Riding along with a big sliced order
The idea: you can see a whale feeding a huge order into the market in pieces, so get in beside him and ride. It did not survive a long test: what is left is smaller than the fees, and all of the profit turned out to be one single month.
Why it sounded right
Out loud the logic is flawless. Somebody announced in advance that they will be buying for the next hour, and they cannot stop halfway without paying for it. So the flow in one direction is known ahead of time. That is about as rare as it gets: normally you are guessing who will buy, and here you were told. Stand next to him, let him push, get out before he does.
The previous test props this up too. The push from the slices does not come back, it keeps going. So the big buyer leaves a mark in the price. And if the mark stays, riding with him should beat standing against him. That is not a guess, it follows directly from what was measured.
The third argument is size. It seems obvious that a bigger order leaves a bigger mark, so there should be a ladder: small orders give nothing, big ones give a lot. A ladder like that looks like proof of a mechanism rather than a coincidence. The first run showed one.
How we tested it
We stretched the window. The first run covered a few days; the repeat covered several months and thousands of orders across dozens of coins. Then we did three things.
We broke the result down by month. That is a simple and very nasty control: if the rule is real, it pays roughly similarly every month, unevenly but present. If all the profit sits in one month, the rule matched the weather, not a mechanism. Separately we checked the ladder by size: we sorted the orders into groups from small to largest and looked at whether the payoff grew. And we checked whether a ranking of the traders behind them carries forward: we picked the best from the first half of the period and looked at what they gave in the second.
How it ended
Nothing is left. The average move in the direction of the order over the first few minutes is smaller than a round trip costs in fees — the trade loses money before you even get to the fine points of execution.
There is no ladder by size. The largest orders were not the best, they were among the worst, and the order of the groups is scrambled. The old "big ones give a lot" rested on a few dozen cases over three days, which is enough for nothing.
All of the profit is one month. Every other month lands at zero or below. And the ugliest part: the first test was run in exactly that best month. The trader ranking does not carry forward either — the top group of wallets did worse going forward than the bottom one, so the order flipped. The old huge spread "by wallet" was noise on one-off cases.
What this teaches
- Break the result down by month before you celebrate. If the profit is collected in one month, you do not have a rule, you have a description of that month.
- Check whether the period of the first, successful test overlaps with that month. In our case it did — and it has now happened on several different ideas.
- Fees first, everything else second. If the average move is smaller than a round trip costs, discussing filters is pointless.
- A ladder by signal strength is a good check on a mechanism, but only on a big sample. On a small one it assembles itself out of thin air and looks more convincing than the result.
- A ranking of "smart participants" has to be tested for carrying forward: pick on the past, look at the future. Without that test, any such ranking is just a list of who got lucky.