We thought we had a low-turnover options strategy. It bought a one-week call when a short-term signal fired, held it for a day or two, then exited. The backtest showed a handful of option orders each week and costs that looked small beside the gross P&L.
Then we counted the underlying trades used to keep the position near its target delta. They outnumbered option trades by more than ten to one. In a volatile session, a single option position generated six hedge orders before lunch. The strategy had not found a cheap way to trade options; its backtest had mostly forgotten that the hedge existed.
That distinction matters because an option’s delta changes with the underlying price, time, and volatility. A delta-neutral position at entry can become a directional bet minutes later. Whether you hedge continuously, on a schedule, or only after delta moves past a threshold changes both risk and turnover.
Our example position was long ten calls, each with a 0.45 delta, on a contract multiplier of one. At entry, that meant a short hedge of 4.5 units of the underlying. The backtest rounded to five and placed the hedge at the same midprice used to value the option.
The market moved up. The call delta rose to 0.60, so the target hedge became six units short. The strategy sold one more unit. When the price pulled back, delta fell to 0.51 and it bought one unit back. Every adjustment looked modest. Together, they paid the spread twice before the option position was even closed.
Where did the missing cost come from?
The first problem was the trigger. The simulation recalculated the hedge after every bar, but executed the adjustment at that bar’s midpoint. The midpoint is a useful mark; it is not a fill promise. For a marketable hedge order, the relevant price is closer to the ask when buying and the bid when selling, with additional impact if the order is large relative to available liquidity.
We replayed the same hedge with a simple quote-side model. For each adjustment, we used the best available bid or ask, added a small impact estimate based on order size and displayed depth, then charged the venue’s taker fee. It wasn’t a universal execution model. It was enough to expose how much the midprice assumption had hidden.
The second problem was that “rebalance every bar” had been treated as a harmless implementation detail. It wasn’t. A one-minute schedule produced frequent small hedge trades. A five-minute schedule traded less, but let delta drift further between checks. A threshold rule waited until the absolute hedge error exceeded one underlying unit, cutting some churn while accepting larger temporary exposure.
| Hedge rule | What it did in the replay | Trade-off |
|---|---|---|
| Every minute | Most adjustments; lowest measured delta drift | More spread, fees, and sensitivity to noisy quotes |
| Every five minutes | Fewer adjustments; wider swings between checks | Less turnover, more exposure to short-term moves |
| One-unit threshold | Skipped small changes until hedge error crossed the threshold | Lower churn, with a deliberate band of residual delta |
There was a small practical wrinkle: our historical quote snapshots were uneven around the busiest minutes. We could estimate costs for those gaps, but shouldn’t pretend that a missing quote contains a reliable executable price. The replay marked those adjustments as uncertain and reported them separately.
On the test period, moving from midpoint fills to quote-side fills with fees and impact reduced the strategy’s net P&L. The one-minute rule also traded substantially more underlying than the option blotter suggested. The threshold rule cut hedge turnover, though its delta exposure spent more time away from zero. Neither result says which policy is better in general; that depends on the strategy’s risk limits, venue liquidity, and how much residual exposure it can tolerate.
What should an options backtest record for its hedge?
At minimum, keep a separate hedge ledger. Each row should show the option position, target delta, current hedge, adjustment size, decision time, assumed execution time, price basis, fees, and impact. If the hedge crosses venues or instruments, record that choice too. A result that combines option and hedge trades into one “options cost” number is hard to audit when it looks suspiciously good.
Also compare policies on the same option signals and price history. If the hedge schedule changes between runs, report the resulting turnover and exposure alongside performance. Otherwise a lower-cost curve may simply be carrying more unhedged risk.
Our original strategy idea survived the accounting change as a research question. Its first backtest did not survive as evidence. That’s a useful outcome: the next run had a clear ledger, an explicit hedge rule, and a cost model we could challenge one assumption at a time.
← All posts
