I recently completed an experiment where I divided the market into eight asset classes including international and REITs. I used a database from years 1989 through 2006. In the first part of the experiment, equal percentages were invested in each asset class. The asset classes were permitted to "run" without rebalancing.
In part 2 of the experiment, the different asset classes were brought back into balance (12.5% of total portfolio for each asset class) at the end of each year. Rebalancing did enhance portfolio performance.
Part 3 of the experiment was designed to see if mean variance optimization (MVO) analysis would increase or decrease portfolio performance. Thirteen five-year rolling time periods were selected to determine what asset allocation would be used for the following year. For example, data from 1989 through 1993 was fed into the MVO software (I used MVOPlus from Efficient Solutions) in order to set the future asset allocation percentages for the 1994 portfolio. The performance of the "standard portfolio" was compared with the MVO Portfolio. Then the next five-year data set, 1990 through 1994, was used to determine the asset allocations for 1995. Performance results for 1995 were then compared. This process was carried out for thirteen different periods.
Constraints were placed on the different asset classes so the MVO software did not generate portfolios that defied common sense. If anyone wishes further information, just ask.
Ten thousand dollars invested in each asset class ($80,000) in 1989 grew to over $585,000 by the end of 2006, if the asset classes were rebalanced. The "MVO Portfolio" grew to move than $950,000 through
2006.Physlab