Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd Logo
Upload

Welcome to Scribd!

  • 34 views

    Bhoumik - Econ 7800 - PS3

    Uploaded by

    Lopa Bhowmik
    This document contains Problem Set 3 for an econometrics course. It includes three analytical problems and two R problems. The analytical problems involve calculating the transition probabilities of a Markov chain over time, finding the stationary distribution, and determining conditional probabilities for a Gibbs sampler. The R problems demonstrate constructing a Gibbs sampler and diagnosing MCMC samples for a Cauchy distribution model in Stan, examining traceplots, autocorrelations, and Geweke statistics.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Bhoumik - Econ 7800 - PS3

    Uploaded by

    Lopa Bhowmik
    34 views7 pages
    This document contains Problem Set 3 for an econometrics course. It includes three analytical problems and two R problems. The analytical problems involve calculating the transition probabilities of a Markov chain over time, finding the stationary distribution, and determining conditional probabilities for a Gibbs sampler. The R problems demonstrate constructing a Gibbs sampler and diagnosing MCMC samples for a Cauchy distribution model in Stan, examining traceplots, autocorrelations, and Geweke statistics.

    Original Description:

    Working on R problem sets

    Original Title

    Bhoumik_Econ 7800_PS3

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    This document contains Problem Set 3 for an econometrics course. It includes three analytical problems and two R problems. The analytical problems involve calculating the transition probabilities of a Markov chain over time, finding the stationary distribution, and determining conditional probabilities for a Gibbs sampler. The R problems demonstrate constructing a Gibbs sampler and diagnosing MCMC samples for a Cauchy distribution model in Stan, examining traceplots, autocorrelations, and Geweke statistics.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    34 views7 pages

    Bhoumik - Econ 7800 - PS3

    Uploaded by

    Lopa Bhowmik
    This document contains Problem Set 3 for an econometrics course. It includes three analytical problems and two R problems. The analytical problems involve calculating the transition probabilities of a Markov chain over time, finding the stationary distribution, and determining conditional probabilities for a Gibbs sampler. The R problems demonstrate constructing a Gibbs sampler and diagnosing MCMC samples for a Cauchy distribution model in Stan, examining traceplots, autocorrelations, and Geweke statistics.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    You are on page 1of 7

    Problem Set 3

    Econ 7800
    Lupa Bhoumik

    Analytical

    1. A) For a matrix to be a left stochastic matrix, each column must sum to 1. So, for the
    given (4x4) matrix;

    𝑃! = 0.1
    𝑃" = 0.8
    𝑃# = 0.3
    𝑃$ = 0.3
    Will make this matrix a proper left-stochastic transition matrix.

    b) The probability that the chain state has state value i at time t + 1 is given by the
    Chapman-Kolmogorov equation, that sums over the probability of being in a particular
    state at the current step and the transition probability from that state into state i . So, after
    t periods the transition matrix will be 𝑝% :

    p [t] = 𝑃% p [0]

    So, at t = 10 and p [0] = (0, 0.3, 0.6, 0.1) we have, the probability mass function of the
    system as,

    p [10] = 𝑃!& * [0, 0.3, 0.6, 0.1]


    0.1550387 0.1550375 0.1550166 0.1550480 0 0.1550260
    0.2652444 0.2652821 0.2652597 0.2652446 0.3 0.2652649
    $ 0𝑥$ 0 = $ 0
    0.2864115 0.2863055 0.2864036 0.2863958 0.6 0.2863734
    0.2933054 0.2933749 0.2933202 0.2933117 0.1 0.2933357

    c) Staionary Distribution:

    A Markov chain reaches a stationary distribution with transition matrix P is some vector,
    such that p∗ = P. p∗ The above matrix has a stationary distribution where it converges to
    (0.1550260, 0.2652649, 0.2863734, 0.293357).
    R

    1.

    a) Constructing a Gibbs Sampler:

    Given that the banker has two specified conditional distributions:


    ∗ ∗ ∗
    P [q|𝑥"& ] ~ Beta [k+ 𝑥"& + 1, n- (k+𝑥"& )+2]
    ∗
    P[𝑥"& |q] ~ Bernoulli [q∗ ]

    We can generate a Gibbs Sampler from these two distributions,

    n <− 20
    k <− 14
    nsim <− 10000
    theta <− c ( )
    s <− c ( )
    theta [ 1 ] <− 1 . 5
    s [ 1 ] <− 1
    for ( i in 2 :Nsim){
    the ta [ i ] <− rbeta ( 1 , k+s+1,n−(k+s )+2)
    s [ i ] <− rbernoulli ( 1 , theta )
    }

    b) Diagnosing the sampler using traceplots and autocorrelation plots for q:

    We can see from the figure below that the distribution of q converges around 0.68 after the burn-
    in period. Again, the 2nd figure shows that the plot dies out very rapidly, which means that
    autocorrelation in our sampling is negligible.
    c ) The posterior mean and 95% HPD region of q is illustrated in the figure below:
    d) Comparing the histogram of theta bias and that of burn, we see some differences, because we
    calculated the distribution of bias theta excluding the 𝑥"& , i. e., with n = 19. So, the ratio of k to n
    changes. We have also found that mean of biased estimate is 0.715 which is slightly higher than
    the mean of the theta distribution, 0.68.
    2. a)

    Setting up the model in STAN


    modelString =
    data {
    int<lower=0> N ;
    real x [N] ;
    }
    parameter s {
    real mu ;
    real<lower=0> sigma ;
    }
    model {
    x ˜ cauchy (mu, sigma ) ;
    }
    x <− rcauchy ( 1 0 0 , 1 , 1 )
    N <− length ( x )
    dataList = list ( x = x , N = N)
    stanFit = stan (model_code = modelString , data=dataList ,
    chains=4 , iter =10000 , warmup=200 , thin=1)

    b) Diagnosing samples using traceplots, autocorrelation plots, and Geweke statistics:

    Figure 1 below shows that the Traceplot for mu and sigma converges for all the four chains.
    Figure 2 again shows that the auto-correlation died out very quickly and stayed low in every
    chain for both mu and sigma. The 3rd figure illustrates the Geweke statistics which lies between
    -2 to 2. From the plot we can see that all of our values for every chain are in that range as we
    wanted. Figure 4 shows the density functions for both of the statistics of mu and sigma, where
    the density function for the both are approximately the same for all of the four chains.
    Figure 1: Traceplot of 𝜇 and 𝜎

    Figure 2: Autcorrelation of 𝜇 and 𝜎


    Figure 3: Geweke diagnostics

    Figure 4: Density functions for mu and sigma

    You might also like