Typically, predictive models are applied across groups, or baskets, of securities that share the characteristics found to have the predictive ability. These patterns of information might be as simple as adding "top company in industry" and "154 representative industry groups," limiting the basket to 500 names, and predicting that over long periods of time, the basket has a good shot at going up.
(You could even put a label on it, like "
," and market it as an index. Then you'd predict only that that index will more or less do the same or a little better than the rest of the market in the hopes of making it the standard by which all others are judged -- you get the picture.)
To make any money by actually investing in the basket rather than selling access to the names in it, it is not enough to simply predict its movements, which may be accomplished by analyzing large and constantly improving databases, you must also profit when your predictions turn out to be accurate. This means that you have to have capital committed, long or short, to a very accurate representation of your basket. This requires efficient trade-execution systems so you can get all of your capital committed, or released, from the basket, within a short enough period of time so that market movement "noise" doesn't wreak havoc with the price assumptions built into your predictive model.
Moving money in and out of baskets, of course, was a logistical pain in the neck until the advent of decent basket-trading software. This has been has been available for less than a generation, and for most of that time its main utility has been the ability to get a whole bunch of trades done simultaneously, if not well. The issue of best execution on each individual security was, and is, more complex.