On Threshold ARCH Models with Gram- Charlier Density Xuan Zhou and W. K. Li Department of Statistics and Actuarial Science,HKU.

Slides:



Advertisements
Similar presentations
Cointegration and Error Correction Models
Advertisements

Chapter 4 Inference About Process Quality
Previous Lecture: Distributions. Introduction to Biostatistics and Bioinformatics Estimation I This Lecture By Judy Zhong Assistant Professor Division.
Threshold Autoregressive. Several tests have been proposed for assessing the need for nonlinear modeling in time series analysis Some of these.
Principal Component Analysis Based on L1-Norm Maximization Nojun Kwak IEEE Transactions on Pattern Analysis and Machine Intelligence, 2008.
Introduction to Algorithmic Trading Strategies Lecture 8 Risk Management Haksun Li
Juan P. Cajigas Centre for Econometric Analysis
Primbs, MS&E 345, Spring The Analysis of Volatility.
Class 2 Statistical Inference Lionel Nesta Observatoire Français des Conjonctures Economiques CERAM February-March-April.
Portfolio Selection With Higher Moments Authors:Campbell R. Harvey, John C. Liechty Merrill W. Liechty and Peter M¨uller Reporter:You-sheng Liu
Introduction to Volatility Models From Ruey. S. Tsay’s slides.
1 Indirect Estimation of the Parameters of Agent Based Models of Financial Markets Peter Winker Manfred Gilli.
KYIV SCHOOL OF ECONOMICS Financial Econometrics (2nd part): Introduction to Financial Time Series May 2011 Instructor: Maksym Obrizan Lecture notes II.
Descriptive statistics Experiment  Data  Sample Statistics Experiment  Data  Sample Statistics Sample mean Sample mean Sample variance Sample variance.
Bootstrap in Finance Esther Ruiz and Maria Rosa Nieto (A. Rodríguez, J. Romo and L. Pascual) Department of Statistics UNIVERSIDAD CARLOS III DE MADRID.
PREDICTABILITY OF NON- LINEAR TRADING RULES IN THE US STOCK MARKET CHONG & LAM 2010.
Statistical Inference and Regression Analysis: GB Professor William Greene Stern School of Business IOMS Department Department of Economics.
The generalized Additive Nonparametric GARCH Model --With application to the Chinese stock market Ai Jun Hou Department of Economics School of Economics.
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.
Inference about a Mean Part II
Measuring market risk:
INFERENTIAL STATISTICS – Samples are only estimates of the population – Sample statistics will be slightly off from the true values of its population’s.
Stress testing and Extreme Value Theory By A V Vedpuriswar September 12, 2009.
Time-Varying Volatility and ARCH Models
Identification of Misfit Item Using IRT Models Dr Muhammad Naveed Khalid.
Normalization of the Speech Modulation Spectra for Robust Speech Recognition Xiong Xiao, Eng Siong Chng, and Haizhou Li Wen-Yi Chu Department of Computer.
Review of Statistical Inference Prepared by Vera Tabakova, East Carolina University.
Chapter 5 Sampling and Statistics Math 6203 Fall 2009 Instructor: Ayona Chatterjee.
Review of Statistical Inference Prepared by Vera Tabakova, East Carolina University ECON 4550 Econometrics Memorial University of Newfoundland.
Fundamentals of Data Analysis Lecture 7 ANOVA. Program for today F Analysis of variance; F One factor design; F Many factors design; F Latin square scheme.
Alignment and classification of time series gene expression in clinical studies Tien-ho Lin, Naftali Kaminski and Ziv Bar-Joseph.
Soft Sensor for Faulty Measurements Detection and Reconstruction in Urban Traffic Department of Adaptive systems, Institute of Information Theory and Automation,
Elements of Financial Risk Management Second Edition © 2012 by Peter Christoffersen 1 Distributions and Copulas for Integrated Risk Management Elements.
Book Review: ‘Energy Derivatives: Pricing and Risk Management’ by Clewlow and Strickland, 2000 Chapter 3: Volatility Estimation in Energy Markets Anatoliy.
University of Ottawa - Bio 4118 – Applied Biostatistics © Antoine Morin and Scott Findlay 08/10/ :23 PM 1 Some basic statistical concepts, statistics.
Fundamentals of Data Analysis Lecture 10 Management of data sets and improving the precision of measurement pt. 2.
An Empirical Likelihood Ratio Based Goodness-of-Fit Test for Two-parameter Weibull Distributions Presented by: Ms. Ratchadaporn Meksena Student ID:
Maximum Likelihood Estimator of Proportion Let {s 1,s 2,…,s n } be a set of independent outcomes from a Bernoulli experiment with unknown probability.
Various topics Petter Mostad Overview Epidemiology Study types / data types Econometrics Time series data More about sampling –Estimation.
Measuring and Forecasting Portfolio Risk on the Romanian Capital Market Supervisor: Professor Moisa ALTAR MSc student: Stefania URSULEASA.
No-Arbitrage Testing with Single Factor Presented by Meg Cheng.
Lecture 4: Statistics Review II Date: 9/5/02  Hypothesis tests: power  Estimation: likelihood, moment estimation, least square  Statistical properties.
Generalised method of moments approach to testing the CAPM Nimesh Mistry Filipp Levin.
EC 532 Advanced Econometrics Lecture 1 : Heteroscedasticity Prof. Burak Saltoglu.
Robert Engle UCSD and NYU and Robert F. Engle, Econometric Services DYNAMIC CONDITIONAL CORRELATIONS.
Empirical Likelihood for Right Censored and Left Truncated data Jingyu (Julia) Luan University of Kentucky, Johns Hopkins University March 30, 2004.
STAT 497 LECTURE NOTE 9 DIAGNOSTIC CHECKS 1. After identifying and estimating a time series model, the goodness-of-fit of the model and validity of the.
Review of Statistical Inference Prepared by Vera Tabakova, East Carolina University ECON 4550 Econometrics Memorial University of Newfoundland.
Point Estimation of Parameters and Sampling Distributions Outlines:  Sampling Distributions and the central limit theorem  Point estimation  Methods.
REGIME CHANGES AND FINANCIAL MARKETS Prepared for Topics in Quantitative Finance | Abhishek Rane - Andrew Ang and Allan Timmermann.
S TOCHASTIC M ODELS L ECTURE 4 P ART II B ROWNIAN M OTIONS Nan Chen MSc Program in Financial Engineering The Chinese University of Hong Kong (Shenzhen)
GARCH Models Þættir í fjármálum Verkefni 1-f Bjartur Logi Ye Shen
13 th AFIR Colloquium 2003 The estimation of Market VaR using Garch models and a heavy tail distributions The dynamic VaR and The Static VaR The Garch.
Review of Statistical Inference Prepared by Vera Tabakova, East Carolina University.
Tail Dependence in REITs Returns Kridsda Nimmanunta Kanak Patel ERES Conference 2009, Stockholm, Sweden 24 –
Computacion Inteligente Least-Square Methods for System Identification.
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.
The Performance of GARCH Models with Short-memory, Long-memory, or Jump Dynamics: Evidence from Global Financial Markets Yaw-Huei Wang National Central.
MODELING VOLATILITY BY ARCH- GARCH MODELS 1. VARIANCE A time series is said to be heteroscedastic, if its variance changes over time, otherwise it is.
Analysis of financial data Anders Lundquist Spring 2010.
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.
Vera Tabakova, East Carolina University
ARCH/GARCH Modelling of Exchange Rates in Turkey
also Gaussian distribution
Stochastic Hydrology Hydrological Frequency Analysis (II) LMRD-based GOF tests Prof. Ke-Sheng Cheng Department of Bioenvironmental Systems Engineering.
Inference about the Slope and Intercept
Inference about the Slope and Intercept
Review of Statistical Inference
Threshold Autoregressive
Presentation transcript:

On Threshold ARCH Models with Gram- Charlier Density Xuan Zhou and W. K. Li Department of Statistics and Actuarial Science,HKU

Outline  Introduction to Threshold Model  Introduction to Gram Charlier Density  Threshold Model with Gram Charlier density  Estimation method  Testing method  Empirical Results

Motivation  Some financial time series are asymmetric. Investors are more nervous when the market is falling than when it is rising. Negative shocks entering the market lead to a larger return volatility than positive shocks of a similar magnitude. Many models have been proposed to investigate this asymmetric feature, such as the ARCH-M model by Engle (1987) and the EGARCH model by Nelson (1990).  Innovation are believed to be non-Gaussian

Threshold AR Model  Tong ’ s (1978) threshold autoregressive model  is the delay parameter, r is the threshold value.

Threshold ARCH Model  Li and Lam (1995) Threshold ARCH model

Conditional density  It is generally accepted that the conditional distribution of the asset return is not Gaussian (Mills(1995)).  The leptokurtosis has been found in most financial time series.

Introduction to Gram-Charlier (GC) density  The Gram-Charlier density is is the standard normal density function, and

Properties of GC density  The mean and variance are  The skewness and excess the kurtosis are and.  We use GC(u,h,s,k) to denote a GC density with mean u, variance h, skewness s, and excess kurtosis k.

Why Gram-Charlier density?  It nests Gaussian density.  It has explicit skewness and kurtosis.

Threshold ARCH Model with Gram Charlier Density is the indicator of regimes. Skewness and excess kurtosis will vary with time. The structure if different in different regimes.

Double Threshold ARCH model with GC density (DTARCHSK) is the indicator function

 Several problems in the estimation. The number of parameters is large. As pointed out by Bond (2001), MLE estimation is quite sensitive to initial parameters, therefore it's necessary to search over a wide set of initial parameters before selecting the model with the highest likelihood value. Estimation

Estimation method: ECM  Step 1: For a given value of the skewness and kurtosis, fit the model by MLE.  Step 2: Conditional on the estimates, calculate the residuals. Find the maximum likehood estimates of the skewness and the kurtosis of the residuals.  Step 3: Repeat until all the parameters converge. Group two Group one

 The convergence is fast. Almost every simulation converges in three iterations.  The first step of the ECM method is a quasi-maximum likelihood estimation. It converges when the third and fourth moments are assumed finite. Therefore, the assumed value for the skewness and kurtosis would not affect Step 1 much. The parameters for mean and variance structure converge fast. All estimates converged within three iterations.

Lagrange Multiplier Test  The threshold structure and GC density both can help explain the asymmetric features, combination of them will definitely enhance the model's ability to capture asymmetry.  On the other hand, they will also interact with each other and prevent us to distinguish them.

Example  Consider the model  When the previous data is positive, the conditional density is skewed to the positive side. When the previous data is negative, the conditional density is still skewed to the positive side. Therefore, the behavior of the series is asymmetric even the mean structure is symmetric.

 Wong and Li ’ s (1997) test does not take into account the conditional density. Their nominal 5% test on the model reject 8% of the experiments.

Lagrange Multiplier test  The null hypothesis is:  The conditional likelihood function is:

Lagrange Multiplier test  The fish information matrix is  Score function is

Because the conditional density is symmetric, we can show that (Engle ’ s (1982) theorem), The information matrix is block diagonal. Therefore, we can drop the second block of the matrix which is not related to the threshold structure in the test.

As the time series is stationary and ergodic, by the martingale central limit theorem, we can show

Supreme Lagrange Multiplier Test  If r is given, define the Lagrange-Multiplier test statistic as  If r is unknown, we define the supreme Lagrange- Multiplier test statistic as  The distribution of S was proved to be related to an Ornstein-Uhlenbeck (O-U) process (Chan(1990)).

Test Comparison

Effect of the skewness in the testing  When the skewness is included in the GC density, the information matrix is no longer a block diagonal matrix. Therefore we can not just drop the second block of the matrix. As a result, the form of the Lagrange Multiplier test will be more complicated to handle.  However, the critical value of the test for the models with skewed density are almost the same as the test of the corresponding models with non-skewed density.

Simulation Models

Experimental results 10%5%2.5%1.0% Model Model Model Model Model Model Model Model Sample size = 400, replication = Estimate of r is obtained by searching between 10% and 90% quantiles of the data.

Empirical results  We apply our model to several foreign exchange rates series, including British Pound (GBP/USD), Japanese Yen/USD (JPY/USD), German Mark (GEM/USD) from Jan 2, 1990 to Dec 29,  Fit the data with four models: ARCH, ARCHSK, DTARCH and DTARCHSK model.

Test results  The test of DTARCHSK and the test of DTARCH generate very different conclusions on the existence of threshold structure. The test of DTARCH is more likely to reject the null hypothesis while the proposed one prefers the null hypothesis. This is because the ARCHSK model has captured most of the asymmetric features and need not to further assign threshold structure.

Q&A

Some Conditional density Model  Engle and Gonzalez- Rivera (1991) Semiparametric ARCH models.  Harvey and Siddique (1999) Autoregressive conditional skewness  Brooks et al.(2005) Autoregressive conditional kurtosis  The skewness and the kurtosis of Student ’ s t-density have to be calculated from the distributional parameters. How skewed and how leptokurtic the density is can not be conveyed directly.