This is a follow-up to the "volatility is not risk" post from last week, found here: http://tfideas.blogspot.com/2010/10/volatility-is-not-risk.html
This viewpoint makes research harder because you can't really use fixed time-period returns data that well anymore - one year returns are sort of meaningless in a flexible timeframe environment, because I don't care what the one year returns are, I care what the annualized return between now and the time I choose to sell is. There's a continuous option to sell in the interim, so one year returns aren't that helpful.
It's "one year returns" (or some other fixed time horizon) that make volatility work, which is probably why people got sidetracked onto that path.
As an alternative, perhaps we can look at the price arc after purchase- something like "time to reach estimated intrinsic value" (or, more accurately, some discount to estimated intrinsic value, the discount being relative to uncertainty surrounding intrinsic value.). If you needed to model that, I suppose you could model with something stochastic, with a drift towards expected value whose magnitude is as a percent of the distance to expected value? Something like that? That lets one mathematically model what is more anecdotally obvious.
One can model "discount to estimated intrinsic value", as well - just as we care about having positive one year returns, we care about a narrowed discount to estimated intrinsic value - but that doesn't account for the amount of time it takes you to narrow that discount.
Just food for thought as a replacement for "one year return" data as a Y variable (where you'd probably want to transform "time to reach estimated intrinsic value" so that better investments are more positive - but there are a number of easy transforms for that).