As JJ test is a parametric test and ARDL is non-parametric then each one of the two tests is based on different assumptions, but despite that which result is the true answer.
DATA SERIES CHECKING: Check you data series to see whether you have I(0) and I(1). The two tests: Johansen and ARDL cointegration tests are not used intergeably.
JOHANSEN TEST: This is used for testing cointegration of several I(1) time series. If you have several time series that are integrated as I(1), then using the Johansen test is appropriate. This test is generally used in vector auregressive model VAR(p) where VAR(p) is given by:
Yt = C + A1Yt-1 + A2Yt-2 + ... + ApYt-p + et
... C = k X 1 vector of constants (intercepts); Ai = time invariant k X k matrix and et = k X 1 vector error.
ARDL COINTEGRATION TEST: Under this approach, the test incorporate I(0) and I(1) in the same estimate. If the variables are I(0) only then use OLS. it is easier and simpler. If the variables are I(1) only then do VECM (Johansen test). If the variables are a mixed of I(0) and I(1) then use ARDL test.
Check your time series to see whether you have a mixed of I(0) and I(1) or either one without the other---if so, follow above suggestions.
REFERENCES:
(1) Johansen, Søren (1991). "Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian Vector Autoregressive Models". Econometrica 59 (6): 1551–1580. JSTOR 2938278.
(2) Hatemi-J, A. (2004). "Multivariate tests for autocorrelation in the stable and unstable VAR models". Economic Modelling 21 (4): 661–683. doi:10.1016/j.econmod.2003.09.005.
i run both cointegration test for the same data in which all the variable are integrated of order. Johanson results shows that there is no cointegration but ARDL shows there is.