In this test, when we refuse the null hypotesis for the serial auotocorrelation of the error, such as:

$e_t = \varepsilon_t + \beta_1 \varepsilon_{t-1} + \dots+ \beta_p \varepsilon_{t-p} $;

it means that the residuals follow an auto-regressive model of order (p).

If i want to avoid this problem, i must add a certain number of lags of the response variable $y$, but in some case this method is not useful. Is there other options to consider?