I can't find the source for this anywhere online. My experience with over 100 hours of practice in front of a mirror (don't ask) and the research I've read suggests otherwise. And even if you can voluntarily contract your orbicularis oculi, you'll need to have the proper onset/offset times and constant intensities for muscle contraction, which are just as hard to fake and easy for a trained observer to spot.
I suggest the "crow's feet" from a Duchenne Smile is still a reliable indicator of feeling happy for all those reasons, if observed appropriately. It was far easier to fake a real smile by thinking of something happy, which is why facial recognition is still much better for emotion detection than lie detection.