I got into insider-buying feeds because they looked like free information, and then spent a long time being disappointed by them. You pull every Form 4 with a purchase code, you rank by dollar size, and you sit on the biggest buys expecting them to lead somewhere. Most of the time they lead nowhere. The problem is not that insider buying is useless. It is that a raw feed treats a director rubber-stamping a routine grant the same as a CFO writing a personal check for stock a month before earnings, and those two events have almost nothing in common except the form number they show up on.
So the interesting question is not whether insiders predict returns. It is which insiders, in which context, and how much to weight each one. That is where the research and the practical work actually line up.
The role hierarchy of who knows what
The intuition is boring but it holds up. The people closest to the numbers tend to trade on the numbers. A CFO sees revenue recognition, margin pressure, and the shape of the quarter before almost anyone. A CEO sees the whole picture but is also the most watched, the most constrained by trading windows, and the most likely to be transacting for reasons that have nothing to do with a view on the stock. A director sits on the board, sees things at the altitude of a quarterly meeting, and often trades because the compensation plan handed them shares, not because they formed an opinion.
Academic work on this has been fairly consistent for a long time. Officer purchases, and CFO purchases in particular, tend to show more predictive power for future returns than director purchases do. The usual read is that the CFO has the sharpest view of near-term fundamentals and the least excuse for buying at a bad price. When a finance chief puts personal money into the stock, it is hard to explain that away as diversification or a routine plan. They know exactly what the next print looks like, and they bought anyway.
Directors are the noisiest end. Some directors are genuinely sharp signals, especially large outside shareholders who join a board and keep adding. But a lot of director activity is plumbing. Retainers paid in stock, deferred-comp elections, small routine top-ups. If you weight a director buy the same as a CFO buy, you are letting the plumbing drown out the signal.
Routine versus opportunistic is the split that matters
The role hierarchy gets you part of the way, but the bigger lever is separating routine insiders from opportunistic ones. This distinction, which comes out of a well-known line of research, ended up being more useful to me than the title on the form.
The idea is simple. Look back at an insider's own history. Some people trade on a schedule, roughly the same month every year, in a way that looks mechanical. Those are routine traders, and their trades carry very little information because the timing was never a choice. Other insiders trade at irregular times, clustered around no obvious calendar event, and those opportunistic trades are where the predictive content lives. The same CFO can be a routine seller through a plan and an opportunistic buyer with a checkbook, and only the second one should move your model.
The cleanest tell for a routine trade sits right on the form. A purchase or sale executed under a pre-arranged plan is supposed to be flagged, and there is a checkbox on Form 4 for exactly that. A trade made under a plan set up months earlier tells you nothing about what the insider thinks today, because today's insider did not decide to make it. So the first thing any weighting model should do is heavily discount anything flagged as planned, and lean toward discretionary buys made in open windows.
Building a role-weighted score
Here is roughly how I think about turning a Form 4 stream into something rankable. None of this needs to be fancy. It needs to encode the two facts above, that role matters and that routineness matters, and then get out of the way.
- Start from open-market purchases only. Filter to the buy transaction code that means an open-market or private purchase. Ignore option exercises, grants, gifts, and anything that is compensation machinery rather than a decision to own more stock.
- Assign a base weight by role. Give CFO and other principal financial officers the top weight, other named officers just below, the CEO in the upper-middle, and directors the lowest base. The CEO sits below the CFO here on purpose, since CEO buying is more often symbolic or window-constrained.
- Apply a routineness multiplier. Pull each insider's trade history and cut the weight hard for anyone whose buys land on a predictable calendar. Discount planned trades toward zero. Give a modest boost to a first-ever open-market buy by that person, which historically tends to carry more signal than a repeat.
- Scale by conviction, not just dollars. Size relative to the insider's own holdings or their known compensation says more than raw dollar amount. A director spending a large multiple of their annual retainer is louder than a billionaire founder adding a rounding error.
- Reward clustering. Several distinct insiders buying inside a short window is a stronger signal than one person buying a lot. Cluster buys across multiple officers are the closest thing to a real tell that a feed will give you.
Combine those into a single score per filing, then aggregate to a per-company signal over a trailing window. The output is not a price target. It is a ranking that puts a clustered set of opportunistic officer buys near the top and a lone planned director top-up near the bottom, which is exactly the ordering a raw dollar sort gets backwards.
The failure modes that will bite you
The most common way this goes wrong is trusting the size field. A giant dollar buy from a founder who already owns a huge slug of the company is often meaningless, because it barely moves their exposure and they may be doing it for optics. Normalize against existing holdings and a lot of the headline-grabbing buys shrink back to nothing.
The second trap is treating sells as the mirror image of buys. They are not. Insiders sell for a hundred boring reasons, taxes, a house, diversification, plan schedules, so sell signals are far weaker and far noisier than buy signals. If you build a symmetric long-short off this, the short side will mostly feed you garbage. Buys are the informative half.
Third, watch the reporting delay. Form 4 has to be filed within a couple of business days of the trade, which is fast, but the trade itself may have happened before that, and the market often moves on the news the moment it hits. You are not front-running the insider. You are trying to identify which insiders, weighted correctly, tend to be early enough and right enough that following the informative subset still pays over a horizon of weeks to months.
If you only take one rule from this, make it the routineness filter. Splitting insiders into routine and opportunistic, and downweighting anything flagged as planned, does more to clean up an insider feed than any amount of clever role scoring on its own. The role weighting is the polish. Throwing out the plumbing is the part that actually moves your hit rate.