A monthly momentum strategy can rank assets correctly and still report a fantasy trading cost. The missing step is the rebalance: converting yesterday’s holdings and today’s target weights into actual orders. Here’s a small teardown using four liquid crypto spot assets, with numbers simple enough to audit by hand.
Assume a $100,000 portfolio, long-only, rebalanced monthly. It holds the top two assets by trailing return, equally weighted. At the prior rebalance it owned $50,000 each of A and B. One month later, after prices move, the portfolio is worth $110,000: A is now worth $60,000 and B $50,000. The new ranking puts B and C in the top two. C is not held yet.
The ranking row: who belongs in the portfolio?
The signal says hold B and C at 50% each. Applied to current portfolio value, that means target holdings of $55,000 apiece. The rank change tells us which assets to own; it doesn’t tell us how much to trade. That depends on the holdings carried into the rebalance.
| Asset | Current value | Target value | Order |
|---|---|---|---|
| A | $60,000 | $0 | Sell $60,000 |
| B | $50,000 | $55,000 | Buy $5,000 |
| C | $0 | $55,000 | Buy $55,000 |
| Total | $110,000 | $110,000 | $120,000 traded |
Notice the quiet detail: B stays in the portfolio, but it still needs a $5,000 order to restore its weight. A backtest that trades only additions and removals misses that drift correction.
The order row: how much turnover did it create?
Gross traded notional is the sum of buys and sells: $60,000 + $5,000 + $55,000 = $120,000. Dividing by the $110,000 portfolio gives 109.1% gross turnover for this rebalance. If the strategy reports one-way turnover, it may instead divide by two, counting the sell and reinvestment as a single rotation. Either convention can work; label it clearly and use the matching cost formula.
Suppose all orders are taker executions with a 0.10% fee per side. Fees are $120. At an assumed 0.15% price impact per side, the estimated impact is another $180. The total rebalance cost is $300, or about 27 basis points of portfolio value. If you subtract 0.25% once from gross traded notional, you get $300 here too. That shortcut works only because both the fee and impact assumptions are applied consistently to every traded dollar.
These are assumptions, not fills. Impact usually depends on order size relative to market volume, and one flat rate can badly misprice the largest orders. A $55,000 buy in a deep market may be easy; the same order in a thin token may walk the book. If the strategy sizes by rank alone, it can accidentally route most of its trading into the least liquid names.
The timing row: what price can the strategy actually get?
There’s another trap hiding in the example. If the ranking uses month-end closing prices, those values are only final after the close. The strategy cannot see the completed ranking and also trade at that same closing price. One defensible simulation calculates targets at the close and executes at the next session’s open, charging costs on the resulting orders. For continuously traded crypto, define a cutoff and a later execution window; “monthly close” needs a timezone and an actual order time.
That delay can change the holdings and therefore the orders. If A rises further before execution, the sell notional grows; if C jumps, its buy gets more expensive. Recalculate target quantities from the portfolio value and executable prices at the time orders are placed. Keeping the close’s target dollars fixed while moving execution forward silently changes the target weights.
The ledger row: does the cost reduce the portfolio?
After execution, charge $300 to the cash balance or equity. The portfolio now has $109,700 before subsequent market moves. If a simulator subtracts a cost from returns but leaves cash untouched, later sizing can spend dollars that were already consumed. Track orders, fills, fees, and holdings in the same ledger; otherwise the next rebalance may start from a portfolio that never existed.
For this example, I’d store the signal timestamp, target weights, pre-trade holdings, order notional, execution price, fee, and impact estimate for each asset. Then I’d compare the calculated turnover with a direct sum of fill notionals. That audit catches the familiar bugs: omitting retained-asset drift, charging costs on half-turnover twice, and using target weights as though they were fills.
A rebalance is where a portfolio rule becomes a set of trades. Write down each conversion, preserve the ledger, and make the turnover convention explicit. The resulting backtest may look less elegant. It will at least be describing orders the strategy could have placed.
← All posts
