Wired, Jan. 2011 at 92, describes a stock trader who identified seven key factors, such as revenue and earnings growth, most predictive of a stock’s performance. He then used a publicly available program from UC Berkeley called the differential evolution optimizer to wade through huge amounts of data on these factors about the best performing stocks at a given time. The program looked at historical data to see how well the weights it was testing predicted the stock’s actual performance. The program kept picking stocks, checking weightings, and testing predictions until it had generated thousands of such weightings. From that set the analyst selected the 10 best-performing weightings.
“The optimizer then mated those weightings – combining them to create 100 or so offspring weightings.” It kept at that process of mating and testing, mating and testing for dozens of iterations until the analyst had a set of weightings to screen stocks.
This approach and software might eventually sort out which law department benchmarks tell us the most about good performance. At some point, if we take lower-than-normal total legal spending as a share of revenue as the measure of good performance (analogous to share price appreciation), with data over a period of years and within an industry or even narrower sector, it sounds like this software could churn through and weight the various benchmark metrics in regard to how well they predict total legal spend.