SlideShare une entreprise Scribd logo
1  sur  38
Télécharger pour lire hors ligne
Advanced Econometrics: Doctoral Course
Marco Delogu
Monte Carlo Experiments
Intro to MonteCarlo Experiment 2
The last topic that I want to cover with you is Montecarlo
Simulation in STATA.
What is a Simulation by Monte Carlo? In broad terms is a
useful and powerful methodology for investigating the
properties of economic estimators and tests.
The power comes from the fact that we can define the
characteristics of the Data Generating Process (DGP) and thus
study its properties like if we were in a laboratory.
Monte Carlo experiments employ pseudo random numbers that
mimic the theoretical properties of realizations of random
variables
Subject of the lecture 3
We learn how to generate pseudo random numbers using Stata.
Then we learn how to create simple programs in STATA. Then
we set Monte Carlo experiment to verify
1. the Weak Law of Large Numbers
2. the Central Limit Theorem
3. Properties of the OLS estimator
After this lecture you will learn how to use some other very
important STATA commands:
1. seed to set the pseudo random generator (important to
reproduce your findings)
2. postfile
3. program and simulate
Generating Random Numbers (1) 4
I am quite sure that you have already encountered the Weak
Law of Large Numbers (If you are interested in the
mathematical proof please let me know). I report the definition
in the next slides. Now we prove the Theorem using the Monte
Carlo method.
First we need to set the seed to generate random numbers in
STATA
set seed 12345
Pseudorandom-number generators use deterministic devices to
produce long chains numbers, with the Seed we can reproduce
the generation of our random numbers in any other pc with
STATA installed (same version).
Law of Large numbers 5
Let f (., θ) the density of a given random variable X. We don’t
know the exact value of θ but we assume that the random
variable X follows the distribution f(, ).
Question: can a random sample give us reliable information
about the value θ?
Yes! We can make this claim thanks to the Weak Law of
Large Numbers. This theorem allows to make reliable
inference about the unknown parameter θ. Stated differently:
Denote by µ the expected value E [X] of X. Using only a finite
number values of X we can make inference about the value of
E (X) because the LLN allows us to construct confidence
intervals.
Weak Law of Large Numbers 6
Let’s state in words the claim of the Weak Law of Large
numbers. A positive integer n can be determined such that if a
random sample of size n, or larger, is draw the probability that
the sample mean ¯X will deviate from µ less than a previously
specified quantity can be made as much close to 1 as desired.
Let’s try to state the claim using the mathematical language.
Taken two numbers ε > 0 and 0 < δ < 1 there exists an integer
n such that if a random sample of size n is taken the sample
mean ¯Xn will deviate from µ by less than ε with probability
larger than 1 − δ. In symbols:
For any ε > 0 and 0 < δ < 1 there exists an integer n such that
for any m ≥ n
P | ¯Xm − µ| < ε ≥ 1 − δ
Weak Law of Large numbers 7
Theorem Weak Law of Large numbers
Let f (.) be a density with mean µ and finite variance σ2. Let
¯Xn be the sample mean of a random sample of size n.
Let ε and δ be any two specified numbers satisfying ε > 0 and
0 < δ < 1.
If n is any integer greater than
σ2
ε2δ
Then
P −ε < ¯Xn − µ < ε ≥ 1 − δ
Generating Random Numbers (2) 8
Before proving the LLN through a Monte Carlo experiment we
need to generate some random numbers. Remind that we
already generate numbers during our previous lecture (go back
to the first set of slides). Imagine that we want to generate 30
observations draw from the uniform distribution.. it is very easy
in stata
clear all
set seed 123456
set obs 30
generate x=runiform()
As you can check in STATA now you have 30 observations
stored in the variable x which are independents draws from the
Uniform distirbution.
Coin tosses (1) 9
Consider a random variable that takes value 1 with probability
1
2 and 0 with probability 1
2 . Let’s generate 30, indipendent,
random variables of such type:
clear
set obs 30
quietly generate y = runiform()
quietly generate x=1 if (y > 0.5)
quietly replace x=0 if (y <= 0.5)
You can check that x takes either value 1 or value 0.
Coin Tosses (2) 10
You can check that x takes either value 1 or value 0. Notice
that, the theoretical mean and variance of x are given by:
E[X] = 1 ∗ 1
2 + 0 ∗ 1
2 = 1
2
and V [X] = E X2 − E [X]2
= 12 ∗ 1
2 + 0 ∗ 1
2 − 1
2
2
=
1
2 − 1
4 = 1
4
If you are in trouble understanding please let me know. Remind
that, in general terms the weak law of large numbers tells you
that the empirical mean approaches the theoretical one as long
as we increase the sample size. Namely, as long as we increase
the number of coin tosses the empirical mean shall converge to 1
2
Postfile 11
Postfile command: declares the memory object in which the
results are stored, the names of the variables in the result
dataset and the name of the results dataset file. In our example
we write:
postfile sim mean1 trials using sim1000, replace
Everyting will be clearer shortly. We generate a new dataset
called sim1000 where we store the results of our experiments.
Simulation 12
forvalues i = 1/100000 {
quietly capture drop y x
quietly set obs ‘i’
quietly generate y =runiform()
quietly generate x=1 if (y > 0.5)
quietly replace x=0 if (y <= 0.5)
summarize x
quietly post sim (r(mean)) (r(N))
}
postclose sim clear
use sim1000
forvalues 13
forvalues: i = 1/100000
means that Stata repeats the instructions between the curly
brackets 10000 times
In our example for every i we construct a new dataset that is
made by i observations.
Finally:
quietly post sim (r(mean)) (r(N))
saves in our external dataset, for each simulation trial, both
value of the mean and the number of observations.
first example 14
Weak Law of Large Numbers 15
The law of large numbers tells us that knowing we can
determine a confidence interval such that for a given sample size
the empirical mean belongs to an interval centered at the
theoretical mean with a certain probability. Setting interval
equal to 0.1 and the confidence interval equal to 95% and
taking advantage of the theorem
1
4
1
(0.1)2
(0.05)
= 500
So the theorem tells that at least the 95% of the means shall
fall in this range....Let’s check making another experiment
Code 16
forvalues i = 1/1000 {
quietly capture
drop y x
quietly set obs 500
quietly generate y = runiform()
quietly generate x=1 if (y>0.5)
quietly replace x=0 if (y<=0.5)
summarize x
quietly post sim (r(mean))
} postclose sim
Check 17
clear
use weaklaw2.dta
* here we generate one variable that take gen test=1 if
((mean1>0.6) — (mean1<0.4))
summarize test
gen obs1 =n
scatter mean1 obs1
Weak Law 2nd graph 18
Weak Law of Large number 19
Please refer to the dofile for practical clarifications.
In our experiment we could not find any experiment for which
the sample mean does not belong to the bounds predicted by
the theorem.
Therefore, our experiments does not reject the theorem!
To check that is not artifact change number of observations
(trial) and make some experiments by yourself.
Central Limit Theorem (1) 20
Theorem: Central-limit theorem: Let f (.) be a density
with mean µand finite variance σ2. Let ¯Xn be the sample mean
of a random sample of size n from f (.). Let the random
variable Zn be defined by
Zn =
¯Xn − E ¯Xn
var ¯Xn
=
¯Xn − µ
σ√
n
The distribution of Zn approaches the standard normal
distribution as n approaches to infinity
Central Limit theorem (2) 21
The CLT says that the limiting distribution of Zn which is ¯Xn
standardized is the standard normal distribution. Namely, this very
important theorem tells us that ¯Xn itself is approximately, or asymptotically
distributed as a normal distribution with mean µ and variance σ2
n
Nothing is assumed about the form of the original density function
The degree of approximation depends on the sample size and on the
particular density f(.)
The idea is that you take many samples of size n. Then the central limit
theorem shows that the distribution of the mean of such samples will
approximate to a normal as the sample size increases
Homework: determine the exact distribution of the sample mean for a
sample from a Poisson distribution (you should know what is a Poisson
distribution, and its properties, feel free to come back to me to have
references that will help you to solve the exercises)
CLT and LLN 22
From Angrist Pischke book (Mastering Metrics):
How can we decide wether statistical result constitute strong
evidence or merely a lucky draw, unlikely to be replicated in
repeated samples?
LLN and CLT are the foundations of formal statistical
inference. As Angrist and Pischke write:
Quantifying sampling uncertainty is a necessary step in any
empirical projects and on the road to understanding statistical
claims made by others
Central Limit Theorem. Simulation 23
Now we try to prove the CLT by means of a Monte Carlo
Experiment:
Central Limit theorem: the sample mean is approximately
normally distribute as N approaches infinity. Let consider a
random variable that has the uniform distribution and a sample
size of 30. Let’s check that when we draw a single sample of
uniform draws the graph that we obtain as nothing to do with a
bell shaped curve:
quietly set obs 30
set seed 10101
generate x=runiform()
summarize x
quietly histogram x, width(0.1)
xtitle(”x from one sample”)
Recap: Uniform Distribution (1) 24
A random variable X has with values a ≤ X ≤ b has a unform
distirbution if its probability density function is given by
f (x|a, b) =
1
b − a
for a ≤ x ≤ b and zero otherwise
now, let a = 0 and b = 1
E [X] =
1
0
x
1
1 − 0
dx =
x2
2
1
0
=
1
2
V (X) =
1
0
x2 1
1 − 0
dx −
1
2
2
=
x3
3
1
0
−
1
4
=
1
3
−
1
4
=
1
12
Recap: Uniform Distribution (2) 25
Program 26
We will do simulation but using a different command. In place
of postfile we write a program. Notice that we put rclass. This
tell STATA that we want that the program return belongs to
the rclass .
program onesample, rclass
drop all
quietly set obs 30
generate x=runiform()
summarize x
return scalar meanforonesample=r(mean)
end
onesample is our first user written program. It generates 30
observations drawn fro the uniform distribution. It returns the
mean of this sample as an rclass object.
Simulate 27
set seed 10101
onesample
return list
We employ the command simulate to run the program as many
time as we wish. Understand that our r.v. has expected value
equal to 0.5 and s.d. equal to 0.0527. Each round we obtain a
new value of the sample mean stored in the xbar variable.
simulate xbar=r(meanforonesample), seed(10101)
reps(1000) nodots:onesample
Reps tells STATA how many simulations we want to perform.
In our case is 1000. We can check now our experiment.
summarize xbar quietly histogram xbar, normal xtitle(”xbar
from many samples”)
Central Limit Theorem (Graph) 28
OLS regression 29
When we do OLS we employ the OLS method to determine
estimates of the True parameters.
Monte Carlo experiment can be used to estimate the reliability
of our estimators.
Let’s apply the MonteCarlo methodology to check some of the
well known OLS properties.
We verify that when the DGP meets the OLS assumptions the
OLS estimates are unbiased.
So what we do? We define a DGP that meets OLS assumptions
and then we veritfy that OLS estimator will have all that good
properties that makes it so widespread used!
Unbiasdness 30
Unbiasdness of the Sample Mean:
E ¯Yn = E [Yi] = µ
What it means: The sample mean should not be expected to
bang on the corresponding population mean: the sample
average in one sample can be too big; while in another sample
can be too small.
Unbiasdness tells us that deviations are not systematically up
or down; rather; in average samples they average out to zero.
Unbiasdness is distinct from the LLN, which says that the
sample mean gets closer and closer to the population mean as
the sample size grows. Unbiasdness of the sample mean holds
for samples of any size, for any n!
Model 31
Let assume that we have some data such that:
E [Y |X] = 100 + 10 ∗ 10 for the first 20 observations
and
E [Y |X] = 100 + 10 ∗ 20 for the remaining 20 observations
We assume that the error term follows a Normal distribution
with mean zero and variance equal to 2500, N (0, 50), where 50
is the standard deviation.
This equation define the theoretical model of our process. In
the real world once we do OLS we just have 1 sample! and we
want to estimate both the parameters, in our case equal to 100,
b1 and 10, b2
Setting the experiment 32
Before to define the simulation we save in global macros the key
input of our DGP.
clear all
global numobs 40 // sample size
global beta1 100 // intercept parameter
global beta2 10 // slope parameter
global sigma 50 // standard error
Now we can define the DGP
Data Generating Process 33
set seed 1234567
set obs $numobs
generate x=10
replace x=20 if n > $numobs/2
generate y = $beta1 + $beta2*x + rnormal(0,$sigma)
Now, given our data we can estimate the parameters through
OLS regression
regress y x
Notice that we do not get neither 1000 nor 10. The result of
one regression is not enough to say something about
the OLS properties!
OLS experiment 34
Look at the program in the dofile. we store for each experiment
the estimate of:
intercept
slope
variance.
return scalar b2 = b[x]
return scalar b1 = b[ cons]
return scalar sig2 = (e(rmse))2
Making a series of experiment 35
Taking advantage of STATA we can generate as many samples
of the DGP process. Then we can evaluate whether OLS
estimates are good or bad:
simulate b1r=r(b1) b2r=r(b2) sig2r=r(sig2), reps (1000) nodots
nolegend seed(1234567): regression1
We have just run 1000 regression. We can compare the
average of our estimates with the actual values.
OLS works! (b1) 36
OLS works! (b2) 37
Exercise 38
Consider a biased coin, probability of getting head =0.3.
Define and perform a MonteCarlo Experimento to obtain a
reliable estimate of the empirical mean
Consider a random variable that follows a Poisson
distribution with mean 1.5 (rpoisson). Determine the
expected value analytically (1.5) and by means of a
MonteCarlo Experiment evaluate the distribution of the
empirical mean considering 10000 samples for each one of
the following 3 cases: sample size=5, sample size=15,
sample size=50. Discuss your result in light of the CLT
consider the OLS simulation. Discuss how the empirical
estimate of V(b2), from the MonteCarlo experiment, differ
from the actual one. What happens when n increases?

Contenu connexe

Tendances

Introduction to random variables
Introduction to random variablesIntroduction to random variables
Introduction to random variables
Hadley Wickham
 
Chapter 2 discrete_random_variable_2009
Chapter 2 discrete_random_variable_2009Chapter 2 discrete_random_variable_2009
Chapter 2 discrete_random_variable_2009
ayimsevenfold
 
Solution of non-linear equations
Solution of non-linear equationsSolution of non-linear equations
Solution of non-linear equations
ZunAib Ali
 

Tendances (18)

Talk 2
Talk 2Talk 2
Talk 2
 
Assignment 2 solution acs
Assignment 2 solution acsAssignment 2 solution acs
Assignment 2 solution acs
 
Statistics Coursework Help
Statistics Coursework HelpStatistics Coursework Help
Statistics Coursework Help
 
Es272 ch3a
Es272 ch3aEs272 ch3a
Es272 ch3a
 
Math Exam Help
Math Exam HelpMath Exam Help
Math Exam Help
 
Introduction to random variables
Introduction to random variablesIntroduction to random variables
Introduction to random variables
 
Lecture 04 newton-raphson, secant method etc
Lecture 04 newton-raphson, secant method etcLecture 04 newton-raphson, secant method etc
Lecture 04 newton-raphson, secant method etc
 
Chapter 2 discrete_random_variable_2009
Chapter 2 discrete_random_variable_2009Chapter 2 discrete_random_variable_2009
Chapter 2 discrete_random_variable_2009
 
Welcome to International Journal of Engineering Research and Development (IJERD)
Welcome to International Journal of Engineering Research and Development (IJERD)Welcome to International Journal of Engineering Research and Development (IJERD)
Welcome to International Journal of Engineering Research and Development (IJERD)
 
Probability Assignment Help
Probability Assignment HelpProbability Assignment Help
Probability Assignment Help
 
2019 PMED Spring Course - SMARTs-Part II - Eric Laber, April 10, 2019
2019 PMED Spring Course - SMARTs-Part II - Eric Laber, April 10, 2019 2019 PMED Spring Course - SMARTs-Part II - Eric Laber, April 10, 2019
2019 PMED Spring Course - SMARTs-Part II - Eric Laber, April 10, 2019
 
Discrete Random Variables And Probability Distributions
Discrete Random Variables And Probability DistributionsDiscrete Random Variables And Probability Distributions
Discrete Random Variables And Probability Distributions
 
Solution of non-linear equations
Solution of non-linear equationsSolution of non-linear equations
Solution of non-linear equations
 
Numerical Method 2
Numerical Method 2Numerical Method 2
Numerical Method 2
 
Chap05 continuous random variables and probability distributions
Chap05 continuous random variables and probability distributionsChap05 continuous random variables and probability distributions
Chap05 continuous random variables and probability distributions
 
Into to prob_prog_hari (2)
Into to prob_prog_hari (2)Into to prob_prog_hari (2)
Into to prob_prog_hari (2)
 
2019 PMED Spring Course - Introduction to Nonsmooth Inference - Eric Laber, A...
2019 PMED Spring Course - Introduction to Nonsmooth Inference - Eric Laber, A...2019 PMED Spring Course - Introduction to Nonsmooth Inference - Eric Laber, A...
2019 PMED Spring Course - Introduction to Nonsmooth Inference - Eric Laber, A...
 
2 random variables notes 2p3
2 random variables notes 2p32 random variables notes 2p3
2 random variables notes 2p3
 

Similaire à Montecarlophd

Applications to Central Limit Theorem and Law of Large Numbers
Applications to Central Limit Theorem and Law of Large NumbersApplications to Central Limit Theorem and Law of Large Numbers
Applications to Central Limit Theorem and Law of Large Numbers
University of Salerno
 
law of large number and central limit theorem
 law of large number and central limit theorem law of large number and central limit theorem
law of large number and central limit theorem
lovemucheca
 
Introduction to Bootstrap and elements of Markov Chains
Introduction to Bootstrap and elements of Markov ChainsIntroduction to Bootstrap and elements of Markov Chains
Introduction to Bootstrap and elements of Markov Chains
University of Salerno
 

Similaire à Montecarlophd (20)

Applications to Central Limit Theorem and Law of Large Numbers
Applications to Central Limit Theorem and Law of Large NumbersApplications to Central Limit Theorem and Law of Large Numbers
Applications to Central Limit Theorem and Law of Large Numbers
 
Advanced Econometrics L5-6.pptx
Advanced Econometrics L5-6.pptxAdvanced Econometrics L5-6.pptx
Advanced Econometrics L5-6.pptx
 
Talk 3
Talk 3Talk 3
Talk 3
 
A bit about мcmc
A bit about мcmcA bit about мcmc
A bit about мcmc
 
8 random variable
8 random variable8 random variable
8 random variable
 
probability assignment help (2)
probability assignment help (2)probability assignment help (2)
probability assignment help (2)
 
Talk 5
Talk 5Talk 5
Talk 5
 
random variation 9473 by jaideep.ppt
random variation 9473 by jaideep.pptrandom variation 9473 by jaideep.ppt
random variation 9473 by jaideep.ppt
 
Excel Homework Help
Excel Homework HelpExcel Homework Help
Excel Homework Help
 
Lecture: Monte Carlo Methods
Lecture: Monte Carlo MethodsLecture: Monte Carlo Methods
Lecture: Monte Carlo Methods
 
Prob distros
Prob distrosProb distros
Prob distros
 
Mayo Slides: Part I Meeting #2 (Phil 6334/Econ 6614)
Mayo Slides: Part I Meeting #2 (Phil 6334/Econ 6614)Mayo Slides: Part I Meeting #2 (Phil 6334/Econ 6614)
Mayo Slides: Part I Meeting #2 (Phil 6334/Econ 6614)
 
1249320870000 asgn 1-jm (1)
1249320870000 asgn 1-jm (1)1249320870000 asgn 1-jm (1)
1249320870000 asgn 1-jm (1)
 
law of large number and central limit theorem
 law of large number and central limit theorem law of large number and central limit theorem
law of large number and central limit theorem
 
Statistics Homework Help
Statistics Homework HelpStatistics Homework Help
Statistics Homework Help
 
DSP_FOEHU - MATLAB 02 - The Discrete-time Fourier Analysis
DSP_FOEHU - MATLAB 02 - The Discrete-time Fourier AnalysisDSP_FOEHU - MATLAB 02 - The Discrete-time Fourier Analysis
DSP_FOEHU - MATLAB 02 - The Discrete-time Fourier Analysis
 
Introduction to Bootstrap and elements of Markov Chains
Introduction to Bootstrap and elements of Markov ChainsIntroduction to Bootstrap and elements of Markov Chains
Introduction to Bootstrap and elements of Markov Chains
 
The axiomatic power of Kolmogorov complexity
The axiomatic power of Kolmogorov complexity The axiomatic power of Kolmogorov complexity
The axiomatic power of Kolmogorov complexity
 
Lec. 10: Making Assumptions of Missing data
Lec. 10: Making Assumptions of Missing dataLec. 10: Making Assumptions of Missing data
Lec. 10: Making Assumptions of Missing data
 
Chapter-6-Random Variables & Probability distributions-3.doc
Chapter-6-Random Variables & Probability distributions-3.docChapter-6-Random Variables & Probability distributions-3.doc
Chapter-6-Random Variables & Probability distributions-3.doc
 

Dernier

MASTERING FOREX: STRATEGIES FOR SUCCESS.pdf
MASTERING FOREX: STRATEGIES FOR SUCCESS.pdfMASTERING FOREX: STRATEGIES FOR SUCCESS.pdf
MASTERING FOREX: STRATEGIES FOR SUCCESS.pdf
Cocity Enterprises
 
Call Girls in Tilak Nagar (delhi) call me [🔝9953056974🔝] escort service 24X7
Call Girls in Tilak Nagar (delhi) call me [🔝9953056974🔝] escort service 24X7Call Girls in Tilak Nagar (delhi) call me [🔝9953056974🔝] escort service 24X7
Call Girls in Tilak Nagar (delhi) call me [🔝9953056974🔝] escort service 24X7
9953056974 Low Rate Call Girls In Saket, Delhi NCR
 

Dernier (20)

Pension dashboards forum 1 May 2024 (1).pdf
Pension dashboards forum 1 May 2024 (1).pdfPension dashboards forum 1 May 2024 (1).pdf
Pension dashboards forum 1 May 2024 (1).pdf
 
7 tips trading Deriv Accumulator Options
7 tips trading Deriv Accumulator Options7 tips trading Deriv Accumulator Options
7 tips trading Deriv Accumulator Options
 
Webinar on E-Invoicing for Fintech Belgium
Webinar on E-Invoicing for Fintech BelgiumWebinar on E-Invoicing for Fintech Belgium
Webinar on E-Invoicing for Fintech Belgium
 
2999,Vashi Fantastic Ellete Call Girls📞📞9833754194 CBD Belapur Genuine Call G...
2999,Vashi Fantastic Ellete Call Girls📞📞9833754194 CBD Belapur Genuine Call G...2999,Vashi Fantastic Ellete Call Girls📞📞9833754194 CBD Belapur Genuine Call G...
2999,Vashi Fantastic Ellete Call Girls📞📞9833754194 CBD Belapur Genuine Call G...
 
Vip Call Girls Rasulgada😉 Bhubaneswar 9777949614 Housewife Call Girls Servic...
Vip Call Girls Rasulgada😉  Bhubaneswar 9777949614 Housewife Call Girls Servic...Vip Call Girls Rasulgada😉  Bhubaneswar 9777949614 Housewife Call Girls Servic...
Vip Call Girls Rasulgada😉 Bhubaneswar 9777949614 Housewife Call Girls Servic...
 
Q1 2024 Conference Call Presentation vF.pdf
Q1 2024 Conference Call Presentation vF.pdfQ1 2024 Conference Call Presentation vF.pdf
Q1 2024 Conference Call Presentation vF.pdf
 
CBD Belapur((Thane)) Charming Call Girls📞❤9833754194 Kamothe Beautiful Call G...
CBD Belapur((Thane)) Charming Call Girls📞❤9833754194 Kamothe Beautiful Call G...CBD Belapur((Thane)) Charming Call Girls📞❤9833754194 Kamothe Beautiful Call G...
CBD Belapur((Thane)) Charming Call Girls📞❤9833754194 Kamothe Beautiful Call G...
 
Turbhe Fantastic Escorts📞📞9833754194 Kopar Khairane Marathi Call Girls-Kopar ...
Turbhe Fantastic Escorts📞📞9833754194 Kopar Khairane Marathi Call Girls-Kopar ...Turbhe Fantastic Escorts📞📞9833754194 Kopar Khairane Marathi Call Girls-Kopar ...
Turbhe Fantastic Escorts📞📞9833754194 Kopar Khairane Marathi Call Girls-Kopar ...
 
Thane Call Girls , 07506202331 Kalyan Call Girls
Thane Call Girls , 07506202331 Kalyan Call GirlsThane Call Girls , 07506202331 Kalyan Call Girls
Thane Call Girls , 07506202331 Kalyan Call Girls
 
Mahendragarh Escorts 🥰 8617370543 Call Girls Offer VIP Hot Girls
Mahendragarh Escorts 🥰 8617370543 Call Girls Offer VIP Hot GirlsMahendragarh Escorts 🥰 8617370543 Call Girls Offer VIP Hot Girls
Mahendragarh Escorts 🥰 8617370543 Call Girls Offer VIP Hot Girls
 
W.D. Gann Theory Complete Information.pdf
W.D. Gann Theory Complete Information.pdfW.D. Gann Theory Complete Information.pdf
W.D. Gann Theory Complete Information.pdf
 
Benefits & Risk Of Stock Loans
Benefits & Risk Of Stock LoansBenefits & Risk Of Stock Loans
Benefits & Risk Of Stock Loans
 
Explore Dual Citizenship in Africa | Citizenship Benefits & Requirements
Explore Dual Citizenship in Africa | Citizenship Benefits & RequirementsExplore Dual Citizenship in Africa | Citizenship Benefits & Requirements
Explore Dual Citizenship in Africa | Citizenship Benefits & Requirements
 
MASTERING FOREX: STRATEGIES FOR SUCCESS.pdf
MASTERING FOREX: STRATEGIES FOR SUCCESS.pdfMASTERING FOREX: STRATEGIES FOR SUCCESS.pdf
MASTERING FOREX: STRATEGIES FOR SUCCESS.pdf
 
Significant AI Trends for the Financial Industry in 2024 and How to Utilize Them
Significant AI Trends for the Financial Industry in 2024 and How to Utilize ThemSignificant AI Trends for the Financial Industry in 2024 and How to Utilize Them
Significant AI Trends for the Financial Industry in 2024 and How to Utilize Them
 
GIFT City Overview India's Gateway to Global Finance
GIFT City Overview  India's Gateway to Global FinanceGIFT City Overview  India's Gateway to Global Finance
GIFT City Overview India's Gateway to Global Finance
 
Kopar Khairane Cheapest Call Girls✔✔✔9833754194 Nerul Premium Call Girls-Navi...
Kopar Khairane Cheapest Call Girls✔✔✔9833754194 Nerul Premium Call Girls-Navi...Kopar Khairane Cheapest Call Girls✔✔✔9833754194 Nerul Premium Call Girls-Navi...
Kopar Khairane Cheapest Call Girls✔✔✔9833754194 Nerul Premium Call Girls-Navi...
 
Call Girls in Tilak Nagar (delhi) call me [🔝9953056974🔝] escort service 24X7
Call Girls in Tilak Nagar (delhi) call me [🔝9953056974🔝] escort service 24X7Call Girls in Tilak Nagar (delhi) call me [🔝9953056974🔝] escort service 24X7
Call Girls in Tilak Nagar (delhi) call me [🔝9953056974🔝] escort service 24X7
 
Collecting banker, Capacity of collecting Banker, conditions under section 13...
Collecting banker, Capacity of collecting Banker, conditions under section 13...Collecting banker, Capacity of collecting Banker, conditions under section 13...
Collecting banker, Capacity of collecting Banker, conditions under section 13...
 
logistics industry development power point ppt.pdf
logistics industry development power point ppt.pdflogistics industry development power point ppt.pdf
logistics industry development power point ppt.pdf
 

Montecarlophd

  • 1. Advanced Econometrics: Doctoral Course Marco Delogu Monte Carlo Experiments
  • 2. Intro to MonteCarlo Experiment 2 The last topic that I want to cover with you is Montecarlo Simulation in STATA. What is a Simulation by Monte Carlo? In broad terms is a useful and powerful methodology for investigating the properties of economic estimators and tests. The power comes from the fact that we can define the characteristics of the Data Generating Process (DGP) and thus study its properties like if we were in a laboratory. Monte Carlo experiments employ pseudo random numbers that mimic the theoretical properties of realizations of random variables
  • 3. Subject of the lecture 3 We learn how to generate pseudo random numbers using Stata. Then we learn how to create simple programs in STATA. Then we set Monte Carlo experiment to verify 1. the Weak Law of Large Numbers 2. the Central Limit Theorem 3. Properties of the OLS estimator After this lecture you will learn how to use some other very important STATA commands: 1. seed to set the pseudo random generator (important to reproduce your findings) 2. postfile 3. program and simulate
  • 4. Generating Random Numbers (1) 4 I am quite sure that you have already encountered the Weak Law of Large Numbers (If you are interested in the mathematical proof please let me know). I report the definition in the next slides. Now we prove the Theorem using the Monte Carlo method. First we need to set the seed to generate random numbers in STATA set seed 12345 Pseudorandom-number generators use deterministic devices to produce long chains numbers, with the Seed we can reproduce the generation of our random numbers in any other pc with STATA installed (same version).
  • 5. Law of Large numbers 5 Let f (., θ) the density of a given random variable X. We don’t know the exact value of θ but we assume that the random variable X follows the distribution f(, ). Question: can a random sample give us reliable information about the value θ? Yes! We can make this claim thanks to the Weak Law of Large Numbers. This theorem allows to make reliable inference about the unknown parameter θ. Stated differently: Denote by µ the expected value E [X] of X. Using only a finite number values of X we can make inference about the value of E (X) because the LLN allows us to construct confidence intervals.
  • 6. Weak Law of Large Numbers 6 Let’s state in words the claim of the Weak Law of Large numbers. A positive integer n can be determined such that if a random sample of size n, or larger, is draw the probability that the sample mean ¯X will deviate from µ less than a previously specified quantity can be made as much close to 1 as desired. Let’s try to state the claim using the mathematical language. Taken two numbers ε > 0 and 0 < δ < 1 there exists an integer n such that if a random sample of size n is taken the sample mean ¯Xn will deviate from µ by less than ε with probability larger than 1 − δ. In symbols: For any ε > 0 and 0 < δ < 1 there exists an integer n such that for any m ≥ n P | ¯Xm − µ| < ε ≥ 1 − δ
  • 7. Weak Law of Large numbers 7 Theorem Weak Law of Large numbers Let f (.) be a density with mean µ and finite variance σ2. Let ¯Xn be the sample mean of a random sample of size n. Let ε and δ be any two specified numbers satisfying ε > 0 and 0 < δ < 1. If n is any integer greater than σ2 ε2δ Then P −ε < ¯Xn − µ < ε ≥ 1 − δ
  • 8. Generating Random Numbers (2) 8 Before proving the LLN through a Monte Carlo experiment we need to generate some random numbers. Remind that we already generate numbers during our previous lecture (go back to the first set of slides). Imagine that we want to generate 30 observations draw from the uniform distribution.. it is very easy in stata clear all set seed 123456 set obs 30 generate x=runiform() As you can check in STATA now you have 30 observations stored in the variable x which are independents draws from the Uniform distirbution.
  • 9. Coin tosses (1) 9 Consider a random variable that takes value 1 with probability 1 2 and 0 with probability 1 2 . Let’s generate 30, indipendent, random variables of such type: clear set obs 30 quietly generate y = runiform() quietly generate x=1 if (y > 0.5) quietly replace x=0 if (y <= 0.5) You can check that x takes either value 1 or value 0.
  • 10. Coin Tosses (2) 10 You can check that x takes either value 1 or value 0. Notice that, the theoretical mean and variance of x are given by: E[X] = 1 ∗ 1 2 + 0 ∗ 1 2 = 1 2 and V [X] = E X2 − E [X]2 = 12 ∗ 1 2 + 0 ∗ 1 2 − 1 2 2 = 1 2 − 1 4 = 1 4 If you are in trouble understanding please let me know. Remind that, in general terms the weak law of large numbers tells you that the empirical mean approaches the theoretical one as long as we increase the sample size. Namely, as long as we increase the number of coin tosses the empirical mean shall converge to 1 2
  • 11. Postfile 11 Postfile command: declares the memory object in which the results are stored, the names of the variables in the result dataset and the name of the results dataset file. In our example we write: postfile sim mean1 trials using sim1000, replace Everyting will be clearer shortly. We generate a new dataset called sim1000 where we store the results of our experiments.
  • 12. Simulation 12 forvalues i = 1/100000 { quietly capture drop y x quietly set obs ‘i’ quietly generate y =runiform() quietly generate x=1 if (y > 0.5) quietly replace x=0 if (y <= 0.5) summarize x quietly post sim (r(mean)) (r(N)) } postclose sim clear use sim1000
  • 13. forvalues 13 forvalues: i = 1/100000 means that Stata repeats the instructions between the curly brackets 10000 times In our example for every i we construct a new dataset that is made by i observations. Finally: quietly post sim (r(mean)) (r(N)) saves in our external dataset, for each simulation trial, both value of the mean and the number of observations.
  • 15. Weak Law of Large Numbers 15 The law of large numbers tells us that knowing we can determine a confidence interval such that for a given sample size the empirical mean belongs to an interval centered at the theoretical mean with a certain probability. Setting interval equal to 0.1 and the confidence interval equal to 95% and taking advantage of the theorem 1 4 1 (0.1)2 (0.05) = 500 So the theorem tells that at least the 95% of the means shall fall in this range....Let’s check making another experiment
  • 16. Code 16 forvalues i = 1/1000 { quietly capture drop y x quietly set obs 500 quietly generate y = runiform() quietly generate x=1 if (y>0.5) quietly replace x=0 if (y<=0.5) summarize x quietly post sim (r(mean)) } postclose sim
  • 17. Check 17 clear use weaklaw2.dta * here we generate one variable that take gen test=1 if ((mean1>0.6) — (mean1<0.4)) summarize test gen obs1 =n scatter mean1 obs1
  • 18. Weak Law 2nd graph 18
  • 19. Weak Law of Large number 19 Please refer to the dofile for practical clarifications. In our experiment we could not find any experiment for which the sample mean does not belong to the bounds predicted by the theorem. Therefore, our experiments does not reject the theorem! To check that is not artifact change number of observations (trial) and make some experiments by yourself.
  • 20. Central Limit Theorem (1) 20 Theorem: Central-limit theorem: Let f (.) be a density with mean µand finite variance σ2. Let ¯Xn be the sample mean of a random sample of size n from f (.). Let the random variable Zn be defined by Zn = ¯Xn − E ¯Xn var ¯Xn = ¯Xn − µ σ√ n The distribution of Zn approaches the standard normal distribution as n approaches to infinity
  • 21. Central Limit theorem (2) 21 The CLT says that the limiting distribution of Zn which is ¯Xn standardized is the standard normal distribution. Namely, this very important theorem tells us that ¯Xn itself is approximately, or asymptotically distributed as a normal distribution with mean µ and variance σ2 n Nothing is assumed about the form of the original density function The degree of approximation depends on the sample size and on the particular density f(.) The idea is that you take many samples of size n. Then the central limit theorem shows that the distribution of the mean of such samples will approximate to a normal as the sample size increases Homework: determine the exact distribution of the sample mean for a sample from a Poisson distribution (you should know what is a Poisson distribution, and its properties, feel free to come back to me to have references that will help you to solve the exercises)
  • 22. CLT and LLN 22 From Angrist Pischke book (Mastering Metrics): How can we decide wether statistical result constitute strong evidence or merely a lucky draw, unlikely to be replicated in repeated samples? LLN and CLT are the foundations of formal statistical inference. As Angrist and Pischke write: Quantifying sampling uncertainty is a necessary step in any empirical projects and on the road to understanding statistical claims made by others
  • 23. Central Limit Theorem. Simulation 23 Now we try to prove the CLT by means of a Monte Carlo Experiment: Central Limit theorem: the sample mean is approximately normally distribute as N approaches infinity. Let consider a random variable that has the uniform distribution and a sample size of 30. Let’s check that when we draw a single sample of uniform draws the graph that we obtain as nothing to do with a bell shaped curve: quietly set obs 30 set seed 10101 generate x=runiform() summarize x quietly histogram x, width(0.1) xtitle(”x from one sample”)
  • 24. Recap: Uniform Distribution (1) 24 A random variable X has with values a ≤ X ≤ b has a unform distirbution if its probability density function is given by f (x|a, b) = 1 b − a for a ≤ x ≤ b and zero otherwise now, let a = 0 and b = 1 E [X] = 1 0 x 1 1 − 0 dx = x2 2 1 0 = 1 2 V (X) = 1 0 x2 1 1 − 0 dx − 1 2 2 = x3 3 1 0 − 1 4 = 1 3 − 1 4 = 1 12
  • 26. Program 26 We will do simulation but using a different command. In place of postfile we write a program. Notice that we put rclass. This tell STATA that we want that the program return belongs to the rclass . program onesample, rclass drop all quietly set obs 30 generate x=runiform() summarize x return scalar meanforonesample=r(mean) end onesample is our first user written program. It generates 30 observations drawn fro the uniform distribution. It returns the mean of this sample as an rclass object.
  • 27. Simulate 27 set seed 10101 onesample return list We employ the command simulate to run the program as many time as we wish. Understand that our r.v. has expected value equal to 0.5 and s.d. equal to 0.0527. Each round we obtain a new value of the sample mean stored in the xbar variable. simulate xbar=r(meanforonesample), seed(10101) reps(1000) nodots:onesample Reps tells STATA how many simulations we want to perform. In our case is 1000. We can check now our experiment. summarize xbar quietly histogram xbar, normal xtitle(”xbar from many samples”)
  • 28. Central Limit Theorem (Graph) 28
  • 29. OLS regression 29 When we do OLS we employ the OLS method to determine estimates of the True parameters. Monte Carlo experiment can be used to estimate the reliability of our estimators. Let’s apply the MonteCarlo methodology to check some of the well known OLS properties. We verify that when the DGP meets the OLS assumptions the OLS estimates are unbiased. So what we do? We define a DGP that meets OLS assumptions and then we veritfy that OLS estimator will have all that good properties that makes it so widespread used!
  • 30. Unbiasdness 30 Unbiasdness of the Sample Mean: E ¯Yn = E [Yi] = µ What it means: The sample mean should not be expected to bang on the corresponding population mean: the sample average in one sample can be too big; while in another sample can be too small. Unbiasdness tells us that deviations are not systematically up or down; rather; in average samples they average out to zero. Unbiasdness is distinct from the LLN, which says that the sample mean gets closer and closer to the population mean as the sample size grows. Unbiasdness of the sample mean holds for samples of any size, for any n!
  • 31. Model 31 Let assume that we have some data such that: E [Y |X] = 100 + 10 ∗ 10 for the first 20 observations and E [Y |X] = 100 + 10 ∗ 20 for the remaining 20 observations We assume that the error term follows a Normal distribution with mean zero and variance equal to 2500, N (0, 50), where 50 is the standard deviation. This equation define the theoretical model of our process. In the real world once we do OLS we just have 1 sample! and we want to estimate both the parameters, in our case equal to 100, b1 and 10, b2
  • 32. Setting the experiment 32 Before to define the simulation we save in global macros the key input of our DGP. clear all global numobs 40 // sample size global beta1 100 // intercept parameter global beta2 10 // slope parameter global sigma 50 // standard error Now we can define the DGP
  • 33. Data Generating Process 33 set seed 1234567 set obs $numobs generate x=10 replace x=20 if n > $numobs/2 generate y = $beta1 + $beta2*x + rnormal(0,$sigma) Now, given our data we can estimate the parameters through OLS regression regress y x Notice that we do not get neither 1000 nor 10. The result of one regression is not enough to say something about the OLS properties!
  • 34. OLS experiment 34 Look at the program in the dofile. we store for each experiment the estimate of: intercept slope variance. return scalar b2 = b[x] return scalar b1 = b[ cons] return scalar sig2 = (e(rmse))2
  • 35. Making a series of experiment 35 Taking advantage of STATA we can generate as many samples of the DGP process. Then we can evaluate whether OLS estimates are good or bad: simulate b1r=r(b1) b2r=r(b2) sig2r=r(sig2), reps (1000) nodots nolegend seed(1234567): regression1 We have just run 1000 regression. We can compare the average of our estimates with the actual values.
  • 38. Exercise 38 Consider a biased coin, probability of getting head =0.3. Define and perform a MonteCarlo Experimento to obtain a reliable estimate of the empirical mean Consider a random variable that follows a Poisson distribution with mean 1.5 (rpoisson). Determine the expected value analytically (1.5) and by means of a MonteCarlo Experiment evaluate the distribution of the empirical mean considering 10000 samples for each one of the following 3 cases: sample size=5, sample size=15, sample size=50. Discuss your result in light of the CLT consider the OLS simulation. Discuss how the empirical estimate of V(b2), from the MonteCarlo experiment, differ from the actual one. What happens when n increases?