of losing 500$ in that case). If yes, then ask again, what about
2000$, or 10,000$ or more. The more the number you set as winning number, half of that
number can be a loss too. For example, if you set X = 50,000$, then 50% chance that you will
win 50,000$ and 50% chance that you will lose 50,000$/2 = 25,000$. Are you willing to play such
a lottery where you can lose 25,000$ for the hope of winning 50,000$? If so, that winning value
50,000 is the value that you can use for R in your exponential utility function. If you think, you can
afford to lose even more for the hope of gaining more, i.e., you can lose 100,000$ for the hope of
gaining 200,000$. If so, then your Risk Tolerance value can be 200,000. In a word, think about
the highest number that can be used for X so that you will be motivated to play the lottery.

Exact method (Certainty Equivalent Based)

Ok, we have demonstrated, how quickly we can approximate the value of the Risk Tolerance "R"
by answering a question asked by the above decision tree. Now, let's see how we can get an
Exact value of R mathematically. Consider a decision tree again, where, instead of answering the
highest value of a winning payoff, you can answer the Certainty Equivalent of a given lottery.
That means, say you can make a decision about playing a lottery with a 50% chance of winning
the value of X and a 50% chance of losing a value Y. Or you can choose to receive a confirmed
(Certain) amount of money, which can be very less than the possible highest outcome of the
lottery. This certain amount of payoff is called "Certainty Equivalent".

Anyway, let's consider the following decision tree. You have two options. One option is to play a
lottery where you can Win "W" with a 50% probability, and you can lose "L" with a 50%
probability. Another option is to receive a certain amount of money without playing the lottery. For
example, Your friend is asking if you will like to sell that lottery for a certain amount of money.
How much money you will ask for selling that lottery? Say, you asked "CE". Then, that CE is your
certainty equivalent.

Now, let's see how we can use the concept of Certainty Equivalent to calculate an exact risk
tolerance of an Exponential Utility Function.
According to the theory of Expected Utility, and Von Neumann–Morgenstern utility theorem, if
you can define a utility function, then your Expected Utility for the given gamble will be equal to
your Certainty Equivalent. Because you will be indifferent between two options only if the
Expected Utility of the tow options are the same. Here two options are either play the lottery or
not to play the lottery for a certain amount of money.

The probabilities of winning and losing are both 0.5.

So, let's formulate the equation, where the EU stands for Expected Utility.

According to the theory of expected utility, the Expected utility of playing the lottery is equal to the
expected utility of the Certainty Equivalent (CE).

Our Utility Function is the Exponential Utility Function which is

So, lets plugin this function to the above equation, after simplifying, we get,

Here, W is the Winning amount from the lottery, L is the loss amount from the lottery, and CE is
the certainty equivalent. All of these 3 values are constant. The only variable is R (Risk
Tolerance). In order to find that value of R, the above equation needs to be solved. Solving such
an equation is not very straight forward. It is a little complicated. But, when you have our
Decision Analysis Software (Decision Tree Software or Rational Will), you won't have to worry
about solving such an equation by yourself. Our software will solve that for you and tell you the
exact value of "R". We will show you how to do that on this page, please keep continue reading.

Calculating Certainty Equivalent

So far, we have developed an equation for finding the Risk Tolerance. We can use the same
equation to find the Certainty Equivalent of an Exponential Utility Function if all W, L, and R are
known. In that case, it will be very easy to solve the equation. Because the left-hand side will just
become a constant value. Just take the natural log of both sides and simplify as shown here.

Scaling Parameters

In real-life scenarios, you may want a utility function equation where the maximum payoff from an
investment or lottery will yield the highest utility value (i.e. 1) and the minimum payoff (or loss)
will give the lowest utility value. By introducing 2 parameters "a" and "b", the exponential utility
function can be scaled such that,

That means, for different lottery or different scenarios, the value of "a" and "b" will be different.

How to find out the value of "a" and "b"?

Say, your highest utility is 1. For the given investment, your highest possible gain can be H and
the lowest possible gain (or loss) can be L.

Then, your utility function with scaling parameters "a" and "b" can be mathematically derived as
Say, in a lottery, you can gain a maximum of 1000$, and lose 500$. Then, setting these values,
you can get "a" and "b" as

In the above equation, R is the Risk Tolerance, as usual. When you use the SpiceLogic Decision
Analysis Software (Decision Tree Software or Rational Will), these scaling parameters will be
evaluated automatically based on your other inputs.

Marginal Utility

Marginal Utility is a measure that indicates how much a person's utility changes by a little change
of payoff. Mathematically, if you differentiate the Utility Function U(x) with respect to the payoff x,
then you get the Marginal Utility Function.

Here, the "b" is a scaling parameter and R is the Risk Tolerance, x is the real-life payoff and U(x)
is the utility value for the given payoff x.

Risk Aversions
Risk Aversion is a mathematical function that indicates how risk-averse a decision-maker is. The
risk aversion function can be derived from the Utility function. As we explained in the Utility
Function chapter that, the absolute risk aversion is

and the relative risk aversion is

If we apply these operations on a scaled Utility Function equation, we get,

Notice that, the absolute risk aversion of an exponential utility function is a constant (1/R), that is
irrespective of wealth. Therefore, the exponential utility function is most appropriate for people
whose risk attitude does not change according to the amount of wealth they have. Many
individuals might be less risk-averse if they had more wealth, which is the idea of the Bernoulli
Utility Function.