Title: PETROLEUM ENGINEERING 689 Special Topics in Unconventional Resource Reserves Lecture 4 Expected Valu
1PETROLEUM ENGINEERING 689Special Topics
inUnconventional Resource ReservesLecture
4Expected Value Decision TreesTexas AM
University - Spring 2007
2Expected Value and Decision Trees
3Learning Objectives
- You will be able to
- Calculate the expected value and standard
deviation of a random variable - Calculate expected monetary value (EMV) using a
payoff table - Determine sensitivity of EMV analysis results to
probabilities used
4Learning Objectives
- You will be able to
- Define expected profitability index, EPI
- Use EMV and EPI to screen investments
- Use EMV and EPI to rank investment alternatives
- Define and explain performance index, I
- Define and use expected opportunity loss, EOL, to
select investments
5Learning Objectives
- You will be able to
- Apply mean-standard deviation screening method to
select investments - Define and explain use of deterministic dominance
in investment selection - Define and explain use of stochastic dominance in
investment selection - Interpret what expected value represents in
applications
6Expected Value of Random Variable
- where
- EX expectation operator, read expectation
of - P(xi) P(Xxi), unconditional probability
associated with variable x - E(x) often referred to as mean of X
7Standard Deviation of Random Variable
- Variance of discrete random variable given by
- where
- S2X variance of X
8Sums of Expected Values, Variances
- When sum of expected values or variances has to
be determined, process straightforward when
variables are both independent and random - EX Y EX EY
- and
- s2X Y s2X s2Y
9Example Calculation of Expected Value
- Expected results from drilling prospect
- 30 chance of 20 MSTB
- 50 chance of 60 MSTB
- 20 chance of 95 MSTB
- What are mean, variance, and standard deviation
of expected reserves?
10Example Calculation of Expected Value
11Example Calculation of Expected Value
- Mean, or expected value, of reserves 55.0 MSTB
- Variance 700.0 MSTB2
- Standard deviation 26.5 MSTB (v700)
- Interpretation
- Over large number of similar trials, expect to
recover 55 MSTB with 67 confidence result will
lie between 28.5 (55-26.5) and 81.5 (5526.5) MSTB
12Expected Monetary Value
- When random variable in expected value is
monetary value, calculated expected value called
expected monetary value, EMV - EMV is weighted average of possible monetary
values (usually NPVs), weighted by respective
probabilities - Monetary values can be undiscounted or
undiscounted - EMV of NPVs called expected present value profit
13Expected Monetary Value
- Best choice of mutually exclusive investment
alternatives is one with largest EMV - For screening investment alternatives, all
alternatives with EMVgt0 are acceptable - When monetary values represent opportunity
losses, alternative with smallest expected value
is optimal choice
14Structural Elements in EMV Calculations
- Acts or strategies, Aj alternatives for action
available - Outcome states, Si different results that may
occur - Consequences or payoffs, Cij gains, rewards,
losses, etc. associated with jth act that results
in ith outcome state
15Structural Elements in EMV Calculations
- Outcome state probabilities, P(Si) probabilities
assigned to outcome states - Criterion Basis decision makers use for most
appropriate course of action from among the
alternatives - Two ways to display structural elements
- Payoff table (tabular)
- Decision tree (graphical)
16Generalized Payoff Table
17Example Payoff Table Application
- Outcomes likely for drilling prospect
- Dry hole probability 65, loss 250,000
- Successful well probability 35, NPV of future
net revenues 500,000 - Can farm out prospect, remove exposure to
drilling expenditure, retain overriding royalty
interest, NPV 50,000 - Determine whether to drill or farm out
18Example Payoff Table Application
19Example Payoff Table Application
- Since EMV of farm-out (17.5M)gtEMV of drilling,
we should farm out - Result highly sensitive to probability of
producer - If increased from 35 to 36, drillling better
option - Sensitivity analysis useful if unsure about
probabilities - Variance of drill option much greater than
variance of farm-out option (drilling much more
risky)
20Example Payoff Table Application
- A company with 100 acres leased wants to drill a
well on 160-acre prospect area - We can join unit by leasing remaining 60 acres in
unit - Evaluation assumes we acquire acreage
- Gross well cost (with equipment) 110M
- Gross dry hole cost 80M
21Example Payoff Table Application
- We have identified 3 options and determined NPVs
for several outcomes - Participate in drilling with 37.5 non-operating
WI (60/160x100 37.5) - Farm out acreage and retain 1/8-th of 7/8-ths
royalty interest on 60 acres - Be carried with back-in privilege (37.5 WI)
after investing parties have recovered 150 of
investment
22Example Payoff Table Application
23Example Payoff Table Application
- Answer questions
- Should we lease adjacent land (mineral rights)?
- If so, what maximum amount should we pay?
- If we lease adjacent land, which option will be
most valuable to us?
24Example Payoff Table Application
25Example Payoff Table Application
- Back-in has largest EMV and is best option
- Maximum value of additional acreage is 25,375/60
423 per acre - If acreage acquired for exactly 25,375, rate of
return will be 10 (discount rate used to
determine NPVs) - Rate of return increases as we pay less to lease
land
26Sensitivity Analysis on Probabilities
- Probabilities used in EMV analysis usually most
uncertain parameters - We need to determine influence of changes in
probabilities on apparent optimal decision to
improve our decision making - Consider earlier example with two acts (drill or
farm out) and two events (dry hole or producer)
27Example Sensitivity Analysis on Probabilities
28Example Sensitivity Analysis on Probabilities
- Let p probability of dry hole
- (1-p) probability of producer
- Then
- EVdrill p(-250) (1-p)(500)
- -750p 500
- EVfarmout p(0) (1-p)(50)
- -50p 50
29Example Sensitivity Analysis on Probabilities
- Decision maker indifferent if EV of two
alternatives equal - Probability at point of indifference given by
- EMVDrill EMV Farmout
- -750p 50 -50p 50
- p 0.6429 or 64.29
- Farm-out optimal for pgt0.6429, but results highly
sensitive to change in probability - We must do our best to ensure probability correct
30Example Sensitivity Analysis on Probabilities
31Sensitivity Analysis on Probabilities
- For previous example with three alternatives
(drill, farm out, back-in), graphical method
easier to implement
32Sensitivity Analysis on Probabilities
33Expected Profitability Index
- Expected profitability index, EPI, is ratio of
EPV of net operating revenue to EPV capital
investment - For non-mutually exclusive investments, we will
maximize corporate NPV if we maximize EPI
provides capital budgeting strategy when we dont
have enough funds to invest in all economically
viable (EMVgt0 or EPIgt1) alternatives - To develop budget, rank investments in descending
order of EPI, start at top and select investments
until investment budget fully allocated - Method fails for farm-out and royalty strategies
because investment is zero (and EPI is infinity)
34Expected Profitability Index Example
- Information for 3 drilling prospects given in
following table - Calculate EPI for each prospect
- Select economic optimum prospect, using decision
rule of maximizing EMV - Assume prospects mutually exclusive
- Select optimal prospects if we have only 150M to
invest and prospects are non-mutually exclusive
35Expected Profitability Index Example
36Expected Profitability Index Example
- Mutually exclusive prospects, no capital
constraint - Prefer prospect B largest EMV
37Expected Profitability Index Example
- Non-mutually exclusive prospects, capital
constraint of 150M - Cannot select prospect B, ECAPEX 198.75gt150
- Choose A and C, total ECAPEX 145,250
- If both projects successful, budget overspent by
15M (120M 45M 165M completed cost) - A success, C dry, budget matched exactly
38Performance Index
- Definition of performance index, I
- I EMV/s expected monetary
value/standard deviation - Maximum I maximizes economic return at given
level of risk - Minimum I sometimes used as threshold for
screening investments - Unlike EMV criterion, takes into account risk
preferences of investor
39Expected Opportunity Loss
- Definition EOL is difference between actual
profit or loss and profit or loss that would have
resulted if decision maker had had perfect
information at time decision made - Example choose to drill well, turns out to be
dry hole, lose 30M - Farm-out would have had zero loss
- EOL 30M 0 30M
40Expected Opportunity Loss
- EOL minimization rule can be used in place of EMV
maximization rule as basis for decision making - Result same with either rule
- EMV easier to work with in complex situations
41Example Expected Opportunity Loss
- Analyze data in following table using EOL
criterion (for drill, farm-out, back-in
alternatives)
42Example Expected Opportunity Loss
43Example Expected Opportunity Loss
- Construct opportunity loss table
- Identify maximum value entry in each row in
previous table - Subtract each entry in same row from maximum
value - Compute expected values by multiplying
probabilities of outcomes by conditional
opportunity losses - Results in following table
44Example Expected Opportunity Loss
45Example Expected Opportunity Loss
- Best choice is back-in alternative, which has
minimum EOL - Same decision as with maximum EMV criterion
46Summary of Decision Criteria
- Choose alternative with largest EMV when profit
is payoff variable and alternatives are mutually
exclusive - Choose alternative with smallest EOL when cost is
payoff variable and alternatives are mutually
exclusive
47Summary of Decision Criteria
- Rank alternatives from largest to smallest EPI
when profit is payoff variable and alternatives
are non-mutually exclusive - When capital is limited, select alternatives with
largest EPI for budget and stop when expected
capital expenditures equal or exceed capital
budget
48Mean-Variance Criterion
- Risk-averse decision maker may use EMV and
standard deviation to screen or rank alternatives - Mean-variance approach favored by some
risk-averse decision makers seeks to choose
alternative that yields highest expected return
with lowest variance
49Mean-Standard Deviation Screening Method
50Mean-Standard Deviation Screening Method
- Investments 1, 7, 8, 9, and 10 provide greater
EMV for given level of risk than other investment
opportunities - Determines efficient frontier
- Choice between investments depends on how risk
averse decision maker is - Investment 10 offers greatest reward, but is
highest risk
51Mean-Standard Deviation Screening Method
- Approach more appropriate when probability
distribution of each alternative can be
represented by mean and standard deviation - Normal distribution good example of appropriate
distribution - When distribution not described completely by
mean and standard deviation, best to compare
distributions themselves
52 Dominance
- When we can compare pdf and cdf of alternatives,
we can use dominance rules to choose between
alternatives - Situations include
- Deterministic dominance pdfs, cdfs, dont
intersect, one alternative always better - First degree stochastic dominance pdfs
intersect, cdfs dont, one alternative still
clear choice over other - Second degree stochastic dominance less clear
53Deterministic Dominance
54First Degree Stochastic Dominance
55More Complex Situation
56Second-Degree Stochastic Dominance
- Compare areas before and after cross-over of
cdfs - Alternative with larger area dominates other
alternative - Called second-degree stochastic dominance
57Application of Dominance Rules
58Application of Dominance Rules
- Assume normal distributions for each case
- Allows use of EMV, standard deviation to generate
cdf
59Application of Dominance Rules
60Application of Dominance Rules
- Back-in option dominates drill option for
0ltEMVlt27M - Drill option dominates back-in option for
EMVgt27M - Area dominated by back-in option slightly larger
than area dominated by drill option - Back-in option has second-degree stochastic
dominance over drill option
61Interpretation of Expected Value
- Expected value is average value per decision
realized when the alternative is repeated over
many trials - What EV is not
- Not the most probable outcome of selecting an
alternative - Not the number which we expect to equal or exceed
50 of the time
62Accomplishments
- You can now
- Calculate the expected value and standard
deviation of a random variable - Calculate expected monetary value (EMV) using a
payoff table - Determine sensitivity of EMV analysis results to
probabilities used
63Accomplishments
- You can now
- Define expected profitability index, EPI
- Use EMV and EPI to screen investments
- Use EMV and EPI to rank investment alternatives
- Define and explain performance index, I
- Define and use expected opportunity loss, EOL, to
select investments
64Accomplishments
- You can now
- Apply mean-standard deviation screening method to
select investments - Define and explain use of deterministic dominance
in investment selection - Define and explain use of stochastic dominance in
investment selection - Interpret what expected value represents in
applications
65Expected Value and Decision Trees
66Expected Value and Decision Trees
67Learning Objectives
- You will be able to
- Construct a decision tree
- Solve a decision tree to determine optimal
decision and payoff - Calculate expected value, variance, and standard
deviation for optimal outcomes from decision tree - Use Excel to simplify arithmetic in value,
variance, standard deviation calculations - Use PrecisionTree to create and solve trees
68Decision Trees Described
- Decision Tree is diagrammatic representation of
decision situation - Value of decision trees
- Help decision maker develop clear understanding
of structure of problem - Make it easier to determine possible scenarios
that can result if particular course of action
chosen - Help decision maker judge nature of information
needed for solving given problem - Help decision maker identify alternatives that
maximize EMV - Serve as excellent communication medium
69Example Decision Tree Drill or Dont Drill
- From left to right
- Typically start with a decision to be made
- Proceed to other decisions or chance events in
- chronological order
70Conventions on Decision Tree
- Decision Nodes
- Represented by square ?
- Point at which we have control
- and must make a choice
- Assigned sequential numbers (D1 here)
- May be followed by another decision node
- or chance node
- Branches emanating from square called
- decision forks, correspond to choices
- available
71Conventions on Decision Tree
- Chance Nodes
- Represented by circle o, numbered
- sequentially (C1 in example)
- Point at which we have no control,
- chance determines outcome
- Chance event probabilistic
- May be followed by series of decision
- nodes or chance nodes
- Branches emanating from circle called
- chance forks, represent possible outcomes
72Conventions on Decision Tree
- Probability or chance
- Likelihood of possible outcomes happening
- End, terminal, or payoff node
- Payoff deterministic financial outcome of
decision - Node represented by triangle (not on example)
- Has no branches following, returns payoff and
probability for associated path
73Guidelines for Designing Trees
- Tree construction is iterative we can change
our minds as we learn more - We should keep trees as simple as possible
- Define decision nodes so we can choose only one
option (but we should describe every option)
74Guidelines for Designing Trees
- We should design chance nodes so they are
mutually exclusive and collectively exhaustive - Tree should proceed chronologically from left to
right - Sum of probabilities should equal one at each
chance node - Remember that often we can draw a tree in number
of different ways that look different but that
are structurally equivalent
75Solving Decision Trees
- Decision analysis on tree can produce expected
value of model, standard deviation, and risk
profile of optimum strategy - Method of calculating optimum path called folding
back or rolling back tree - Solve from right to left consider later
decisions first
76Solving Decision Trees
- Chance node reduction
- Calculate expected value of rightmost chance
nodes and reduce to single event - Decision node reduction
- Choose optimal path of rightmost decision nodes
and reduce to single event (choose maximum ECi
at decision node) - Repeat
- Repeat procedure until you arrive at final,
leftmost, decision node
77Example Decision Tree
- Aggie Oil Company wants to decide whether to
drill new prospect - Geologists and engineers expect
- Probability of dry hole 60, NPV -65M
- Probability of 60M STB 30, NPV 120M
- Probability of 90M STB 10, NPV 180M
78Example Decision Tree
79Example Decision Tree
80Another Example Decision Tree
- Aggie Oil Company plans to drill a well, wants to
determine EMV of drilling - 35 chance of dry hole, NPV -65M
- 65 chance of producer if successful
- 60 chance of 30M STB, NPV 60M
- 30 chance of 60M STB, NPV 120M
- 10 chance of 90M STB, NPV 180M
81Another Example Decision Tree
82Alternative Approach Collapse Tree
83Alternative Approach Collapse Tree
Too much collapsing obscures important detail
not recommended
84Constructing Risk Profiles
- Risk profile is distribution function describing
chance associated with every possible outcome of
decision model - Steps to generate risk profile
- Reduce chance nodes (collapse tree)
- Reduce decision nodes consider only optimal
branches
85Steps in Constructing Risk Profiles
- Repeat steps 1 and 2 until tree is reduced to
single chance node with set of values and
corresponding probabilities - Generate risk profile
- Final set of payoff and probability pairs defines
discrete probability distribution used to
generate risk profile - Can graph risk profile as discrete cumulative
density distribution or scatter diagram
86Steps in Constructing Risk Profile
- 5. Calculate expected value, variance, and
standard deviation, as in example
87Spreadsheet Applications
- Excel built-in functions simplify calculation of
EMV, variance, standard deviation - Palisades PrecisionTree assists us in
constructing and solving decision trees
88Excel SUMPRODUCT Function
89PrecisionTree
- Part of Palisades suite
- Add-in to Microsoft Excel
- Allows us to create and solve decision trees in
Excel - Also capable of performing sensitivity analysis,
displaying results as spider graphs and tornado
charts
90Running Precision Tree
- Pages 212 to 226 of Mian, Vol. II, serve as a
tutorial for using PrecisionTree to create and
solve decision trees - Be sure that you can reproduce Examples 3-10,
3-11, and 3-12
91Learning Objectives
- You can now
- Construct a decision tree
- Solve a decision tree to determine optimal
decision and payoff - Calculate expected value, variance, and standard
deviation for optimal outcomes from decision tree - After study and practice, you will be able to
- Use Excel to simplify arithmetic in value,
variance, standard deviation calculations - Use PrecisionTree to create and solve trees
92Expected Value and Decision Trees