tstats and pvalues using fmincon

I used fmincon to solve a MLE problem. I thus found the estimated coefficients, but I'd like to also have a tstat and p-value of each estimated parameter.

So, I used

sqrt(diag(inv(hessian)))

(where "hessian" is an output of fmincon) to get the standard errors of each estimated parameter.

Now, to get the t-stats, I simply divided my estimated coefficients by their own SE. However, sometimes the sign of the estimated parameter is negative, but the SE is positive, so the tstat turns negative and the p-value is consequently very high. Should I take the absolute value of the tstat?

Am I doing something wrong?