1 Outline II Volatility analysis with models –First generation univariate ARCH, GARCH Estimation (QML with bounds and simulated annealing)

Slides:



Advertisements
Similar presentations
Statistical Time Series Analysis version 2
Advertisements

Cointegration and Error Correction Models
Scott Nelson July 29, Outline of Presentation Introduction to Quantitative Finance Time Series Concepts Stationarity, Autocorrelation, Time Series.
ARCH (Auto-Regressive Conditional Heteroscedasticity)
A. The Basic Principle We consider the multivariate extension of multiple linear regression – modeling the relationship between m responses Y 1,…,Y m and.
Part II – TIME SERIES ANALYSIS C5 ARIMA (Box-Jenkins) Models
STAT 497 APPLIED TIME SERIES ANALYSIS
Lecture 7: Principal component analysis (PCA)
DYNAMIC CONDITIONAL CORRELATION : ECONOMETRIC RESULTS AND FINANCIAL APPLICATIONS Robert Engle New York University Prepared for CARLOS III, MAY 24, 2004.
Linear Methods for Regression Dept. Computer Science & Engineering, Shanghai Jiao Tong University.
Simulation Modeling and Analysis
Data mining and statistical learning, lecture 4 Outline Regression on a large number of correlated inputs  A few comments about shrinkage methods, such.
Modelling volatility and correlation
1 Ka-fu Wong University of Hong Kong Volatility Measurement, Modeling, and Forecasting.
Bootstrap in Finance Esther Ruiz and Maria Rosa Nieto (A. Rodríguez, J. Romo and L. Pascual) Department of Statistics UNIVERSIDAD CARLOS III DE MADRID.
Lecture 6: Multiple Regression
RISK MANAGEMENT GOALS AND TOOLS. ROLE OF RISK MANAGER n MONITOR RISK OF A FIRM, OR OTHER ENTITY –IDENTIFY RISKS –MEASURE RISKS –REPORT RISKS –MANAGE -or.
The generalized Additive Nonparametric GARCH Model --With application to the Chinese stock market Ai Jun Hou Department of Economics School of Economics.
Prediction and model selection
1 MULTIVARIATE GARCH Rob Engle UCSD & NYU. 2 MULTIVARIATE GARCH MULTIVARIATE GARCH MODELS ALTERNATIVE MODELS CHECKING MODEL ADEQUACY FORECASTING CORRELATIONS.
Chapter 11 Multiple Regression.
Multivariate volatility models Nimesh Mistry Filipp Levin.
Analysis of Variance & Multivariate Analysis of Variance
Forecasting Revenue: An Example of Regression Model Building Setting: Possibly a large set of predictor variables used to predict future quarterly revenues.
Time-Varying Volatility and ARCH Models
1 Forecasting BET Index Volatility MSc.: Răzvan Ghelmeci Supervisor: Prof. Moisă Altăr.
Lecture 7: Simulations.
DISSERTATION PAPER Modeling and Forecasting the Volatility of the EUR/ROL Exchange Rate Using GARCH Models. Student :Becar Iuliana Student :Becar Iuliana.
1 Least squares procedure Inference for least squares lines Simple 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.
Forecasting Revenue: An Example of Regression Model Building Setting: Possibly a large set of predictor variables used to predict future quarterly revenues.
Elements of Financial Risk Management Second Edition © 2012 by Peter Christoffersen 1 Distributions and Copulas for Integrated Risk Management Elements.
Realized Volatility Distribution in Japanese Stock Market Tetsuya Takaishi Hiroshima University of Economics To be published in Evolutionary and Institutional.
Downside Risk: Implications for Financial Management Robert Engle NYU Stern School of Business Tilburg April 22, 2004.
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.
Literature Review ZHU Cai AMA. Contents Paper-Searching by Journals 2 Paper -Searching by Authors 3 Results of Elementary Analysis 4 Suggestions 5 Introduction.
Various topics Petter Mostad Overview Epidemiology Study types / data types Econometrics Time series data More about sampling –Estimation.
Academy of Economic Studies DOCTORAL SCHOOL OF FINANCE AND BANKING Bucharest 2003 Long Memory in Volatility on the Romanian Stock Market Msc Student: Gabriel.
Measuring and Forecasting Portfolio Risk on the Romanian Capital Market Supervisor: Professor Moisa ALTAR MSc student: Stefania URSULEASA.
By Nafiu Bashir Abdussalam Department of Economics Bayero University, Kano And Jamaladeen Abubakar Department of.
Ray Y Chou Institute of Economics Academia Sinica
Elements of Financial Risk Management Second Edition © 2012 by Peter Christoffersen 1 Simulating the Term Structure of Risk Elements of Financial Risk.
Univariate Linear Regression Problem Model: Y=  0 +  1 X+  Test: H 0 : β 1 =0. Alternative: H 1 : β 1 >0. The distribution of Y is normal under both.
Robert Engle UCSD and NYU and Robert F. Engle, Econometric Services DYNAMIC CONDITIONAL CORRELATIONS.
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.
SEM Basics 2 Byrne Chapter 2 Kline pg 7-15, 50-51, ,
Chapter 23 Volatility. Copyright © 2006 Pearson Addison-Wesley. All rights reserved Introduction Implied volatility Volatility estimation Volatility.
Testing for unit roots in Eviews
Dynamic Models, Autocorrelation and Forecasting ECON 6002 Econometrics Memorial University of Newfoundland Adapted from Vera Tabakova’s notes.
GARCH Models Þættir í fjármálum Verkefni 1-f Bjartur Logi Ye Shen
September 28, 2000 Improved Simultaneous Data Reconciliation, Bias Detection and Identification Using Mixed Integer Optimization Methods Presented by:
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.
OPTIONS PRICING AND HEDGING WITH GARCH.THE PRICING KERNEL.HULL AND WHITE.THE PLUG-IN ESTIMATOR AND GARCH GAMMA.ENGLE-MUSTAFA – IMPLIED GARCH.DUAN AND EXTENSIONS.ENGLE.
Learning Theory Reza Shadmehr Distribution of the ML estimates of model parameters Signal dependent noise models.
1 Lecture Plan : Modelling volatility and correlation in energy markets Time Varying Volatility and Correlation in Energy Markets Models for.
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.
Analysis of financial data Anders Lundquist Spring 2010.
1 VaR Models VaR Models for Energy Commodities Parametric VaR Historical Simulation VaR Monte Carlo VaR VaR based on Volatility Adjusted.
Robert A. Yaffee, Ph.D. Timberlake Consultancy, Ltd. December 6, 2007
Kakhramon Yusupov June 15th, :30pm – 3:00pm Session 3
Ch8 Time Series Modeling
Applied Risk Analysis with OxMetrics
STAT 497 LECTURE NOTE 9 DIAGNOSTIC CHECKS.
Predictive distributions
Applied Econometric Time Series Third Edition
Forecasting the Return Volatility of the Exchange Rate
Regression Forecasting and Model Building
The Examination of Residuals
Presentation transcript:

1 Outline II Volatility analysis with models –First generation univariate ARCH, GARCH Estimation (QML with bounds and simulated annealing) Diagnostics Forecasting (Simulated confidence intervals) Forecast Evaluation –Second generation univariate GARCH-in-mean EGARCH GJR GARCH Leverage effects and volatility smiles

2 Outline III Long-Memory GARCH –APARCH –FIGARCH FIGARCH- BBM FIGARCH-Chung FIEGARCH –FIAPARCH FIAPARCH-BBM FIAPARCH-Chung –Davidson’s HYGARCH –VaR

3 Outline IV Multivariate GARCH –Multivariate BEKK models Factor garch: OGARCH, GOGARCH Dynamic correlations: CCC, DCC

4 Risk Analysis with We analyze volatility of indicators and assets with What is new about 5? It contains most of the multivariate Garch models One can obtain the Ox Code for the menu model just run One can model outliers and predictors in the mean and variance models Estimation models has been improved. Simulated annealing option included. Simulation of models is now possible Functions to detect high frequency jumps have been included.

5 Load and Examine Nasdaq Returns Notice the 1987 crash. We construct a dummy variable for Oct 19, 1987

6 We want record of all variable constructions so I do this with the algebra code

7 This constructs our dummy variable

8 Correlograms reveal an AR(1) and possibly some seasonality We take note of the AR(1) process in the mean

9 Basic Pre-Model Analysis

10 Select the variable

11 Define the sample

12 Specify the preliminary tests

13 Choose the Stationarity tests We select ADF test

14 Choose the Long-Memory Test We select this Geweke Porter- Hudak test

15 Misspecification test results

16 Nonstationarity and Long-Memory Results Long memory parameter is weak—should be.2 to.5 for persistence. Nonstationary

17 Pre-Model Analysis The Jarque-Bera tests suggests nonnormality--- we should probably try a t distribution The ARCH tests suggest ARCH effects The Portmanteau tests suggest autocorrelation The Nasdaq returns are nonstationary and there is long memory

18 Variable Selection

19 Baseline model parameter selection AR(1) GARCH(1,1) normal distribution is our baseliine model

20 Normal GARCH(1,1)

21 Normal GARCH(1,1) model output You may need more constraints or a different distribution

22 Selecting Post-Estimation tests

23 Test Results I Box-Pierce Q tests on standardized residuals confirm proper modeling Engle’s ARCH LM test Jarque Bera test

24 Test Results II Sign Bias test for Leverage Nyblom Hansen stability tests: Critical values are.75 for 1% and.47 for 5% levels. Null is parameter stability Integrated effect Theoretical v actual innovation Fit needs improvement

25 Test Results III Properly modeled ARCH effects Basel II Kupiec Tests Expected shortfall amounts

26 Test Results IV Quantile Regression VAR

27 AR(1) GARCH(1,1) sk(t) Distribution is a skewed t distribution

28 AR(1) GARCH(1,1) sk(t)

29 AR(1)-GARCH(1,1) sk(t) output

30 Graphical Analysis

31 Graph selection

32 Graphical Output

33 Model Comparison

34 Simulation of CEV confidence intervals The model just run can be simulated from the Ox Code. The simulations generates multiple replications. Means and standard errors can be computed. These CEV means and standard errors can be graphed.

35 Simulation of GARCH models

36 Simulation menu

37 Simulations file created

38 Alt-O invokes the Ox Code of the model just run

39 Remainder of Ox code for simulation

40 Graphed simulated confidence intervals around the Conditional Error Variance

41 Subset Models Constraining parameters to be zero. We perform an ARCH(12)-t on NQ. We find that the a t-8 =0 We wish to eliminate that from the model, so we constrain it to be zero.

42 We set up an AR(1) ARCH(12) t model

43 We opt for Matrix form starting values– this lets us fix values

44 We set ARCH(8)=0

45 The Subset Model does not contain ARCH(8)

46 Asymmetric GARCH EGARCH Exponential GARCH (Nelson) GJR Threshold GARCH (Glosten, Jagannathan, Runkle) APARCH Asymmetric Power GARCH (Ding, Engle, and Granger)

47 Volatility Smile

48 Engle and Ng Asymmetry tests

49 Asymmetry tests

50 GJR Asymmetric GARCH(1,1)

51 Model Comparison

52 Forecasts Conditional mean, with confidence intervals Conditional variance –Intervals can be simulated VaR intervals serve as confidence intervals

53 Click on the test icon

54 Forecast selection

55 Forecasts graphed from GJR model

56 Forecasts printed

57 Forecast Evaluation

58 Outlier Modeling Mean model outliers Variance model outliers Cross-model outliers

59 Value at Risk Normal APARCH (1,1) APARCH has long memory capabilities and threshold capabilities built in. Is usually used with a skewed t distribution. In this case I use an APARCH(1,1) with a t distribution to generate the Value at Risk. It does handle the fat-tails but in this case there is no appreciable asymmetry.

60 VaR Robust standard errors

61 Nonstationary GARCH Riskmetrics IGARCH

62 Long Memory Models APARCH (Ding, Engle, and Granger,1993) FIGARCH-Baillie, Bollerslev, and Mikkelsen(BBM) FIGARCH-Chung FIAPARCH (Tse, 1998) FIAPARCH-Chung FIEGARCH (Bollerslev and Mikkelsen, 1996) Davidson’s Hyperbolic GARCH

63 Long-Memory Processes We substitute this function for (1-L) in FIGARCH, etc.

64 Long-Memory Models We run the basic descriptives test on it –And find that it has long memory with a GPH – d =.2885 with p = –Therefore we try a long-memory model. –A FIGARCH - Chung model

65 Asymmetric and Long Memory models Long memoryLong memory asymmetricasymmetric

66 Ar(1) -Chung’s Method with normal distribution

67 AR(1) Chung Model sk(t)

68 Ox Programs for Realized and Integrated Volatility diffusion models

69 GARCH type output

70 Mean Spot and Integrated Volatility Mean GARCH volatility

71 Graphical Output

72 Ox Diffusion Models Other diffusion models include diffusion models for estimation of realized volatility with jumps. Lee-Mykland’s statistical test for detecting jumps at ultra-high-frequency. Estimation of integrated volatility with jumps. estimation of microstructure noise. Estimation of intraday seasonality with flexible fourier functional form filter.

73 Multivariate GARCH Baba,Engle, Kraft, Kronger (BEKK) models –Scalar –Diagonal RiskMetrics MGARCH Factor GARCH –Carol Alexander’s Orthogonal GARCH –GOGARCH Generalized Orthogonal GARCH (ML, NLS)

74 Multivariate GARCH menu

75 BEKK(p,q) Model C,A, and G are nxn, but C is upper triangular

76 Problem-number of parameters ARCH and GARCH BEKK(1,1) models have N(5*N+1)/2 parameters. This is a lot. To reduce the number of parameters, constraints have to be imposed.

77 Diagonal BEKK Matrices C and G are diagonal to restrict the number of parameters.

78 Scalar BEKK Another way to reduce the number of parameters is to run a Scalar BECKK. Matrices A and G are matrices of ones multiplied by a scalar.

79 RiskMetrics MGARCH

80 OGARCH(1,1,m) Alexander and Chibumba(1997) Laurrent, S. (2007) Estimating and Forecasting ARCH models using Timberlake Consultants, Ltd.,p. 177.

81 OGARCH-cont’d Ibid, 178.

82 OGARCH-cont’d Alexander warns that high dimensional factor estimation can grind to a halt. She suggests low order dimensional component extraction. QMLE is used. ARFIMA can be specified in the mean model.

83 Select OGARCH

84 OGARCH

85 Select AR(1)-GJR-GARCH(1,1) with scree plot and standard GARCH output for 2 components

86 Select 2 components

87 Scree plot suggests 1 component

88 Mean model estimates

89 Mean model estimates-cont’d

90 PCA I

91 PCA II

92 Univariate GARCH model for PC(1)

93 Univariate GARCH model for PC(2)

94 Graphs of the Raw Series

95 Graphs of Standardized Residuals

96 Conditional Variances

97 Graph of the Conditional Correlations

98 Forecasts Prints conditional mean forecasts Prints conditional variance forecasts Prints v-c forecasts Prints conditional correlation forecasts

99 Printed forecasts

100 Printed forecasts-cont’d.

101 Multivariate tests

102 Generalized OGARCH (van der Weide, 2002) One can test whether the correlations between the components are really zero. This model outperforms the OGARCH sometimes. The log-likelihood may be lower. The orthogonality assumption between OGARCH components is relaxed. Rather the Z matrix in is assumed to be square and invertible only. (Laurent, 2007, class notes).

103 GOGARCH - (Laurent notes, con’td) where P and L are defined as the eigenvectors and eigenvalues, and U is the product of N(N-1)/2 rotation matrices:

104 Specification tests (Laurent notes cont’d). The specification tests (univariate and multivariate) are used to assess the fit and specification of the model. Univariate tests are applied to each u it Univariate tests are applied to each z it. Univariate tests are applied to each u it u jt to assess the covariance specification. Multivariate tests are applied to the vector z t as a whole.

105 Rotation matrices For the trivariate case, the rotation matrices are There are N(N-1)/2 rotation angles are the parameters to be Estimated.

106 Conditional Correlation Models Constant conditional correlation is not time-varying The Correlation between the conditional variances is made time-varying Forecasts are possible Graphs of conditional correlations are possible

107 Conditional Correlation Models Bollerslev’s Constant Conditional Correlation Tse and Tsui(2002) Dynamic Conditional Correlation Engle(2002) Dynamic Conditional Correlation

108 Constant Conditional correlation (Bollerslev, 1990) Nonlinear combinations of conditional variances from different GARCH models

109 Constant conditional correlation- cont’d Originally, the CCC model had a GARCH(1,1) specification for each Conditional variance in D t: The CCC model has N(N+5)/2 parameters. Ht is positiive definite if and only if all N conditional variances are positive and R is positive definite. The unconditional variances are easy to obtain but unconditional covariances are difficult to cacluate because of nonlinearity.

110 Dynamic Conditional Correlation (Tse and Tsui, 2002)

111 Tse and Tsui’s DCC

112 Tse and Tsui’s DCC

113 Dynamic Conditional Correlation (Engle, 2002)

114 Engle’s DCC

115 Engle DCC output

116 Engle’s DCC

117 Engle’s Dynamic Conditional Correlation

118 Forecasts

119 Forecasts of conditional correlatioin

120 Graphical Conditional Correlation