http://www.nature.com/nsu/030106/030106-9.html Someone might have posted this but here it is... It deals with Supply and Demand...
One of the authors is Doyne Farmer. He is a physicist who specialized in complex systems. He started a business with Norman Packard to create a large-scale automated trading system. There is a book by Thomas Bass which tells the story, The Predictors. It's one of my all-time favorite trading books. (for entertainment puposes, though)
Yes prices move faster on small company issues than big ones, but you can't put as much position size on the smaller capitalized issues. If you have a small fund, go with the smaller capitalized issues. Some of those are very thinly traded and you have to discover the lower limit on liquidity however, it's just going to be a shares per bar figure, depending on what bar interval you watch. These researchers always like to go with millions of data points, makes their work look irrefutable I guess. Statistically, you can get results good enough for trading with a really small number of samples as far as I can tell. Speed of movement of issues has been a favorite research subject of mine at times. I downloaded history on 1300 or so issues at one point and then did some number crunching and I got a bunch of spectrums of speed of movement versus all sorts of other parameters, very interesting work and I recommend it. From that you can develope a spectrum of your universe of issues and can do a backtest on just a few issues at different places in the spectrum and apply the results to the whole universe. Not too much need for a backtester that does a large folio at that point, just backtest on a few issues and you pretty much have your result for your whole basket of issues. People are going with backtesters that do a whole large folio of issues, I've always wondered one thing; what the heck do they do with all the data they collect? The issues have to have a common thread running through all of them for the results to be useful and if they do have that thread then there is not much reason to test on all of them. Maybe it's a case of thinking the computer will substitute for reasoning. Then again maybe it's just that I have gone to a lot of trouble to generate a work-around for my backtest-on-a-single-issue-at-a-time software. Just my two and a half cents worth, remember, everybody's opinion is worth about what you pay for it, true in this case as well. Max