Advanced Econometrics - Lecture 5 Univariate Time Series Models.

Slides:



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

Autocorrelation Functions and ARIMA Modelling
Copyright(© MTS-2002GG): You are free to use and modify these slides for educational purposes, but please if you improve this material send us your new.
Economics 20 - Prof. Anderson1 Stationary Stochastic Process A stochastic process is stationary if for every collection of time indices 1 ≤ t 1 < …< t.
Using SAS for Time Series Data
Model Building For ARIMA time series
Part II – TIME SERIES ANALYSIS C5 ARIMA (Box-Jenkins) Models
STATIONARY AND NONSTATIONARY TIME SERIES
Nonstationary Time Series Data and Cointegration Prepared by Vera Tabakova, East Carolina University.
Non-stationary data series
Financial Econometrics
Unit Roots & Forecasting
Slide 1 DSCI 5340: Predictive Modeling and Business Forecasting Spring 2013 – Dr. Nick Evangelopoulos Lecture 7: Box-Jenkins Models – Part II (Ch. 9) Material.
Time Series Building 1. Model Identification
R. Werner Solar Terrestrial Influences Institute - BAS Time Series Analysis by means of inference statistical methods.
10 Further Time Series OLS Issues Chapter 10 covered OLS properties for finite (small) sample time series data -If our Chapter 10 assumptions fail, we.
STAT 497 APPLIED TIME SERIES ANALYSIS
Non-Seasonal Box-Jenkins Models
1 Ka-fu Wong University of Hong Kong Modeling Cycles: MA, AR and ARMA Models.
Economics Prof. Buckles1 Time Series Data y t =  0 +  1 x t  k x tk + u t 1. Basic Analysis.
Modeling Cycles By ARMA
1 Ka-fu Wong University of Hong Kong Volatility Measurement, Modeling, and Forecasting.
ARIMA Forecasting Lecture 7 and 8 - March 14-16, 2011
Financial Econometrics
Economics 20 - Prof. Anderson
Modern methods The classical approach: MethodProsCons Time series regression Easy to implement Fairly easy to interpret Covariates may be added (normalization)
1 Ka-fu Wong University of Hong Kong Some Final Words.
Non-Seasonal Box-Jenkins Models
L7: ARIMA1 Lecture 7: ARIMA Model Process The following topics will be covered: Properties of Stock Returns AR model MA model ARMA Non-Stationary Process.
12 Autocorrelation Serial Correlation exists when errors are correlated across periods -One source of serial correlation is misspecification of the model.
AR- MA- och ARMA-.
ARMA models Gloria González-Rivera University of California, Riverside
STAT 497 LECTURE NOTES 2.
Linear Stationary Processes. ARMA models. This lecture introduces the basic linear models for stationary processes. Considering only stationary processes.
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.
Data analyses 2008 Lecture Last Lecture Basic statistics Testing Linear regression parameters Skill.
FAME Time Series Econometrics Daniel V. Gordon Department of Economics University of Calgary.
K. Ensor, STAT Spring 2005 Model selection/diagnostics Akaike’s Information Criterion (AIC) –A measure of fit plus a penalty term for the number.
Week 21 Stochastic Process - Introduction Stochastic processes are processes that proceed randomly in time. Rather than consider fixed random variables.
Econometrics 2 - Lecture 3 Univariate Time Series Models.
K. Ensor, STAT Spring 2004 Memory characterization of a process How would the ACF behave for a process with no memory? What is a short memory series?
Time Series Basics Fin250f: Lecture 8.1 Spring 2010 Reading: Brooks, chapter
3.Analysis of asset price dynamics 3.1Introduction Price – continuous function yet sampled discretely (usually - equal spacing). Stochastic process – has.
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.
MULTIVARIATE TIME SERIES & FORECASTING 1. 2 : autocovariance function of the individual time series.
Auto Regressive, Integrated, Moving Average Box-Jenkins models A stationary times series can be modelled on basis of the serial correlations in it. A non-stationary.
Dynamic Models, Autocorrelation and Forecasting ECON 6002 Econometrics Memorial University of Newfoundland Adapted from Vera Tabakova’s notes.
EC208 – Introductory Econometrics. Topic: Spurious/Nonsense Regressions (as part of chapter on Dynamic Models)
1 Chapter 5 : Volatility Models Similar to linear regression analysis, many time series exhibit a non-constant variance (heteroscedasticity). In a regression.
Previously Definition of a stationary process (A) Constant mean (B) Constant variance (C) Constant covariance White Noise Process:Example of Stationary.
MODELS FOR NONSTATIONARY TIME SERIES By Eni Sumarminingsih, SSi, MM.
TESTING FOR NONSTATIONARITY 1 This sequence will describe two methods for detecting nonstationarity, a graphical method involving correlograms and a more.
TESTING FOR NONSTATIONARITY 1 This sequence will describe two methods for detecting nonstationarity, a graphical method involving correlograms and a more.
Components of Time Series Su, Chapter 2, section II.
Univariate Time series - 2 Methods of Economic Investigation Lecture 19.
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.
Stationarity and Unit Root Testing Dr. Thomas Kigabo RUSUHUZWA.
Analysis of financial data Anders Lundquist Spring 2010.
1 Lecture Plan : Statistical trading models for energy futures.: Stochastic Processes and Market Efficiency Trading Models Long/Short one.
Review of Unit Root Testing D. A. Dickey North Carolina State University (Previously presented at Purdue Econ Dept.)
Analysis of Financial Data Spring 2012 Lecture 4: Time Series Models - 1 Priyantha Wijayatunga Department of Statistics, Umeå University
Financial Econometrics Lecture Notes 2
CHAPTER 16 ECONOMIC FORECASTING Damodar Gujarati
STATIONARY AND NONSTATIONARY TIME SERIES
Unit Root & Augmented Dickey-Fuller (ADF) Test
Lecturer Dr. Veronika Alhanaqtah
BOX JENKINS (ARIMA) METHODOLOGY
Presentation transcript:

Advanced Econometrics - Lecture 5 Univariate Time Series Models

Advanced Econometrics - Lecture 5 Time Series Stochastic Processes Stationary Processes The ARMA Process Deterministic and Stochastic Trends Models with Trend Unit Root Tests Estimation of ARMA Models ARCH and GARCH Models April 16, 2010 Hackl, Advanced Econometrics, Lecture 5 2

Private Consumption April 16, 2010 Hackl, Advanced Econometrics, Lecture 5 3 Private consumption in Austria (in Bn EUR), prices at basis 1995

Private Consumption, cont’d April 16, 2010 Hackl, Advanced Econometrics, Lecture 5 4 Growth of private consumption in Austria (in Bn EUR), prices at basis 1995

Disposable Income April 16, 2010 Hackl, Advanced Econometrics, Lecture 5 5 Disposable income in Austria (in Mio EUR)

Time Series April 16, 2010 Hackl, Advanced Econometrics, Lecture 5 6 Is a time-ordered sequence of observations of a random variable Examples: Annual values of private consumption Changes in expenditure on private consumption Quarterly values of personal disposable income Monthly values of imports Notation: Random variable Y Sequence of observations Y 1, Y 2,..., Y T Deviations from the mean: y t = Y t – E{Y t } = Y t – μ

Components of a Time Series April 16, 2010 Hackl, Advanced Econometrics, Lecture 5 7 Components or characteristics of a time series are Trend Seasonality Irregular fluctuations Time series model: represents the characteristics as well as possible Purpose of modeling Describing the time series Forecasting the future Example: Y t = βt + Σ i γ i D it + ε t with D it = 1 if t corresponds to i-th quarter, D it = 0 otherwise for describing the development of the disposable income

Advanced Econometrics - Lecture 5 Time Series Stochastic Processes Stationary Processes The ARMA Process Deterministic and Stochastic Trends Models with Trend Unit Root Tests Estimation of ARMA Models ARCH and GARCH Models April 16, 2010 Hackl, Advanced Econometrics, Lecture 5 8

9 Stochastic Process Time series: realization of a stochastic process Stochastic process is a sequence of random variables Y t, e.g., {Y t, t = 1,..., n} {Y t, t = -∞,..., ∞} Joint distribution of the Y 1,..., Y n : p(y 1, …., y n ) Of special interest Evolution of the expectation  t = E{Y t } over time Dependence structure over time Example: Extrapolation of a time series as a tool for forecasting April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 10 AR(1)-Process States the dependence structure between consecutive observations as Y t = δ + θY t-1 + ε t, |θ| < 1 with ε t : white noise, i.e., serially uncorrelated, mean zero, V{ε t } = σ² “Autoregressive process of order 1” From Y t = δ + θY t-1 + ε t = δ+θδ +θ²δ +… +ε t + θε t-1 + θ²ε t-2 +… follows E{Y t } = μ = δ(1-θ) -1 In deviations from μ, y t = Y t - , the model is y t = θy t-1 + ε t Autocovariances γ k = Cov{ Y t, Y t-k } k=0: γ 0 = V{Y t } = θ²V{Y t-1 } + V{ε t } = … = Σ i θ 2i σ² = σ²(1-θ²) -1 k=1: γ 1 = Cov{Y t,Y t-1 } = E{(θy t-1 +ε t )y t-1 } = θV{y t-1 } = θσ²(1-θ²) - 1 In general: γ k = Cov{Y t,Y t-k } = θ k σ²(1-θ²) - 1 Depends of k, not of t! April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 11 MA(1)-Process States the dependence structure between consecutive observations as Y t = μ + ε t + αε t-1 with ε t : white noise, V{ε t } = σ² Moving average process of order 1 E{Y t } = μ Autocovariances γ k = Cov{Y t,Y t-k } k=0: γ 0 = V{Y t } = σ²(1+α²) k=1: γ 1 = Cov{Y t,Y t-1 } = ασ² γ k = 0 for k = 2, 3, … Depends of k, not of t! April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 12 AR-Representation of MA- Process The AR(1) can be represented as MA-process of infinite order y t = θy t-1 + ε t = Σ ∞ i=0 θ i ε t-i given that |θ| < 1 Similarly, the AR representation of the MA(1) process y t = αy t-1 – α²y t-2 + … ε t = Σ ∞ i=0 (-1) i α i+1 y t-i-1 + ε t given that |α| < 1 April 16, 2010

Advanced Econometrics - Lecture 5 Time Series Stochastic Processes Stationary Processes The ARMA Process Deterministic and Stochastic Trends Models with Trend Unit Root Tests Estimation of ARMA Models ARCH and GARCH Models April 16, 2010 Hackl, Advanced Econometrics, Lecture 5 13

Hackl, Advanced Econometrics, Lecture 5 14 Stationary Processes Refers to the joint distribution of Y t ’s, in particular to second moments A process is called strictly stationary if its stochastic properties are unaffected by a change of the time origin The joint probability distribution at any set of times is not affected by an arbitrary shift along the time axis Covariance function: γ t,k = Cov{Y t, Y t+k }, k = 0, ±1,… Properties: γ t,k = γ t,-k Weak stationary process: E{Y t } = μ for all t Cov{Y t, Y t+k } = γ k, k = 0, ±1, … for all t and all k Also called covariance stationary process April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 15 AC and PAC Function Autocorrelation function (AC function, ACF) is independent of the scale of Y for a stationary process: ρ k = Corr{Y t,Y t-k } = γ k /γ 0, k = 0, ±1,… Properties: |ρ k | ≤ 1 ρ k = ρ -k ρ 0 = 1 Correlogram: graphical presentation of the AC function Partial autocorrelation function (PAC function, PACF): θ kk = Corr{Y t, Y t-k |Y t-1,...,Y t-k+1 }, k = 0, ±1, … θ kk is obtained from Y t = θ k0 + θ k1 Y t θ kk Y t-k Partial correlogram: graphical representation of the PAC function April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 16 AC and PAC Function, cont‘d Examples for the AC and PAC functions White noise ρ 0 = θ 00 = 1 ρ k = θ kk = 0, if k ≠ 0 AR(1) process, Y t = δ + θY t-1 + ε t ρ k = θ k, k = 0, ±1,…  θ 00 = 1, θ 11 = θ, θ kk = 0 for k > 1 MA(1) process, Y t = μ + ε t + αε t-1  ρ 0 = 1, ρ 1  α/(  α 2 ), ρ k = 0 for k > 1 PAC function: damped exponential if  > 0, otherwise alternating and damped exponential April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 17 AC and PAC Function: Estimates Estimating of AC and PAC function Estimator for ρ k : Estimator for θ kk : coefficient of Y t-k in the regression of Y t on Y t-1, …, Y t-k April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 AR(1) Processes, Verbeek, p.274 April 16,

Hackl, Advanced Econometrics, Lecture 5 MA(1) Processes, Verbeek, p.275 April 16,

Advanced Econometrics - Lecture 5 Time Series Stochastic Processes Stationary Processes The ARMA Process Deterministic and Stochastic Trends Models with Trend Unit Root Tests Estimation of ARMA Models ARCH and GARCH Models April 16, 2010 Hackl, Advanced Econometrics, Lecture 5 20

Hackl, Advanced Econometrics, Lecture 5 21 The ARMA(p,q) Process Generalization of the AR and MA processes: ARMA(p,q) process y t = θ 1 y t-1 + … + θ p y t-p + ε t + α 1 ε t-1 + … + α q ε t-q with ε t : white noise Lag (or shift) operator L (Ly t = y t-1, L 0 y t = Iy t = y t, L p y t = y t-p ) ARMA(p,q) process on operator notation θ(L)y t = α(L)ε t with operator polynomials θ(L) and α(L) θ(L) = I - θ 1 L - … - θ p L p, α(L) = I + α 1 L + … + α q L q April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 22 Lag Operator Lag (or shift) operator L Ly t = y t-1, L 0 y t = Iy t = y t, L p y t = y t-p Algebra of polynomials in L like algebra of variables Examples: (I - ϕ 1 L)(I - ϕ 2 L) = I – ( ϕ 1 + ϕ 2 )L + ϕ 1 ϕ 2 L 2 (I - θL) -1 = Σ ∞ i=0 θ i L i MA(∞) representation of the AR(1) process y t = (I - θL) -1 ε t the infinite sum needs (e.g., finite variance) |θ| < 1 MA(∞) representation of the ARMA(p,q) process y t = [θ (L)] -1 α(L)ε t similarly the AR(∞) representations; invertibility condition: restrictions on parameters April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 23 Invertibility of Lag Polynomials Invertibility condition for I - θL: |θ| < 1 Invertibility condition for I - θ 1 L - θ 2 L 2 : θ(L) = I - θ 1 L - θ 2 L 2 = (I - ϕ 1 L)(I - ϕ 2 L) with ϕ 1 + ϕ 2 = θ 1 and - ϕ 1 ϕ 2 = θ 2 Invertibility conditions: both (I – ϕ 1 L) and (I – ϕ 2 L) invertible; | ϕ 1 | < 1, | ϕ 2 | < 1 Characteristic equation: θ(z) = (1- ϕ 1 z) (1- ϕ 2 z) = 0 Characteristic roots: solutions z 1, z 2 from (1- ϕ 1 z) (1- ϕ 2 z) = 0 Invertibility conditions: |z 1 | > 1, |z 2 | > 1 Can be generalized to lag polynomials of higher order Unit root: a characteristic root of value 1 Polynomial θ(z) evaluated at z = 1: θ(1) = 0, if Σ i θ i = 1 Simple check, no need to solve characteristic equation April 16, 2010

Advanced Econometrics - Lecture 5 Time Series Stochastic Processes Stationary Processes The ARMA Process Deterministic and Stochastic Trends Models with Trend Unit Root Tests Estimation of ARMA Models ARCH and GARCH Models April 16, 2010 Hackl, Advanced Econometrics, Lecture 5 24

Hackl, Advanced Econometrics, Lecture 5 25 Types of Trend Trend: The expected value of a process Y t increases or decreases with time Deterministic trend: a function f(t) of the time, describing the evolution of E{Y t } over time Y t = f(t) + ε t, ε t : white noise Example: Y t = α + βt + ε t describes a linear trend of Y; an increasing trend corresponds to β > 0 Stochastic trend: Y t = δ + Y t-1 + ε t or ΔY t = Y t – Y t-1 = δ + ε t, ε t : white noise  describes an irregular or random fluctuation of the differences ΔY t around the expected value δ  AR(1) – or AR(p) – process with unit root  “random walk with trend” April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 26 Example: Private Consumption Private consumption, AWM database; level values (PCR) and first differences (PCR_D) April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 27 Trends: Random Walk and AR Process Random walk: Y t = Y t-1 + ε t ; random walk with trend: Y t = 0.1 +Y t-1 + ε t ; AR(1) process: Y t = Y t-1 + ε t ; ε t simulated from N(0,1) April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 28 Random Walk with Trends The random walk with trend Y t = δ + Y t-1 + ε t can be written as Y t = Y 0 + δt + Σ i≤t ε i δ: trend parameter Components of the process Deterministic growth path Y 0 + δt Cumulative errors Σ i≤t ε i Properties: Expectation Y 0 + δt is not a fixed value! V{Y t } = σ²t becomes arbitrarily large! Corr{Y t,Y t-k } = √(1-k/t) Non-stationary April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 29 Random Walk with Trends, cont’d From Corr{Y t,Y t-k } = √(1-k/t) follows For fixed k,Y t and Y t-k are the stronger correlated, the larger t With increasing k, correlation tends to zero, but the slower the larger t (long memory property) Comparison of random walk with the AR(1) process Y t = δ + θY t-1 + ε t AR(1) process: ε t-i has the lesser weight, the larger i AR(1) process similar to random walk when θ is close to one April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 30 Non-Stationarity: Consequences AR(1) process Y t = θY t-1 + ε t OLS Estimator for θ: For |θ| < 1: the estimator is Consistent Asymptotically normally distributed For θ = 1 (unit root) θ is underestimated Estimator not normally distributed Spurious regression problem April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 31 Spurious Regression Random walk without trend: Y t = Y t-1 + ε t, ε t : white noise Y t is a non-stationary process, stochastic trend? V{Y t }: a multiple of t Specified model: Y t = α + βt + ε t Deterministic trend Constant variance Misspecified model! Consequences for OLS estimator for β t- and F-statistics: wrong critical limits, rejection probability too large R 2 is about 0.45 although Y t random walk without trend Granger & Newbold, 1974 April 16, 2010

Advanced Econometrics - Lecture 5 Time Series Stochastic Processes Stationary Processes The ARMA Process Deterministic and Stochastic Trends Models with Trend Unit Root Tests Estimation of ARMA Models ARCH and GARCH Models April 16, 2010 Hackl, Advanced Econometrics, Lecture 5 32

Hackl, Advanced Econometrics, Lecture 5 33 How to Model Trends? Specification of Deterministic trend, e.g., Y t = α + βt + ε t : risk of wrong decisions Stochastic trend: analysis of differences ΔY t if a random walk, i.e., a unit root, is suspected Consequences of spurious regression are more serious Consequences of modeling differences: Autocorrelated errors Consistent estimators Asymptotically normally distributed estimators HAC correction of standard errors April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 34 Elimination of a Trend In order to cope with non-stationarity Trend stationary process: the process can be transformed in a stationary process by subtracting the deterministic trend Difference stationary process, or integrated process: stationary process can be derived by differencing Integrated process: stochastic process Y is called integrated of order one if the first differences yield a stationary process: Y ~ I(1) integrated of order d, if the d-fold differences yield a stationary process: Y ~ I(d) April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 35 Trend-Elimination: Examples Random walk Y t = δ + Y t-1 + ε t with white noise ε t ΔY t = Y t – Y t-1 = δ + ε t Y t is a stationary process A random walk is a difference-stationary or I(1) process Linear trend Y t = α + βt + ε t Subtracting the trend component α + βt provides a stationary process Y t is a trend-stationary process April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 36 Integrated Stochastic Processes Random walk Y t = δ + Y t-1 + ε t with white noise ε t is a difference- stationary or I(1) process Many economic time series show stochastic trends From the AWM Database ARIMA(p,d,q) process: d-th differences follow an ARIMA(p,q) process April 16, 2010 Variabled YERGDP, real1 PCRConsumption, real1-2 PYRHousehold's Disposable Income, real1-2 PCDConsumption Deflator2

Advanced Econometrics - Lecture 5 Time Series Stochastic Processes Stationary Processes The ARMA Process Deterministic and Stochastic Trends Models with Trend Unit Root Tests Estimation of ARMA Models ARCH and GARCH Models April 16, 2010 Hackl, Advanced Econometrics, Lecture 5 37

Hackl, Advanced Econometrics, Lecture 5 38 Unit Root Test AR(1) process Y t = δ + θY t-1 + ε t with white noise ε t OLS Estimator for θ: Distribution of DF If |θ| < 1: approximately t(T-1) If θ = 1: critical values of Dickey & Fuller DF test for testing H 0 : θ = 1 against H 1 : θ < 1 θ = 1: characteristic polynomial has unit root April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 39 Dickey-Fuller Critical Values Monte Carlo estimates of critical values for DF 0 : Dickey-Fuller test without intercept DF: Dickey-Fuller test with intercept DF τ : Dickey-Fuller test with time trend April 16, 2010 T p = 0.01p = 0.05p = DF DF DF τ DF DF DF τ N(0,1)

Hackl, Advanced Econometrics, Lecture 5 40 Unit Root Test: The Practice AR(1) process Y t = δ + θY t-1 + ε t with white noise ε t can be written with π = θ-1 as ΔY t = δ + πY t-1 + ε t DF tests H 0 : π = 0 against H 1 : π < 0 DF test statistic Distribution of DF Two steps: 1.Regression of ΔY t on Y t-1 : OLS-estimator for π = θ Test of H 0 : π = 0 against H 1 : π < 0 based on DF; critical values of Dickey & Fuller April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 41 Unit Root Test: Extensions DF test for model with intercept: ΔY t = δ + πY t-1 + ε t DF test for model without intercept: ΔY t = πY t-1 + ε t DF test for model with intercept and trend: ΔY t = δ + γt + πY t-1 + ε t DF tests in all cases H 0 : π = 0 against H 1 : π < 0 Test statistic in all cases Critical values depend on cases April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 42 ADF Test Extended model according to an AR(p) process: ΔY t = δ + πY t-1 + β 1 Δy t-1 + … + β p Δy t-p+1 + ε t Example: AR(2) process Y t = δ + θ 1 Y t-1 + θ 2 Y t-2 + ε t can be written as ΔY t = δ + (θ 1 + θ 2 - 1)Y t-1 – θ 2 ΔY t-1 + ε t the characteristic equation (1 - ϕ 1 L)(1 - ϕ 2 L) = 0 has roots θ 1 = ϕ 1 + ϕ 2 and θ 2 = - ϕ 1 ϕ 2 a unit root implies ϕ 1 = θ 1 + θ 2 =1: Augmented DF (ADF) test Test of H 0 : π = 0 against H 1 : π < 0 Needs its own critical values Extensions similar to the DF-test Phillips-Perron test: alternative method; uses HAC-corrected standard errors April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 43 Example: Price/Earnings Ratio Data set PE: annual time series data on price index and the composite earnings index of the S&P500, Price/earnings ratio Mean 14.6 Min 6.1 Max 36.7 Std 5.1 April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 44 Price/Earnings Ratio, cont’d Extended model according to an AR(2) process gives: ΔY t = – 0.136Y t Δy t Δy t-2 with t-statistics (Y t-1 ), (Δy t-1 ) and (Δy t-2 ) and p-values 0.014, and p-value of the DF statistic 0.121; 1% critical value: % critical value: % critical value: Non-stationarity cannot be rejected for the log PE ratio Unit root test for first differences: DF statistic -7.31, p-value (1% critical value: -3.48) log PE ratio is I(1) However: for sample : DF statistic -3.52, p-value April 16, 2010

Advanced Econometrics - Lecture 5 Time Series Stochastic Processes Stationary Processes The ARMA Process Deterministic and Stochastic Trends Models with Trend Unit Root Tests Estimation of ARMA Models ARCH and GARCH Models April 16, 2010 Hackl, Advanced Econometrics, Lecture 5 45

Hackl, Advanced Econometrics, Lecture 5 46 ARMA Models: Application Application of the ARMA(p,q) model in data analysis: Three steps 1.Model specification, i.e., choice of p, q (and d if an ARIMA model is specified) 2.Parameter estimation 3.Diagnostic checking April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 47 Estimation of ARMA Models The estimation methods are OLS estimation ML estimation AR models: the explanatory variables are Lagged Y t Uncorrelated with ε t OLS estimation April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 48 MA Models: OLS Estimation MA models: Minimization of sum of squared deviations is not straightforward E.g., for an MA(1) model, S(μ,α) = Σ t [Y t - μ - αΣ j=0 (- α) j (Y t-j-1 – μ)] 2  S(μ,α) is a nonlinear function of parameters  needs Y t-j-1 for j=0,1,…, i.e., historical Y s, s < 0 Approximate solution from minimization of S*(μ,α) = Σ t [Y t - μ - αΣ j=0 t-2 (- α) j (Y t-j-1 – μ)] 2 Nonlinear minimization, grid search ARMA models combine AR part with MA part April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 49 ML Estimation Needs an assumption on the distribution of ε t ; usual normality Log likelihood function, conditional on initial value log L(α,θ,μ,σ²) = - (T-1)log(2πσ²)/2 – (1/2) Σ t ε t ²/σ² ε t are functions of the parameters AR(1): ε t = y t - θ 1 y t-1 MA(1): ε t = Σ j=0 t-1 (- α) j y t-j Initial values: y 1 for AR, ε 0 = 0 for MA Extension for exact ML estimator Again, estimation for AR models easier ARMA models combine AR part with MA part April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 50 Model Specification Based on the form of Autocorrelation function (ACF) Partial Autocorrelation function (PACF) Structure of AC and PAC functions typical for AR and MA processes Example: MA(1) process: ρ 0 = 1, ρ 1 = α/(1-α²); ρ i = 0, i = 2, 3, … AR(1) process: ρ k = θ k, k = 0, ±1,… April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 51 ARMA(p,q)-Processes Condition for AR(p) θ(L)Y t = ε t MA(q) Y t = α (L) ε t ARMA(p,q) θ(L)Y t = α (L) ε t Stationarity roots z i of θ(z)=0: | z i | > 1 always stationary roots z i of θ(z)=0: | z i | > 1 Invertibilityalways invertible roots z i of α (z)=0: | z i | > 1 AC functiondamped, infinite  k = 0 for k > q damped, infinite PAC function  kk = 0 for k > p damped, infinite April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 52 Empirical AC and PAC Function Estimation of the AC and PAC functions AC ρ k : PAC θ kk : coefficient of Y t-k in regression of Y t on Y t-1, …, Y t-k MA(q) process: standard errors for r k, k > q from √T(r k – ρ k ) → N(0, v k ) with v k = 1 + 2ρ 1 ² + … + 2ρ k ² test of H 0 : ρ 1 = 0: compare √Tr 1 with critical value from N(0,1), etc. AR(p) process: test of H 0 : ρ k = 0 for k > p based on asymptotic distribution April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 53 Diagnostic Checking ARMA(p,q): Adequacy of choices p and q Analysis of residuals from fitted model: Correct specification: residuals are realizations of white noise Portmanteau test: for a ARMA(p,q) process follows the Chi-squared distribution with K-p-q df Overfitting Starting point: a general model Comparison with a model with reduced number of parameters: AIC or BIC AIC: tends to result asymptotically in overparameterized models April 16, 2010

Advanced Econometrics - Lecture 5 Time Series Stochastic Processes Stationary Processes The ARMA Process Deterministic and Stochastic Trends Models with Trend Unit Root Tests Estimation of ARMA Models ARCH and GARCH Models April 16, 2010 Hackl, Advanced Econometrics, Lecture 5 54

Hackl, Advanced Econometrics, Lecture 5 55 ARCH Processes Autoregressive Conditional Heteroskedasticity (ARCH): Special case of heteroskedasticity Error variance: autoregressive behavior Allows to model successive periods with high, other periods with small volatility Typical for asset markets Example: y t = x t ’θ+ ε t with ε t = σ t v t, v t ~ NID(0,1)  the conditional error variance, given the information I t-1, is σ t ²  ARCH(1) process σ t ² = E{ε t ²| I t-1 } = ϖ + αε t-1 ²  I t-1 is the information set containing all past including ε t-1 April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 56 The ARCH(1) Process ARCH(1) process describes the conditional error variance, i.e., the variance conditional on information dated t-1 and earlier σ t ² = E{ε t ²| I t-1 } = ϖ + αε t-1 ² I t-1 is the information set containing all past including ε t-1 Conditions for σ t ² ≥ 0: ϖ ≥ 0, α ≥ 0 A big shock at t-1, i.e., a large value |ε t-1 |,  Induces high volatility, i.e., large σ t ²  makes large values |ε t | more likely at t (and later) ARCH process does not imply correlation! The unconditional variance of ε t is σ² = E{ε t ²} = ϖ + αE{ε t-1 ²} = ϖ /(1 - α) given that 0 ≤ α < 1 The ε t process is stationary April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 57 More ARCH Processes Various generalizations ARCH(p) process σ t ² = ϖ + α 1 ε t-1 ² + … α p ε t-p 2 = ϖ + α(L)ε t-1 ² with lag polynomial α(L) of order p-1 Conditions for σ t ² ≥ 0: ϖ ≥ 0; α i ≥ 0, i = 1,…p Condition for stationarity: α(1) < 1 GARCH(p,q) process „Generalized ARCH“ Similar to the ARMA representation of levels σ t ² = ϖ + α 1 ε t-1 ² + … α p ε t-p 2 + β 1 σ t-1 ² + … + β q σ t-q ² = = ϖ + α(L)ε t-1 ² + β(L)σ t-1 ² E.g., GARCH(1,1): σ t ² = ϖ + αε t-1 ² + βσ t-1 ²; with “surprises“ v t = ε t-1 ² - σ t ²: ε t ² = ϖ + (α + β)ε t-1 ² + v t - βv t-1 ², i.e. ε t ² follow ARMA(1,1) April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 58 Test for ARCH Processes Null hypothesis of homoskedasticity, to be tested against the alternative ARCH(q) 1.Estimate the model of interest using OLS: residuals e t 2.Auxiliary regression of squared residuals e t 2 on a constant and q lagged e t 2 3.Test statistic TR e 2 with R e 2 from the auxiliary regression, p-value from the chi squared distribution with q df April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 59 More ARCH Processes, cont’d EGARCH or exponential GARCH log σ t ² = ϖ + βσ t-1 ² + γε t-1 /σ t-1 + α|ε t-1 |/σ t-1 Asymmetric if γ ≠ 0  γ < 0: positive shocks („good news“) reduce volatility April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 60 Time Series Models in GRETL Model > Time Series > ARIMA Estimates an ARMA model, with or without exogenous regressors Model > Time Series > ARCH Estimates the specified model allowing for ARCH: (1) model estimated via OLS, (2) auxiliary regression of the squared residual on its own lagged values, (3) weighted least squares estimation Model > Time Series > GARCH Estimates a GARCH model, with or without exogenous regressors April 16, 2010

Hackl, Advanced Econometrics, Lecture 5 61 Exercise Answer questions a. to e. of Exercise 8.2 of Verbeek data from the data sets “SP500” containing daily returns on Standard & Poor's 500 index from January 1981 to April 1991, computed as the change in log index April 16, 2010