Rob Arnott, Chairman of Research Affiliates LLC, is causing shock-waves in the index fund and ETF world. He published a paper in 2004 arguing that market-cap weighted indexes are intrinsically inefficient and lead to underperformance, and should be replaced by indexes weighted by companies' fundamentals, such as sales, revenues, book value, earnings, or number of employees. Here's his argument in his own words, and two links that allow you to read more on this:
If we could see the future, we would know the true fair value of every company and we may see that share prices are far removed from true fair value. Some companies are going to be way above true fair value, and other companies, where the future is brighter than people think, will be way below true fair value.
One thing we know about capitalization weighting is that a capitalization-weighted index overweights every single stock that is trading above fair value and underweights every single stock that is trading below true fair value. These errors create a drag on the return of capitalization-weighted indexes. However, since we can’t know what the true fair value is for any particular stock, we need to construct an index that randomizes these weighting errors.
Equal-weighting a portfolio is one good way to randomize the errors. Half of the stocks will be overweight and half will be underweight. Some that are above true fair value will be underweight; some that are below true fair value will be overweight. So equal-weighting fixes the problem, but it introduces its own, very serious problems. Equal-weighted indexes are capacity constrained, high-volatility, and result in high turnover in some of the least liquid companies in the market. You just can’t use equal weighting on an institutional scale.
In our research we found that if you weight a portfolio based on objective measures of company size, the errors will be randomized. You will not structurally overweight all of the overvalued stocks and underweight all of the undervalued stocks. It’s a lot better to be wrong 50% of the time than 100% of the time.