Empirical
Finance:Analysis of non-stationary processes II: Estimating and testing long-run relationships in systems of equations
Module Leader: Dr Stuart Fraser
stuart.fraser@wbs.ac.uk
代写留学生论文Room D1.18 (
Social Studies)Warwick Business School 2
Today
1. Key problems with Engle-Granger 2 Step Estimator.
2. VAR and VECM models.
3. ‘Johansen’: A systems approach to testing for cointegration.
Seminar 8: Testing PPP using Johansen
Warwick Business School 3
Problems with EG 2 step: 1. Presence of multiple cointegratingrelationships.A key problem with EG 2 step is that it can only identify a single cointegratingrelationship.Single equation OLS is used to estimate the long-run (Step 1).However with n variables there can be up to n-1cointegratingrelationships (see Appendix 1):If EG 2 Step is used in this context we may end up estimating some unidentified linear combinationof all the cointegratingrelationships.1−≤nr‘r’is standard notation for the number of cointegratingrelationshipsNeed at least one I(1) linear combinationwhich corresponds to the common stochastic trend(s):n-r≥1If r=n then all the variables must be individually I(0) (see Appendix 1). ⇒with n>2 there can be multiple cointegratingrelationships (r>1)
Warwick Business School 4
Example: Testing the Expectations Hypo
thesis of the term structure (see Brooks 7.12 and lecture 7).A test of the EH could be based on testing for cointegrationamongst pairs of yieldsof different maturities.However, a more comprehensive (and powerful) test could be based on a vector of yieldsfrom across the maturity spectrumNormalizing the spreads on the one period spot rate, EH implies there are n-1 (linearly independent) spreads which are I(0): ()()()()1,1~1,1~1001010100112CIyCIRRRtntttβ′⇒⎟⎟⎟⎟⎟⎠⎞⎜⎜⎜⎜⎜⎝⎛⎟⎟⎟⎟⎟⎠⎞⎜⎜⎜⎜⎜⎝⎛−−−MKMMKK()()njCIRRjtt,,2 ,1,1~K=−(). theare ) of (rows of columns Thematrix. 1an is vectors ingcointegratβββ′−×nnIn matrix form ⇒()()()()()1,1~,,2,CIRRnjRRktjtjttjt−⇒=+Τ+=KεThe EH (+rational expectations)implies the yield spreads arecointegrated.Warwick Business School 5
Problems with EG 2 step: 2. EG 2 step is an inefficient estimator On one level the distinction between exogenous and endogenous variables is unimportant with cointegratedvariables.–Super-consistency still holds even if X is endogenous:So in effect it doesn’t matter which variable (Y or X) we make the ‘dependent variable’:ββεββ⎯→⎯+=ΣΣ−−piiixTxTˆˆ211With cointegratedvariables super-consistency still holdsin the long run equation even if X is endogenous. Why? Because the sample variance of X tends to ∞(X is I(1)).This is in sharp contrast to the CLRM where Y and X are assumed to be I(0). In that case endogeneityimplies the OLS estimator is inconsistent (see Lecture 4).***ttttttYXXYεβμεβμ++=++=Either regression will yield super-consistent estimates of the long run parameters.For example, with PPP we ‘normalized’the relationshipon log(S) (log nominal exchange rate). However we might just as well have normalized on log(P) or log(P*).Warwick Business School 6
Problems with EG 2 step: inefficiencyHowever this also means we can write an ECM for X. Starting from the ADL for X……the corresponding ECM for X is:Therefore Y and X form a systemof
本论文由英语论文网提供整理,提供论文代写,英语论文代写,代写论文,代写英语论文,代写留学生论文,代写英文论文,留学生论文代写相关核心关键词搜索。