Start with the number that should anchor everything else. Vanguard ran a 60/40 stock and bond portfolio through 92 years of data, January 1926 to December 2018. Rebalanced once a year with a 5% band, it returned 8.19% annually after tax. Never rebalanced at all, it returned 8.74%.
So rebalancing didn't add return over that stretch. It cost roughly half a percentage point a year. What it bought was risk control. Annualised volatility fell from 14.0% to 11.4%, and the Sharpe ratio rose from 0.46 to 0.51.
The never-rebalanced portfolio also finished with an average equity weight of 85%, not 60%. It wasn't a better 60/40. It was a different, riskier portfolio still wearing the 60/40 label. That distinction is where most of the confusion about the rebalancing bonus lives, and it's worth separating the two effects properly.
What the bonus is supposed to be
The term "diversification return" comes from Booth and Fama in 1992. The idea is simple enough. A portfolio held at constant weights compounds faster than the weighted average of its assets' own compound returns. That gap is the diversification return.
Scott Willenbrock, writing in the Financial Analysts Journal in 2011, gives a clean worked example. Take 50% S&P 500 and 50% Barclays US Long Treasury from 1 January 2000, rebalanced each year end. Over the decade to 2009 the equity leg compounded at minus 0.95% and the Treasury leg at 7.59%. The blended portfolio compounded at 4.44%, against a weighted average of the two legs of 3.32%. The 1.12% difference is the diversification return.
That's a real gap, and it isn't a rounding error. But look hard at what it's measured against. The benchmark isn't the same portfolio left alone. It's a synthetic blend of compound returns that nobody actually holds.
The original bonus was 0.49%, and its author retracted it
William Bernstein coined the phrase "rebalancing bonus" in a 1996 piece on Efficient Frontier. He used US common stock and long-term corporate bonds over 1926 to 1994, which returned 10.19% and 5.51% a year. The arithmetic blend of the two, which he calls the Markowitz return, is 7.85%. A 50/50 portfolio rebalanced annually returned 8.34%. The bonus: 0.49%.
Half a percentage point. That's the headline figure from the paper that named the effect, and it's already modest. The follow-up is the part almost nobody quotes.
In the next issue, under the title "When Doesn't It Pay to Rebalance?", Bernstein was blunt about his own number. He wrote that the excess return was illusory. Equal amounts in stocks and bonds on 1 January 1926, left alone and untaxed, returned 9.17% a year. That beats the 8.34% from rebalancing. The unrebalanced portfolio simply drifted into being mostly equities, and for the last 40 years of the period it was more than 90% stock.
So the bonus was measured against a benchmark that flattered it, and it vanished against the benchmark that mattered. The same drift mechanism runs on much shorter timescales too, as our look at how far a 60/40 portfolio wanders in a single year shows.
Where the extra compound return actually comes from
Willenbrock's main contribution is to kill a popular explanation. The diversification return is usually attributed to variance reduction, because the approximation g ≈ r − σ²/2 makes lower variance look like higher compound growth. He argues the mathematics is being read backwards.
The real source is the trading. Rebalancing forces you to sell whatever has risen in relative weight and buy whatever has fallen. That's a contrarian rule, and the rule is what produces the return. Lower variance is necessary but not sufficient.
His test case makes the point sharply. A buy-and-hold portfolio also has lower variance than the weighted average of its holdings. It earns no diversification return at all. Only a rebalanced portfolio does, because only a rebalanced portfolio trades. Willenbrock calls diversification the free lunch of finance and the diversification return the "free dessert", earned while holding the risk profile constant.
Whether the dessert is worth much depends entirely on what's being rebalanced. In an equally weighted commodity futures index, where the average constituent has a standard deviation near 30%, Willenbrock argues that most of the 4.52% excess return Gorton and Rouwenhorst measured is diversification return. Volatile assets, similar expected returns, low correlation: that's the recipe. A 60/40 of global equities and hedged bonds is nothing like it.
The case against calling it a bonus at all
The sharpest critique comes from Donald Chambers and John Zdanowicz in 2014. They argue that the whole concept rests on a badly chosen benchmark and a confusion between rates and money.
Their reasoning runs like this. Expected portfolio value is governed by arithmetic mean returns, not geometric ones. Volatility drags down the compound rate without reducing expected wealth. A lower-volatility portfolio will therefore show a higher geometric mean than a higher-volatility one with the same arithmetic mean. That looks like added return when nothing has been added.
They state three conclusions. Rebalancing does tend to raise geometric mean returns even when returns are serially uncorrelated, but those higher geometric means don't cause higher expected portfolio values. Rebalancing raises expected value when asset prices are mean-reverting, and the gain comes from the mean reversion rather than from diversification or variance reduction. And the higher geometric mean of a low-volatility portfolio can't be arbitraged against a high-volatility one when both share the same arithmetic mean.
Their reframing is useful even if the full argument doesn't convince you. Call it rebalancing return rather than diversification return, they suggest, and note that it should be positive when prices mean-revert and negative when they trend. That makes it a bet on a return pattern, not a free dessert. It's the same reason volatility and drawdown answer different questions about risk: the measure you pick decides what looks like a gain.
When the bet loses
Bernstein's own formula makes the conditions explicit. For two assets the bonus equals X₁X₂(Var₁/2 + Var₂/2 − Cov₁,₂), which is the difference between the pair's mean variance and their covariance. Higher volatility raises it. Lower correlation raises it. A large gap in long-run returns destroys it, and destroys it further the longer the gap persists.
He put a rough threshold on that last condition. Over 1970 to 1994, rebalancing pairs of national equity markets almost always beat leaving them alone. Only when long-run return differences exceeded about 5 percentage points did the unrebalanced pair win, and then only by carrying more risk. Rebalancing between industry groups has a much worse record, he noted, because whole industries shrink permanently while others grow.
A German study puts numbers on the losing case. Frieder Meyer-Bullerdiek tested an equally weighted portfolio of 15 stocks that had been in the DAX since 1988, using weekly data from January 2006 to December 2015. Over the full ten years, rebalancing produced negative rebalancing returns at every frequency tested. Weekly rebalancing compounded at 0.1167% a week against 0.1485% for buy-and-hold, a shortfall of about 0.032% a week or roughly 1.8 percentage points a year.
The decomposition is the instructive part. The volatility return was positive, at 0.0546% a week. It was swamped by a dispersion discount of 0.0864%, because a few stocks trended hard and rebalancing kept selling them. Strip out the five stocks with the most extreme end weights and the rebalancing returns turn positive. That isn't a strategy. That's hindsight.
What happens when you test it properly
Hubert Dichtl, Wolfgang Drobetz and Martin Wambach ran the most careful statistical test in this literature. They applied a double block bootstrap to monthly stock and government bond data for the US, the UK and Germany, from January 1982 to December 2011. Three families of rebalancing were tested: periodic, threshold and range.
Their finding on returns is the one that matters here. The simulations gave only weak evidence that buy-and-hold returned more than rebalancing, and no significant economic difference in net asset values either. Neither mean reversion nor momentum in the data was strong enough to produce superior returns for either strategy. In the US, the confidence interval around the return difference included zero.
On risk the result is emphatic. Rebalancing at every frequency showed significantly lower volatility than buy-and-hold, and significantly higher Sharpe ratios. Monthly rebalancing produced significantly lower Sharpe ratios than quarterly, in all three countries and at every horizon. That's evidence that trading too often costs something even before commissions and tax. Their conclusion is that risk reduction explains why rebalancing is so widely used.
Vanguard's table says the same thing from a different angle. Thirteen different rebalancing rules produced annualised returns between 8.19% and 8.39%, with volatility between 11.4% and 11.8%. The choice of rule barely mattered. The choice to have one did. Our comparison of annual rebalancing against 5% threshold bands works through those rules in detail.
Bernstein reached the same place in 2000 with a different dataset. Across 24 overlapping 28-year periods starting in 1969, a 40/15/15/30 portfolio returned 12.030% a year rebalanced monthly and 12.267% rebalanced every four years. The average difference between quarterly and four-yearly rebalancing was 18 basis points.
So how big is the bonus, really?
Pull the published figures together and a range emerges. Bernstein's 1926 to 1994 stock and bond pair: plus 0.49% against the arithmetic blend, negative against buy-and-hold. Willenbrock's 2000 to 2009 stock and Treasury pair: plus 1.12% against the strategic return, in a decade when equities went nowhere. Vanguard's 92 years: minus 0.55% against buy-and-hold. Meyer-Bullerdiek's German equities: minus 1.8% a year.
Every positive figure there is measured against a synthetic benchmark. Every figure measured against the portfolio you'd otherwise be holding is small or negative. The honest summary is that the return effect is close to a coin flip. Its sign depends on whether the assets mean-revert over your holding period, and its size is usually well under a percentage point either way. Costs and tax then eat into whatever is left.
None of that touches the risk case, which is much stronger and much less disputed. Rebalanced portfolios held their target risk; unrebalanced ones drifted upward in equity weight and got hit harder in drawdowns. That effect shows up clearly in what rebalancing did through the 2008 and 2020 crashes, and it's the reason the practice survives despite a return story this thin.
What would change the conclusion
Three findings would move this materially.
First, evidence of reliable mean reversion at the asset-class level. Both Chambers and Zdanowicz and Meyer-Bullerdiek make the bonus conditional on it. Dichtl and co-authors found it too weak to matter across 30 years and three countries. A dataset showing persistent negative autocorrelation at typical rebalancing horizons would turn the bet into an edge.
Second, a different asset mix. The stock-bond evidence is dominated by a 90-year equity risk premium, which is exactly the return gap that kills the bonus. Rebalancing between assets with genuinely similar long-run returns and 30% volatility is where the arithmetic favours it, and where Willenbrock's estimate runs to several percentage points.
Third, a proper cost accounting. Vanguard's figures are tax-adjusted, but the German study ignores transaction costs entirely and says so. Realistic costs would push the return effect further negative while leaving the risk benefit untouched.
The one thing that wouldn't change the conclusion is another backtest showing a rebalanced portfolio with a higher Sharpe ratio. Critics and advocates already agree on that. It's a statement about risk, and it says nothing about a bonus.