[plc]

All right, enough. The amount of ignorance out there about HFT is huge.

I developed HFT algorithms in a previous life at a very large, well known bank. The majority of trades out there are in fact market making related. The ones that are market taking are usually at the expense of OTHER HFT algorithms, the ones that are slow and showing out dated prices.

And yes, HFT algorithms most DEFINITELY hold inventory. Some don't, but most do. It's too expensive to get hit on a bid and then try to get rid of it right away. They often hold inventory for days, weeks, even months. I know this for a fact. I can't tell you know much time is spent on worrying about and managing the inventory HFT systems accumulate.

HFT is immensely useful for large mutual funds. They replaced a broken system of high school dropouts who worked on the floor of the NYSE. Mutual funds HATED calling the floor, waiting an hour for execution, not knowing what was happening, and inevitably being taken for a ride. This is all public...go ahead and read a lot of the articles after Thain took over as president of the NYSE and gave tours of the floor to the large mutual fund managers...they were not impressed, to put it mildly.

With HFT, you have transparency and immediacy. You put in an order, and you get that price. In the old days, mutual funds called in an order, and had no idea what they would get, it would be whatever fill the NYSE trader gave them. Executing orders throughout the day makes perfect sense, and is cheaper execution in the long term. You simply get a bad price if you try to sell a billion shares of MSFT in an instant, so instead you spread it over time.

HFT does not front run in any way that is different than trading since the beginning of time. They are trying to understand if their is future demand, and adjusting their prices accordingly. You don't think that after being hit on a bid, the old NYSE floor traders didn't change their prices? They're doing the same exact thing as HFT does.

There's a ton more to say about it, but these are just to counter some of your points.

[maxerickson]

If the HFT was hugely beating the buy mechanisms that mutual funds were using, I would expect the mutual funds to start looking to buy HFT services.

I wonder how much of the dislike for HFT comes from applying intuition about 2 party trades to a many party market. Mostly, I think people ignore that the HFT systems are competing with each other, not just arbitrarily stepping into the middle of transactions. The latter really violates intuition about fairness.

[plc]

Mutual funds do buy HFT services in executing their long term investments...they do so by going to the open market, where HFT dominates. They pay HFT services by paying the typical 1 cent spread.

One of the greatest examples of how nothing nefarious is going on is from a large multi billion dollar hedge fund I used to work at. They have both an equity investment group, as well as a HFT equity liquidity providing platform. I worked on HFT, and knew the equity guys. They always wanted to execute with us in the open market. It was so much better than what they used to have to do.

Why would one of the most sophisticated equity investors in the world want to run an operation that "skims" from the other? Because that's not what HFT does, and of course they know that. They use their own product (their HFT system) to execute their own buy orders. If that's not a testimonial, I don't know what is.

[prodigal_erik]

There are just two parties per share. The HFT bought each share from one seller and immediately sold it to one buyer; their optimal holding time is two round trips to the exchange. If the seller and buyer were both in market within a fraction of a second of each other and the trade would have gone through anyway, the HFT isn't adding any liquidity but parasitically attacking a flaw in the way the exchange matches and clears trades.

[plc]

What are you talking about? If a non HFT buyer/seller both want to buy/sell at 20, then they will go to the market and that will happen. In fact, that happens all the time.

This is precisely the point about HFT. If you are a seller, there isn't always a buyer. HFT is there when there ISN'T the other side. The scenario you described is absurd. How often do a buyer / seller want to interact within the same millisecond?

They can of course. HFT does nothing to stop that. Two people can trade in the open market with each other whenever they want. The reason most of the time HFT is involved is HFT gives the best price. That's why it's become so big. They provide the best price to their customer, not attacking some "flaw" in the exchange.

[maxerickson]

Part of what I am saying is that there isn't much room to do that. Take some publicly traded company and examine their order book. For any company with decent volume, the bid ask spread will likely be $0.01. For companies with less volume, buyers and sellers won't waltz into the market at the same second.

Here's an exciting company with a spread of $0.01 (at the moment anyway):

http://finance.yahoo.com/q/ecn?s=WEC+Order+Book

My selection criteria was to try to find a S&P 500 component that I wouldn't expect to be among the most highly traded S&P 500 components.

[brazzy]

> The ones that are market taking are usually at the expense of OTHER HFT algorithms, the ones that are slow and showing out dated prices.

Isn't that exactly the point the GP was making?

[plc]

How is that even close to the point he was making? HFT vs HFT algo trades happen infrequently. They do happen, but it's rare. But guess what? This forces each other to be better, just like any other competitive industry. The ones that do poorly consistently are just being told by the market that they should leave and go find something else to do, because they're not any good at this.

If it's bad that one firm capitalizing on the situation of another firm making a mistake is the point, then I think he should really rethink his participation in capitalism.

[brazzy]

Quote "robots trading with robots without any regard to the stocks they are trading." end quote.

The problem with the competition between HFTs is that it's turned into a ridiculous arms race to shave microseconds off the response time - how is that in any way productive?

[plc]

Without any regards to the stocks they are trading? Ironically, Mark Cuban shoots himself in the foot here when he complains about macro events dominating micro events.

It turns out to price an asset, like a stock, a wide variety of skills is needed. Some people focus on macro trends, like analyzing political events, oil prices, etc. Some people talk to management of the company, count cars at Wal Mart, etc. These are the micro trends.

And some people try to estimate the current supply and demand for a certain stock. That is what HFT does. So yes, it has "regards to the stocks they are trading." Just different regards. That's why specialization works so well. HFT worries about current supply/demand issues, and tries to estimate a fair price to set the market so that when the average investor comes in, and looks at a stock with a 1 cent spread, he is going to get a fair price. Over the long term, people who analyze micro and macro factors will trade with HFT, which will then force the algorithms to change the stock price to reflect their views.

Is an arms race to shave milliseconds off of trading times the most important thing in the world? Absolutely not. But is gaming the Apple app store to get your mobile app higher in the rankings any different? At least with HFT, milliseconds do matter in more situations than you can imagine. If one day you buy a stock that had bad news announced just milliseconds before, and HFT did not update that quickly so you paid a price way, way too high, you would be upset. Again, not the most important thing in the world but it does help.

The millisecond thing is highly overrated. Anyone in the industry knows that it only matters to a small degree…understanding supply demand and adjusting accordingly is the most important skill. That's why the most profitable high frequency firms don't care about what millisecond they execute in…that is one of the great ironies of this--you guys are concerned about something that not many other people who actually do this for a living are concerned about. But you guys latch on this minor point as the major one, and that is a mistake. Do developers worry about how quickly their apps load? Yes. Does it mean their product will be a success? Not necessarily. That is the same thing here. HFT shops work to make sure they aren't too behind, but focusing on that aspect is just trivial compared to the actual work they do.

[tptacek]

No.