3.2 Testing for cointegration The testing procedure is carried out in two phases. Using, e.g.

ADF, in the rst step we test whether both of the series are I (1). In the second step the cointegration of the series is tested.
Remark 3.2: There are several other tests for testing the unit root. E.g. Phillips-Perron test is one which is frequently used. Furthermore, modications for the testing procedure must be carried out if there is trend in the series.

10

Johansens test procedure: In the Johansen testing there are two test statistics; the trace statistics and the maximum eigenvalue statistic. The trace statistic tests the null hypothesis: there are at most r cointegrating relations against the alternative of m cointegrating relations (i.e., the series are stationary), r = 0, 1, . . . , m 1. The maximum eigenvalue statistic test the null hypohesis: there are r coinegrating relations against the alternative: there are r + 1 cointegrating relations.

Johansen,

Sren (1991). Estimating and testing cointegration vectors in Gaussian vector autoregressive models. Econometrica, 59, 15511580.
11

Example 3.3: Test the integration of the three stock indexes shares price series. (EViews output). ADF with four lags.
-------------------Stock ADF Test Exchange Statistic -------------------Finland -2.51 Norway -1.07 Sweden -1.35 -------------------1% 5% 10% Critical Value* Critical Value Critical Value -3.4418 -2.8658 -2.5691

*MacKinnon critical values for rejection of hypothesis of a unit root.

12

None of the test statistics are statistically signicant. Hence all (logarithmic) stock indexes are integrated, and a further check of dierences reveals that each index is integrated of order one.

If the series are integrated of the same order. Then the next step is to test whether the series are cointegrated.

13

Example 3.4: Test pair wise possible cointegration of the three stock index series
Included observations: 737 Test assumption: Linear deterministic trend in the data Lags interval: 1 to 4 ---------------------------------------------------------Likelihood Critical Values Hypothesized Eigenvalue Ratio 5 Percent 1 Percent No. of CE(s) ---------------------------------------------------------Series: LFIN LNOR 0.011019 8.818672 15.41 20.04 None 0.000885 0.652729 3.76 6.65 At most 1 Series: LFIN LSWE 0.009158 7.464918 15.41 20.04 None 0.000928 0.684529 3.76 6.65 At most 1 Series: LNOR LSWE 0.026201 20.76955 15.41 20.04 None ** 0.001629 1.201820 3.76 6.65 At most 1 ---------------------------------------------------------*(**) denotes rejection of the hypothesis at 5%(1%) significance level Normalized Cointegrating Coefficients: Norway and Sweden LNOR 1.000000 LSWE -0.938233 (0.04471) C -0.381887

The test results suggest that Sweden and Norway are cointegrated.

14

Example 3.5: Consider the case series in Example 3.2.

Unit root tests: (ADF with intecept) ============================== Series ADF p-val -----------------------------x -2.201151 0.2066 y -2.445337 0.1306 -----------------------------diff(x) -15.35371 0.0000 diff(y) -15.31091 0.0000 ==============================

ADF-tests indicate strong evidence that both series are I (1). However, the Johansen CI-tests below indicate that there is no sign of cointegration
Unrestricted Cointegration Rank Test (Trace) ============================================================== Hypothesized Trace 0.05 No. of CE(s) Eigenvalue Statistic Critical Value Prob.** -------------------------------------------------------------None 0.050303 11.54141 15.49471 0.1803 At most 1 0.003082 0.651297 3.841466 0.4196 ============================================================== Trace test indicates no cointegration at the 0.05 level * denotes rejection of the hypothesis at the 0.05 level **MacKinnon-Haug-Michelis (1999) p-values

15