Costco (COST) does not cause cancer. Let's get that out of the way. But a method used in a set of Harvard cancer studies says it does. Deric Tilson and Adam Stein took that method and aimed it at the warehouse chain. Out came more than 2.2 million cancer deaths. The Journal's Notable & Quotable column on Thursday, Oct. 15 ran their findings.
Tilson and Stein write at the Ecomodernist, a Substack blog. Earlier this year they argued that studies tying nuclear power plants to higher cancer rates were nonsense. Then they spent months rebuilding the analysis. Their finding is simple. Pick any landmark you like and the method shows higher cancer risk around it. The studies, in their words, “simply do not prove anything about cancer incidence.”
The examples make the point. Living near a private four-year university shows a 15-fold increase in cancer mortality compared with living near a nuclear plant. Costco had the largest effect of anything they tested. The method pins more than 20% of all cancer deaths from 2000 to 2018 on the chain. Nobody believes that. That's the point. A tool that convicts everyone has told you nothing about anyone.
A method that can't say no
Here's the flaw they found. The original studies compared data points across circles 400 kilometers wide. That's about 250 miles across, wider than many states. The researchers found a link at nuclear plants and wrote it up. They never pointed the same tool at other landmarks to see if it lit up there too. So they had no way to know the result was coming from the method, not the plants. Tilson and Stein say they have yet to get a negative result from it. They think it may be unable to produce one.
Our read
Behavioral. We'd call this a hold, and the holding part is easy. Costco is held at weight in 20 of our books. Nothing was added this week, and this column isn't about the company anyway. It's about a habit of mind. A test only means something if it can fail. If a method finds what you're looking for no matter where you point it, it's finding itself. Wall Street runs on this error. A strategy gets tested on the past, it works, and nobody checks whether it would have worked on random noise too.
Here's where this touches your own money. Somewhere in your files is a chart that sold you something. It might be the backtest in a fund brochure, the screen that powers a stock newsletter, or the performance line in an annuity's sales sheet. Pull one of them out this week. Then ask whether that method has ever said no. If every version of it beats the market, you're looking at Costco and cancer. The honest ones show the years they lost, and the ideas they turned down. You check the method on a calm week, with nothing riding on the answer. If you'd like a second pair of eyes on that chart, bring it to your next review.