If a backtest improves when you rebalance more often, the first question is not whether the signal got smarter. It’s whether the extra trades were charged what they would cost to execute.
Rebalancing frequency changes two things at once: how quickly a portfolio follows its target, and how much trading it takes to get there. The useful comparison holds the signal and sizing rule constant, then measures the return earned by faster adjustment against the full cost of the additional turnover.
Why does a daily rebalance beat a weekly one in my backtest?
Sometimes it should. If a signal moves quickly, waiting a week can leave the portfolio holding stale weights. A daily schedule can reduce that lag. But it can also trade on small target changes that a weekly schedule ignores, and those changes are exactly where costs can overwhelm a thin edge.
Consider a $100,000 portfolio whose target weights move by 1% each day. That means about $1,000 of orders per day, before accounting for offsetting buys and sells. At a 10 basis point one-way cost, trading $1,000 costs roughly $1. If the signal truly benefits from each adjustment, that may be cheap. If the target wiggles around a stable level and the benefit is only a few cents, the daily schedule is paying to chase noise.
In a backtest, the weekly portfolio may look worse because it reacts late. The daily one may look better because it gets a more current target. Neither result is persuasive until both include comparable fills, fees, spread, and impact.
How do I compare daily and weekly rebalancing fairly?
Keep the underlying signal, universe, sizing, and test period fixed. Change only the schedule. Use the same execution assumptions for both, and calculate turnover from the orders each schedule actually generates, after netting changes across the portfolio.
| Measure | What it tells you | Common mistake |
|---|---|---|
| Gross return | How the target portfolio performed before trading costs | Calling it the strategy's result |
| Turnover | How much portfolio value changed hands | Counting buys but forgetting corresponding sells |
| Net return | What remains after fees, spread, and estimated impact | Applying a single fee rate while ignoring order size |
| Tracking error to target | How far the held portfolio lagged its desired weights | Treating perfect target matching as free |
One practical convention is to define one-way turnover as the lesser of total buys or total sells divided by portfolio value. State the convention in the report. Some systems use total traded notional instead, which can be about twice as large for a fully funded rotation. Either can work; silently switching between them makes comparisons useless.
Then report the incremental result: how much net return the faster schedule added, and how much additional cost it incurred. A daily schedule that adds 0.4 percentage points gross, costs 0.7 points more, and ends 0.3 points behind the weekly version has answered the question.
How should I model the cost of rebalancing?
Start with the venue and instrument. Spot and futures fee schedules differ; maker and taker rates differ; funding can apply to positions held through a payment window. For a simple first pass, apply the actual fee tier and a spread cost to each order. For larger orders, add impact as a function of order size and liquidity.
Suppose a strategy crosses the spread to buy and later crosses it again to sell. If the spread is 4 basis points, the round trip costs about 4 basis points in spread, not 2: each crossing pays roughly half the spread. Add taker fees on both sides. A “10 bp per trade” assumption is meaningless unless “trade” and the cost components are defined.
Impact should rise with the order’s size relative to available liquidity. Charging the same flat rate for a $500 order and a $500,000 order makes high-frequency rebalancing look artificially scalable. When the data can’t support a credible impact estimate, show a sensitivity range and make clear where the schedule stops working.
Should I use a rebalancing band instead of a fixed schedule?
A band can prevent tiny target changes from generating orders. For example, rebalance only when a holding drifts more than 2 percentage points from its target. This trades some tracking accuracy for fewer orders.
But a band is another parameter. If you try 0.5%, 1%, 2%, and 5% bands across many assets and keep the best result, you have added a search dimension. Choose a small set of economically motivated bands, record every trial, and evaluate the chosen rule on periods that did not select it.
A useful report shows the schedule, band, turnover convention, cost model, and the results before and after costs. If you can’t reconstruct why an order fired, you can’t tell whether a better schedule made the strategy better or just busier.
When is more frequent rebalancing worth it?
When the value of faster adjustment survives realistic costs and a modest change in assumptions. Compare net results across schedules, inspect turnover by asset and market condition, and check whether the advantage depends on one unusually liquid period.
And keep the paper strategy on the same schedule. A backtest that refreshes targets at every close while the paper system batches orders hourly is testing two different execution policies. The point of the comparison is to learn what the strategy asks the account to do, and what that activity costs.
← All posts
