how do you know if a strategy is actually viable?
how do you know what term you want to sell those options in typically?
first note - this post is geared towards more "serious" options traders. clarifying that for those hoping for a quick win or easy apply so you don't waste your time here.
**the mission for this post is to provide a framework on how to use AI to literally turn you into a one man research machine with ZERO human coding.**
*why? because this kind of thing literally has life changing potential for people.*
a huge part of my trading performance comes down to research and data analysis. i have spent over six figures buying high fidelity options data from CBOE - that's no shit. i also needed to teach myself how to code with no technical background.
for context because it normally comes up, ive maintained just over a 30% CAGR from 2007 into 2025. it's good but nothing record breaking, which is really by design. i value consistency heavily.
for my dabase and research work, it was massively inefficient, painful, and slow. regardless, it was worth it's weight in gold. i was genuinely able to move from taking best guesses to conducting thorough research to actually quantify and qualify things.
my strategy development process has 4 major steps:
1. profit mechanism (PM) research. this is the underlying market effect were attempting to monetize. most options traders start with focusing on structure: they learn a vertical spread and go try to trade it. this is wrong. there is no edge in a vertical (or any structure). edge lies in the profit mechanism and your ability to exploit it. an example of a profit mechanism is variance risk premia through earnings releases. in aggregate, options markets price rich volatility relative to whats realized in earnings - this can be monetized. we need to actually measure this.
\- how much?
\- how often?
\- does it present in specific names vs others?
\- etc.
2. signal overlay. after the PM is understood, the next step is to search for signals that reliably help me measure it, track it, predict it, and monitor it. for the example above, things like how does the nearest term vol change coming into the ER. how does it trend relative to realized through the release, etc.
3. structure testing. once i understand what the PM looks like, how to track and predict it, i now need to see HOW to capture it. is a straddle best? strangle? iron condor? what term? what deltas? when should I exit it? etc.
4. strategy development. i take the outputs from the prior 3 and build the strategy outline to codify the details. this is also where i include notes on the PM traits so I can monitor them isolated to validate how effective the PM edge is and trends.
holy wall of text, whats the point?
first, for those that didn't have one before, you now have a framework to build strategies. more importantly and the cool part, AI has closed the gap on every testing element above.
while AI is cool for automating, i think people dont realize how valuable it is to conducting research. a shitty strategy automated is still a bad strategy.
on monday, i randomly decided to start a project (which im working on as i type hence the time to write the novel). i wanted to see what I could accomplish using AI to handle ALL coding, infrastructure building, etc. i am honestly blown away. this is genuinely accessible to everyone interested.
a quick overview of the process.
AI plan > infrastructure build (VS Code and extentions (python, jupyter, R, etc), mini conda, git bash, duckdb > API locate (orats, thetadata, massive) > environment build (establish database) > scope target ingest (what tickers, what fidelity, etc. this is fueled by what you want to explore) > AI writes code and executes. you monitor.
why a database? it allows for effective iteration without running into API overhead constraints.
as it sits, my AI database has over 600M rows with options data back to 2007 for all optionable underlyings that I trade, with stock data for a wider universe (smaller data footprint), with insider trading, sec filings, news, earnings, guidance. i have all datapoints from FRED (think economic reports, FOMC, CPI, PPI, etc), BLS, etc.
the data is expensive to get and build the foundational data for the DB but going forward, much of it can be gotten for FREE from your broker and free data sources. the APIs don't need to be maintained long-term, they're a startup cost.
all with zero code. all available to any trader that wants to build their own following the same model.
a few notes:
1. storage is important. im using RAID HDDs for cold storage and backup. I'm using pooled SSDs for warm storage and a solo SSD for hot storage. total size is 46TB. you don't need anywhere near this much btw. scope what you want with AI first to get a sense.
2. the process isn't excessively human intensive but does require oversight. if you have no coding background, a good practice is to take the code from one (claude or codex) and feed it to the other to review.
3. your computer is going to run for a while straight - longer depending on how robust the db is that you're building.
try it out - AI is making stuff like this accessible to everyone. you have nothing to lose and you'll learn a lot.