Modelling 0DTE options execution

Same-day expiry options are frequently modelled as ordinary options with a small time to expiry. The arithmetic supports that description and the execution problem does not. Nearly every quantity that matters behaves differently when expiry is measured in hours, and the differences run in the direction that makes a naive backtest optimistic.

Spread is not a small cost here

For a longer-dated option a bid-offer of five cents against three dollars of premium is under two percent. For a same-day option with twenty cents of premium the same five cents is twenty five percent, paid on entry and again on exit.

That single fact reorders the problem. A strategy needing to capture a move must overcome a round trip that can approach half its premium, which means the entry has to be right about direction, magnitude and timing simultaneously. Strategies that look viable on mid prices frequently have negative expectancy the moment they are evaluated on the prices at which trades actually occur.

The consequence for research is that mid-price backtesting is not a mild approximation in this instrument. It is the difference between a result and no result, and any study of 0DTE that does not state which side of the spread it assumed has not reported its most important assumption.

Gamma dominates everything else

As expiry approaches, gamma for near-the-money strikes rises sharply while vega collapses. The position's behaviour becomes almost entirely a function of spot, and almost entirely non-linear.

Two practical consequences follow. Delta is unstable: a hedge ratio computed a few minutes ago may be materially wrong, so any strategy that assumes a stable delta is assuming something the instrument does not provide. And the risk profile is genuinely asymmetric in a way linear approximations do not capture, so a risk system reporting delta and vega without the second-order terms is reporting the least relevant part of the exposure. This is the cross-derivative argument from the firm's method in its most acute form: the diagonal terms are small and changing fast, and the curvature is the position.

Theta is the other side of the same coin. Decay on the final day is not linear in time. It accelerates through the session and is concentrated in the final hours, which means holding period is a first-order decision. An hour of being right and early can cost more than being wrong would have.

Pin risk and assignment are execution problems

Near expiry, a position at a strike close to spot faces genuine uncertainty about whether it finishes in the money. For cash-settled index options this resolves into a settlement price. For physically settled equity options it resolves into a delivered position, and the holder does not learn the outcome until after the close.

The exposure is real: an option believed to expire worthless can be assigned, leaving an unhedged underlying position overnight. A backtest that assumes clean expiry at intrinsic value has removed a risk that exists, and removed it in the specific scenario where it is largest.

Settlement mechanics differ enough between products that they cannot be modelled generically. Cash versus physical, morning versus afternoon settlement, and how the settlement price is itself calculated all change the answer, and each needs to be looked up for the specific contract rather than assumed.

Liquidity is not where it looks

Displayed size on near-dated options is thin relative to the volume that trades, because much of the liquidity is provided on request rather than resting. Quotes widen sharply around events and in the final period before the close, precisely when strategies want to act.

Depth also varies dramatically across strikes within the same expiry. Round strikes and strikes near spot are liquid; adjacent strikes may be effectively untradeable in size. A backtest that treats the chain as uniformly available has assumed a book that does not exist.

Why end-of-day option data cannot test these strategies

This is the most common structural error in 0DTE research, and it is fatal rather than approximate.

An option that opens and expires within the same session has no meaningful end-of-day history. A daily bar for it is one observation of an instrument whose entire life was that session. Reconstructing an intraday strategy from daily data means interpolating a path that was never recorded, and the interpolation will be smooth where the actual path was not.

Evaluating these strategies requires intraday quote and trade data for the option chain, timestamped, with both sides of the market. That data is substantially larger and more expensive than daily summaries, and there is no adequate substitute. A study conducted without it is describing an instrument it has not observed, however careful the modelling downstream.

What a defensible model requires

• Both sides of the quote, at the decision timestamp. Buy at the offer, sell at the bid, with no mid-price assumption anywhere.
• Quoted size respected, with anything beyond it either rejected or walked up the chain at worse prices.
• Realistic latency, since a delay of a second on a high-gamma position is a real price difference rather than a rounding error.
• Explicit settlement handling for the specific contract, including assignment where physical delivery is possible.
• Time-of-day conditioning throughout. Behaviour at the open, mid-session and into the close are different regimes, and pooling them produces an average that describes none of them.

The general shape of the fill problem is covered in what a backtest gets wrong about fills. What makes same-day expiry distinctive is that every one of those errors is amplified by an instrument whose premium is small, whose spread is wide relative to it, and whose sensitivity to spot is at its maximum. The errors that are tolerable elsewhere are decisive here.