Realistic Fill Assumptions for Historical Option Backtests
Summary
The discussion explains why filling historical option trades at the midpoint of quoted bid and ask can make a backtest too optimistic. A market maker generally quotes a spread to compensate for the risk of trading against a one-sided order, so the reply recommends testing conservative fills at the relevant side of the spread. It also notes that multi-leg option spreads may sometimes execute inside the quoted spread, especially when their combined risk is lower or a market maker can offset the position against other flow.
The answer cautions that larger orders can move implied volatility as market makers respond to order flow, causing execution prices to worsen. For more reliable analysis, it recommends obtaining trade and quote data, ideally quotes recorded at trade time. The document does not specify how to reconstruct missing or implausible quotes, nor does it provide a detailed fill model; its guidance is qualitative and depends on liquidity, order size, and execution circumstances.
Key ideas
- Midpoint fills can overstate performance because they assume execution inside the quoted spread.
- A conservative backtest should account for the side of the spread an order would have to cross.
- Multi-leg option orders may receive better prices when their combined risk is lower or can be offset.
- Larger orders can shift implied volatility and worsen the price available for entry.
- Trade and quote records observed together can support more realistic execution assumptions.
Tags
Full text
# Backtesting on historical option data # Backtesting on historical option data I have downloaded some daily historical option data for a timespan of 10 years and want to perform trading backtests with them. Data are European index options, on ODAX. My question is about realistic assumptions on backtest fillings: a) valid bid/asks: I would assume the mean of bid and ask as filling price b) bid and ask=zero: This happens mostly for some of the ITM options, the underlying beeing more than 300 points away from ATM-strike. Not sure what to do: my idea - market maker has to take counterparty, so I assume realistic quotes and realistic bid ask spread and again mean of bid and ask. c) bid and ask unrealisticly high/unrealisticly low: happens both ITM and OTM, same procedure as in b) Are those realistic filling rules? If not, how should the rules be? ## Answer by Nathan S. (score 2) https://quant.stackexchange.com/a/16904 The mean of bid and ask for a fill is not realistic and this will impact your analysis a lot in options. The market maker is quoting the spread because that's how they intend to get paid for the risk if you bring them one leg. Use the worst pricing assumption on this, even though you should probably still rest orders in practice for anything that has a spread of significance. Option Spread Trades If you bring a spread trade with lower risk then market maker is more likely to do business inside of the bid/ask spread. If the market maker sees you resting an order and has an opportunity to spread (risk management options position type spread - not bid/ask) your trade against somebody else resting in the book then they're also more inclined to do business inside of the bid/ask spread. Market Impact If you go with size (what size is depends on liquidity of the products you trade, but it can be surprisingly low in some 'popular' products) in one trade or a number of trades then market makers in the product will adjust implied volatility in response to order flow. This means that good prices for establishing your position will be getting away from you. More Data is Always Better Try to get trade and quote data or at least get trade data where quotes are included at trade time. This is a lot of data and it's never cheap. But options are complicated and the liquid markets are very well arbitraged.
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.