14 Vector Autoregressions, Unit Roots, and Cointegration.

Slides:



Advertisements
Similar presentations
Cointegration and Error Correction Models
Advertisements

Autocorrelation Functions and ARIMA Modelling
Introduction Describe what panel data is and the reasons for using it in this format Assess the importance of fixed and random effects Examine the Hausman.
Financial Econometrics
Dynamic panels and unit roots
Multiple Regression and Model Building
Structural modelling: Causality, exogeneity and unit roots Andrew P. Blake CCBS/HKMA May 2004.
Using SAS for Time Series Data
Use of Business Tendency Survey Results for Forecasting Industry Production in Slovakia Use of Business Tendency Survey Results for Forecasting Industry.
Nonstationary Time Series Data and Cointegration Prepared by Vera Tabakova, East Carolina University.
Christopher Dougherty EC220 - Introduction to econometrics (chapter 13) Slideshow: cointegration Original citation: Dougherty, C. (2012) EC220 - Introduction.
Non-stationary data series
Economics 20 - Prof. Anderson1 Testing for Unit Roots Consider an AR(1): y t =  +  y t-1 + e t Let H 0 :  = 1, (assume there is a unit root) Define.
Advanced Time Series PS 791C. Advanced Time Series Techniques A number of topics come under the general heading of “state-of-the-art” time series –Unit.
1 MF-852 Financial Econometrics Lecture 11 Distributed Lags and Unit Roots Roy J. Epstein Fall 2003.
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.
Regression with Time-Series Data: Nonstationary Variables
Time Series Building 1. Model Identification
Vector Error Correction and Vector Autoregressive Models
FITTING MODELS WITH NONSTATIONARY TIME SERIES 1 Detrending Early macroeconomic models tended to produce poor forecasts, despite having excellent sample-period.
Empirical study of causality between Real GDP and Monetary variables. Presented by : Hanane Ayad.
13 Introduction toTime-Series Analysis. What is in this Chapter? This chapter discusses –the basic time-series models: autoregressive (AR) and moving.
Economics Prof. Buckles1 Time Series Data y t =  0 +  1 x t  k x tk + u t 1. Basic Analysis.
Chapter 9 Simultaneous Equations Models. What is in this Chapter? In Chapter 4 we mentioned that one of the assumptions in the basic regression model.
Prediction and model selection
ARIMA Forecasting Lecture 7 and 8 - March 14-16, 2011
Chapter 11 Multiple Regression.
Economics 20 - Prof. Anderson
Topic 3: Regression.
12 Autocorrelation Serial Correlation exists when errors are correlated across periods -One source of serial correlation is misspecification of the model.
Chapter 15 Forecasting Copyright © 2011 Pearson Addison-Wesley. All rights reserved. Slides by Niels-Hugo Blunch Washington and Lee University.
ECON 6012 Cost Benefit Analysis Memorial University of Newfoundland
Hypothesis Testing in Linear Regression Analysis
Regression Method.
#1 EC 485: Time Series Analysis in a Nut Shell. #2 Data Preparation: 1)Plot data and examine for stationarity 2)Examine ACF for stationarity 3)If not.
Various topics Petter Mostad Overview Epidemiology Study types / data types Econometrics Time series data More about sampling –Estimation.
Chapter 4 Linear Regression 1. Introduction Managerial decisions are often based on the relationship between two or more variables. For example, after.
Cointegration in Single Equations: Lecture 6 Statistical Tests for Cointegration Thomas 15.2 Testing for cointegration between two variables Cointegration.
The Properties of Time Series: Lecture 4 Previously introduced AR(1) model X t = φX t-1 + u t (1) (a) White Noise (stationary/no unit root) X t = u t i.e.
Multiple Regression. Simple Regression in detail Y i = β o + β 1 x i + ε i Where Y => Dependent variable X => Independent variable β o => Model parameter.
1 Another useful model is autoregressive model. Frequently, we find that the values of a series of financial data at particular points in time are highly.
How do we identify non-stationary processes? (A) Informal methods Thomas 14.1 Plot time series Correlogram (B) Formal Methods Statistical test for stationarity.
3.Analysis of asset price dynamics 3.1Introduction Price – continuous function yet sampled discretely (usually - equal spacing). Stochastic process – has.
Correlation & Regression Analysis
EC208 – Introductory Econometrics. Topic: Spurious/Nonsense Regressions (as part of chapter on Dynamic Models)
The Box-Jenkins (ARIMA) Methodology
Module 4 Forecasting Multiple Variables from their own Histories EC 827.
Cointegration in Single Equations: Lecture 5
Components of Time Series Su, Chapter 2, section II.
Univariate Time series - 2 Methods of Economic Investigation Lecture 19.
MBF1413 | Quantitative Methods Prepared by Dr Khairul Anuar 8: Time Series Analysis & Forecasting – Part 1
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.
Advanced Statistical Methods: Some more topics in time series Gunjan Malhotra
Dr. Thomas Kigabo RUSUHUZWA
An Introduction to Error Correction Models (ECMs)
Time Series Econometrics
Financial Econometrics Lecture Notes 4
Nonstationary Time Series Data and Cointegration
An Introduction to Macroeconometrics: VEC and VAR Models
VAR models and cointegration
Ch8 Time Series Modeling
CHAPTER 16 ECONOMIC FORECASTING Damodar Gujarati
COINTEGRATION AND ERROR CORRECTION MODELS
Unit Roots 31/12/2018.
Unit Root & Augmented Dickey-Fuller (ADF) Test
Introduction to Time Series
Vector AutoRegression models (VARs)
Presentation transcript:

14 Vector Autoregressions, Unit Roots, and Cointegration

What is in this Chapter? This chapter discusses work on time- series analysis starting in the 1980s. – First there is a discussion of vector autoregression models, –Next we talk of the different unit root tests. –Finally, we discuss cointegration, which is a method of analyzing long-run relationships between nonstationary variables. We discuss tests for cointegration and estimation of cointegrating relationships

14.2 Vector Autoregressions In previous sections we discussed the analysis of a single time series When we have several time series, we need to take into account the interdependence between them One way of doing this is to estimate a simultaneous equations model as discussed in Chapter 9 but with lags in all the variables Such a model is called a dynamic simultaneous equations model

14.2 Vector Autoregressions However, this formulation involves two steps: –first, we have to classify the variables into two categories, endogenous and exogenous, –second, we have toimpose some constraints on the parameters to achieve identification. Sims argues that both these steps involve many arbitrary decisions and suggests as an alternative, the vectorautoregression (VAR) approach. This is just a multiple time-series generalization of the AR model. The VAR model is easy to estimate because we can use the OLS method

14.2 Vector Autoregressions

14.3 Problems with VAR Models in Practice We have considered only a simple model with two variables and only one lag for each. In practice, since we are not considering any moving average errors, the autoregressions would probably have to have more lags to be useful for prediction Otherwise, univariate ARMA models would do better.

14.3 Problems with VAR Models in Practice Suppose that we consider say six lags for each variable and we have a small system with four variables Then each equation would have 24 parameters to be estimated and we thus have 96 parameters to estimate overall This overparameterization is one of the major problems with VAR models.

14.3 Problems with VAR Models in Practice One such model that has been found particularly useful in prediction is the Bayesian vectorautoregression (BVAR) In BVAR we assign some prior distributions for the coefficients in the vector autoregressions In each equation, the coefficient of the own lagged variable has a prior mean 1, all others have prior means 0, with the variance of the prior decreasing as the lag length increases For instance, with two variables y1t and y2t and four lags for each, the first equation will be

14.3 Problems with VAR Models in Practice

14.4 Unit Roots

14.5 Unit Root Tests

The Low Power of Unit Root Tests –Schwert (1989) first presented Monte Carlo evidence to point out the size distortion problems of the commonly used unit root tests: the ADF and PP tests. – Whereas Schwert complained about size distortions, DeJong et al. complained about the low power of unit root tests –They argued that the unit root tests have low power against plausible trend-stationary alternatives

14.5 Unit Root Tests –They argue that the PP tests have very low power (generally less than 0.10) against trend-stationary alternatives but the ADF test has power approaching 0.33 and thus is likely to be more useful in practice. –They conclude that tests with higher power need to be developed

14.5 Unit Root Tests

Structural Change and Unit Roots –In all the studies on unit roots, the issue of whether a time series is of the DS or TS type was decided by analyzing the series for the entire time period during which many major events took place –The Nelson-Plosser series, for instance, covered the period ,which includes the two world wars and the Depression of the 1930s

14.5 Unit Root Tests If there have been any changes in the trend because of these events, the results obtained by assuming a constant parameter structure during the entire period will be suspect Many studies done using the traditional multiple regression methods have included dummy variables (see Sections 8.2 and 8.3) to allow for different intercepts (and slopes) Rappoport and Richlin (1989) show that a segmented trend model is a feasible alternative to the DS model.

14.5 Unit Root Tests Perron (1989) argues that standard tests for the unit root hypothesis against the trend-stationary (TS) alternatives cannot reject the unit root hypothesis if the time series has a structural break. Of course, one can also construct examples where, for instance –y1, y2,, ym is a random walk with drift –ym+1,..., ym+n is another random walk with a different drift –and the combined series is not the DS type

14.5 Unit Root Tests Perron's study was criticized on the argument that he "peeked at the data" before analysis— that after looking at the graph, he decided that there was a break But Kim (1990), using Bayesian methods, finds that even allowing for an unknown breakpoint, the standard tests of the unit root hypothesis were biased in favor of accepting the unit root hypothesis if the series had a structural break at some intermediate date.

14.5 Unit Root Tests When using long time series, as many of these studies have done, it is important to take account of structural changes. Parameter constancy tests have frequently been used in traditional regression analysis.

14.6 Cointegration

In the Box-Jenkins method, if the time series is nonstationary (as evidenced by the correlogram not damping), we difference the series to achieve stationarity and then use elaborate ARMA models to fit the stationary series. When we are considering two time series, yt and Xt say, we do the same thing. This differencing operation eliminates the trend or long-term movement in the series

14.6 Cointegration However, what we may be interested in is explaining the relationship between the trends in yt and Xt We can do this by running a regression of Yt on Xt, but this regression will not make sense if a long-run relationship does not exist. By asking the question of whether yt and Xt are cointegrated, we are asking whether there is any long-run relationship between the trends in yt and xt.

14.6 Cointegration The case with seasonal adjustment is similar Instead of eliminating the seasonal components from y and x and then analyzing the de-seasonalized data, we might also be asking whether there is a relationship between the seasonals in y and x This is the idea behind "seasonal cointegration. Note that in this case we do not considerfirst differences or I(1) processes For instance, with monthly data we consider twelfthdifferences yt-yt-12 Similarly, for Xt we consider Xt-Xt-12

14.7 The Cointegrating Regression

14.9 Cointegration and Error Correction Models If xt and yt are cointegrated, there is a long-run relationship between them Furthermore, the short-run dynamics can be described by the error correction model (ECM) This is known as the Granger representation theorem If Xt ~ I(1), yt ~I(1), and zt = yt -βxt is I(0), then x and y are said to be cointegrated The Granger representation theorem says that in this case Xt and yt may be considered to be generated by ECMs:

14.9 Cointegration and Error Correction Models

14.10 Tests for Cointegration