A First Peek at the Extremogram: a Correlogram of Extremes 1. Introduction The Autocorrelation function (ACF) is widely used as a tool for measuring Serial.

Slides:



Advertisements
Similar presentations
FINANCIAL TIME-SERIES ECONOMETRICS SUN LIJIAN Feb 23,2001.
Advertisements

Advanced topics in Financial Econometrics Bas Werker Tilburg University, SAMSI fellow.
Autocorrelation Functions and ARIMA Modelling
SAMSI RISK WS Heavy Tails and Financial Time Series Models Richard A. Davis Columbia University Thomas Mikosch University.
Threshold Autoregressive. Several tests have been proposed for assessing the need for nonlinear modeling in time series analysis Some of these.
1 12. Principles of Parameter Estimation The purpose of this lecture is to illustrate the usefulness of the various concepts introduced and studied in.
Model Building For ARIMA time series
Model specification (identification) We already know about the sample autocorrelation function (SAC): Properties: Not unbiased (since a ratio between two.
Time Series Building 1. Model Identification
Pair-copula constructions of multiple dependence Workshop on ''Copulae: Theory and Practice'' Weierstrass Institute for Applied Analysis and.
STAT 497 APPLIED TIME SERIES ANALYSIS
Output analyses for single system
Copula approach to modeling of ARMA and GARCH models residuals Anna Petričková FSTA 2012, Liptovský Ján
13 Introduction toTime-Series Analysis. What is in this Chapter? This chapter discusses –the basic time-series models: autoregressive (AR) and moving.
Econometric Details -- the market model Assume that asset returns are jointly multivariate normal and independently and identically distributed through.
Resampling techniques Why resampling? Jacknife Cross-validation Bootstrap Examples of application of bootstrap.
On Threshold ARCH Models with Gram- Charlier Density Xuan Zhou and W. K. Li Department of Statistics and Actuarial Science,HKU.
KYIV SCHOOL OF ECONOMICS Financial Econometrics (2nd part): Introduction to Financial Time Series May 2011 Instructor: Maksym Obrizan Lecture notes II.
Bootstrap in Finance Esther Ruiz and Maria Rosa Nieto (A. Rodríguez, J. Romo and L. Pascual) Department of Statistics UNIVERSIDAD CARLOS III DE MADRID.
Evaluating Hypotheses
PREDICTABILITY OF NON- LINEAR TRADING RULES IN THE US STOCK MARKET CHONG & LAM 2010.
HDD and CDD Option Pricing with Market Price of Weather Risk for Taiwan Hung-Hsi Huang Yung-Ming Shiu Pei-Syun Lin The Journal of Futures Markets Vol.
The generalized Additive Nonparametric GARCH Model --With application to the Chinese stock market Ai Jun Hou Department of Economics School of Economics.
Geometric Approaches to Reconstructing Times Series Project Outline 15 February 2007 CSC/Math 870 Computational Discrete Geometry Connie Phong.
Measuring market risk:
Elec471 Embedded Computer Systems Chapter 4, Probability and Statistics By Prof. Tim Johnson, PE Wentworth Institute of Technology Boston, MA Theory and.
Time-Varying Volatility and ARCH Models
Introduction to AEP In information theory, the asymptotic equipartition property (AEP) is the analog of the law of large numbers. This law states that.
Correlation and Linear Regression
The Examination of Residuals. The residuals are defined as the n differences : where is an observation and is the corresponding fitted value obtained.
Section Copyright © 2014, 2012, 2010 Pearson Education, Inc. Lecture Slides Elementary Statistics Twelfth Edition and the Triola Statistics Series.
The Examination of Residuals. Examination of Residuals The fitting of models to data is done using an iterative approach. The first step is to fit a simple.
Various topics Petter Mostad Overview Epidemiology Study types / data types Econometrics Time series data More about sampling –Estimation.
Autoregressive Integrated Moving Average (ARIMA) Popularly known as the Box-Jenkins methodology.
Measuring and Forecasting Portfolio Risk on the Romanian Capital Market Supervisor: Professor Moisa ALTAR MSc student: Stefania URSULEASA.
K. Ensor, STAT Spring 2005 Model selection/diagnostics Akaike’s Information Criterion (AIC) –A measure of fit plus a penalty term for the number.
PROBABILITY AND STATISTICS FOR ENGINEERING Hossein Sameti Department of Computer Engineering Sharif University of Technology Principles of Parameter Estimation.
FAT TAILS REFERENCES CONCLUSIONS SHANNON ENTROPY AND ADJUSTMENT OF PARAMETERS AN ADAPTIVE STOCHASTIC MODEL FOR RETURNS An adaptive stochastic model is.
Week 21 Stochastic Process - Introduction Stochastic processes are processes that proceed randomly in time. Rather than consider fixed random variables.
A generalized bivariate Bernoulli model with covariate dependence Fan Zhang.
Estimation Method of Moments (MM) Methods of Moment estimation is a general method where equations for estimating parameters are found by equating population.
Sampling and estimation Petter Mostad
Seasonal ARMA forecasting and Fitting the bivariate data to GARCH John DOE.
GARCH Models Þættir í fjármálum Verkefni 1-f Bjartur Logi Ye Shen
1 Chapter 5 : Volatility Models Similar to linear regression analysis, many time series exhibit a non-constant variance (heteroscedasticity). In a regression.
Standard and Poor index (S&P500) Standard and Poor index (S&P500) : Returns (first difference)
Correlogram - ACF. Modeling for Forecast Forecast Data The Base Model Linear Trend Logistic Growth Others Models Look for a best approximation of the.
Introduction to stochastic processes
Time Series Analysis PART II. Econometric Forecasting Forecasting is an important part of econometric analysis, for some people probably the most important.
Lecture 8 Stephen G. Hall ARCH and GARCH. REFS A thorough introduction ‘ARCH Models’ Bollerslev T, Engle R F and Nelson D B Handbook of Econometrics vol.
STATISTICS People sometimes use statistics to describe the results of an experiment or an investigation. This process is referred to as data analysis or.
MODEL DIAGNOSTICS By Eni Sumarminingsih, Ssi, MM.
1 A NEW SPECIFICATION TOOL FOR STOCHASTIC VOLATILITY MODELS BASED ON THE TAYLOR EFFECT A. Pérez 1 and E. Ruiz 2 1 Dpto. Economía Aplicada Universidad de.
K. Ensor, STAT Spring 2004 Volatility Volatility – conditional variance of the process –Don’t observe this quantity directly (only one observation.
Estimating standard error using bootstrap
Lecture Slides Elementary Statistics Twelfth Edition
Vera Tabakova, East Carolina University
Hypotheses and test procedures
Time Series Analysis and Its Applications
Ch8 Time Series Modeling
Chapter 6: Autoregressive Integrated Moving Average (ARIMA) Models
Nonlinear Structure in Regression Residuals
Model Building For ARIMA time series
Sampling Distribution
Sampling Distribution
Stochastic Volatility Models: Bayesian Framework
Statistical tests for replicated experiments
The Examination of Residuals
Lecturer Dr. Veronika Alhanaqtah
CH2 Time series.
Presentation transcript:

A First Peek at the Extremogram: a Correlogram of Extremes 1. Introduction The Autocorrelation function (ACF) is widely used as a tool for measuring Serial dependence in a time series. However, for non-Gaussian and nonlinear time series the ACF provides little information into the dependence structure of the process. It is also of little use if one is only interested in the extremes. We consider an analog of the autocorrelation function, the extremogram, developed by Davis and Mikosch, which measures the lagged extremal dependency of the extreme values in the sequence. 1.1 Motivation The motivation of the research comes from the fact that GARCH and Stochastic Volatility processes exhibit very similar features using the routine time series diagnostic tools. However, by observing the asymptotic behavior of the extremes of GARCH and Stochastic Volatility models it is possible to tell them apart. Davis and Mikosch (1998) show that GARCH processes exhibit extremal clustering, while SV processes lack this form of clustering. 1.2 Background Leadbetter et al. introduced the extremal index, θ in (0,1], for a stationary time series which is a measure of clustering in the Extremes. For a GARCH process the extremal index of θ < 1 indicates clustering while for a SV process θ = 1 indicates a lack of clustering. Although explicit formulae for the extremal index exist for distributions such as ARMA and GARCH, they are in general very complicated and difficult to simulate. So the problem of finding probabilistically reasonable and statistically estimable measures of extremal dependence in a strictly stationary sequence is to some extent an open one. Hence Davis and Mikosch take a different tack and study extremal dependence structure of general strictly stationary vector- valued time series X t. 2. The Extremogram Definition: For two sets A and B bounded away from 0, the extremogram is defined as ρ A,B (h ) =lim n→∞ P(a n -1 X 0 A, a n -1 X h B)/ P(a n -1 X 0 A) In many examples, this can be computed explicitly. If one takes A=B=(1,∞), then ρ A,B (h) = lim x→∞ P(X h >x, | X 0 >x) = λ(X 0,X h ) often called the extremal dependence coefficient (l = 0 means independence or asymptotic independence). 2.1 The Empirical Extremogram To estimate the extremogram, we use the empirical extremogram defined as following And in our simulation study we choose the levels A and B in the form of =(m, ∞ ), where m is some large sample quartiles. 3. Empirical Simulations In this section, we simulated two time series. One from GARCH(1,1) model, one from the Stochastic Volatility model (SV). The parameters were chosen so that the marginal distributions had the same index of regular variation and had roughly the same dependence characteristics for the ACF of the squares. For GARCH(1,1), we choose α 1 =0.22, β 1 =0.4, α 0 =0.3. Through the algorithm given by Davis and Mikosch, we can get the regular variation parameter α= Thus restrict our SV model to be driven by the noise from t-distribution with α degrees of freedom. We adopted the following parameterization for the SV model {X t }, X t =σ t σε t where ε t ~t(α)/sqrt(α/(α-2)) And if we let η t =log(σ t 2 ), we simulate η t from an ARMA(1,1) model with mean 0, η t =φη t-1 +Z t +θZ t-1 where Z t ~N(0, τ 2 ) The parameters are chosen to be φ=0.85, θ=0.95, τ 2 =0.07. And σ will be determined by σ 2 *exp(0.5γ(0))=α 0 /(1-α 1 -β 1 ) where γ(0)=(1+ θ 2 +2φθ) τ 2 /(1-φ 2 ). Thus the two series will have the same index of regular variation and approximately have the same magnitude. Further, they appear quite similar in both ACF of the absolute values and ACF of the squares. Ivor Cribben; Li Song; Chun-Yip Yau Richard A. Davis: Advisor Department of Statistics, Columbia University, New York, New York Each simulated series has 500,000 data points. The following graph shows the estimated extremogram of the two series under four different levels. We can see that the GARCH model appears to have a much more severe clustering of extremes than the SV model as we increased the level. The following graph shows the empirical distribution of the extremogram based on simulations. The parameterizations are the same as we defined above. (Each realization is a 100,000-point time series and each model is being replicated for 1000 times.) As we can see, the extremograms of GARCH and SV behave differently for small lags, which indicates that we expect to see more clustering of extremes of the GARCH than that of the SV. 4. Applications to model selection In this section, we illustrate how the extremogram plays a role in model selection between a GARCH and a SV model. We analyze the log return data of IBM stock from 1962 to A GARCH model and a SV model are fitted to the data. Both models give reasonable fit in terms of the behavior of the residuals. To select the model from the point of view of extremal dependency, we compare the empirical extremogram of the data and the simulated empirical extremograms using the parameter estimates of the fitted models. Here the series length is and the level is taken to be the 99.5 percentile. The extremogram of the simulated SV model is more similar to the empirical extremogram than that of the GARCH model. Thus we may conclude that SV model gives a better description to the data in terms of extremal dependency. Reference: Davis, R.A. and Mikosch, T. (1998). The Sample ACF of Heavy--Tailed Stationary Processes with Applications to ARCH. Ann.~Statist Davis, R.A. and Mikosch, T. (2008). The Extremogram: a Correlogram for Extreme Events. (Submitted.) Leadbetter, M.R., Lindgren, G. and Rootzen, H. (1983) Extremes and Related Properties of Random Sequences and Processes. Springer, Berlin. C OLUMBIA U NIVERSITY IN THE CITY OF NEW YORK