Methodology

Following insider buying by hand is mostly elimination. You pull a screen, open each company, look at whether the buying is unusual for that company, work out who bought and whether it means anything, discard nine of ten, and repeat next week. The reading is the easy part; the collecting and discarding is what takes the hours.

This does that pass for you. It reads every Form 3, 4 and 5 as the SEC publishes them, applies the eliminations below, and presents what survives with the figures you would otherwise have assembled by hand — how unusual the buying is for that company, how much it moved each buyer’s own position, how long they have been an insider, and what their previous purchases did.

SEC Forms 3, 4 and 5 — the filings insiders are legally required to submit within two business days of trading. History comes from the SEC’s DERA bulk datasets; everything since is fetched from EDGAR directly and polled every ten minutes, so a filing appears here minutes after the SEC publishes it. Prices come from public market data, adjusted for splits and dividends.

Reading insider filings and acting on them is entirely legal. The filings exist precisely so that anyone can.

Selling is not ingested at all, and that is a deliberate asymmetry rather than an omission. Insiders are paid largely in stock, so selling is how they turn compensation into money — a house, a tax bill, tuition, or simply not holding their entire net worth in one company. There are many reasons to sell and they mostly say nothing about the business.

Buying is the opposite. An insider who buys on the open market is choosing to increase exposure to a company they already depend on for their salary, with their own money, when they could have bought anything else. Collecting the sell side would produce a constant stream of red flags that mean nothing, and the noise would bury the buying.

Most of the work is here rather than in the ranking. Raw insider data is dominated by activity that looks like buying and is not.

• Fund activity

  Anyone holding 10% or more files as an insider, so index funds and asset managers appear alongside the CFO. They usually take a board seat too, which means the filing says Director and the relationship flag cannot be trusted — the filer's name has to decide it. A fund adding to a position is following a mandate, not acting on a view.

• Routine buyers

  Companies whose insiders buy in nine or more of the last 24 months. Somebody who buys every other month tells you nothing by buying again; the information is in the departure from habit, which is why the whole score is built around a company's own baseline rather than around dollar amounts.

• Option exercises sold the same day

  A code M acquisition paired with a code S disposal is a cash-out, not a purchase. Aggregators that read the acquisition leg on its own report these as insider buying, which is how a routine compensation event becomes a false signal.

• Rule 10b5-1 plan trades

  Scheduled months in advance, so they carry no information about today — the opposite of the opportunistic buying worth looking for. Form 4 has flagged these with a checkbox since April 2023, so they can now be excluded outright rather than guessed at.

• 401(k)-sized purchases

  Below $20,000, a purchase is usually an automatic payroll contribution or a discounted-stock match rather than a decision. An executive acting on conviction puts in a meaningful fraction of their pay, not a rounding error.

• Amended filings

  A Form 4/A restates an earlier Form 4 and both documents remain public. Counting both double-counts the transaction, inflating both the cluster and the buyer's apparent conviction. Not a judgement call — a correctness fix.

• Mega caps

  Above $100bn in market value, a $500,000 purchase says nothing about a company that size, and insiders under that much scrutiny do not step in front of news. Flagged rather than removed: it is a judgement you may not share.

• All buyers newly appointed

  An insider of under a year buying for the first time is usually establishing an expected holding or signalling confidence on arrival, not reacting to something they have learned. Also flagged rather than removed.

These operate at three different levels, and the distinction matters when reading a result:

• Buyers removed from the cluster: funds, purchases under $20,000, and same-day exercise-and-sell. These do not count toward the buyer count or the total.
• Clusters hidden from default results: no qualifying buyers left after the above, every buyer a fund, every trade a 10b5-1 plan trade, or the largest single buyer under $20,000.
• Flags that inform rather than exclude: routine buying, mega cap, and all-buyers-newly-appointed. These stay fully visible with the flag beside them.

Excluded buyers are still listed on the company page with the reason next to them. The filtering is visible, never silent — if you disagree with a call, you can see that it was made.

Everything below is a figure you would otherwise assemble by hand, one company at a time.

• How unusual the buying is for this company

  Cluster volume against the company's own prior buying, institutions excluded from both sides. This is the question the eye is doing when it looks at an insider-activity chart for a flat stretch followed by a spike — done arithmetically, so a quiet company that suddenly sees buying stands out from one where somebody always buys.

• Conviction, as a share of what they already held

  A $2m purchase by someone already holding 10.7m shares is a 0.5% addition and means little; a $158k purchase taking a holding from 35,000 to 40,000 shares is a real decision. Raw dollars flatter the wealthy and hide the meaningful. A position opened from nothing has no percentage to report and is treated as its own case — it is the strongest version of the pattern, not a missing number.

• How long they have been an insider

  A Form 3 is the initial statement of beneficial ownership, filed within ten days of becoming an insider — so its date is the start of the relationship, straight from the filer. This replaces searching for someone's name and their company and hoping an appointment press release turns up. A long-tenured insider buying for the first time on record is a different event from a new appointee's first purchase.

• What their previous purchases did

  Each past purchase is measured over a fixed six-month window from the trade, compared against the market over the same window. Fixed, because measuring every old buy to today's close makes anyone who bought years ago in a rising market look prescient and anyone who bought recently look useless.

• Which roles bought

  A general counsel is risk-averse and paid less than the C-suite; if they cannot resist buying, that is worth noticing. A CEO is paid largely in stock and part of their job is supporting the share price, so their purchases carry the most theatre and the least information.

• Whether the move has already happened

  The price change since the cluster, shown next to it. A cluster that filed five weeks ago on a stock that has already run is history rather than an opportunity, and this is the fastest way to see that.

The score combines these into a single ordering so the list can be worked from the top. Departure from the company’s own baseline and conviction relative to the existing position carry the most weight; the number of buyers carries less than its prominence suggests. Every input is shown next to the score, so the ranking never has to be taken on trust.

Nothing. An insider may only buy while not holding material non-public information, so the stretch in which they are most confident about the business is often exactly the stretch in which they are barred from trading on it. A company that shows no cluster here has not failed a test — it has not taken one.

Weeks in which nothing survives the eliminations are normal and expected. The point of running the scan often is catching a cluster within days of the filing, not producing a candidate every week.

• It does not size a purchase against the buyer’s wealth. $1m from someone earning $1m a year is conviction; $1m from someone worth $300m is pocket change. That needs compensation data from proxy statements, which is not collected here — so it stays your judgement.
• It does not read the business. Nothing here looks at valuation, debt, competition or what the company actually does. Unusual insider buying is a reason to open the file, not a substitute for reading it.
• It cannot beat the filing deadline. Insiders have two business days to report. Over a weekend that is five calendar days, and nothing on this side shortens it.
• Joint filings are attributed to a single primary owner, so a filing made by several people counts as one buyer.
• Companies without usable price history — delistings, symbols the price sources do not carry — keep their filing data but lose the price-derived figures, which are then shown as unavailable rather than as zero.

Section 16(b) requires insiders to surrender any profit on stock bought and sold within six months, so the people generating this signal are locked in for at least that long. What you are reading is a decision made on a six-month-plus horizon. Trading it over a few weeks is a different bet from the one the insider made.

The second tool has nothing to do with the first. It covers every US-listed operating company — around 5,600 of them, funds and ETFs excluded — whether or not an insider has ever touched it, because most companies worth researching have no insider buying at all.

It answers the question that starts any research: is this a good business, and is the price reasonable. It does not answer whether to buy. What it saves is the hour of pulling figures together before you can even form an opinion.

The method is borrowed and credited, not invented here. It comes from a published body of work by the Investor52 channel, and it frames a company as three separate questions:

• Growth — sales up at least 50% over five years, expected earnings growth above 10% a year, and no fall in revenue in the most recent year.
• Financial strength — debt below equity, positive cash flow, and a return on equity above 15%.
• Valuation — a PEG under 2, which is fair or better.

Each is three-valued. A company whose figures are not all reported comes out unknown rather than passing, because a missing number is not a pass. Passing all three is the sweet spot; passing two is a state the method treats as worth attention on its own, and it is counted separately rather than hidden.

A price-to-earnings ratio on its own says very little. A low one is as often a weak business as a cheap one, and screening on it tends to collect companies that are cheap for good reason. So P/E is shown as context and never used as a signal by itself.

The ratio that decides is PEG: the P/E divided by the earnings growth analysts expect over the next five years. Under 1 the price asks for less growth than is already expected. Between 1 and 2 is fair, 2 to 3 expensive, above 3 the business has to far exceed expectations for the price to make sense.

It is computed here rather than taken from the data source. The source publishes a PEG built on the forward P/E; the method defines it on the trailing P/E, and the two disagree for nearly half the companies the source calls undervalued. Both are shown side by side so the gap is visible rather than hidden.

Because analyst estimates miss low about 40% of the time, the same ratio is also computed on the median expected growth for the company’s industry. Where a company’s own estimate is depressed, that second figure is often the more useful read, and both appear on the company page.

Roughly half the market is left explicitly unclassified, and that is stated rather than quietly hidden:

• No earnings — no positive P/E, so no PEG. These are ranked instead on where their price-to-sales sits among their industry peers, which is what the method reaches for when a company is still loss-making.
• No estimate — earnings exist, but nobody publishes a five-year growth forecast. PEG needs an analyst, and much of the market below the large caps has none.
• Shrinking — an estimate exists and is negative. That forms no ratio either, but it is a finding rather than a gap.

All of them still have a page, with whatever is known about them, and all of them are reachable from the All tab. None are dropped.

The thresholds come from a source that screens large caps and the S&P 500. The insider screener deliberately covers everything below $100B, where these numbers are not validated. They are published here as the source’s, not as ours.

Earnings are as reported. The author of the method strips one-off tax benefits by hand before computing anything — a single quarter’s tax credit can make a P/E look half what it is — and no pipeline can do that reliably. A ratio flattered by a one-off will read as cheap here.

Basic materials, energy, utilities, real estate and financial services are flagged as hard to read, because their value turns on the oil price, what a mine still holds, or an interest rate, none of which a generic ratio can see. They are still scored and still listed, marked rather than hidden.

Estimates are estimates. A PEG under 1 is a reason to look, not a reason to buy.

Company fundamentals are collected once a night. Annual revenue history comes from SEC XBRL filings, which is what makes it possible to tell steady growth from one good year — a single annualised rate cannot. A day of data is only published once its row count and the completeness of every column look plausible, so a partial or broken collection leaves the previous day in place rather than half-replacing it. Each company page shows the date its figures were taken.