I beat my own backtests trading 0DTE SPX options

The common consensus seems to be:

"Backtests always overstate returns."

But that's only really if you suck at backtesting.

If your backtest framework is actually solid, you can actually estimate returns pretty accurately.

Common pitfalls and how I avoided them:

1. Overfitting: Every parameter value was chosen as a range, not a single point. Picking a single value is arbitrary, picking a range is much more robust. Further, each parameter has an actual justification, not just from the data. For example: "skip when IV is extreme" is based on how markets actually work, not just cherry picked, even if the exact value is based on the data. Every component of the strategy MUST have some rationale.
2. Parameter soup: The temptation is strong to keep adding complexity. But, to make something robust, you must resist. The best strategies have very little parameters or complexity. Actual market behavior isn't as complicated as you think.
3. Fill modeling: Using custom code or data often causes this one. My simple solution was to just pay the ~$20/m for QuantConnect, which includes their SPX options data as well. Honestly, pretty cheap compared to how expensive losing strategies are.
4. Transaction fees: Used custom fill model in QuantConnect to model the ACTUAL fees that tastytrade charges me (1.74 per leg of the vertical).

Results: Find the full breakdown here. The live results have been even better than the backtests for the last 3 months.

Hope this will be useful to anyone designing their own strategy.

https://preview.redd.it/yurm8v9d5frg1.png?width=1398&format=png&auto=webp&s=ebada4d9c0c9c7fd15f496ce791d31f2cd0017eb