Thanks for the offer. I would like to know what kind of factors your algorithms consider. Are they mostly indicators that traditional investors look for, like product announcements and earnings statements, but HFT aims to react faster than others, or do you put in variables day traders might ignore? Not just Twitter trends, but obscure things like weather events, stock price proximity to multiples of ten, ease of pronunciation of company names.

I can't comment on anything I've specifically been involved in and even that is a small subset of the motivation of all the market participants. The price discovery that I described in mk's answer above is a combination of everyone coming together with their own strategies. These strategies come from a wide variety of ideas and time horizons. Trading on P/E on a month to month basis might be one. While trading on twitter traffic in a minute to minute horizon may be another. Trading on order flow on a microsecond by microsecond basis is another. Predicting how the weather is going to affect Oil and Utility companies is probably something else going on.

Ha it's interesting that you mentioned stock price proximity to multiples of ten because that is something I wouldn't expect someone outside the industry to even know about. Are you a shark trying to extract info? It's true stock prices do weird things when they get close to whole numbers. I've pictured it a matter of psychology, I think it's the same reason that queues at restaurants get busy at 12:00 and 1:00pm but not at 12:23. It's because everyone told their friends to meet at the restaurant at 12:00 not 12:23. I think people do the same thing for stocks. They say "OK, I will finally sell TSLA if it hits 180." Not "I will sell TSLA if it hits 179.97." except me that's exactly what I say because I know that's going to get my order executed before everyone at 180. Some stocks can even trade along 1 penny from a whole number all day long. It's weird and I don't get it completely.

IMO as I described above, in the smaller time horizon space the simpler and more fundamental of a signal you can act on the lower risk you are going to have and the quicker you are going to be able to act on that signal.

The article also mentions that transaction fees approach zero. This seems a critical factor too for a high-frequency trader. How close to zero, and how significant are transaction costs overall?

Here are the costs of all the regular lit exchanges NYSE, NASDAQ, BATS, and Direct Edge. It's less than a penny to to take liquidity of a liquid stock and you also get paid to provide liquidity of that same stock. How much you get paid for taking less how much you get paid for providing is clearly positive resulting in profit for the exchange. Additionally you might be interested to know that every big lit market has an inverted exchange where you get paid for taking and have to pay for providing. There tends to be less trading on these exchanges.

Executing a trade, whether that trade came from a signal or a client is all about finding the cheapest liquidity in the market. If you can provide liquidity and get paid by NASDAQ then do that first. If that doesn't exist then go to the dark pool where you might be able to match between the Best bid and ask. If that doesn't work then go to a lit exchange and take liquidity. You can see from the cost lists above that it's not going to be more than a fraction of a penny to execute in any of these areas.

As far as their significance even though its small how much you are going to get paid for a trade from your client is also pretty small. So you want to be as frugal as possible. Trading is a numbers game. You want to make each fraction of a penny of profit you can make in a transaction as big as possible while doing as many of those transactions as possible while also taking on as little risk as possible.