The arithmetic and cost of rebalancing

Chapter 1 noted that weights move without you. This chapter is about whether to move them back, and it reaches a conclusion that most explanations of rebalancing do not.

The drift

Two sleeves compounding at different rates diverge, and the arithmetic is unforgiving. A 60/40 portfolio with equity at 11% a year and debt at 7%:

Year	Equity weight
0	60.0%
5	64.3%
10	68.4%
15	72.2%
20	75.8%
30	81.9%

After twenty years the "60/40 portfolio" is a 76/24 portfolio. Nobody decided that. It is what a 4-point difference in compounding rates does over two decades.

And in a world with randomness the drift is far wider. Simulating twenty years of correlated returns, a never-rebalanced 60/40 ends with a median equity weight of 73.8%, and a range from 50.7% at the 5th percentile to 88.5% at the 95th.

That spread is the real argument for rebalancing. It is not that the drifted portfolio is worse — it is that you no longer know what you own. An investor who chose 60/40 for a reason has, without acting, ended up somewhere between 51% and 89% equity depending on what markets did.

Does rebalancing add return?

This is where most explanations go wrong, so it is worth testing rather than asserting. Twenty thousand simulated twenty-year paths, equity at 11% and 14% volatility, debt at 7% and 4%, correlation 0.2:

Strategy	Mean CAGR	Spread of CAGR	5th percentile	95th percentile	Turnover
Never rebalance	9.67%	2.56%	5.82%	14.17%	0
Rebalance annually	9.48%	2.18%	5.96%	13.13%	170
Rebalance on a 5% band	9.49%	2.20%	5.95%	13.19%	104

Rebalancing lowered the mean return by about 0.19 percentage points a year. It did not add anything.

The mechanism is not subtle. Rebalancing a portfolio whose assets have different expected returns means systematically selling the higher-returning asset to buy the lower-returning one. Over time that must reduce expected return, because you are holding less of the thing that returns more.

What it bought instead is in the other columns. The spread of outcomes narrowed from 2.56% to 2.18%, and the bad case improved: the 5th percentile rose from 5.82% to 5.96%. Rebalancing trimmed the top and lifted the bottom — which is exactly what a risk control should do, and exactly not what a return enhancer would do.

So the honest statement is: rebalancing is a risk-management device that costs a small amount of expected return. The "rebalancing bonus" that appears in sales material is a special case requiring assets with similar expected returns and strong mean reversion; it is not a general property, and it is not what the arithmetic above produces.

This also disposes of a common objection to rebalancing — that it means "selling your winners". It does mean that, and that is the point, and it has a cost, and the cost is worth paying for the control. All three clauses are true together.

Calendar or bands

Two ways to decide when.

Calendar. Rebalance on a fixed date — annually, or semi-annually as the Nifty 50 itself does.

Threshold bands. Rebalance when a weight strays more than a set amount from target, say five percentage points.

The simulation says the outcomes are nearly identical — 9.48% against 9.49%, with similar spreads. The difference is in the turnover column: 104 against 170, so bands did the same job with about 39% less trading.

Why bands are more efficient. A calendar rule trades on a date whether or not anything has drifted, and ignores a large drift that happens the week after. A band rule trades when there is something to fix. The same risk control for less cost is a straightforward improvement, and it is why institutional policies are usually written as bands with a calendar review rather than as calendar trades.

The practical hybrid: check on a schedule, act only if a band is breached.

What it costs

The simulation above was costless. Real rebalancing is not, and the costs are in a specific order of size for an Indian individual.

Tax is usually the largest. Rebalancing realises gains that would otherwise have stayed unrealised, and brings tax forward from some distant date to this year. Bringing a tax payment forward is a real cost even if the rate is unchanged, because the money paid is money no longer compounding — the Tax subject's chapter on tax and your real return makes this argument, and its chapter on capital gains covers how the heads work under the Income-tax Act 2025.

Transaction costs are second, and smaller than people expect. On a ₹10 lakh portfolio:

Drift to correct	Amount traded	Cost at 0.1%	Cost at 0.3%
3%	₹30,000	₹30	₹90
5%	₹50,000	₹50	₹150
10%	₹1,00,000	₹100	₹300

A 5% band breach on a ₹10 lakh portfolio costs somewhere between ₹50 and ₹150 to correct. That is not the reason to rebalance less often; the tax is.

Exit loads are third, where they apply — the Mutual funds subject covers them, and a load charged on units held under a year can exceed every other cost combined.

And for anyone using derivatives to adjust exposure, note that the cost went up. STT on the sale of a futures contract rose to 0.05% from 0.02% with effect from 1 April 2026 — a 150% increase on the instrument most often used to shift exposure without selling the underlying.

The cheapest rebalancing is not a trade

Three ways to move weights without selling anything, in descending order of usefulness.

Direct new contributions. A monthly SIP into whichever sleeve is below target rebalances continuously at zero cost and zero tax. For anyone still accumulating, this handles most drift before it becomes a band breach, and it is the single most useful idea in this chapter.

Direct distributions. Dividends, coupons and withdrawals can be taken from the overweight sleeve.

Rebalance inside a single fund where possible. A fund that holds both equity and debt rebalances internally, and the investor realises nothing. The same economic act is taxable across two funds and not taxable inside one — which is a genuine structural advantage of a balanced fund, and one of the few places where the product wrapper changes the answer rather than just the fee. The Mutual funds subject covers what such a fund charges for it, which is the other half of the comparison.

Working the problem

60/40, never rebalanced, twenty years, equity at 11% and debt at 7%.

Step 1 — compound each sleeve. Starting from ₹60 and ₹40 per ₹100:

$\text{equity} = 60 \times 1.11^{20} = ₹483.7, \qquad \text{debt} = 40 \times 1.07^{20} = ₹154.8$

$w_{\text{equity}} = \frac{483.7}{483.7 + 154.8} = 75.8\%$

Step 2 — would annual rebalancing have made you richer? No. The simulation says mean CAGR falls from 9.67% to 9.48%. Rebalancing would have made you poorer, on average, by about 0.19 points a year — before any tax or transaction cost, which make it worse.

Over twenty years that compounds to a real difference. ₹100 growing at 9.67% reaches ₹636; at 9.48% it reaches ₹614. About 3.5% less final wealth, as the price of the control.

Step 3 — what it would have done instead. Three things.

Kept the portfolio at the risk you chose. The unrebalanced portfolio ends at 76% equity in the deterministic case, and anywhere from 51% to 89% in the simulated ones. The rebalanced one is 60% equity on every path, which is the whole product being bought.

Narrowed the range of outcomes. The spread of twenty-year CAGRs fell from 2.56% to 2.18%, and the fifth-percentile outcome improved from 5.82% to 5.96%.

Removed a decision you would otherwise face at the worst time. A portfolio that has drifted to 85% equity will fall much harder in a crash, and that is precisely when an investor discovers their allocation and sells. The Behavioural finance subject's measured gap is largest in exactly that situation.

Step 4 — the conclusion I would actually draw. Rebalance, but for the right reason and with the cheap method.

The reason is risk, not return. Anybody who tells you rebalancing increases returns is selling something; the arithmetic says it does the opposite, and the simulation measures by how much.

The method is bands, funded by contributions. A five-point band checked annually, corrected first out of new money and only then by selling, captures essentially all the control for a fraction of the turnover and most of the tax.

And the 0.19 points is the honest price. Knowing what you own costs about a fifth of a percentage point a year. That is cheap for what it buys, and it is not free, and both halves should be said.

The point

Different compounding rates make weights drift: a 60/40 portfolio at 11% and 7% becomes 76/24 in twenty years, and across simulated paths the equity weight lands anywhere between 51% and 89%. Rebalancing fixes that, and over twenty thousand simulated twenty-year paths it reduced mean CAGR from 9.67% to 9.48% while narrowing the spread of outcomes and lifting the fifth percentile — a risk control that costs return, not a source of it, since it systematically sells the higher-returning asset. Threshold bands achieve the same with about 39% less turnover than calendar rebalancing. The dominant cost for an individual is tax brought forward rather than transaction charges, which is why the cheapest rebalancing is done with new contributions, and why a fund that rebalances internally does it without realising anything.