Turnover and Rebalancing

A backtest computes returns as if you could snap the portfolio to its ideal weights for free, every day, forever. Reality charges you every time you trade. Turnover measures how much trading a strategy actually demands, and rebalancing is the policy that governs when you do it. Getting this right is often the difference between a strategy that looks great on paper and one that survives contact with real costs, because turnover is the multiplier that turns a per-trade cost into a drag on your whole return.

Measuring turnover

Turnover is the fraction of the portfolio you replace over a period. The common one-way definition sums the changes in each asset's weight and halves it (halving avoids double-counting the sell and the matching buy):

$$\text{Turnover} = \frac{1}{2}\sum_{i=1}^{N} \big| w_i^{\text{new}} - w_i^{\text{old}} \big| ,$$

where $w_i^{\text{old}}$ and $w_i^{\text{new}}$ are asset $i$'s weight before and after the rebalance. A turnover of 0.20 means you turned over 20% of the book. Multiply annual turnover by your round-trip Transaction Costs to get the cost drag:

$$\text{annual cost drag} \approx \text{annual turnover} \times \text{round-trip cost}.$$

Worked example

A three-asset portfolio drifts and you rebalance back to target:

Asset	Old weight	Target weight	Change
Stocks	0.66	0.60	−0.06
Bonds	0.24	0.30	+0.06
Gold	0.10	0.10	0.00

The absolute changes sum to $0.06 + 0.06 + 0.00 = 0.12$. One-way turnover is $\tfrac12 \times 0.12 = \textbf{0.06}$, you trade 6% of the portfolio. On a $50,000,000 book that's $3,000,000 of trading; at 15bp round-trip, the rebalance costs $4,500. Do this monthly and annual turnover is roughly $12 \times 6\% = 72\%$, costing about $0.72 \times 0.15\% \approx 0.11\%$ a year, small here because the drift is small, but a fast signal that fully rebuilds the book each month can hit 1000%+ turnover and bleed several percent annually.

Rebalancing policies

You almost never want to chase the target continuously, that maximizes cost for minimal benefit. The main policies trade off tracking error against turnover:

Policy	Rule	Turnover	Trade-off
Calendar	Rebalance every month/quarter	Predictable	May trade when unnecessary, or drift too long between dates
Threshold (bands)	Rebalance an asset only when it drifts past ±X%	Lower	Trades only when it matters; needs monitoring
No-trade band	Trade back to the edge of the band, not the target	Lowest	Minimizes cost; accepts small permanent tracking error

The no-trade band is the practitioner's favorite. You define a tolerance around each target weight and do nothing while the weight stays inside it; when it breaches the band you trade back only to the band's edge, not all the way to the center. This exploits a deep result: because a small deviation from optimal weights costs you almost nothing (the objective is flat near its optimum) but every trade costs real money, the optimal policy is to tolerate a band of "good enough" and trade only at its boundary.

Where it misleads

Two more traps. First, larger portfolios can't assume costs stay linear, big rebalances move the market (Market Impact), so high turnover caps how much capital a strategy can hold (Portfolio Capacity). Second, turnover interacts with taxes for taxable accounts: frequent trading realizes short-term gains, an extra drag backtests never model.

The discipline is simple to state and hard to practice: every trade must earn its cost. Rebalance because the drift genuinely hurts your risk or return, not because the calendar told you to, and always net the paper alpha against the turnover it demands before believing a backtest.