Turnover and the Cost of Changing States

A label switch is not a trade

Two labels can map to identical holdings, producing no trade when the label changes. Conversely, a portfolio can trade while its label stays unchanged because weights drift, constituents change, or a scheduled rebalance occurs. Keep three records distinct: classification changes, target allocation changes, and executed or simulated trades. Counting state changes alone cannot establish implementation costs. A cost study should begin with the allocation policy and a chronological holdings ledger, including occasions when an intended action was delayed or could not occur.

Declare the turnover denominator and convention

For a fully invested, self-financing rebalance, gross traded weight can be defined as the sum of absolute changes from pre-trade to target weights. A common one-way turnover convention is half that sum. State which convention is used and whether cash, flows, derivatives, and leverage enter it. Compare targets with drifted pre-trade weights, not automatically with the previous target. The half-sum convention does not by itself describe cash flows or derivative notional exposure correctly. Report the underlying buys and sells so readers can reproduce the measure.

For short positions or derivatives, preserve contract quantities, direction, multipliers, and the chosen exposure measure alongside normalized turnover. Margin posted is not interchangeable with traded notional. A position reversal includes closing the old exposure and establishing the new one; measuring only a change in absolute exposure can conceal the transaction entirely.

Match each cost assumption to its base

A cost quoted per unit traded must multiply traded notional, not the entire portfolio unless the entire portfolio trades. Distinguish one-side from round-trip assumptions. Separate explicit charges, spread crossing, market impact, financing, and any modeled delay effect. Do not add a separate spread deduction when the execution model already includes it. Sharpe's arithmetic of active management distinguishes returns before and after costs; a research report needs the same clarity without treating an aggregate accounting argument as evidence about one particular strategy.

Check a fictional rebalance line by line

The invented example moves Asset X from 60% to 30% and Asset Y from 40% to 70%, with no intervening drift or external flow. It sells 30% and buys 30% of the same pre-trade net asset value. Gross traded weight is 60%; one-way turnover is 30%. An invented charge of 8 basis points per unit traded on each side produces 2.4 basis points of portfolio cost per leg, or 4.8 in total. The arithmetic ignores fee-induced weight renormalization and all other costs.

Separate an assumption from an estimate

Label a flat cost parameter as an assumption unless it has been estimated from relevant execution evidence. A constant charge cannot automatically represent different order sizes, liquidity conditions, venues, or urgency. Show a predeclared sensitivity range and the cost components it excludes. A break-even cost is the assumption that erases a specified gross advantage, not a forecast of achievable execution. In the fictional example, a separately invented 20-basis-point gross one-period advantage would leave 15.2 basis points after the illustrated 4.8-basis-point charge only.

Evaluate stabilization as a different policy

Hysteresis, confirmation, minimum holding periods, and no-trade bands can alter which trades occur. Research by Novy-Marx and Velikov examines transaction-cost mitigation, including different conditions for entering and continuing to hold a position. That motivates examining such mechanisms, not importing their empirical results into a regime classifier. Every added buffer changes delay and exposure as well as turnover. Declare candidate rules before evaluation, retain the original comparison, and report missed or delayed changes rather than scoring the mechanism solely by reduced trading.

Keep costs attached to the original decisions

Charge trades generated by the decision-time label archive, not a cleaned-up sequence rebuilt with revised inputs. Preserve canceled orders, partial actions, and stale-input decisions in the research ledger. Reconcile gross return, each modeled deduction, and net return on a common accounting base; disclose approximations when costs are simply subtracted. Report trading concentration near uncertain boundaries because an average can conceal clustered implementation demands. These examples teach cost accounting and uncertainty, not investment advice, observed transaction costs, or evidence that a particular switching policy is economical.

Inspect the comparison

Primary references

References support the stated methods and data conventions. Original fictional examples are not observations from the cited sources.

• William F. Sharpe: The Arithmetic of Active Management
• Novy-Marx and Velikov: A Taxonomy of Anomalies and Their Trading Costs

Continue the investigation