AI is now at a point where in many aspects it could be even smarter than me. Especially proving out market hunches.

So recently, I wanted to see where I could improve my options strategies in my book.

There are decades of academic papers on options trading. The hard part is figuring out which ideas hold up when you actually test them.

Ken Griffin described Citadel doing exactly this: reproduce the analysis in a finance paper, then check whether the finding holds on different data.

In his Goldman Sachs interview, he said AI had cut work that used to take weeks to a few hours.

The Transcript @TheTranscript_ Citadel Founder Ken Griffin at the Stanford Leadership Forum 2026 discussing how AI agents compress elite analytical work from months into hours: "It has been interesting to watch...work that we would usually do with people with master's and PhDs in finance over the course of 6:15 PM • May 16, 2026 108 Retweets 551 Likes Read 17 replies

​

That’s the kind of work I wanted to try with my own trading research.

Could I take ideas from these papers, test them against real options data, and find something useful?

My Methodology

I connected the research workflow to Unusual Whales data: about 6.6 million options alerts across roughly 730 trading sessions, from 2023 to 2026, alongside stock prices, option quotes, insider filings and earnings data.

I reviewed about 60 academic papers and set up about 30 strategy and comparison tests.

The starting question was specific: does this activity predict a stock move, or a profitable options trade? And what can I learn from these hypotheses?

Let's check out the findings below...

1. Which call option sweeps performed better?

Million-dollar call sweeps with three to five weeks left until expiration had stronger month-long stock returns than calls with one to three weeks left.

A sweep is an order filled across multiple exchanges. We filtered for calls costing at least $1 million, buying near the seller’s asking price, a strike above the current stock price, and unusually high trading volume. We excluded trades flagged as part of a larger options package.

For the group with three to five weeks left, the stocks averaged about 12% above SPY over a month in the earlier period and 7% in the later period. The shorter-expiry group averaged about 1% earlier and −2% later over the same holding period.

The later result for the longer-expiry group developed slowly: the stocks lagged the market by about 1% in the first week, then led by about 7% over a month.

That monthly average came from only about 30 later events across 20 stocks. Removing the three biggest winners reduced it to about 2%; removing five made it roughly flat.

2. Insider buying performed better when options priced in less movement

I looked at stocks where at least two officers or directors reported purchases of $25,000 or more within ten days. Each buyer had gone a year without another qualifying purchase in the available records.

Then I compared two groups: stocks whose options implied relatively little movement compared with their own recent history, and stocks whose options implied relatively large movement. This measure is called implied volatility, or IV.

The low-IV group averaged about 5% above SPY over a month, versus about 1% for the high-IV group.

3. Stocks whose options implied more movement than usual had a small return advantage

This test compared each stock’s current IV with its own recent average, then ranked the stocks by that difference.

Perhaps this is because high implied vol means the move may be priced in?

Over the following week, the highest-ranked group outperformed the lowest-ranked group on average in both periods. We measured which stocks performed better after the signal.

The comparison bought the highest fifth of stocks and sold short the lowest fifth. Its average weekly advantage was less than 1%. Restricting the later test to stocks that were easier to trade, then correcting the entry timing, cut that advantage by about 70%.

4. Big put sweeps were weak bearish signals and may include hedging

Stocks after large put sweeps averaged about 1% above SPY over the next week in both periods. The stock result went in the opposite direction from a simple bearish reading of the alert.

So takeaway here may be that puts are less signal than calls.

A put buyer may be protecting shares they already own. The alert shows the options trade, but it doesn’t show the rest of the buyer’s portfolio. Hedging is a plausible explanation for the weak bearish signal; this study didn’t measure how many buyers were hedging.

5. Heavy buying in both calls and puts didn’t make buying both profitable

I tested buying a call and a put after heavy activity on both sides of the same stock. The pair can benefit from a large move in either direction, but the move still has to pay for both options.

Buying at the quoted asking price and selling at the quoted bid lost about 12% of the premium paid in the earlier period and 7% in the later period, on average.

Personal

Btw, my Wall Street Prompt cofounder Dave is speaking at a free webinar with AlphaSense this Thursday, October 15, at 12pm ET!

He’ll be covering:

• How 1,000+ decision-makers are using AI at their companies and funds
• How you can get more out of AI tools
• Stories from working with some of the top AI users in finance
• Live Q&A

Can’t join live? Sign up and you’ll get the recording in your inbox afterwards.