How portfolio rebalancing works

A portfolio has a target mix: the share of its value meant to sit in each asset, chosen once and written down. Sixty percent stocks and forty percent bonds is the usual shorthand, though the weights can cover cash, property or anything else held.

Rebalancing is the set of trades that puts the actual weights back onto the target ones. Because the weights always add to one, it means the same thing every time: sell part of what has risen above its target and buy what has fallen below it.

Drift is not a decision anyone makes. It is arithmetic. If asset $i$ starts at weight $w_i$ and returns $r_i$ over a period, its new weight is

$$w_i' = \frac{w_i(1 + r_i)}{\sum_j w_j(1 + r_j)}$$

which rises whenever $r_i$ beats the portfolio's own return. Over a long run the direction is one way: whichever asset compounds fastest keeps taking a larger share, and each larger share makes its next good year count for more.

At 9 percent a year for stocks and 4 percent for bonds, a 60/40 mix left completely alone drifts like this:

Those two rates are an assumption picked to show the mechanism, not a forecast, and they are nominal, before inflation. That last point costs nothing here: a weight depends only on the ratio between the two growth factors, so deflating both by the same inflation rate leaves every row of the table exactly where it is. Real drift is lumpier than the table and runs backwards for years at a time.

One thing needs less correcting than it looks. A fund that weights its holdings by market value needs no trades to hold those weights in line, because they move with prices by construction. It still trades for index changes, corporate actions and its own inflows and outflows, but not to undo drift. The drift to fix is the drift between the funds you hold rather than inside them.

Drift is silent. No trade appears, no statement flags it, and the portfolio still feels like the one that was set up, because nobody set up anything else. What changed is how much of the risk sits in one place.

Risk does not add up in proportion to the weights. For two assets, portfolio volatility is

$$\sigma_p = \sqrt{w^2\sigma_s^2 + (1 - w)^2\sigma_b^2 + 2w(1 - w)\rho\,\sigma_s\sigma_b}$$

Put stock volatility at 15 percent a year, bond volatility at 5 percent, and the correlation between them at 0.1. All three are illustrative, and the correlation is the least stable of them: stocks and bonds have spent long stretches falling together rather than offsetting, and a higher correlation lifts every volatility figure below. A 60/40 portfolio on those inputs has a volatility of about 9.4 percent a year. The same portfolio after drifting to 80/20 has about 12.1 percent, a rise of 29 percent, with no decision taken anywhere.

The split of the risk is starker than the split of the money. On those figures, once each asset carries its own variance plus its share of the covariance between the two, stocks supply about 93 percent of the variance of a 60/40 portfolio and about 98 percent of an 80/20. A balanced fund is not balanced in risk terms and never was. Drift makes it less so.

The consequence shows up in a fall. If stocks drop 40 percent while bonds hold flat, a 60/40 portfolio is down 24 percent and an 80/20 is down 32 percent. Recovering the first takes a gain of about 32 percent; recovering the second takes about 47 percent, which is how drawdown arithmetic works. That is not a market shock behaving unusually. It is the same shock landing on a larger stock weight than the plan called for.

The mirror image gets less attention. Drift pushes the stock weight to its highest after a long run of gains and to its lowest after a severe fall, so an untouched portfolio carries its largest stock weight out of a boom and its smallest out of a crash. That is a statement about exposure and not a forecast. It says the weight sits furthest from the chosen one at exactly the two moments the choice matters most, not that a reversal is due.

There are two families of rule, and one hybrid that most practice settles on.

Calendar rebalancing trades on a fixed date: every year, every half year, every quarter. It is simple, it can be diarised, and it asks for no monitoring at all. It also ignores what markets did, so it trades when nothing has moved and waits when everything has.

Threshold rebalancing trades when a weight leaves a band around its target, say five percentage points either side of 60 percent. It responds to size rather than to the calendar, so it acts when there is something to act on. It needs the weights watched, and in a violent stretch it can call for trades repeatedly.

Band width is where the detail lives. Five percentage points around a 60 percent target is a 55 to 65 range, roughly 8 percent of the target itself. The same five points around a 10 percent holding is half the position, so an absolute band that is tight on the largest sleeve is loose on every small one. A relative band, such as a quarter of the target weight, scales with position size instead, and the trade-off is visible in the numbers: a quarter of a 10 percent holding is 2.5 points, tighter than five, while a quarter of a 60 percent holding is 15 points, far looser. Neither shape is right everywhere, which is why bands are often set in points on the large sleeves and in proportions on the small ones.

The hybrid is to check on a schedule and trade only if a weight sits outside its band. That caps monitoring at a few dates a year and skips the trades a pure calendar rule would make to correct a drift of half a point.

Two further choices. Rebalancing all the way back to target, or only to the near edge of the band, which cuts turnover and leaves a little of the run in place. And the frequency itself, which attracts more agonising than it repays: studies that line monthly, quarterly and annual rules up against each other find different winners in different periods, and gaps between sensible rules that are small next to the gap between having a rule and having none.

The cheapest rebalancing trade is the one that is never a sale.

• New contributions. A deposit sent whole to whichever sleeve sits furthest below target pulls the weights back much harder than the same deposit split in target proportions.
• Income. Dividends and interest taken as cash instead of reinvested where they arose can then land wherever the mix is short.
• Withdrawals. In the spending phase, money taken from whatever sits above target does the work of the rebalancing trade and the work of the withdrawal at once.

The arithmetic is worth seeing once. A portfolio has drifted to 65 percent stocks against a 60 percent target, and receives a contribution worth 5 percent of its value, all of it directed at bonds. The stock holding has not changed while the total has grown, so the stock weight falls to about 61.9 percent. A five point drift becomes a drift of under two points with nothing sold. Split the same contribution in target proportions instead and the weight only slips to about 64.8 percent, which is the difference the direction makes.

This works best early. A contribution is large next to a small portfolio and small next to a large one, so the deposit that holds the mix in place in the first decade barely moves it in the third. Cash flow does the job for a while, and then trades have to finish it.

Where several accounts are involved, the sheltered ones are the cheaper place to sell. The mix that matters is the one measured across everything owned, so a trade inside a retirement account can correct a drift caused by a taxable one, and in the United States that trade creates no tax at the time it is made. Sheltered wrappers in other countries often behave the same way, though not all do: some are taxed on gains realised inside the wrapper. Tax-advantaged accounts sets out which wrappers do what. Which asset belongs in which account is a separate question, and one the tax code answers differently in different countries.

Inside a tax-sheltered account, rebalancing is close to free: sell, buy, and nothing falls due until money leaves the wrapper, if anything falls due at all. Inside a taxable account, the same two trades are a taxable event.

In the United States, selling a holding for more than it cost realises a capital gain, taxed in the year of the sale. How long the holding was owned decides which schedule applies, and both the schedules and the thresholds are set by legislation and revised, so the durable part is the mechanism rather than any rate: rebalancing pulls a future tax bill into the present, and tax paid now is money that stops compounding. Elsewhere the treatment differs enough to change the answer. Some countries tax gains only above an annual allowance, some at a flat rate on disposal whatever the holding period, some not at all on assets held long enough. The size of the friction is local. Its existence is not.

Trading costs are usually the smaller of the two frictions in a taxable account, and the only one inside a sheltered account: the bid-ask spread paid on every trade, dealing fees where a platform charges them, and transaction taxes such as the stamp duty the United Kingdom charges on purchases of shares in UK-incorporated companies.

What the friction changes is the rule, not whether to have one. It argues for wider bands, less frequent trading, and using contributions and income before sales. Two specifics for a taxable account in the United States. Realising a loss on the underweight side can offset a gain taken on the overweight side, which is tax-loss harvesting. And the wash sale rule blocks a loss if a substantially identical security is bought back inside 30 days either side of the sale, which is easy to trip by rebalancing straight back into the fund just sold. The loss is not destroyed: it is added to the cost of the replacement and arrives on the eventual sale instead, so what the rule takes away is the timing.

Two forces pull in opposite directions here, and they are measured against different baselines. Keeping the baselines apart is most of the work.

Against it. Selling the faster grower to buy the slower one lowers growth for as long as the faster grower keeps winning. Take the volatility out: stocks return exactly 9 percent every year, bonds exactly 4. A 60/40 rebalanced yearly then returns exactly 7.00 percent a year, forever. Buy the same two and never trade, and the compound rate is 7.24 percent over 10 years, 7.49 percent over 20 and 7.70 percent over 30, because the winner grows into a larger share. After 20 years the untouched portfolio is about 9.5 percent larger. Nothing in that example wobbles, so nothing is being collected from wobble: it is what selling the winner costs. Swapping the two rates over changes nothing, because the same asset still leads the whole way. What changes the sign is the assets changing places: the general two-period result is that leaving a portfolio alone wins whenever the same asset leads in both periods, and rebalancing wins whenever they swap.

For it. Put the volatility back, and a rebalanced blend compounds faster than the weighted average of its own parts, by roughly half the variance that spreading removed:

$$\text{extra} \approx \tfrac{1}{2}\left(\sum_i w_i\sigma_i^2 - \sigma_p^2\right)$$

On the 15 and 5 percent volatilities from earlier, that is about 0.28 percentage points a year for a 60/40. For two assets of equal 20 percent volatility with zero correlation, held half and half, it is about 1 point.

Two warnings travel with that figure, and skipping them is where most of the overselling of rebalancing happens. It is measured against the weighted average of the parts, not against leaving the portfolio alone, and those are different questions with different answers. And it does not depend on the assets taking turns: for two assets the bracket works out to $w(1-w)$ times the variance of the gap between their returns, which is positive whenever the two differ at all. So it stays positive in exactly the runs where leaving the portfolio alone still finishes ahead. It is not a number to set against the 9.5 percent above.

Which force wins against leaving the portfolio alone is a fact about the period, not about the rule, so the return effect is small and inconsistent. The risk effect is neither, because it needs no forecast: a mix held near target keeps the exposure chosen for it, and diversification holds only while the weights do.

Three things this is not. Not market timing: the rule is fixed in advance and needs no view of what comes next. Not a reason to buy more of one falling company, since the argument assumes a broad asset class rather than a single holding, and a single holding can reach zero and stay there. Whole national markets have been wiped out too, so breadth moves the odds rather than removing the risk. Not the same decision as changing the target, which is a separate judgement about how much risk suits a particular person. This page is educational material about how the mechanism works, not financial advice about your own money.