Derek

Ex-hedge fund trader. 20+ yrs trading options. Building AI option trading models. Join the lab 👇

[Derek @derekdevore]

The exciting part of an AI trading system is the model.

The part that decides whether it works is boring: unadjusted splits, missing expirations, stale quotes at 3:59.

Nobody sees that work.
Every result depends on it.

[Derek @derekdevore]

A 0DTE option costs pennies. The futures hedge behind it can be huge.

But hedging has no fixed direction.
Dealers long gamma: they sell rallies, buy dips, moves pin.
Short gamma: they chase, moves stretch.

via Nicholas Crown on YouTube

https://t.co/RWEunxg1uD

[Derek @derekdevore]

Meta plans to raise $95 billion for AI infrastructure.

Nvidia, AMD and Broadcom may buy up to $5 billion of those shares.

The suppliers are helping fund their customer's build-out.

Long $META calls plus semis? That can be one bet on AI capex.

per CNBC

[Derek @derekdevore]

@lalovestrump I agree. Plus at my location the manager isn't stingy with extra salsa's, etc.

[Derek @derekdevore]

The unglamorous part of building a trading system:

Running it live at 1-lot size for three months.

Logging every fill. Checking it against the backtest. Watching nothing exciting happen.

That boring stretch is how you earn the right to size up.

[Derek @derekdevore]

A chart signal isn't a trade.

Bearish cross = close long? Or open short? Same indicator, two strategies, two different backtests.

"Compiled and verified" means it runs. It doesn't prove an edge.

Backtest both mappings.

via Smritikapoor on Medium

https://t.co/sCKgWoR61a

[Derek @derekdevore]

@storm1news Yes

[Derek @derekdevore]

@theofinr I just accept it as a grind and keep going..

[Derek @derekdevore]

One limit: shuffling assumes trades are independent.

Short-premium losses tend to cluster in volatile stretches. A plain shuffle breaks those streaks apart and can understate risk.

Shuffle blocks of consecutive trades too.

Not financial advice.

[Derek @derekdevore]

Your backtest's max drawdown is one number from one ordering of trades.

Shuffle the same trades and that number moves. Sometimes a lot.

Here's how to find the drawdown you should actually plan for, in 4 steps: 🧵

[Derek @derekdevore]

If you have AI write the script, check one thing before trusting it:

Every shuffle should end at the exact same total P&L.

If the totals differ, it's sampling with replacement or dropping trades. That's a different test.

[Derek @derekdevore]

@ADsXe I thought Red Lobster was straight-up luxury.

[Derek @derekdevore]

@OptionEdgeHQ Nice work

[Derek @derekdevore]

A quant interview dice problem is really an exit rule.

Stop when your roll beats the expected value of rolling again.
1 roll: worth 3.5. 2 rolls: stop on 4+, worth 4.25.

Holding a winner? Ask what continuing is worth.

via Amjad Saidam on Medium

https://t.co/XcioFV7sn8

[Derek @derekdevore]

A trade that followed every rule and lost money is a good trade.

A trade that broke every rule and made money is a bill that hasn't arrived yet.

Grade the process, not the P&L.

Over 200 trades, the process becomes the P&L.

[Derek @derekdevore]

4. Do the cushion math.

Per 100 trades at 95%: +$38 in credits, -$23 in losses. Net +$15 per share.

Each extra loser costs ~$5.

Three more bad days per 100 trades and the edge is gone.

High win rate. Thin margin. Test the tail first.

Not financial advice.

[Derek @derekdevore]

2. Count the gap days in your sample.

1DTE means you hold overnight. You're exposed to the open, not just the close.

List the biggest overnight down gaps in the test window. How many had a trade on?

If it's a handful, the tail is barely sampled.

[Derek @derekdevore]

3. Inject the shock.

Add a synthetic gap through your short strike at the open, say once a year.

Fill it at max loss, not the mid. Spreads blow out on those mornings.

If one bad open erases a year of credits, the edge lived in the sample, not the strategy.

[Derek @derekdevore]

1. Split win rate from payoff.

Say you sell a 5-wide put spread for $0.40.
Max loss: $4.60.

Breakeven win rate = 4.60 / 5.00 = 92%.

So a 95% win rate isn't a big edge. It's a 3-point cushion.