VAR Models and Cointegration 1

Transcription

1 VAR Models and Cointegration 1 Sebastian Fossati University of Alberta 1 These slides are based on Eric Zivot s time series notes available at:

2 The Cointegrated VAR In a series of important papers Soren Johansen firmly roots cointegration and error correction models in a vector autoregression framework. Consider the levels VAR(p) for the (n 1) vector Y t Y t = ΦD t + Π 1 Y t Π p Y t p + ε t, t = 1,..., T D t = deterministic terms 2 / 64

3 The VAR(p) model is stable if det(i n Π 1 z Π p z p ) = 0 has all roots outside the complex unit circle. If there are roots on the unit circle then some or all of the variables in Y t are I (1) and they may also be cointegrated. If Y t is cointegrated then the VAR representation is not the most suitable representation for analysis because the cointegrating relations are not explicitly apparent. 3 / 64

4 The cointegrating relations become apparent if the levels VAR is transformed to the vector error correction model (VECM) Y t = ΦD t + ΠY t 1 + Γ 1 Y t Γ p 1 Y t p+1 + ε t Π = Π Π p I n p Γ k = Π j, k = 1,..., p 1 j=k+1 In the VECM, Y t and its lags are I (0). The term ΠY t 1 is the only one which includes potential I (1) variables and for Y t to be I (0) it must be the case that ΠY t 1 is also I (0). Therefore, ΠY t 1 must contain the cointegrating relations if they exist. If the VAR(p) process has unit roots (z = 1) then det(i n Π 1... Π p ) = 0 det(π) = 0 Π is singular 4 / 64

5 If Π is singular then it has reduced rank; that is rank(π) = r < n. There are two cases to consider: 1. rank(π) = 0. This implies that Π = 0 Y t I (1) and not cointegrated The VECM reduces to a VAR(p 1) in first differences Y t = ΦD t +Γ 1 Y t Γ p 1 Y t p+1 + ε t 5 / 64

6 2. 0 < rank(π) = r < n. This implies that Y t is I (1) with r linearly independent cointegrating vectors and n r common stochastic trends (unit roots). Since Π has rank r it can be written as the product Π = (n n) α β (n r)(r n) where α and β are (n r) matrices with rank(α) = rank(β) = r. The rows of β form a basis for the r cointegrating vectors and the elements of α distribute the impact of the cointegrating vectors to the evolution of Y t. The VECM becomes Y t = ΦD t + αβ Y t 1 + Γ 1 Y t Γ p 1 Y t p+1 + ε t, where β Y t 1 I (0) since β is a matrix of cointegrating vectors. 6 / 64

7 Non-uniqueness It is important to recognize that the factorization Π = αβ is not unique since for any r r nonsingular matrix H we have αβ = αhh 1 β = (ah)(βh 1 ) = a β Hence the factorization Π = αβ only identifies the space spanned by the cointegrating relations. To obtain unique values of α and β requires further restrictions on the model. 7 / 64

8 Example: A bivariate cointegrated VAR(1) model Consider the bivariate VAR(1) model for Y t = (y 1t, y 2t ) The VECM is Y t = Π 1 Y t 1 + ɛ t Y t = ΠY t 1 + ε t Π = Π 1 I 2 Assuming Y t is cointegrated there exists a 2 1 vector β = (β 1, β 2 ) such that β Y t = β 1 y 1t + β 2 y 2t I (0) Using the normalization β 1 = 1 and β 2 = β the cointegrating relation becomes β Y t = y 1t βy 2t This normalization suggests the stochastic long-run equilibrium relation y 1t = βy 2t + u t 8 / 64

9 Since Y t is cointegrated with one cointegrating vector, rank(π) = 1 so that ( ) ( ) Π = αβ α ( ) 1 α 1 α 1 β = 1 β = α 2 β α 2 The elements in the vector α are interpreted as speed of adjustment coefficients. The cointegrated VECM for Y t may be rewritten as α 2 Y t = αβ Y t 1 + ε t Writing the VECM equation by equation gives y 1t = α 1 (y 1t 1 βy 2t 1 ) + ε 1t y 2t = α 2 (y 1t 1 βy 2t 1 ) + ε 2t The stability conditions for the bivariate VECM are related to the stability conditions for the disequilibrium error β Y t. 9 / 64

10 It is straightforward to show that β Y t follows an AR(1) process β Y t = (1 + β α)β Y t 1 + β ε t or u t = φu t 1 + v t, u t = β Y t φ = 1 + β α = 1 + (α 1 βα 2 ) v t = β ε t = u 1t βu 2t The AR(1) model for u t is stable as long as φ = 1 + (α 1 βα 2 ) < 1 For example, suppose β = 1. Then the stability condition is φ = 1 + (α 1 α 2 ) < 1 which is satisfied if α 1 α 2 < 0 and α 1 α 2 > / 64

11 Johansen s Methodology for Modeling Cointegration The basic steps in Johansen s methodology are: Specify and estimate a VAR(p) model for Y t. Construct likelihood ratio tests for the rank of Π to determine the number of cointegrating vectors. If necessary, impose normalization and identifying restrictions on the cointegrating vectors. Given the normalized cointegrating vectors estimate the resulting cointegrated VECM by maximum likelihood. 11 / 64

12 LR Tests for the Number of Cointegrating Vectors The unrestricted cointegrated VECM is denoted H(r). The I (1) model H(r) can be formulated as the condition that the rank of Π is less than or equal to r. This creates a nested set of models H(0) H(r) H(n) H(0) = non-cointegrated VAR H(n) = stationary VAR(p) This nested formulation is convenient for developing a sequential procedure to test for the number r of cointegrating relationships. 12 / 64

13 Remarks: Since the rank of the long-run impact matrix Π gives the number of cointegrating relationships in Y t, Johansen formulates likelihood ratio (LR) statistics for the number of cointegrating relationships as LR statistics for determining the rank of Π. These LR tests are based on the estimated eigenvalues ˆλ 1 > ˆλ 2 > > ˆλ n of the matrix Π. These eigenvalues also happen to equal the squared canonical correlations between Y t and Y t 1 corrected for lagged Y t and D t and so lie between 0 and 1. Recall, the rank of Π is equal to the number of non-zero eigenvalues of Π. 13 / 64

14 Johansen s Trace Statistic Johansen s LR statistic tests the nested hypotheses H 0 (r) : r = r 0 vs. H 1 (r 0 ) : r > r 0 The LR statistic, called the trace statistic, is given by n LR trace (r 0 ) = T ln(1 ˆλ i ) i=r 0+1 If rank(π) = r 0 then ˆλ r0+1,..., ˆλ n should all be close to zero and LR trace (r 0 ) should be small since ln(1 ˆλ i ) 0 for i > r 0. In contrast, if rank(π) > r 0 then some of ˆλ r0+1,..., ˆλ n will be nonzero (but less than 1) and LR trace (r 0 ) should be large since ln(1 ˆλ i ) << 0 for some i > r / 64

15 The asymptotic null distribution of LR trace (r 0 ) is not chi-square but instead is a multivariate version of the Dickey-Fuller unit root distribution which depends on the dimension n r 0 and the specification of the deterministic terms. Critical values for this distribution are tabulated in Osterwald-Lenum (1992) for n r 0 = 1,..., / 64

16 Determining the Number of Cointegrating Vectors A sequential procedure is used to determine the number of cointegrating vectors: 1. First test H 0 (r 0 = 0) against H 1 (r 0 > 0). If this null is not rejected then it is concluded that there are no cointegrating vectors among the n variables in Y t. 2. If H 0 (r 0 = 0) is rejected then it is concluded that there is at least one cointegrating vector and proceed to test H 0 (r 0 = 1) against H 1 (r 0 > 1). If this null is not rejected then it is concluded that there is only one cointegrating vector. 3. If the H 0 (r 0 = 1) is rejected then it is concluded that there is at least two cointegrating vectors. 4. The sequential procedure is continued until the null is not rejected. 16 / 64

17 Johansen s Maximum Eigenvalue Statistic Johansen also derives a LR statistic for the hypotheses H 0 (r 0 ) : r = r 0 vs. H 1 (r 0 ) : r 0 = r The LR statistic, called the maximum eigenvalue statistic, is given by LR max (r 0 ) = T ln(1 ˆλ r0+1) As with the trace statistic, the asymptotic null distribution of LR max (r 0 ) is not chi-square but instead is a complicated function of Brownian motion, which depends on the dimension n r 0 and the specification of the deterministic terms. Critical values for this distribution are tabulated in Osterwald-Lenum (1992) for n r 0 = 1,..., / 64

18 Specification of Deterministic Terms Following Johansen (1995), the deterministic terms in are restricted to the form ΦD t = µ t = µ 0 + µ 1 t If the deterministic terms are unrestricted then the time series in Y t may exhibit quadratic trends and there may be a linear trend term in the cointegrating relationships. Restricted versions of the trend parameters µ 0 and µ 1 limit the trending nature of the series in Y t. The trend behavior of Y t can be classified into five cases: 18 / 64

19 1. Model H 2 (r): µ t = 0 (no constant): Y t = αβ Y t 1 + Γ 1 Y t Γ p 1 Y t p+1 + ε t and all the series in Y t are I (1) without drift and the cointegrating relations β Y t have mean zero. 2. Model H 1 (r): µ t = µ 0 = αρ 0 (restricted constant): Y t = α(β Y t 1 + ρ 0 ) + Γ 1 Y t Γ p 1 Y t p+1 + ε t the series in Y t are I (1) without drift and the cointegrating relations β Y t have non-zero means ρ Model H 1 (r): µ t = µ 0 (unrestricted constant): Y t = µ 0 + αβ Y t 1 + Γ 1 Y t Γ p 1 Y t p+1 + ε t the series in Y t are I (1) with drift vector µ 0 and the cointegrating relations β Y t may have a non-zero mean. 19 / 64

20 4. Model H (r): µ t = µ 0 + αρ 1 t (restricted trend). The restricted VECM is Y t = µ 0 + α(β Y t 1 + ρ 1 t) + Γ 1 Y t Γ p 1 Y t p+1 + ε t the series in Y t are I (1) with drift vector µ 0 and the cointegrating relations β Y t have a linear trend term ρ 1 t. 5. Model H(r): µ t = µ 0 + µ 1 t (unrestricted constant and trend). The unrestricted VECM is Y t = µ 0 + µ 1 t + αβ Y t 1 + Γ 1 Y t Γ p 1 Y t p+1 + ε t the series in Y t are I (1) with a linear trend (quadratic trend in levels) and the cointegrating relations β Y t have a linear trend. 20 / 64

21 Simulated Y t from Bivariate Cointegrated VECM Case 1: No constant Case 2: Restricted constant Time Case 3: Unrestricted constant Time Case 4: Restricted trend Time Time Case 5: Unrestricted trend Time 21 / 64

22 Simulated β Y t from Bivariate Cointegrated VECM Case 1: No constant Case 2: Restricted constant Time Case 3: Unrestricted constant Time Case 4: Restricted trend Time Time Case 5: Unrestricted trend Time 22 / 64

23 Johansen s Cointegration Tests in R The package urca includes functions for cointegration analysis based on Johansen s methods. > ca.jo function (x, type = c("eigen", "trace"), ecdet = c("none", "const", "trend"), K = 2, spec = c("longrun", "transitory"),...) Note: This function considers 3 specifications for the deterministic term: none corresponds to case 1, i.e. no constant const corresponds to case 2, i.e. restricted constant trend corresponds to case 4, i.e. restricted trend K is the number of lags for the VAR in levels, so K 1 is the number of lags in the VECM representation 23 / 64

24 Example 1: Testing for Cointegration Simulated data set 1: y 1t = y 2t + u t, u t = 0.75u t 1 + ε 1t y 2t = y 2t 1 + ε 2t ε it iid N(0, (0.5) 2 ) for i = 1, 2 > # Example (1) bivariate cointegrated system > nobs <- 250 > e <- rmvnorm(n=nobs,sigma=diag(c(.5,.5))) > u.ar1 <- arima.sim(model=list(ar=.75),nobs,innov=e[,1]) > y2 <- cumsum(e[,2]) > y1 <- y2 + u.ar1 > data1 <- cbind(y1,y2) 24 / 64

25 Example 1: Trace Statistic > # trace statistic > test1 <- ca.jo(data1,ecdet="const",type="trace",k=2,spec="transitory") > summary(test1) Test type: trace statistic, without linear trend and constant in cointeg Eigenvalues (lambda): [1] e e e-20 Values of teststatistic and critical values of test: test 10pct 5pct 1pct r <= r = Test finds 1 cointegrating vector 25 / 64

26 Example 1: Max Eigenvalue Statistic > # max eigenvalue statistic > test2 <- ca.jo(data1,ecdet="const",type="eigen",k=2,spec="transitory") > summary(test2) Test type: maximal eigenvalue statistic (lambda max), without linear tre Eigenvalues (lambda): [1] e e e-20 Values of teststatistic and critical values of test: test 10pct 5pct 1pct r <= r = Test finds 1 cointegrating vector 26 / 64

27 Example 2: Testing for Cointegration Simulated data set 2: y t1 = 0.5y 2t + 0.5y 3t + u t, u t = 0.75u t 1 + ε 1t y 2t = y 2t 1 + ε 2t y 3t = y 3t 1 + ε 3t ε it iid N(0, (0.5) 2 ) for i = 1, 2, 3 > # Example (2) trivariate cointegrated system (1 coint vector) > nobs <- 250 > e <- rmvnorm(n=nobs,sigma=diag(c(.5,.5,.5))) > u.ar1 <- arima.sim(model=list(ar=.75),nobs,innov=e[,1]) > y2 <- cumsum(e[,2]) > y3 <- cumsum(e[,3]) > y1 <-.5*y2 +.5*y3 + u.ar1 > data2 <- cbind(y1,y2,y3) 27 / 64

28 Example 2: Trace Statistic > # trace statistic > test1 <- ca.jo(data2,ecdet="const",type="trace",k=2,spec="transitory") > summary(test1) Test type: trace statistic, without linear trend and constant in cointeg Eigenvalues (lambda): [1] e e e e-18 Values of teststatistic and critical values of test: test 10pct 5pct 1pct r <= r <= r = Test finds 1 cointegrating vector 28 / 64

29 Example 3: Testing for Cointegration Simulated data set 3: y 1t = y 3t + u t, u t = 0.75u t 1 + ε 1t y 2t = y 3t + v t, v t = 0.75v t 1 + ε 2t y 3t = y 3t 1 + ε 3t ε it iid N(0, (0.5) 2 ) for i = 1, 2, 3 > # Example (3) trivariate cointegrated system (2 coint vectors) > nobs <- 250 > e <- rmvnorm(n=nobs,sigma=diag(c(.5,.5,.5))) > u.ar1 <- arima.sim(model=list(ar=.75),nobs,innov=e[,1]) > v.ar1 <- arima.sim(model=list(ar=.75),nobs,innov=e[,2]) > y3 <- cumsum(e[,3]) > y1 <- y3 + u.ar1 > y2 <- y3 + v.ar1 > data3 <- cbind(y1,y2,y3) 29 / 64

30 Example 3: Trace Statistic > # trace statistic > test1 <- ca.jo(data3,ecdet="const",type="trace",k=2,spec="transitory") > summary(test1) Test type: trace statistic, without linear trend and constant in cointeg Eigenvalues (lambda): [1] e e e e-18 Values of teststatistic and critical values of test: test 10pct 5pct 1pct r <= r <= r = Test finds 2 cointegrating vectors 30 / 64

31 Maximum Likelihood Estimation of the VECM If it is found that rank(π) = r, 0 < r < n, then the cointegrated VECM Y t = ΦD t + αβ Y t 1 + Γ 1 Y t Γ p 1 Y t p+1 + ε t becomes a reduced rank multivariate regression. Johansen derived the maximum likelihood estimation of the parametes under the reduced rank restriction rank(π) = r (see Hamilton for details). He shows that ˆβ mle = (ˆv 1,..., ˆv r ), where ˆv i are the eigenvectors associated with the eigenvalues ˆλ i The MLEs of the remaining parameters are obtained by least squares estimation of Y t = ΦD t + α ˆβ mley t 1 + Γ 1 Y t Γ p 1 Y t p+1 + ε t 31 / 64

32 Normalized Estimates of α and β The factorization ˆΠ mle = ˆα mle ˆβ mle is not unique. The columns of ˆβ mle may be interpreted as linear combinations of the underlying cointegrating relations. For interpretations, it is often convenient to normalize or identify the cointegrating vectors by choosing a specific coordinate system in which to express the variables. 32 / 64

33 Johansen s normalized MLE An arbitrary normalization, suggested by Johansen, is to solve for the triangular representation of the cointegrated system. The resulting normalized cointegrating vector is denoted ˆβ c,mle. The normalization of the MLE for β to ˆβ c,mle will affect the MLE of α but not the MLEs of the other parameters in the VECM. Let ˆβ c,mle denote the MLE of the normalized cointegrating matrix β c. Johansen (1995) showed that T (vec( ˆβ c,mle ) vec(β c )) is asymptotically (mixed) normally distributed. ˆβ c,mle is super consistent. 33 / 64

34 Testing Linear Restrictions on β The Johansen MLE procedure only produces an estimate of the basis for the space of cointegrating vectors. It is often of interest to test if some hypothesized cointegrating vector lies in the space spanned by the estimated basis: H 0 : β = (r n) ( β 0 φ β 0 = s n matrix of hypothesized cv s ) φ = (r s) n matrix of remaining unspecified cv s Result: Johansen (1995) showed that a likelihood ratio statistic can be computed, which is asymptotically distributed as a χ 2 with s(n r) degrees of freedom. 34 / 64

35 Example 1 continued... The test statistics found 1 cointegrating vector, i.e. r = 1. The function cajorls (urca package) estimates the restricted VECM. > # estimate restricted VECM > model1 <- cajorls(test1,r=1) > print(model1) $rlm $beta ect1 y1.l y2.l constant Note that the β coefficients closely match those in the DGP 35 / 64

36 Restricted VECM Response y1.d : Coefficients: Estimate Std. Error t value Pr(> t ) ect *** y1.dl y2.dl Residual standard error: on 245 degrees of freedom Multiple R-squared: , Adjusted R-squared: F-statistic: on 3 and 245 DF, p-value: Note that y 1t depends on the cointegrating vector 36 / 64

37 Restricted VECM Response y2.d : Coefficients: Estimate Std. Error t value Pr(> t ) ect y1.dl y2.dl Residual standard error: on 245 degrees of freedom Multiple R-squared: , Adjusted R-squared: F-statistic: 0.56 on 3 and 245 DF, p-value: Note that y 2t does not depend on the cointegrating vector 37 / 64

38 Testing Restrictions on β Hypotheses on the long-run coefficients β can be formulated in terms of linear restrictions on β: H 0 : R β = 0 or β = BΨ where Ψ are the unknown parameters in the cointegrating vectors β, B = R and R is the orthogonal complement of R such that R R = / 64

39 Testing Restrictions on β Suppose we want to test whether the cointegrating vector is β = (1, 1) without imposing any restrictions on the intercept. With the notation β = (β, β 3 ) = (β 1, β 2, β 3 ) the hypothesis becomes β 1 = β 2. Note that the vector β is typically normalized such that β 1 = 1. In terms of the parameterization β = BΨ, the matrix B is given by B = and Ψ = (ψ 1, ψ 2 ) such that BΨ = (ψ 1, ψ 1, ψ 2 ). In terms of the parameterization R β = 0, the matrix R is given by R = (1, 1, 0) which implies β 1 + β 2 = 0 and β 3 unrestricted. Note that R B = / 64

40 Testing Restrictions on β > B1 <- matrix(c(1,-1,0,0,0,1),nrow=3) > B1 [,1] [,2] [1,] 1 0 [2,] -1 0 [3,] 0 1 > R <- Null(B1) > R [,1] [1,] [2,] [3,] The test can be performed by passing the matrix B to the test function 40 / 64

41 Testing Restrictions on β > # test restrictions on beta > test3 <- blrtest(z=test1,h=b1,r=1) > summary(test3) Estimation and testing under linear restrictions on beta Eigenvalues of restricted VAR (lambda): [1] The value of the likelihood ratio test statistic: 0 distributed as chi square with 1 df. The p-value of the test statistic is: 0.96 The restrictions are not rejected 41 / 64

42 Estimate VECM with Restrictions on β > # estimate VECM with beta restrictions > model2 <- cajorls(test3,r=1) > print(model2) $rlm $beta ect1 y1.l y2.l constant The estimated cointegrating vector now satisfies the imposed restriction: β 1 + β 2 = 0 while the intercept remains unrestricted 42 / 64

43 Testing Restrictions on β Suppose we want to test whether the cointegrating vector is β = (1, 1) with no intercept, i.e we want R β = 0 such that β 1 + β 2 = 0 and β 3 = 0. > B2 <- matrix(c(1,-1,0),nrow=3) > B2 [,1] [1,] 1 [2,] -1 [3,] 0 > R <- Null(B2) > R [,1] [,2] [1,] [2,] [3,] / 64

44 Testing Restrictions on β > # test restrictions on beta > test4 <- blrtest(z=test1,h=b2,r=1) > summary(test4) Estimation and testing under linear restrictions on beta Eigenvalues of restricted VAR (lambda): [1] The value of the likelihood ratio test statistic: 0.2 distributed as chi square with 2 df. The p-value of the test statistic is: 0.91 The restrictions are not rejected 44 / 64

45 Estimate VECM with Restrictions on β > # estimate VECM with beta restrictions > model3 <- cajorls(test4,r=1) > print(model3) $rlm $beta ect1 y1.l1 1 y2.l1-1 constant 0 The estimated cointegrating vector now satisfies the imposed restrictions: β 1 + β 2 = 0 and β 3 = 0 45 / 64

46 Testing Restrictions on α Suppose we want to test whether α 2 = 0 without imposing any restrictions on α 1, i.e. we want to test for weak exogeneity of y 2t. This hypothesis can be represented as a linear restriction on α: H 0 : R α = 0 or α = AΨ The matrix R is given by R = (0, 1) which implies α 2 = 0 and α 1 unrestricted. > A1 <- matrix(c(1,0),nrow=2) > R <- Null(A1) > R [,1] [1,] 0 [2,] 1 The test can be performed by passing the matrix A to the test function 46 / 64

47 Testing Restrictions on α > # test weak exogeneity > test5 <- alrtest(z=test1,a=a1,r=1) > summary(test5) The value of the likelihood ratio test statistic: 0.17 distributed as chi square with 1 df. The p-value of the test statistic is: 0.68 Weights W of the restricted VAR: [,1] [1,] [2,] The restrictions are not rejected 47 / 64

48 Testing Restrictions on α and β We can test the restrictions on α and β jointly. > # test restriction on beta and weak exogeneity > test6 <- ablrtest(z=test1,a=a1,h=b2,r=1) > summary(test6) Estimation and testing under linear restrictions on alpha and beta Eigenvalues of restricted VAR (lambda): [1] The value of the likelihood ratio test statistic: 0.51 distributed as chi square with 2 df. The p-value of the test statistic is: 0.78 The restrictions are not rejected 48 / 64

49 Testing Restrictions on α and β Eigenvectors, normalised to first column of the restricted VAR: [,1] [1,] 1 [2,] -1 [3,] 0 Weights W of the restricted VAR: [,1] [1,] [2,] The estimated vectors now satisfy the imposed restrictions: β 1 + β 2 = 0, β 3 = 0, and α 2 = 0 49 / 64

50 Example 3 continued... The test statistics found 2 cointegrating vectors, i.e. r = 2. The function cajorls (urca package) estimates the restricted VECM. > # estimate restricted VECM > model1 <- cajorls(test1,r=2) > print(model1) $rlm $beta ect1 ect2 y1.l e y2.l e y3.l e constant e Note that the β coefficients closely match those in the DGP 50 / 64

51 Restricted VECM Response y1.d : Coefficients: Estimate Std. Error t value Pr(> t ) ect e-05 *** ect y1.dl y2.dl y3.dl Residual standard error: on 243 degrees of freedom Multiple R-squared: , Adjusted R-squared: F-statistic: on 5 and 243 DF, p-value: Note that y 1t only depends on the first cointegrating vector 51 / 64

52 Restricted VECM Response y2.d : Coefficients: Estimate Std. Error t value Pr(> t ) ect ect * y1.dl * y2.dl y3.dl Residual standard error: on 243 degrees of freedom Multiple R-squared: , Adjusted R-squared: F-statistic: on 5 and 243 DF, p-value: Note that y 2t only depends on the second cointegrating vector 52 / 64

53 Restricted VECM Response y3.d : Coefficients: Estimate Std. Error t value Pr(> t ) ect ect y1.dl y2.dl y3.dl Residual standard error: on 243 degrees of freedom Multiple R-squared: , Adjusted R-squared: F-statistic: on 5 and 243 DF, p-value: Note that y 3t does not depend on any of the cointegrating vectors 53 / 64

54 Testing Restrictions on β Suppose we want to test β 11 + β 13 = 0 and β 22 + β 23 = 0 without imposing any restrictions on the intercept. In terms of the parameterization R β = 0, the matrix R is given by R = (1, 1, 1, 0) which implies β 11 + β 13 = 0 and β 22 + β 23 = 0, with β 14 and β 24 unrestricted. Note that the matrix β is typically normalized such that β 11 = β 22 = 1 and β 12 = β 21 = 0. > B1 <- matrix(c(1,0,-1,0,0,1,-1,0,0,0,0,1),nrow=4) > R <- Null(B1) > R [,1] [1,] [2,] [3,] [4,] / 64

55 Testing Restrictions on β > # test restrictions on beta > test3 <- blrtest(z=test1,h=b1,r=2) > summary(test3) Estimation and testing under linear restrictions on beta Eigenvalues of restricted VAR (lambda): [1] The value of the likelihood ratio test statistic: 1.14 distributed as chi square with 2 df. The p-value of the test statistic is: 0.57 The restrictions are not rejected 55 / 64

56 Estimate VECM with Restrictions on β > # estimate VECM with beta restrictions > model2 <- cajorls(test3,r=2) > print(model2) $rlm $beta ect1 ect2 y1.l y2.l y3.l constant The estimated cointegrating vectors now satisfy the imposed restrictions: β 11 + β 13 = 0 and β 22 + β 23 = 0 while the intercepts remain unrestricted 56 / 64

57 Testing Restrictions on α > # test weak exogeneity of y3t > A1 <- matrix(c(1,0,0,0,1,0),nrow=3) > test4 <- alrtest(z=test1,a=a1,r=2) > summary(test4) The value of the likelihood ratio test statistic: 1.53 distributed as chi square with 2 df. The p-value of the test statistic is: 0.47 Weights W of the restricted VAR: [,1] [,2] [1,] [2,] [3,] The restrictions are not rejected 57 / 64

58 Testing Restrictions on α and β We can test the restrictions on α and β jointly. > # test restriction on beta and weak exogeneity > test5 <- ablrtest(z=test1,a=a1,h=b1,r=2) > summary(test5) Estimation and testing under linear restrictions on alpha and beta Eigenvalues of restricted VAR (lambda): [1] The value of the likelihood ratio test statistic: 1.96 distributed as chi square with 2 df. The p-value of the test statistic is: 0.38 The restrictions are not rejected 58 / 64

59 Zivot (2000) continued... > # trace test > data <- cbind(st,ft) > test1 <- ca.jo(data,ecdet="const",type="trace",k=2,spec="transitory") > summary(test1) Test type: trace statistic, without linear trend and constant in cointeg Eigenvalues (lambda): [1] e e e-18 Values of teststatistic and critical values of test: test 10pct 5pct 1pct r <= r = The test finds 1 cointegrating vector 59 / 64

60 Estimate Restricted VECM The test statistics found 1 cointegrating vectors, i.e. r = 1. The function cajorls (urca package) estimates the restricted VECM. > # estimate restricted model > model1 <- cajorls(test1,r=1) > print(model1) $rlm $beta ect1 st.l ft.l constant / 64

61 Restricted VECM Response st.d : Coefficients: Estimate Std. Error t value Pr(> t ) ect * st.dl ft.dl Residual standard error: on 240 degrees of freedom Multiple R-squared: , Adjusted R-squared: F-statistic: on 3 and 240 DF, p-value: Note that s t only depends on the cointegrating vector 61 / 64

62 Restricted VECM Response ft.d : Coefficients: Estimate Std. Error t value Pr(> t ) ect * st.dl ft.dl Residual standard error: on 240 degrees of freedom Multiple R-squared: , Adjusted R-squared: F-statistic: on 3 and 240 DF, p-value: Note that f t only depends on the cointegrating vector 62 / 64

63 Testing Restrictions on β We can test whether the cointegrating vector is β = (1, 1) with no intercept, i.e we want R β = 0 such that β 1 + β 2 = 0 and β 3 = 0. > # test restrictions on beta > B1 <- matrix(c(1,-1,0),nrow=3) > test3 <- blrtest(z=test1,h=b1,r=1) > summary(test3) Eigenvalues of restricted VAR (lambda): [1] The value of the likelihood ratio test statistic: 6.71 distributed as chi square with 2 df. The p-value of the test statistic is: 0.03 The restrictions are not rejected at 1% level 63 / 64

64 Estimate VECM with Restrictions on β > # estimate restricted model > model2 <- cajorls(test3,r=1) > print(model2) $rlm $beta ect1 st.l1 1 ft.l1-1 constant 0 The estimated cointegrating vector now satisfies the imposed restrictions: β 1 + β 2 = 0 and β 3 = 0 64 / 64