UPDATED: “STATISTICALLY INTERESTING, BUT NOT STATISTICALLY SIGNIFICANT” — COINBASE'S OWN DESK TESTED REKTEMBER AND BURIED IT — THREE SITES STILL QUOTED THREE DIFFERENT AVERAGES THIS WEEK

Every August the same chart goes around. September is red, September is cursed, get out before it gets you. So we built the chart ourselves. One price series, thirteen Septembers, no rounding tricks. It comes out to a 2.97% average loss, which is close to the number everyone repeats. Then we went looking for someone to check it against, and it fell apart. Three sites credited the same data provider on three straight days and printed three different averages: 2.86%, 2.87% and 3.08% down. CoinDesk counted five green Septembers since 2013. We count six. Nobody will say which year they left out. A token presale used a fourth number, 2.2%, that matches nothing anybody else publishes, and used it to sell a coin. Move the window three years and the average September swings from down 9.3% to up 5.6%. Same asset, same month, same arithmetic. Coinbase's own research desk tested it last September and wrote that the pattern is “statistically interesting, but not statistically significant.” A year later the chart is still going around. Here is the part worth keeping. September is one day old. Nobody knows what it does, and the people telling you they do cannot agree on what it did. Every year somebody sells real coins to a story about a month, and the coins do not come back. An average you keep rewriting isn't an average.

UPDATE, 6:10 PM ET. Decrypt makes it four. Its explainer went up this afternoon, credits the same data provider, and prints a 2.97% average loss, which matches our number to the decimal and none of the other three. Its median, 2.44% down, matches nobody. It counts eight red Septembers out of thirteen, which agrees with CoinDesk and not with us. Four sites, one source, four averages, two counts. The one line worth keeping is from DYOR's Ben Kurland: “more myth than math.”