If it's completely untrue, then I'd like to hear about how your computerized trades ever "discover" a good deal.

I'm fairly sure this is the answer: They don't -- instead, they're just good at noticing, very quickly, the actions of the market actors who do. Which is just another way of saying that they're about beating others to the deal.

(I'd also guess that they're probably also quite good at handing recreational day-trader's asses to them.)

It's a little less black and white. Suppose I put a guy with a physics PhD in a dark room for a while and he tells me 'with 99% probability, a stock with a price-to-earnings ratio of under 10 will beat the S&P 500 the next year". So whenever a stock's ratio gets below 10, I buy the stock and short the S&P 500 and wait for the bags of money to arrive. In that case, I'm not really relying on noticing the actions of others who have discovered a good deal. Yet, I am keenly interested in beating anyone to the stock when the ratio gets to 9.99. In reality, the market signals are more complicated statistical relationships, they could just as easily arise from someone being very dumb, instead of very smart. Or just the result of random variation that day.

And yet, even for your hypothetical, you use the term "market signals." ;)

If you look at markets through an information-centric lens, it seems possible to draw a clear distinction between actions that introduce external data and those that are pure acts of deduction, no matter how brilliant.

so how do you win? my naive impression is that speed must play an important part, or it wouldn't be called high frequency.

(so speed is important to you, just as water is important to a fish; it may not be what you focus on, but that doesn't mean that it's not critical).

It's crucial, for sure. But there are trades for which it is the only determinant, and those are dangerous waters.

There are market participants who make their money on pure latency arbitrage; it is a 100% speed game (because anyone can figure out that 1 - 1 = 0).

The majority of HFT falls into the "you have to be fast enough to not get run over" bucket.

ok, so i think the original thread here is arguing that "being run over" is just the market doing what it would do anyway, without you playing in the traffic and endangering us all...