The cointegration analysis in hypothesis of heteroschedasticity: the wild bootstrap test