I had always read that the smile never had a presence in markets until the crash of 87'. After that, people didn't want to give free-rides, particularly on the down side as things crash much more rapidly down than up. I read something in 'The Volatility Smile' that I haven't read elsewhere, and thought it interesting: I suppose it is legit, but after going through a hundred books or so, this is the first time I've seen it explained that way. Granted, I'm far from an expert (yet), and he did say it is 'One' of the reasons... I also didn't realize there were so many collars, but if a large amount of fund managers are doing them to lock in profits for the year (so the book mentions), I guess it makes some sense.
Somehow those kinds of arguments don't satisfy me. Basically it is saying (if the effect is big enough) there is your free lunch. Same argument as to why implied volatility ought to be higher than historical: "more demand than supply for long options". If that were true, I'd expect the effect to be rather small, as otherwise it doesn't cost much to increase supply for the players who are already set up to sell volatility.