After a +15% day: 58% close red, yet the average is positiveEthereum / TetherUSBINANCE:ETHUSDTnonsanchartETH closed +17.5% yesterday, BTC +7.1%, and the fear and greed index flipped to greed for the first time since January. So the obvious question is what usually happens next. I pulled every daily candle across 8 majors (BTC, ETH, SOL, XRP, DOGE, BNB, ADA, LINK) that closed 15% or more above its open. 253 instances. Then I measured the following day. Here is the distribution of that next day: below -10%: 13.8% -10% to -5%: 16.5% -5% to -2%: 17.7% -2% to 0%: 9.8% 0% to +2%: 7.5% +2% to +5%: 11.0% +5% to +10%: 8.7% above +10%: 14.6% Red total: 57.9%. More than 5% down: 30.3%. More than 10% down: 13.8%. Now the part that makes this worth posting. The average next-day return is +0.92%, against a +0.21% baseline for any random day in the sample. So on average, the day after a big pump is better than a normal day. The median is -1.51%. Both numbers are correct. They disagree because the distribution is barbell shaped, not centred. Roughly one in seven of those next days rips another 10% or more, and that tail alone drags the mean above zero while most of the individual outcomes sit in the red. This is, I think, the cleanest example I have found of why "what is the average return" is often the wrong question. Two people can quote accurate statistics from the same 253 samples and give you opposite advice. The momentum crowd cites the mean. The fade-the-pump crowd cites the hit rate. Neither is lying. What it changes for me practically: after a vertical day, the modal outcome is a red candle, so entering because "it is clearly going up" is buying the least likely single bucket. But sizing to fade it ignores that 14.6% of the time you are standing in front of another 10% move. I did not test whether this is exploitable, and I doubt it is in this raw form. Base rates are context, not a signal. Method: daily candles, Binance spot. Up day measured close versus open on the same candle. Next-day return measured close to close. Same 8 symbols throughout, full available history per symbol. If someone has run this on equities or on a longer crypto history, I would be interested in whether the barbell holds or whether it is specific to this cycle.