The authors find that better results are obtained by delaying rebalancing as long as four years. In the case of threshold rebalancing, it is better to wait until relative weights become 5% or more out of alignment.
When transaction costs and taxes are considered, the underperformance of monthly, quarterly, and annual (or low threshold) rebalancing is even more noticeable. Smith and Desormeau also find that results improve when rebalancing is tied to the Federal Reserve’s monetary cycle.
The explanation Smith and Desormeau offer in their paper is that markets seem to have a tendency to trend in the short run before mean reversion kicks in. Sitting tight for up to three or four years lets a portfolio ride the wave longer and capture more of the upside (but involves letting risk rise).
This might work in most environments, but I’m not sure if it (or other forms of mechanical rebalancing) will deliver good results in all environments. For example, a portfolio manager rebalancing throughout the 12-year slide in the Nikkei in the 1980s and 1990s may not have kept his or her job.
Selling an asset just because its price went up and buying another just because its price went down seems suboptimal to me. Maybe the decision to rebalance should also be guided by value concepts: sell the rising price asset if the value ratios indicate overvaluation and buy the falling price asset if the ratios signal undervaluation. But then this begins to look like value investing, not rebalancing?