The Insider Buy Signal Everyone Gets Wrong
Five years of Form 4 data shows what actually works and what does not : Rhodie House Options Intelligence | June 2026
Every week on X, someone posts a screenshot of an SEC Form 4 filing like they found the market’s instruction manual. A CEO buys $3 million of his own stock. The post gets 400 likes. The comments fill with “smart money alert” and “following the insiders.” I post the same alerts, I track insider buys and sells in my Trader Tool Kit Dashboard and I even tag my own curated option flows with insider buys. That’s fine as far as it goes because it is informational and the more information you have the more likely you are to make well thought out decisions when trading. The bit you are missing is what I have set out below.
There is a problem. I tested it.
HOW I RAN THE DATA
I pulled five years of insider purchase data from a public SEC filing database, 2021 through mid-2026. Officers and directors only. No pure 10%-shareholder plays. Minimum $50,000 per transaction. Two separate studies.
Study 1: tested the most popular insider thesis on financial social media: CEO open-market purchases in the 12 to 30 day window before the next earnings report. Entry at the next trading day’s open after the filing date, forward returns measured at 30, 60, and 90 calendar days, SPY as benchmark.
Study 2: looked at something different: clusters of three or more distinct officers or directors at the same company buying stock within a 30-calendar-day window. Non-overlapping windows so the same buying event is not counted twice. Entry was the next trading day’s open after the last cluster buy.
Two limitations to state upfront. First, survivorship bias: price history is only available for stocks that are still trading today. Any company that clustered, then went bankrupt or was delisted, is absent from the results. That removes the worst-case outcomes from both studies. The returns shown are cleaner than what a real-money portfolio would have experienced. Second, the entry assumption is slightly optimistic: the back test uses the transaction date from SEC filings, but in practice you would not see a Form 4 until it is filed, which can be up to two business days after the actual trade. Real-world entries would be a day or two later in most cases. Neither of these invalidates the findings, but they mean the numbers should be read as directional, not precise.
A note on subgroup breakdowns: I sliced the data many ways. The overall n=163 supports the headline conclusion. Individual subgroups with fewer than 20 observations should be treated as illustrative, not statistically definitive.
