Scribd Logo
Upload

Welcome to Scribd!

  • 22 views

    Financial Risk Measurement and Management: Volatility Forecast: Beyond Riskmetrics

    Uploaded by

    Pratik Jain
    This document discusses volatility forecasting and risk measurement using GARCH models. It begins with an introduction to GARCH models and their advantages over the RiskMetrics approach. Maximum likelihood estimation is then introduced as a method to estimate the parameters of GARCH models. The document estimates a GARCH(1,1) model on some return data and compares the volatility forecasts to those from RiskMetrics. It finds the GARCH model assigns a higher weight to recent shocks compared to RiskMetrics. Plots of returns, volatility forecasts and VaR from the different models are presented.

    Copyright:

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

    Financial Risk Measurement and Management: Volatility Forecast: Beyond Riskmetrics

    Uploaded by

    Pratik Jain
    22 views7 pages
    This document discusses volatility forecasting and risk measurement using GARCH models. It begins with an introduction to GARCH models and their advantages over the RiskMetrics approach. Maximum likelihood estimation is then introduced as a method to estimate the parameters of GARCH models. The document estimates a GARCH(1,1) model on some return data and compares the volatility forecasts to those from RiskMetrics. It finds the GARCH model assigns a higher weight to recent shocks compared to RiskMetrics. Plots of returns, volatility forecasts and VaR from the different models are presented.

    Original Description:

    Original Title

    Sessions_8

    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 discusses volatility forecasting and risk measurement using GARCH models. It begins with an introduction to GARCH models and their advantages over the RiskMetrics approach. Maximum likelihood estimation is then introduced as a method to estimate the parameters of GARCH models. The document estimates a GARCH(1,1) model on some return data and compares the volatility forecasts to those from RiskMetrics. It finds the GARCH model assigns a higher weight to recent shocks compared to RiskMetrics. Plots of returns, volatility forecasts and VaR from the different models are presented.

    Copyright:

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

    Financial Risk Measurement and Management: Volatility Forecast: Beyond Riskmetrics

    Uploaded by

    Pratik Jain
    This document discusses volatility forecasting and risk measurement using GARCH models. It begins with an introduction to GARCH models and their advantages over the RiskMetrics approach. Maximum likelihood estimation is then introduced as a method to estimate the parameters of GARCH models. The document estimates a GARCH(1,1) model on some return data and compares the volatility forecasts to those from RiskMetrics. It finds the GARCH model assigns a higher weight to recent shocks compared to RiskMetrics. Plots of returns, volatility forecasts and VaR from the different models are presented.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 7

    11/21/2020

    Financial Risk Measurement and Management


    Session 8

    Dilip Kumar 1

    Volatility forecast: Beyond RiskMetrics


    RiskMetrics forecast for t+1 variance can be written as
    = + 1−
    As we saw earlier, RiskMetrics sets  = 0.94, for all asset classes
    You don’t like such “dirty” ad-hoc estimates?
    The above expression for volatility forecast can be written more generally as
    =+ +
    If we set =0,  = 1 - , = , then we get the RiskMetrics model
    The generic model is the simplest GARCH model
    GARCH refers to Generalized Autoregressive Conditional Heteroskedasticity
    The model that we have shown is commonly referred to as GARCH(1,1) model

    Dilip Kumar 2

    1
    11/21/2020

    Introduction to Maximum Likelihood Estimation


    Let us say I give you a set of random numbers. I just tell you that they
    have been drawn from a normal distribution. How do you find the mean
    and std. deviation of the distribution from which they were drawn?
    Obvious solution: Estimate the sample mean and standard deviation.
    Use this as our “estimate” for the true mean and std. deviation of the population
    from which these random numbers were drawn
    Here is another way of estimating the same: Maximum Likelihood
    Estimate
    It answers the basic question: Given a particular sample, what is the distribution
    that is most likely to have given rise to such data?

    Dilip Kumar 3

    Introduction to Maximum Likelihood Estimation


    Concept 1: The notion of likelihood:
    Assume that these random numbers were drawn from a distribution of mean 
    and standard deviation .
    What is the probability that this number is equal to the number drawn, say x?
    Given a sample data point x, if we were to assume that it came from a N(, )
    distribution, we can define the associated likelihood function as
    1 −
    , | = −
    2 2
    Look familiar? Should be! Likelihood function is exactly equal to the pdf
    It is important to understand what L represents: given that x is observed, what is
    the likelihood that it came from a normal distribution with mean  and std. dev 
    It is a common mistake to think of L as the probability of x; it is incorrect!
    You can easily estimate this in Excel
    , | = , , ,
    Dilip Kumar 4

    2
    11/21/2020

    Introduction to Maximum Likelihood Estimation


    Concept 2: The notion of joint likelihood
    We similarly calculate the likelihood values for all random numbers provided
    Since each of these are assumed to be iid, the joint likelihood of having drawn the
    numbers x1, x2, …., xN is merely the product of individual likelihoods
    Concept 3: Optimization
    Remember MLE answers the basic question: Given a particular sample, what is the
    distribution that is most likely to have given rise to such data?
    Now, finding ,  reduces to an optimization problem: What value of , 
    maximizes the likelihood of having drawn this set of random numbers?
    Useful simplification: Instead of maximizing the joint likelihood (which is a
    product), we usually maximize the log of joint likelihood (which is merely
    a sum).

    Dilip Kumar 5

    Introduction to Maximum Likelihood Estimation


    Exercise: In the worksheet “MLE”, you have been provided a set
    of random numbers. These have been drawn from a normal
    distribution of unknown mean and unknown standard
    deviation. Provide an estimate of those.
    Sample mean: 0.17049; sample std.dev: 0.81358
    MLE Estimate: Mean: 0.17049; std. dev: 0.81358
    As an aside, these random numbers were generated from a normal
    distribution with mean 0.2 and std dev of 0.8
    Now, how are we going to use this idea for estimating GARCH
    models?
    Dilip Kumar 6

    3
    11/21/2020

    Volatility forecast: GARCH Estimation


    The GARCH (1,1) model can be summarized as:
    = , ~ 0,1
    =+ +
    The returns are conditionally normally distributed, i.e., return for period
    t+1 follows normal distribution with mean 0 and standard deviation .
    Note, is based on information available up to time t.
    It is also dependent on the three parameters : , ,
    Estimate the GARCH model for the period under consideration
    Assume the following starting values: =0.00005,  = 0.15, =0.75

    Dilip Kumar 7

    Volatility forecast: GARCH Estimation


    The GARCH (1,1) model can be summarized as:
    = , ~ 0,1
    =+ +
    Methodology (Assume the following starting values: =0.00005,  = 0.15, =0.75)
    Take first estimate of variance as a sample variance.
    Starting 2nd day onward, our variance forecast will be a weighted-sum of variance forecasted
    for the previous day and today’s returns as given by the GARCH equation (above) and the
    values for parameters we assumed.
    For each of the given days, we calculate the likelihood that the returns are from a normal
    distribution of mean zero and variance equal to the conditional variance forecasted by
    GARCH model
    We then calculate the sum of (log) likelihood and find that values of , , that maximizes
    the (log) likelihood function!

    Dilip Kumar 8

    4
    11/21/2020

    Volatility forecast: GARCH Estimation


    Results: For the period under consideration, we arrived at the
    following parameter estimates
    =0.000022,  = 0.085152, =0.905315
    So, the estimated GARCH (1,1) model is:
    = , ~ 0,1
    = 0.000022 + 0.085152 + 0.905315
    Now, we can use this equation to forecast variances for next periods.
    Note that, while RiskMetrics gave a weightage of 0.06 to the most recent
    shock, the GARCH model gives a higher weightage : 0.08
    Splitting hairs? How significant a difference is this? (can use larger sample to
    fit GARCH model)
    Let us plot!
    Dilip Kumar 9

    Volatility forecast: GARCH Estimation


    The GARCH (1,1) model can be summarized as:
    = , ~ 0,1
    =+ +
    The returns are conditionally normally distributed, i.e., return for period
    t+1 follows normal distribution with mean 0 and standard deviation .
    Note, is based on information available up to time t.
    It is also dependent on the three parameters : , ,
    Estimate the GARCH model for the period under consideration using R.

    Dilip Kumar 10

    5
    11/21/2020

    Comparison of Variance Forecasts


    0.2 0.003

    0.15 0.0025

    0.1
    0.002
    0.05
    0.0015
    0
    2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
    0.001
    -0.05

    -0.1 0.0005

    -0.15 0
    Return 252D 126D RM GARCH

    Dilip Kumar 11

    Value at Risk: RiskMetrics & GARCH


    800

    600

    400

    200

    0
    2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
    -200

    -400

    -600
    PnL HS VaR 252D VaR 126D VaR RM VaR GARCH VaR

    Dilip Kumar 12

    6
    11/21/2020

    Value at Risk: RiskMetrics & GARCH


    Now, let us compare Value at Risk number obtained using
    various methodologies (for Nifty)
    Comparision of VaR Methodologies: Breaches
    Historical Simulation 146 (out of 2816) 5.185%
    252D Variance 145 (out of 2816) 5.149%
    126D Variance 147 (out of 2816) 5.220%
    RM Model 150 (out of 2816) 5.327%
    GARCH Model 134 (out of 2816) 4.758%

    From the above table it is clear that, for the current example, GARCH
    does a better job in terms of forecasting volatility for VaR estimates
    Note, the results are typical of most scenarios. However, it cannot be taken as a
    statement of fact; it needs to be verified for different settings.
    Dilip Kumar 13

    You might also like