Key takeaways
- From 1972 to 2014 an annual calendar rebalance earned a 0.43 net Sharpe ratio against 0.40 for a 20% tolerance band, at identical five-basis-point trading costs.
- Averaged across five asset classes, autocorrelation was positive at 0.15 for one-month returns and minus 0.33 at three years, so the horizon decides which rule wins.
- A monthly rebalanced 60/40 portfolio hit a maximum drawdown 1.2 times worse than buy-and-hold in the 2007-2009 crisis, a gap of 5 percentage points.
- The same 5% threshold fired 58 times under monthly monitoring but only 34 times under annual monitoring across Vanguard's 92-year sample.
- Widening a no-trade band from 1 to 6 percentage points raised the average equity share from 70% to 71.2% and tracking error from 18 to 49 basis points.
Calendar vs band rebalancing: two rules, one portfolio, one yearly decision
You and a friend hold the same 60/40 portfolio and follow different rules. You trade on a fixed date — every December, say. Your friend leaves the portfolio alone until a holding drifts far enough from its target to trigger a trade. Whose rule is better?
Your friend's sounds obviously better. It reacts to the size of the move rather than to an arbitrary date in your diary.
Over 42 years, it lost. Between 1972 and 2014, a five-asset portfolio rebalanced every December earned a net Sharpe ratio of 0.43. The same portfolio rebalanced whenever a holding drifted 20% away from its target earned 0.40.
Both rules traded roughly once a year. Both cost about five basis points a year in trading. One of them watched the market. The other only watched the calendar. The calendar won.
Those figures come from a 2015 AQR study by Antti Ilmanen and Thomas Maloney, run on a portfolio of US and non-US equities, US and non-US government bonds and commodities. They invert the usual case for tolerance bands. Over those 42 years the smarter-sounding rule didn't work out, and the reason is regime.
The calendar's blind spot is real
Give your friend's band its strongest ground first. A calendar rule can sit through an enormous move and do nothing at all.
Sandy Rattray, Nick Granger, Campbell Harvey and Otto van Hemert show the extreme version. A US stock-bond portfolio that started at 60% equities in 1927 had drifted to 76% by 1929, then down to 32% by 1932. A date in a diary has no view on any of that.
Vanguard's 92-year grid puts numbers on the waiting. Suppose you monitored a 60/40 portfolio once a year against a 5% threshold: you'd have traded just 34 times between 1926 and 2018, roughly once every 2.7 years. Widen the threshold to 10% and it falls to 14 events in 92 years. We've already run the head-to-head between annual rebalancing and 5% threshold bands; this piece is about when each rule breaks rather than which is tidier.
Even large institutions that use bands still check them on a schedule. Norges Bank Investment Management, which runs Norway's sovereign wealth fund, rebalanced when the equity share moved more than 4 percentage points from its strategic target. In August 2018 the Bank advised the Ministry of Finance to narrow that to plus or minus 2 percentage points around a 70% equity share, assessed at each month-end. The band responds to the move. The looking is still discrete.
In a trend, reacting to the move is the problem
Here's the part the tolerance-band story leaves out. Rebalancing isn't a neutral act. It sells whatever went up and buys whatever went down.
Rattray and co-authors formalise it. Relative to buy-and-hold, a rebalanced portfolio behaves like buy-and-hold plus a short straddle on the relative value of the two assets. That's negative convexity. When two assets diverge sharply and keep diverging, you keep feeding the loser.
Their global financial crisis example is blunt. A monthly rebalanced 60/40 portfolio reached a maximum drawdown 1.2 times worse than the buy-and-hold version, a gap of 5 percentage points, at exactly the moment markets were worst. By February 2009 the cumulative return difference between the two had reached 5.3 percentage points. Their data run monthly from 1960 to 2017, using US equity returns from Kenneth French's data library and Treasury data from the Federal Reserve.
A band makes that worse, not better, because it fires precisely when divergence is large. A calendar rule that happens to be looking away trades later, smaller, or not at all. Our piece on what rebalancing did in the 2008 and 2020 crashes covers the drawdown side in more detail.
Forty-two years of calendar vs band rebalancing, and the lazier rule won
The AQR test is the cleanest calendar-versus-band comparison in the public literature, because it holds turnover roughly constant. Its bands are relative: a 20% band means a holding has moved 20% away from its own target weight, not 20 percentage points.
Over 1972 to 2014, net of estimated trading costs, the annual calendar rule returned 9.2% a year with 8.0% volatility, a 0.43 Sharpe ratio and a maximum drawdown of 29.7%. The 20% band returned 9.1% with 8.2% volatility, a 0.40 Sharpe ratio and a 32.8% drawdown. Annual turnover was 9.2% for the calendar and 9.4% for the band. Trading cost was five basis points for both. Discretionary trading is where the bill gets large, and overconfidence in investing is the mechanism usually blamed for it.
Rebalancing monthly was worse again: 9.0% return, 0.39 Sharpe, 32.8% drawdown and 26.0% annual turnover. Buy-and-hold managed 8.7% with a 0.32 Sharpe and a 41.2% drawdown, so doing nothing at all was clearly the worst option on risk. Biennial rebalancing scored best of the lot at 0.44.
Was that one lucky month? No. Ilmanen and Maloney report annual schedules anchored to March, June, September and December separately. All four beat monthly rebalancing over the full sample and in both halves of it. From 1993 to 2014 the March schedule scored 0.50 against 0.43 for monthly.
Rattray and co-authors reach a compatible conclusion from different data. Testing 60% plus or minus 2 and plus or minus 4 percentage point thresholds against monthly, quarterly and annual calendars over 1960 to 2017, they found the threshold rules slightly less potent at reducing drawdowns. They also noted that other band widths didn't materially improve matters. Quarterly and annual calendars improved drawdowns relative to monthly, with one exception: Black Monday in 1987, a drawdown that reversed quickly.
Momentum runs about a year, reversal takes three
So why would waiting help you? Because prices in these samples trend over months and revert over years.
Ilmanen and Maloney measured it directly. Averaged across their five asset classes over 1972 to 2014, autocorrelation was positive at 0.15 for one-month returns, 0.08 at three months and 0.09 at twelve months. At three years it was minus 0.33, and at five years minus 0.31. For non-US government bonds the three-year figure was minus 0.51.
That pattern rewards a specific behaviour. Rebalance often and you fight short-term momentum; rebalance rarely and you harvest long-term reversal. An annual or biennial schedule behaves like buy-and-hold at short horizons and like a rebalancer at long ones. The authors describe it as getting the best of both. They add that three- or five-year rebalancing keeps the advantage, while at ten years the benefit fades because full reversal cycles complete in between.
Both halves of that pattern have independent support, and both come with problems. Tobias Moskowitz, Yao Hua Ooi and Lasse Pedersen documented return continuation over one to twelve months across 58 liquid futures contracts, with reversal beyond. All 58 produced positive time-series momentum returns from 1985 to 2009, and 52 were significant at the 5% level. Run that same time series momentum factor through the six developed-equity crashes since 1990 and it gained in every one.
That's the evidence your patient calendar rule is quietly leaning on, and it hasn't gone unchallenged. Dashan Huang, Jiangyuan Li, Liyao Wang and Guofu Zhou, publishing in the same journal in 2020, report that asset-by-asset regressions show little evidence of the effect, that the pooled t-statistic falls short of bootstrap critical values, and that the overall evidence is weak.
Long-horizon reversal sits on similarly contested ground. James Poterba and Lawrence Summers found positive serial correlation at short horizons and negative correlation at long ones, across US data from 1871 to 1986, seventeen other markets from 1957 to 1985, and individual US firms from 1926 to 1985. Their estimate was that transitory components have a standard deviation of 15% to 25% of value and account for more than half the variance of monthly returns. They were candid that variance ratio tests have little power. With sixty years of data, such tests have less than a one-in-four chance of rejecting the random walk against the alternatives they cared about. They suggested using critical values looser than the conventional 0.05 as a result.
So the regime story rests on real but statistically fragile patterns. That matters for how much weight you can put on it.
Monitoring frequency and band width are one dial, not two
The second half of the question is more tractable, because it's mostly arithmetic. A band can only fire when somebody looks. So how often will you actually look? Change that and you've changed the effective band.
Vanguard's Figure 4 shows it cleanly, holding the threshold fixed and varying the monitoring interval over 1926 to 2018.
| Threshold | Monthly monitoring | Quarterly | Annual |
|---|---|---|---|
| 0%, any deviation at all | 1,116 | 372 | 93 |
| 1% | 426 | 233 | 83 |
| 5% | 58 | 47 | 34 |
Read down the columns instead. At monthly monitoring, moving from a 0% to a 5% threshold cut events by 95%. At annual monitoring the same change cut them by 63%. Your band does less work when you look less often, because infrequent looking is already a filter.
The costs of that choice are small and measurable. Norges Bank's 2018 discussion note simulated no-trade bands of 1 to 6 percentage points on a 70% equity share, using monthly MSCI USA and Bloomberg Barclays US Treasury returns from January 1973 to December 2017, bootstrapped into 1,000 fifty-year samples. Continuous monthly rebalancing generated about 6% one-way turnover a year and about 4 basis points of annual transaction cost. Widening the band raised the average equity share from 70% at 1 percentage point to 71.2% at 6 points, and lifted annualised tracking error against continuous rebalancing from 18 to 49 basis points.
The same note tested whether bands can exploit the regime effect deliberately. Its answer was mostly no. Narrower bands do co-move positively with expected returns when momentum dominates at short horizons. But the effects are, in the authors' words, economically small and statistically weak. They concluded that no-trade bands are an ineffective way to capture time-varying expected returns, and that this belongs low down the list when you're comparing rules.
Where the band still earns its keep
None of this makes bands useless. It makes their benefit a control benefit rather than a return benefit.
In the AQR sample, the average allocation range — how far holdings wandered from target — was 70% for the annual calendar and 52% for the 20% band. A 10% band tightened it to 35%, and monthly rebalancing to 27%. Buy-and-hold was 104%. If what you want is a portfolio that stays close to the risk you designed, tighter bands and more frequent looking deliver exactly that, and the return cost is a few basis points of turnover. What the rule is actually buying is risk control rather than return, which is what the evidence on the rebalancing bonus shows.
Now put money on it. Imagine you'd started 2005 with $100,000 in a 60/40 portfolio, checked it every quarter, and traded whenever it drifted 5% from target. On Vanguard's decade-long case study you'd have finished 2014 with $177,082 after tax. Your neighbour who never rebalanced finished with $172,170, about 5 percentage points of cumulative return behind, having been overweight equities into the crash and underweight through the recovery. It's the same mechanism described in our piece on how far a 60/40 portfolio drifts in a single year.
The counter-argument: a rule you'll actually keep to
The strongest objection to everything above is that it optimises the wrong variable.
Critics of regime-aware rebalancing argue that the calendar's advantage isn't its momentum exposure at all. A fixed date is cheap to run, easy to audit, hard to argue with, and it gets done. A band needs monitoring, a definition of what counts as a breach, and a decision from you at the worst possible moment, when one asset has just fallen hard. Vanguard's own conclusion after 92 years of data was that no specific threshold or frequency consistently outperforms, and that investors do best with a strategy they can stick with.
That objection is largely right, and the evidence above supports it more than it undercuts it. The spread across every sensible rule in the AQR table was 0.39 to 0.44 on Sharpe. In Vanguard's grid it was 8.19% to 8.39% in annualised return. The gap between rebalancing and not rebalancing was far bigger than any gap between rules: 0.32 against 0.43 on Sharpe, and 41.2% against 29.7% on maximum drawdown.
There's a second point in your friend's favour, made by Ilmanen and Maloney themselves. A trigger-based rule removes the element of chance an arbitrary date introduces. When a crisis lands in March and your rebalancing date is 31 December, the calendar's payoff depends on luck nobody can forecast. Anyone who has read the evidence on written investment policy statements and pre-commitment will recognise the trade. A rule that removes discretion is worth something even when its expected return is slightly lower.
Where the objection overreaches is in treating the regime effect as purely theoretical. It isn't. How long would you have sat with a drawdown 5 percentage points deeper in 2008? For anyone who capitulated near the bottom, that gap was real money.
What would change the calendar vs band rebalancing conclusion
Several things, and some of them are already halfway there.
The momentum evidence could fail to hold. Huang and co-authors have argued in print that the asset-by-asset case for time-series momentum is weak. If short-horizon trending isn't reliably present, the calendar's edge in the AQR sample was partly luck, and the honest ranking of rules collapses to a tie.
The samples matter too, and it's worth knowing what they're built from. AQR's results are hypothetical, net of an assumed uniform 0.5% transaction cost and gross of fees, and AQR sells trend-following strategies whose payoff profile the paper favours. The Man Group study carries the same shape of conflict: Rattray was Man Group's chief investment officer, Harvey advises the firm, and their proposed fix is an allocation to trend-following. Vanguard's Figure 4 splices five equity indices and five bond indices across 92 years and adjusts returns at assumed 30% income and 20% capital gains tax rates, so a UK investor's numbers would differ. Norges Bank's results are bootstrap simulations on US data alone.
Costs could change the ranking outright. All these studies assume institutional trading costs of a few basis points. If you're paying spreads on small trades, or realising capital gains outside a tax wrapper, you face a completely different cost curve. That pushes towards fewer, larger trades regardless of regime.
Your own portfolio matters as well. These results cover broad asset classes with roughly comparable long-run returns. A sleeve with three or four times the volatility of everything else breaches any band constantly, and the frequency-versus-width trade changes shape entirely. The width that would make one rule equally strict on a calm sleeve and a fast one is itself calculable, and volatility scaled bands come out a long way from a single 5 point number.
Finally, a longer horizon would help. Ilmanen and Maloney note the calendar advantage fades at ten-year rebalancing intervals, because full mean-reversion cycles complete in between. Nobody has enough independent 42-year samples to settle this. Two studies pointing the same way isn't proof, and the uncertainty bands here are wide — which leaves you and your friend where the evidence leaves everyone. You're choosing between two rules that sit close together, and both a long way ahead of doing nothing.