MODULE – 5

Risk & Decision Analysis - Simulation

01/18/2022 1
Risk & Decision Analysis - Simulation
• Expected Value (EV) computation also called “Monte Carlo
Simulation”
• Most popular management science method
• Computation with probability distribution
• Simulation = Monte Carlo Simulation
• Simulation, option of describing risk & uncertainty of input variables
in probability distribution e.g. CAPEX, OPEX, pay thickness
• Input probability distribution of variables into deterministic formulas
gives probability distribution of output e.g. NPV

01/18/2022 2
Simulation - Outline
• Simulation overview
• Simulation process, & descriptions of key techniques
• Random numbers
• Stopping rules
• Latin hypercube sampling
• Dependency
• Applications & examples
• Paper by Orodu et al. (2011)

01/18/2022 3
Simulation Overview of Methods
• Primary reason, to define distribution of outcome for uncertainty
analysis, including statistical mean (EMV or EV NPV) & other
measures of central tendency & dispersion
• EMV, common criterion for investment decisions
• NPV = f(oil price, CAPEX, OPEX, tax, royalty, discount rate, production
rate); these variables may be interrelated
• Deterministic values of independent variables yields a deterministic
NPV, hence deterministic evaluation.

01/18/2022 4
 Prob. Prob. Prob.

= × ×......

NPV i Po

• Input values commonly not certain at time of decision

• Hence distribution of possible values used for input values
• Outcome; NPV also a distribution

01/18/2022 5
 Why Simulation
• Conversion of probability distribution into probability
density distribution & represented by analytical
functions
• These analytical functions are then required to give
the density function.
• Analytical approach not feasible for economic
evaluation and petroleum risk.
• Use of mean values of the input distribution; mean of
dependent variable not equal to applying the mean of
the independent variables.
01/18/2022 6
Hence simulation is adopted, how:

• Series of scenarios corrected to generate NPV for each scenario.

• In each scenario values are selected from the distribution of each
input variable.
• Distribution created for the dependent variables.
• Each repetitive correlation is called a pass, simulation trial or a trial.
• least number of trials is 100, 1000 or more.
• Details, including required number or trials, sampling values of
distribution, defining distribution of input variables and relationship
between dependent variables

01/18/2022 7
ADVANTAGES OF APPROACH

• Risk and uncertainty described as a range and distribution of possible

values rather single point value (deterministic).
• Number of random variable isn’t limited.
• Distribution not limited to specific forms (normal, log-normal….)
• Expertise of different parameters can provide input on distribution.
• Computer programming of simulation proves it’s straight forward.
• Sensitivity analysis is possible and the most influential variables can
be determined

01/18/2022 8
SIMULATION PROCESS

• STEP 1: Define all the variables.

• STEP 2: Define the variable relationships in the deterministic projection
model “equation”.
• STEP 3: Sort the input variables into two groups.
• Deterministic variables
• Stochastic variables:
• sensitivity analysis is about prioritizing variables.
• Pareto’s principle top 3-6 chance variables account for 80-90% of the uncertainty.
φ = normal distribution; k =log-normal distribution
• STEP 4: Define distributions for the random stochastic variables:
some variables have associations, also called correlations and dependencies that have to be
considered.

01/18/2022 9
SIMULATION PROCESS Cont’d

• STEP 5: Perform the repetitive simulation trials to describe the distribution of

value.
• Sampling scheme
• Monte Carlo sampling, random numbers sampled from 0 to 1 e.g. RAND ()
function in Excel.

01/18/2022 10
SIMULATION PROCESS Cont’d

• STEP 6: Calculate expected value of the outcome value distribution.

EV NPV = EMV
• Uncertainty/risk analysis (NPV)
• Decision analysis (EMV/EV NPV)
Result presentation
• Greater than cumulative probability.
• Confidence interval.

01/18/2022 11
Random Numbers
• Unbiased way/method of sampling values from random variable
distributions.
• Random numbers sampling method is one of the methods for
sampling distributions.
• Are dimensionless, positive numbers sequence of numbers 0 to 1
• No pattern in order of numbers
• Each number equally likely to occur in the sequence
• Practical source of random number is by random number generator.
• An equation-technically, it is a pseudo-random number generator.
• Generates numbers that are uniformly distributed (no discoverable pattern).

01/18/2022 12
Random Numbers Cont’d
• Generator algorithm is standard component of programmable
tools.
• Spreadsheets e.g. Excel, RAND function.
• Good generator possesses the capability to set the STED or starting
value, this enables regeneration of same series of random numbers.
• Use of random numbers
• Example, with 5 random 1 1

variables as input into an

Random Number
equation, 5 different

Cumulative
Probability
random numbers are
required.

0 0
01/18/2022
Random Variable 13
Stopping Rules
• No specific laid down rules for number of trials required.
• For convergence, this depends on:
• Number of distribution sampled for each trial.
• Shape and magnitude of each distribution.
• Relationship amongst variables.
• Minimum number of trials is 100
• Maximum number of trials is 10,000 and more.

01/18/2022 14
Stopping Rules Cont’d
• Brute force methods
• Standard error of the mean

01/18/2022 15
Stopping Rules Cont’d
Brute Force
• Run 100 trials and compute mean.
• Change speed, run 100 trials and compute mean.
• Compute mean, if far apart, repeat step 1 and 2 with 200 trials.
• Compute mean of all runs

01/18/2022 16
Stopping Rules Cont’d
Standard Error of the Mean

S=sample standard deviation

n=trials

01/18/2022 17
LATIN HYPERCUBE SAMPLING (LHS)

• Traditional Monte Carlo sampling use random number generator to

determine sampling points.
• The random number generator should give a uniform distribution, but not so.

1
Frequency

0 0.5 1
Random Numbers
01/18/2022 18
LATIN HYPERCUBE SAMPLING (LHS) Cont’d

• LHS is a hybrid of conventional Monte Carlo sampling and uniform

sampling.
• Uniform sampling by uniformly spacing the sample points along the
cumulative frequency axis.
• LHS is 10 times better than the conventional Monte Carlo sampling.
• Standard-error-mean of 100 trials for LHS sampling is approximately
1000 trials of conventional Monte Carlo sampling.

01/18/2022 19
MONTE-CARLO SAMPLING
• Manual method
• graphical approach
• equation
• Graphical Approach
• Equation
• CDF of distribution stated as RN = f(x)
• Solving for x is a function of RN (inverse function of the CDF)

• Applicable for simple distribution

• Uniform distribution
• Triangular distribution
• Exponential distribution
• not for complex distribution such as Normal, log-Normal etc.
• Basically integration of PDF to set CDF & inverse function of CDF to
01/18/2022
get random variables 20
Basically integration PDF to set CDF & inverse function of CDF to get
random variables
• Triangular distribution sampling T(XL, XM, XH)
CDF: If RN ≤ (XM – XL) / (Xh– XL)
Inverse CDF: X = XL + √ (XM– XL) (Xh– XL)RN
CDF: If RN ≥ (XM – XL) / (Xh– XL)
Inverse CDF X = XH - √ (XH– XM) (Xh– XL)(1-RN)

01/18/2022 21
XL XM XH
• Uniform distribution simply U(Xmin,Xmax)
X = RN (XM AX – XMIN) + (XMIN)

xmin xmax

01/18/2022 22
• Exponential distribution sampling

PDF; f(x) = λe-λx

CDF; P{X≤x} = 1- e-λx
Inverse function CDF; X = In (1-RN) / (-λ)

1/λ = mean of distribution

01/18/2022 23
Excel Function for Simulation
• Use of excel in-built functions
• Fit-fit-purpose applications, possible using Excel’s built-in function in
macro language, Visual Basic for Applications (VBA).
Excel’s spread sheet functions also accessible in VBA
• Functions
• Random numbers
• RAND (); uniform random numbers generation (0 →1)
• EXCEL RAND()
• Recalculation; use F9

01/18/2022 24
Excel Function for Simulation
Probability Distribution – Inverse CDF
• =BETAINV (prob., α, β, A, B) - random
inverse of the cumulative probability function
• =CHINV (prob., degree-freedom)`
inverse one –tailed cumulative probability of chi-squared distribution
• =FINV (prob., degrees-freedom, degrees-freedom2)
inverse of F cumulative probability distribution
• = GAMMAINV (prob., α, β)
inverse of gamma cumulative probability distribution
• =LOGINV (prob., μ, σ)
inverse log-normal cumulative probability distribution
• =NORMINV ( prob., μ, σ)
Inverse normal cumulative probability distribution
use, e.g NORMINV(RAND(),µ,σ)
• = NORMSINV ( prob.)
inverse of standard normal cumulative probability distribution

01/18/2022 25
Further Excel statistical function
 
• PDF & CDF FUNCTIONS 
=BINOMDIST ()
=EXPONDIST ()
=HYGEOMDIST ()
=NEGBINOMDIST ()
=POISSON ()
=WEIBULL ()
 
• PDF & CDF; used with VLOOKUP () function to create random variables
e.g VLOOKUP (RAND(), look up-vector, result- vector)

01/18/2022 26
Dependency
• Total (full) Dependency
• Diffuse (partial) Dependency

01/18/2022 27
Total (Full) Dependency
• Cross-plot with dependence
• Correlation coefficient “r”, is used to access the
the relationship of a variable to the other
• r = ±1 or close to ± 1.0 sufficient for y=f(x)

 xy  xy
r 
 x y  xx yy

 xy    xi  x   yi  y   xx      xi  x 
2
x
2
01/18/2022 28
Diffuse (Partial) Dependency
• Dependency not well formed function
• For variable y, substantial range of values for x
• Hence y-distribution required for each trial of x
• “x” independent random variable for this example
• “y” dependent random variable
• How to sample dependent random variable?

01/18/2022 29
 Decision analysis for petroleum exploration” 2nd Ed. Newendorp & Schuyler
01/18/2022 30
 Decision analysis for petroleum exploration” 2nd Ed. Newendorp & Schuyler
01/18/2022 31
Dependency - Procedure
• Cross plot of random variable X & Y
• Draw envelope around data points
• Representative equations of upper bound & lower bound of envelope
• Equation: Linear or Curvi-linear
• Upper boundary function; Yupper (x)
• Lower boundary function ; Ylower (x)

01/18/2022 32
Dependency – Procedure Cont’d
• Identify shape of dependent variable’s distribution
• Normalize the distribution generated above for each x in x-axis
• Slightly different normalization for log-normal, normal..........

y  ymin
ynorm 
ymax  ymin

01/18/2022 33
Dependency – Procedure Cont’d
• Sample the independent variable X by Monte Carlo sampling or Latin
Hypercube Sampling
• gives x1 -trial value
• Trial value partially dependent variable
• sample normalized distribution f(y)
(include fig.8.19)
normalized value = yni= ynorm
• random variable y
• Y = Ylower(X1) + Yni{Yupper(X1) - Ylower(X1)}

01/18/2022 34
Dependency – Procedure Cont’d
• NON- IDEAL SITUATION
• Different shape of the dependent variables distribution w.r.t. “X” (horizontal
axis)
“density pattern of points y, y vs x in cross- section varies”
• e.g. mode of triangle may change
• function of shape versus x is necessary
• an alternative; different distribution at specified intervals of “x”
• correlation coefficient applicable to model partial dependency

01/18/2022 35
Software Applications
• @Risk by Palisade Corporation
• Oracle Crystal Ball
• Spreadsheet based software as Excel Add-in
• Functionality
• Monte Carlo Simulation
• Monte Carlo Sampling
• Latin Hypercube Sampling
• Dependency
• Sensitivity Analysis: Spider & Tornado plots, rank correlation
• Optimization
• Fit probability distribution to data
01/18/2022 36
Oracle Crystal Ball Application - Example

01/18/2022 37
 Valuation of Oil & Gas Properties
by
Empirical Production Forecast

Valuation Methods
•Actual cost or replacement value
•Sale of Similar Properties
•Capitalization of Income, Engineering Evaluation Method

Economic Analysis and Investment Decisions (Chi U. Ikoku, 1985)

01/18/2022 38
 Present Value of Declining Income
•Exponential production decline q  q e  Dt
i
•D=constant decline rate per
year
•q*=present value production
•(D+j)=exponential declining
production....instantaneous q*  qi e  Dt  e  jt  qi e  ( D  j )t
declining rate
•Q*D=total present value n qi
  
 ( D  j )t ( D  j )n
production Q *
D  q e dt  1  e
0 i (D  j)
•De=discount factor for an
exponentially declining volume, D Q  Q*
e D D
defined as the number with
which the undiscounted qi  1  e  ( D  j ) n 
cumulative production has to be QD 
multiplied to arrive at the D
present value. qi
1  e  ( D  j ) n 
Q* D  D  j    D 1  e ( D  j )n
De   
QD 
qi 1  e  Dn  D  j 1  e  Dn
D
n
De  D ( D  j )
01/18/2022 39
Profitability of a Development

• Prior to authority for expenditure (A.F.E) to drill for field development,

economic evaluation of prospect is carried out by profit indicators (NPV or IRR)
• Well performance for yet to be drilled well obtaining from surrounding wells,
including recoverable reserves.
• Lease Block
• Associated cost prior to drilling (outcome of either dry well or producer)
• Lease bonus, seismic (geological & geophysical), rentals, drilling costs
• Further associated cost for producer
• Well completion, lease royalty, state & local taxes, OPEX, federal income taxes
• Above used for economic analysis

01/18/2022 40
 qi=initial unrestricted flow rate qi  DQ
Q=recoverable reserves

qc=(maximum efficient rate) qi  qc

allowable production rate, prior to tc 
Dqc
production rate decline (for a
period of tc) and declines to
abandonment production rate qa at qi
ta

tc=constant production period

qc qc e  Dt
tD=decline production period

qa
tc ta
td

01/18/2022 41
 Present Value PVOC  Co ( PCFoc )
PVOC=present value of operating cost
(1  e  jta )
(cumulative)  Co
Co=operating cost per year j
PCF= Present Cost Factor

PVRc=net revenue of production for

constant production period PVRc  qc (1  e  jtc )
PVRd=net revenue of production for q
exponentially declining production PVRd  c e  jtc 1  e  ( D  j ) td 
period D j
PVR=present value of composite
production
PVR  PVRc  PVRd
qc   j   jtc 
  1  e  jtc

  e 
1  e  ( D  j ) td  
j   D j 

01/18/2022 42
 PVF: Composite Present
Value Factor
PVR=(Reserves)(PVF)
qc qi  qc qc
Recoverable Reserves (Q) Q  qc tc 
D

1 e  Dtd

D
 
D
1  e Dtd  
qi  qc  Dtd  qi  qa 
 1  e   1  
D  qi  D qi 
qa qi 1.0
PVR=Q (PVF) qi
Q 
D
qi
PVR  ( PVF )
D
 qc  D   
PVF  
 qi
  
j
   
1  e 
 jtd

 j

 D
  jtc

j
e 1 
 e
  D  j  td
 


no period of prorated
 D 
production; qc/qi=1 ,tc=0 PVF   
 1 e
 ( D  j ) td

 D j
01/18/2022 43 43
 Example
Consider the case of one well in an oil lease block. The reservoir data are as follows:
Lease area, A = 160 acres
Net Pay thickness, h = 100ft
Recovery, ER= 80bbl/acre-ft
Decline rate, D= 25%/year
MER rate, Qc= 500 bbl/day [(maximum efficient rate) allowable production rate, prior to production rate decline]
Working interest = 100%
Royalty = 12%
State & Local taxes = 7%
Selling price of crude oil = $30/bbl
Operating cost = $500/month
Drilling & completion costs = $2,000,000
Calculate on a before tax basis!
(a) Net present value of oil to be produced at discovery time.
(b) Net present value at decision time if there is a Six (6) month lag from decision time
to discovery time.

Hint: Production rate at economic limit is linked to operating cost.

01/18/2022 44
 Short cut profitability methods

• Approximate methods to
previous solution for qi
simplicity.
• Solution for this figure, during
period of constant or qc qc e  Dt
allowable production at mid-
year discounting, present
value of allowable production qa
(total)=Q*c tc ta
td

qc=allowable production rate bbl/year   1  i  tc  1 

v=net income per barrel, $/bbl Qc*  qc (v)  
 i  1  i 
i=discount rate, %/year tc  0.5
tc=time of allowable production, years

01/18/2022 45
 Exponential Decline Period
Q D=present value of the declining
*   1  i  0.5  r td 
 td  0.5 
production at the start of the decline   1  i  
period, $ Q* D  QI  
1 i  r
 
 
QI= income during the first year of
decline $  1 r 
QI  QD (v)  t 
1  r d 
r= ratio of production rates of
successive years q1 q2 q
td= number of years of production on   t  e D  r
qi q1 qt 1
exponential decline

QD= amount of recoverable reserves

qi  1  e  Dt 
during the declining production
qi  q
Q*=composite present value QD  
production at discovery time D D
Q* D
Q Q c 
* *

1 i c
t
01/18/2022 46 46
 Example

A lease producing at an allowable rate yields a net income of

$300,000 per year. This lease has a 10-year period of exponential
decline following its five years of allowable production. The lease
will be abandoned when the net income falls to $30,000 per year.
Calculate the net present value of the production from this lease
at discovery time if the discount rate is 15% per year.

01/18/2022 47
 Graphical Techniques
(Approximate Calculation of Profit Indicators)

Brons Nomograph (Exponential and Hyperbolic Decline)

Brons & Sibergh Nomographs

Relates total present profit, discounted cash flow rate of return or internal rate
of return, and payout time for three different income patterns, constant rate,
exponentially declining and hyperbolically declining (b=0.5). These nomographs
are approximate solutions and so does not replace more accurate calculations.

01/18/2022 48