Rebalancing is usually sold as a free lunch: trim the winner, top up the loser, and collect a “rebalancing bonus” without ever having to time anything. The formula most often cited for that bonus is Booth and Fama’s 1992 diversification return - roughly half the variance reduction you get from combining assets that don’t move in lockstep. It’s a real, well-established result. It also isn’t, on its own, an explanation for why rebalancing pays off. Later work correcting that attribution makes the sharper point: the bonus doesn’t come from variance reduction at all. It comes from mean reversion between the assets being rebalanced.
- Booth & Fama (1992): diversification return ≈ ½ × (weighted-average variance − portfolio variance) - commonly taught as rebalancing’s built-in mathematical edge.
- The correction: that arithmetic describes diversification, not rebalancing. Any actual rebalancing bonus is driven by the rebalanced assets reverting toward each other, not by the variance formula itself.
- Perold & Sharpe’s classic result: in a trending market - one direction, sustained - a fixed-mix rebalanced portfolio underperforms buy-and-hold, because rebalancing forces repeated trimming of whichever asset keeps winning.
- Vanguard, December 2024: a 200/175 basis-point threshold band beat monthly calendar rebalancing by roughly 15-22 bps a year in accumulation and 22-25 bps in decumulation, at about a quarter of the transaction costs.
Structurally, rebalancing is a bet on reversion whether anyone labels it that way or not: sell whichever asset has drifted furthest above target, on the expectation it gives some of that move back. Two assets that keep swapping the lead will hand a rebalanced portfolio exactly that pattern. Two assets where one simply keeps winning - not a rare regime; entire decades have looked like this - hand the same mechanical rule the opposite result, and the rule has no way to tell which regime it’s in before the fact.
What rebalancing does deliver, in either regime, is holding risk exposure where it was set. Left alone, a portfolio’s winners grow as a share of the total and its risk profile drifts with them - a 60/40 left untouched for a few strong equity years is no longer a 60/40, and is carrying more equity risk than anyone actually chose. Restoring that split is rebalancing’s one unconditional job. Whether doing so helps or costs returns over any given stretch depends entirely on which of the two regimes above the market happened to be in - and that part isn’t something the rebalancing rule itself can answer.
None of this argues against rebalancing. It argues for being precise about which case is being made. The risk-control case is unconditional and provable. The return case sits on top of it, holds only in specific market regimes, and can’t be verified in advance. Threshold-based rebalancing - acting once drift crosses a band, rather than on a fixed calendar - captures most of the risk-control benefit at a fraction of the trading cost of doing it on a schedule; the Vanguard figures above measure the size of that difference, not a verdict on the return question either way.
Back to Writing