The sub-millisecond timeframe is part of a race to be at the top of the book or a race to pull orders. Very few HFT's literally trade a single security repeatedly on a sub-millisecond timeframe.
The general strategy is "our signal changed - move fast, get to the top of the book in milliseconds". Then you wait, and seconds to minutes later someone will fill your order sitting at the top of the book. Then you play the same game on the opposite side to unload your position.
The game of shaving milliseconds is solely about beating other HFT's to the punch, it's not about actual market movements.
Thanks for the article. It seems to me like a missing piece is a discussion of volatility. That is, I get it that HFT drives spreads lower, but spreads are the wrong direction to make money, even in volume. That is, if pre-HFT spread is that the best buy price is $10.00 and best sell price is $10.05, and post-HFT best buy = $10.03 and best sell = $10.04, you still lose a penny on every trade instead of a nickel -- you can't make that up on volume! :-)
It seems to me that volatility is required for any market maker to make money, and that's a fundamental conflict, isn't it? That is, "outside" traders would prefer the market to be smooth, whereas "inside" traders want it to fluctuate.
I can see the argument that, in actuality, HFT on the whole needs less fluctuation to extract enough profit to provide liquidity, so in theory, it would be expected to be a more stable market maker than human operators. Is that basically equivalent to what you're saying in the article? (And is there any data on that hypothesis?)
The general strategy is "our signal changed - move fast, get to the top of the book in milliseconds". Then you wait, and seconds to minutes later someone will fill your order sitting at the top of the book. Then you play the same game on the opposite side to unload your position.
The game of shaving milliseconds is solely about beating other HFT's to the punch, it's not about actual market movements.