The historical cohort says do not overread the print
Arizona Sonoran Copper Company Inc. insider-trading story">
Our cohort data for the bucket Directeur · Mid is the best reality check in the file. The sample size is 35,794, which is large enough to keep the statistic from being a toy. Over 90 days, the win rate is 44.8%, and the average return is -0.81%. Over 365 days, the average return is 9.77%. That is the right way to read insider filings if you are serious about them. The short-horizon cohort is not especially strong, and in this bucket it is actually negative on average. The longer window is better, but still not a promise and not a forecast. It is historical cohort data for a role and size bucket, not a prediction for Arizona Sonoran or for this trade.
That caveat matters more here than usual because the trade sits inside a deal process. The market is not simply digesting an operating company’s fundamentals. It is digesting transaction terms, approval odds, and timing. A director buy in that setting may reflect confidence in closing mechanics, alignment with the process, or simply a willingness to add exposure at a moment when the stock has already been marked up. The cohort data does not tell you which of those motives is dominant. It only tells you that, in the broader Directeur · Mid bucket, the 90 day edge has been weak. So if you are tempted to treat the filing as a clean bullish edge, stop there. The historical record does not support that kind of certainty.
The cluster is the real tell, not the lone trade
InsiderTrades data flags this as a cluster, and the dossier says there were four distinct insiders involved, with 12 recent declarations in the cluster window. The recent list shows Laing buying on June 25 and Nick Nikolakakis buying on June 24, with multiple declarations for each. That is the kind of pattern that matters more than any single line item. One director buy can be idiosyncratic. A cluster says the behavior is broader. It suggests the board or insider group is not acting in isolation.
The cluster also helps explain why the score sits where it does. Our scoring is not simply rewarding size. It is rewarding size plus role plus clustering. A director buying about EUR 2.29 million is one thing. A director buying that amount while another director is also buying in the same window is another. The second buy at about EUR 1.26 million reinforces the same point. This is not a one-off gesture. It is coordinated enough in timing to matter. Still, you should not overstate the breadth of the cluster. The dossier gives four distinct insiders and 12 recent declarations, but it does not give the full universe of who was active, what each exact allocation was, or whether the buying was balanced across the group. The pattern is real. The granularity is limited.
What the strategy context says, and what it does not
The dossier’s strategy context is useful, but only if you read it as a screen, not a promise. It says the strategy holds for 90 days, caps position size at 0.08%, and has an out of sample Sharpe of 0.56 with a CAGR of 17% in a restricted EU venue universe. It also says the universe win rate is 51.5%. That is decent, and better than random, but the caveat is baked in. These results survive only on a restricted EU venue universe, they do not survive search aware deflation, and the window is short and single regime. That means you do not get to lift the numbers out of context and call them a permanent edge.
Still, the strategy context helps frame why a filing like this gets attention. A 90 day hold is long enough to capture post filing drift without pretending you know the terminal outcome of a deal. A small max position size keeps the process from becoming a hero trade. And the universe win rate above 50% says the screen has some discipline. If you are using the strategy as a filter, this name fits the kind of event it was built for: a live insider cluster in a liquid enough, transaction driven situation where the filing itself is the main object. But again, that is a process note, not a guarantee of return.