11
Co-integration and VECM
YiYi--NungNung YangYangCYCU, TaiwanCYCU, Taiwan
May, 2012May, 2012
-
Page Page 22
Learning objectives
Integrated variablesCo-integrationVector Error correction model (VECM)Engle-Granger 2-step co-integration testJohansen co-integration test
-
Page Page 33
Integrated variables
YYtt is an is an I(nI(n) variable if its ) variable if its nn--thth difference difference nnYYtt ~ I(0)~ I(0)
We call We call YYtt is is integrated of order nintegrated of order n..An I(0) variable is a stationary variableAn I(0) variable is a stationary variable
E.g., E.g., YYtt is an I(1) variable if its 1is an I(1) variable if its 1stst--difference: difference: YYtt ~ I(0)~ I(0)
-
Page Page 44
Summation of I(n) variables
I(mI(m) + ) + I(nI(n) == ) == I(mI(m))if if mmnn ((m,nm,n are integers)are integers)Y ~ I(1), X ~ I(0)Y ~ I(1), X ~ I(0)Let z=Y + X, then z ~I(1) Let z=Y + X, then z ~I(1)
I(nI(n) + ) + I(nI(n) == ) == I(nI(n))Example: Example:
Y ~ I(1), X ~ I(1)Y ~ I(1), X ~ I(1)Let z=Y + X, then z ~I(1) Let z=Y + X, then z ~I(1) The only exception isThe only exception is
If Y and X are If Y and X are cointegratedcointegrated
-
Page Page 55
Co-integration
If Y ~ If Y ~ I(nI(n), X ~ ), X ~ I(nI(n), and n ), and n 11Let z=Let z=11Y + Y + 22 XXif z ~I(0), then we callif z ~I(0), then we callY and X are Y and X are cointegratedcointegrated oror
There is a There is a coco--integration relationshipintegration relationship between these between these two nonetwo none--stationary variables, Y and X.stationary variables, Y and X.
-
Page Page 66
Co-integrationDef:Def:
There exists a There exists a linear combinationlinear combination of the noneof the none--stationary variables (integrated of the same orders) stationary variables (integrated of the same orders) that is stationarythat is stationaryThere could be There could be more than onemore than one such linear such linear combination if there are more than 2 variables combination if there are more than 2 variables involved in studiesinvolved in studies
The basic idea of The basic idea of cointegrationcointegration relates closely to the relates closely to the concept of unit rootsconcept of unit roots
Economic implicationsEconomic implicationsThere exists (longThere exists (long--run) equilibrium relationships run) equilibrium relationships between (among) nonebetween (among) none--stationary variables.stationary variables.
-
Page Page 77
Co-integrating vectors
for example, a Demand of money model:for example, a Demand of money model:m m p = p = 00 + + 11 y + y + 22 r + er + ewherewhere
m: nominal money supplym: nominal money supplyp: price levelp: price levely: incomey: incomer: interest rater: interest ratee: error terme: error term
-
Page Page 88
Co-integrating vectors
In equilibriumIn equilibriumm m p p 11 y y 22 r r 0 0 = e= eThe above model can be rewritten asThe above model can be rewritten asx x = e= e
wherewherex = (m, p, y, r, const)x = (m, p, y, r, const) = (= (11, , --11, , --11, , --22, , --00))
is called is called CoCo--integrating vectorintegrating vector
-
Page Page 99
VAR of order p and VECM
Consider a VAR of order p with a Consider a VAR of order p with a deterministic part given by Adeterministic part given by A00(all in matrix form)(all in matrix form)yytt = A= A00+A+A11yytt11+ A+ A22yytt22+++ + AApp yyttpp + e+ ettwe can rewe can re--write the above as VECMwrite the above as VECM(it is a long story(it is a long story))
yytt=A=A00++yytt11++11yytt11++++pp--11 yyttp+1p+1+ e+ ettPlease refer to Enders (2008)Please refer to Enders (2008)
-
Page Page 1010
VAR of order p and VECM
For example, a VAR(1) model For example, a VAR(1) model yytt = A= A00+A+A11yytt1 1 + e+ ettwe can rewe can re--write the above as VECMwrite the above as VECMyytt=A=A00++yytt11+ e+ ett can be decomposed intocan be decomposed into==
: matrix of short: matrix of short--run adjusting coefficientsrun adjusting coefficients: co: co--integrating vector (matrix)integrating vector (matrix)
-
Page Page 1111
Granger representation theorem
Let Y = (yLet Y = (y1, 1, yy2, 2, ,,yykk), ), Y~I(nY~I(n))If all variables in Y are If all variables in Y are cointegratedcointegrated
CoCo--integrationintegrationThere exist There exist at least oneat least one coco--integrating vector to let:integrating vector to let:Y ~ Y ~ I(nI(n--jj), for any j), for any j1.1.
VECMVECMIf and only if the variables in Y are coIf and only if the variables in Y are co--integrated, integrated, there must exist a VECMthere must exist a VECMCointegrationCointegration VECMVECM
-
Page Page 1212
VAR(1) system as an example
yytt = A= A00+A+A11yytt11 + e+ ett
VECM VECM (if all variables are co(if all variables are co--integrated)integrated)yytt=A=A00++yytt11+ e+ ett
-
Page Page 1313
Key concepts before running software
All nonAll non--stationary variables integrated of the stationary variables integrated of the same order may have cosame order may have co--integration integration relationshiprelationshipSingle equation version of Single equation version of cointegrationcointegration
exampleexampleyytt = a + = a + xxtt + + uutt, , yytt,, xxtt ~ I(1) but u~ I(1) but utt~I(0)~I(0)CoCo--integrating vectorintegrating vector(1, (1, --, , --a)a)ECM (error correction model)ECM (error correction model)yytt = a = a --(y(ytt--11--xxtt--11))++xxtt + + uutt
==a a -- uutt--11++xxtt + + uutt
-
Page Page 1414
Engle-Granger co-integration test
Engle-Granger 2-step co-integration testmake sure orders of integrated variables are the same. Usually, they are all I(1) variableStep 1. run OLS by regressing yt on xts save residuals as a variable, e.g., etStep 2. use ADF test to see whether et is a I(0) variable:
If et ~ I(0), then the variables are cointegrated;otherwise, there is no cointegration among(between) those variables.
-
Page Page 1515
Plots of Y1 and X1
-
Page Page 1616
Case9.1 spurious regression and Engle-Granger co-integration test
1. use gretl to open ex-COINT.wf1.Run regression as follows y1t=a0+a1x1tResults
-
Page Page 1717
Case9.1 spurious regression and Engle-Granger co-integration test
2. Conduct ADF tests on Y1 and X1(you may use or as follows)
StrategiesStrategies(a) show ADF statistics with various lags: (a) show ADF statistics with various lags: no matter how many lags, they all have unit rootsno matter how many lags, they all have unit roots(b) use any use any information criterions to choose the optimal lags and report them in tables
Notes: ADF tests should be done with level and 1stdifferenced variables
-
Page Page 1818
Case9.1 spurious regression and Engle-Granger co-integration test
2.(a) show ADF statistics with various lags: 2.(a) show ADF statistics with various lags: no matter how many lags, they all have unit no matter how many lags, they all have unit rootsroots
Variable ADF-statistic
1% c.v.
5% c.v.
Note: c.v. are from Eviews.
-
Page Page 1919
Case9.1 spurious regression and Engle-Granger co-integration test
2.(b) use any use any information criterions to choose the optimal lags and report them in tables
Suggestions:Sample size 300, use BIC or HQC
-
Page Page 2020
-
Page Page 2121
-
Page Page 2222
Case9.1 spurious regression and Engle-Granger co-integration test
2.(b) use any use any information criterions to choose the optimal lags and report them in tables
0.000.000.000.000.730.730.560.56
PP--valuevalue
135.88135.8800--11.81611.816XX11tt
72.1772.1700--12.18812.188YY11tt
148.62148.62000.1550.155X1X1tt
93.3993.3900--0.3510.351Y1Y1tt
AICAIClagslagsADFADFvariablevariable
-
Page Page 2323
Case9.1 spurious regression and Engle-Granger co-integration test
3(a). Save residuals from OLS in step 1 as e1t, then do ADF test as in step 2(a) or 2(b) on these residuals.
Variable ADF-statistic
1% c.v.
5% c.v.
Note: c.v. are from Eviews.
-
Page Page 2424
Case9.1 spurious regression and Engle-Granger co-integration test
3(b). Use gretl: [model]->[Time series]->[Cointegration test]->[Engle-Granger]
-
Page Page 2525
-
Page Page 2626
9.5 Johansen co-integration test
Johansen co-integration test procedure 1. use VAR lags selection to choose p2. by Johansens method, testing if variables are co-integrated.3. make sure rank() to see how many co-integrating vectors: i.e., based on trace test ormax tests.
4. normalizing integrating vectors and give an interpretations
-
Page Page 2727
Rank and the number of co-integrating vectors
The number of Rank of The number of Rank of = the number of = the number of coco--integrating vectorsintegrating vectors
-
Page Page 2828
trace test ormax tests
Two often used in determine the number of coTwo often used in determine the number of co--integrating vector (number of integrating vector (number of Rank(Rank()
-
Page Page 2929
Normalizing integrating vectors
If If
11Y + Y + 22 X = eX = ethis can be rethis can be re--written aswritten asY + (Y + (22//11)) X = e/X = e/11
This is called normalization (on This is called normalization (on , co, co--integrating vectorintegrating vector))..
-
Page Page 3030
Case 9.2 Johansen co-integration test
1. plot x, y, z
-
Page Page 3131
Case 9.2 Johansen co-integration test
Selection of VAR lagsSelection of VAR lags(e.g., max lag=8)(e.g., max lag=8)Use BIC to select p=1Use BIC to select p=1
-
Page Page 3232
Case 9.2 Johansen co-integration testFill with
p+1
-
Page Page 3333
Case 9.2 Johansen co-integration testresults (I)
-
Page Page 3434
Case 9.2 Johansen co-integration testresults (II)
-
Page Page 3535
Case 9.2 Johansen co-integration test
There is only 1 Cointegrating vectorThere is only 1 Cointegrating vector
==(1, (1, yy, , zz, const, const) ) =(1.000, =(1.000, --0.984, 0.984, --1.018, 1.018, --0.013)0.013)
This implies a longThis implies a long--run relationship:run relationship:eett--11 = X= Xtt--11 --0.984Y0.984Ytt--11 --1.018Z1.018Ztt--11 --0.0130.013