Insider Trading with AI

By Dave Wang

Under the right conditions, insider buys by the executive team is one of the strongest bull signals for stocks.

Specifically, when there is (1) Trades in massive size and (2) A cluster of buys from multiple executives (3) Poor market sentiment .... there could be some information asymmetry that the public doesn't know.

A great example of this is Jamie Dimon where he's bought his own JP Morgan stock 3 times and called the bottom on each....including a $26M purchase at the bottom in 2016

This last week, I noticed UnitedHealth Group checked the boxes of the 3 conditions I like to see for "bullish" signals of executives buying their stock. We saw (1) Huge volume in purchases (2) Multiple insiders buying in clusters (3) Fear around $UNH stock from the reports of the Department of Justice investigating them, among other factors

>$30M in stock purchased by UNH insiders

I used AI as my analyst to investigate the below list of questions I had to determine whether these Form 4 purchases are signal or noise:

  1. Explanation of potential reasons around this purchase - is this an event driven trade where the "insiders" are betting the WSJ reports are nothingburgers? Or is this just a signaling function?
  2. Deep research into the context of the FUD around UNH and if it has merit or exaggerated reaction
  3. Transaction type: Open-market buy, option exercise with “hold,” Rule 10b5-1 plan, or indirect trust purchase
  4. Size context: Dollars bought relative to the annual comp, existing ownership, and daily trading volume
  5. Timing window: Trade date versus Form 4 filing date, distance from earnings blackout, and proximity to material announcements
  6. Insider cluster effect: does this have statistical merit?
  7. Track record weight: Historical hit-rate of this CEO’s past buys versus random timing. Is Hemsley’s historical timing skill above chance?
  8. Compensation alignment: Incentive-comp targets due for reset? A purchase just before targets are set can weaken the inference.​​

Here's our gameplan:

  1. Craft an extremely detailed prompt
  2. Paste this prompt into ChatGPT Deep Research

The Prompt:

(ChatGPT Deep Research)

SYSTEM PROMPT
“You are an elite buy-side analyst trained to produce exhaustive, evidence-backed reports. You think like a 160-IQ hedge-fund PM, obsess over data integrity, and write in a clear, first-person style. Use concise sentences, Markdown headings, bullet lists, and numbered steps. No em dashes. Use emojis sparingly for emphasis only.”
USER PROMPT
I need a deep-dive on the recent May 2025 insider-buying cluster at UnitedHealth Group ($UNH). Deliver a long-form report (2000 - 3000 words) with the sections below.
1 Define the Signal Precisely
- Transaction type: Split each Form 4 into open-market buy, option exercise with “hold,” Rule 10b5-1 plan, indirect trust, or other.
- Size context: Dollars spent versus the insider’s latest annual comp, existing ownership, and average daily dollar volume.
- Timing window: Trade date vs Form 4 file date, days since last earnings blackout, and distance to any known material announcements.
- Cluster intensity: Identify all UNH insider trades ±10 days around 16 May 2025. Flag a “strong cluster” when >3 insiders each spend >$1 m.
2 Hypothesis Framing Explain whether this looks like
1. Event-driven conviction (insiders betting recent WSJ/DOJ headlines are noise), or
2. Pure signaling to calm the market.
Lay out potential motives for each camp and what future datapoints would validate or refute them.
3 Contextual FUD Analysis
- List every negative catalyst in play (Medicare Advantage margin fears, DOJ probe, guidance withdrawal, cost-trend spike).
- For each, assess if sell-side and media coverage is substantive or exaggerated.
- Pull historical analogs where managed-care names dropped >20 % on policy scares then recovered. Summarize win-rate and median rebound.
4 Statistical Back-Tests
1. Insider cluster efficacy:
- Universe: All S&P 500 companies 2002-present.
- Definition: >3 insiders spending >$1 m each within 10 days.
- Regression: 6-month alpha ~ cluster_dummy + prior 1-month drawdown + valuation_quartile.

2. CEO hit-rate:
- Pull every Stephen Hemsley open-market buy 2006-2024.
- Bucket by forward 180-day returns on those purchases
5 Compensation Alignment Check
- Detail UNH incentive-comp reset cadence.
- Highlight whether the May 2025 purchases line up just before a new target year or equity grant.
- Explain how that timing strengthens or dilutes the bullish inference.
6 Risk-Reward Blueprint
- State conditions to enter, add, trim, or exit a UNH long driven by this insider signal.
- Suggest sizing rules tied to forecast idiosyncratic alpha and drawdown limits.
- Include catalyst calendar: Q2 earnings (14 Jul 2025), CMS rate notice draft (Sep 2025), any DOJ update.
7 Key Takeaways
Bullet the final verdict: strength of signal, probabilities, and implementation notes.
Data Requirements
- SEC Form 4 database (2000-present).
- CRSP daily prices and volume, Fama-French factors.
- 13F, Refinitiv ownership, CMS announcements, WSJ article archive.
- Option Metrics or equivalent for skew and OI.
Deliver only the finished report. Cite sources inline in parentheses, e.g. “(Form 4, 16 May 2025).”
No appendices unless critical. End with a one-sentence actionable bottom line.

The Result

https://chatgpt.com/share/68327481-45bc-8001-ba05-63f8b1b41476

The above has the full report on whether the $UNH dip is a buying opportunity based on the insider trading activity last week by executives.

The bottom line conclusion from this report is "The May 2025 insider buying cluster at UNH stands out as a textbook bullish signal in the face of extreme pessimism. Insiders acted aggressively because they likely see UNH’s long-term value intact and the current issues as surmountable. Our analysis concurs that the sell-off was likely exaggerated – a confluence of temporary hits rather than permanent impairment."

A few points I found quite interesting:

  1. The new CEO put 6% of his entire net worth into $UNH despite existing high personal concentration to $UNH
  2. Deep Research found market analogies for this exact situation: (1) WellCare Group dropped 60% in 2007 when the FBI raided their headquarters in 2007, but it did a 10x from those lows in the next decade (2) UNH fell 30% in 2006 from a separate scandal but was back to previous highs in one year
  3. A few academic studies cited show insider cluster buys are a very positive signal

New Posts!

Here are some AI investing resources I put out last week:

If you made it this far, forward this to a friend who might find this helpful

Get one high impact AI prompt every week

Join thousands of smart investors.

2026 — Built by Dave Wang. Not financial advice, only for educational purposes.