Most signals in equity markets are noise. Earnings surprises mean revert. Analyst upgrades chase price. Even free cash flow yields can be optical illusions in capital-light businesses. But one signal has stood up to forty years of academic scrutiny and millions of dollars of hedge-fund infrastructure built to exploit it: open-market buying by a company's own CEO or CFO. When the person who knows the company best — the one with the daily P&L, the customer call summaries, the pipeline meetings — writes a personal check to add to their already enormous, undiversified, and often illiquid position, they are telling you something the market will only figure out later.
The seminal academic work is [Seyhun (1986, Journal of Financial Economics)](https://www.sciencedirect.com/science/article/abs/pii/0304405X86900609), which found that insiders, on average, beat the market by roughly 5% in the year following their open-market purchases. [Lakonishok and Lee (2001)](https://www.nber.org/papers/w6656) updated and expanded this with a much larger dataset and reached the same conclusion. More recently, a long-running study by Jefferies' insider strategy team showed cluster buys (defined as three or more insiders buying within a short window) outperformed the S&P 500 by an annualized 9.4% from 2003 through 2020. The signal has not been arbitraged away because it depends on a behavior that is psychologically and economically costly: executives are already over-exposed to their own employer through restricted stock, options, and human capital. Voluntarily concentrating further is the opposite of what any rational portfolio manager would advise.
Why does it work? Because executives have an information edge that no analyst model can match. They know which large customer is about to sign a renewal, whether the new product launch is hitting internal milestones, how the quarter is shaping up two weeks before close, and whether the cost-cutting initiative has actually moved gross margin. None of this information is illegal to act on as long as the trade is not based on a specific, material, non-public event (an unannounced acquisition, an earnings miss they know about, etc.). The Form 4 filing requirement, which forces insiders to disclose trades within two business days, exists precisely to make this asymmetry visible to the rest of the market.
Not all insider buying is equal. The single most important filter is title. Independent directors often buy as a governance gesture, especially at small caps; their information edge is real but smaller than the CEO's or CFO's. The CEO, with daily P&L visibility, and the CFO, who sees revenue and margin in real time, are the highest-signal filers. Beneficial-owner filings (10%+ holders) are useful but contaminated by activist funds and private equity who may be buying for control rather than fundamental conviction. The dataset we screen against — Form 4 transaction code P (open-market purchase) filtered to CEO and CFO — is the cleanest version of the signal.
Cluster buying is the second filter. When three or more insiders buy within a 30-day window, the signal historically outperforms single-insider buys by a wide margin. The logic is intuitive: it is hard for three independent people to all be wrong about the same company at the same time. Cluster buys at the end of a long downtrend, especially after a sentiment-crushing earnings miss, are the most valuable variant. They imply the insiders believe the market is mispricing a temporary problem.
Size matters less than people think, but direction matters enormously. A $50,000 buy by a CEO already worth $40 million looks small in dollar terms, but it is a deliberate choice to add to an already concentrated position. The same CEO selling $50 million of shares may simply be diversifying or paying tax bills on vested RSUs. This is why most serious insider trackers ignore sells almost entirely. The classic [Lakonishok-Lee paper](https://www.nber.org/papers/w6656) found that insider sales had essentially no predictive value once you controlled for tax planning and option exercises.
Open-market buys versus option exercises is another critical distinction. An insider exercising a stock option and immediately selling is a planned compensation event with no informational content. An insider writing a personal check at the current market price is putting their own after-tax cash at risk. Only the latter is in our dataset. The [SEC's Form 4 transaction code system](https://www.sec.gov/about/forms-3-4-5) makes this filterable: code P (purchase) is the open-market buy, code M (option exercise) is a compensation event, code S (sale) is mostly noise.
Price context matters too. Insider buys at 52-week lows have historically outperformed buys at 52-week highs by a meaningful margin. This is not surprising: it is easier for an insider to have an informational edge when sentiment is poor and the analyst community has thrown in the towel. The classic setup is a name that has been left for dead, earnings have just disappointed, and the CEO files a Form 4 buying $250,000 of stock the following week. That is the signal in its purest form.
Sector matters. Insider buying signals have historically been strongest in small and mid caps, where analyst coverage is sparse and the informational asymmetry is largest. Mega caps generate signals too, but they tend to be smaller in magnitude. In financials, particularly community banks, insider buying has been one of the most reliable predictors of strong forward returns, partly because bank executives have unusually clear visibility into loan-book quality and net interest margin.
What this signal does not do is time the bottom. Insiders are usually early. They buy when valuations look reasonable to them, not when they are at the absolute bottom. The Lakonishok-Lee study found that the average insider buy underperformed for the first one to three months and then materially outperformed over the following six to eighteen months. If you treat insider buying as a 12-month investment thesis rather than a swing trade, the math is dramatically better.
Common failure modes are worth naming. Insiders can be wrong about their own company, particularly about industry-wide demand shocks. The 2008 crisis is full of bank executives who bought their own stock at $30 and watched it go to $3. The COVID crash trapped a lot of insider buyers in early March 2020 before the V-shaped recovery saved them. The signal is not magic; it is a probabilistic edge. Position sizing and diversification across multiple insider buys remain essential.
The right way to use this signal is as a top-of-funnel idea generator. Treat every fresh cluster buy as a research prompt: pull the 10-K, read the most recent earnings call, look at the balance sheet, and figure out what the insiders see. If the answer is a credible thesis backed by the financials, that is when the signal becomes actionable. If the answer is a falling-knife business with no path to recovery, the insider may simply be wrong and the position should be passed on.
One practical note on data: the [SEC's EDGAR system](https://www.sec.gov/edgar) publishes Form 4 filings in close to real time. Our screener pulls these filings, filters to CEO and CFO open-market purchases in the last 90 days, and ranks by total dollar value of the buy. The total-dollar lens biases toward conviction, but you should always check whether the buy is concentrated in one filer (less reliable) or distributed across multiple insiders (more reliable). The accompanying Form 4 URL takes you directly to the filing so you can verify the title, transaction code, and price for yourself.
Insider buying is one of the very few signals that has survived four decades of academic testing and the rise of quant funds. It works because it is grounded in human behavior — executives have to overcome real psychological friction to add to an already concentrated position — and that behavior cannot be arbitraged away by a faster algorithm. Treat it as the starting point of a research process, not the ending point, and it will steer you toward more good ideas than almost any other screen.