Seems ouma performs poorly, though the sign of the effect seems correct. Starting conditions cause this.
Even starting conditions: improves the fit and flips the sign of the result.
More parameters: 
addendum
Longer runs using simulated annealing can still get stuck.
