Believe it or not, my position sizing is actually fairly complex and I don't have a short answer. If you don't mind a waffling (and lengthy) response then here it is:
My equity position sizing is largely due to how my system development process works. Let's assume I've managed to develop 4 equities models. Two models that rely on a mean reverting process (1 long and 1 short) and two models that rely on some sort of trend based element (1 long and 1 short). This is in an effort to have complimentary models that display low correlation to each other over some period of time. Ideally, these systems will act in unison, when the TF isn't working, the MR will be... in an ideal world of course. Nothing is ever ideal, but you get the point.
So now we have 4 models with some decent stats over the last 20 years or so. Each model will have days where it just falls apart, if one was trading an MR long model yesterday for example, it would have lost quite a bit until maybe the last hour or so, and potentially much more if exposure wasn't controlled. Each model will also have days where one position just kills it (I don't use stops). Say you're long and the co. gets sued; I've had positions move 30% against me in a matter of minutes.
So those are the two problem scenarios, single product risk, and model risk. So what I'll do is assign a max position size and max model exposure limit for each model based on historical testing and a bit of common sense.
So back to the 4 models, say they're allocated equally to start, at 25% each, 3% max position size for any one stock in any one model; provided volume, price and volatility will allow that. What does the combined return look like for the entire basket of models working together? Which one (or set of models) account for the worst case day/trade?
What if I assign MR long 40% and MR short 10%, and make the long side max. position size 2x the short position size? What does the return dist. look like here? Did my results improve? Rinse and repeat until an acceptable risk/return ratio is found across the entire data set.
So those are the initial steps in the process and they are somewhat straight forward. Next comes the interesting part, building hedging systems. Systems that one wouldn't trade on their own, but, when added to the basket of models, will allow one to increase position size and increase leverage. This has a lot to do with portfolio selection as well as common sense. Does shorting 1 ES contract for every 50K in long equity exposure make sense over the testing period? Maybe.
So the end result from this round about process is that having lots of small positions is better. Spreading your equity across many products will allow one to use easy hedges (like ES for a basket of equities), and, will to some degree allow one to absorb some of that eventual single product crap as a 30% adverse move on a 2% position is not too painful, and, might not matter if other positions are working well that day.
In terms of max. size of any individual position, it depends on one's portfolio and the type of order required to execute. Like you mention, ADV is a good way to go. Using a combination of ECN's, staggering orders over a time period, there are all sorts if methods one can use. I usually wait for a narrower spread and decent quotation size when an entry is hit. This does occasionally cost me, but, it really depends on what type of orders one is using.
.