The pressure a sliced order puts on the book
A large order that a bot chops into pieces and feeds into the market really does push the book around. You just cannot get paid for that push: the move does not come back, and what is left to take is smaller than the fees.
Why it sounded right
You can watch the mechanism happen, and that is the strongest part of the idea. On some venues a large sliced order is public: you see the coin, the side, the size, and the window it has to be filled in. Then you watch the book and see each slice eat the top levels, the price jerk, and market makers refill a few seconds later. This is not theory. It happens on your screen.
The obvious conclusion follows. If a slice is bigger than the liquidity sitting above it, it pushes the price further than it should go. So right after a slice the price is too high, and you can stand against it — sell into where a buy just pushed. The idea rests on real microstructure, not on a pretty picture, which is exactly why it is easy to believe.
A second class of observation adds to the confidence. Sometimes the bot stops halfway through, and the price falls apart right after. That looks like a ready-made trade: the bot gave up, the bid is gone. One or two of those in a row convince you harder than any table.
How we tested it
First we collected the slices ourselves. For every sliced order we recorded the book and every trade, split the fill into slices, and for each one measured how far it pushed the middle of the market, how much of the liquidity above it it ate, and whether the price came back after half a minute, a minute, five minutes.
The main control was to strip out the market's own move. The first raw run showed a strong lopsidedness: standing against the sells paid, standing against the buys did not. That looked like a discovery. But most of the sample sat in one coin, and that coin was going up that day. So "standing against sells pays" just meant "being long on an up day." To subtract that, we stopped comparing against zero and started comparing against the same coin's move over the same window, counting it so that buys and sells cancel the general drift: if the effect is real, you see it on both sides. If you only see it on one, that is the market moving.
How it ended
The push is there. The snap-back is not. The bigger the slice relative to the liquidity available, the harder the price jerked — that held up firmly. But the return was negative: after a slice the price does not come back, it keeps going the same way. Which means the founding thought, "it got pushed, so it will bounce," is wrong at the root.
Once the market's own move was removed, what was left from standing against the slices was real but tiny — around the cost of a round trip in fees, or below it. On top of that, every slice has a dozen professional liquidity providers on the other side of you doing the same thing cheaper. A separate check also killed the pretty theory about the bot that stops: orders the bot carried through to the end did slightly better than the ones cut off halfway. Exactly the opposite of what the vivid case suggested.
What this teaches
- An effect you can see and an effect that pays are different things. The price impact here is measurable and real, and there is no money in it, because it is smaller than the costs.
- Before you count the money, check whether the move comes back at all. If it keeps going, the whole logic of standing against it collapses, even when the trade sometimes wins.
- If the result only shows up on one side — only in buys or only in sells — it is almost always the market moving, not a discovery. Count both sides and demand that both be positive.
- When one coin fills the sample, you are measuring that coin, not the rule.
- The vivid case where everything worked is the worst thing to build a theory on. On the bot that stops, the data said the opposite.
- The effect still has a use, not as a trade of its own but as a correction: where to place your passive orders if you are already sitting in the book.