HFT Myths

H

hft

Got some time to kill this weekend. I've been working in HFT for the last 7 years. Futures, FX, some equities, mainly market-making. I've always been curious about supposed front-running allegations and other myths that are assumed/spread by people outside the industry. So...

I'm not here to debate subjective opinions about HFT. What I'm willing to do is answer any concrete questions about the industry that I can (and admit when I can't), within reason of confidentiality and competitive advantage of course.

Maybe I'm opening up a can of worms, or maybe no one cares to even respond/ask. We'll see.
 
How do HFT algos front run SMART routing orders?

Here's an example.

There's 500 shares available at the ask price. 100 shares each at ARCA, NSDQ, BATS, NSX, and, EDGE. I want to buy all 500 so i send a 500 buy order via my broker's smart routing system at the ask price. But I only get filled on 100 shares. The remaining 400 shares either get bought by someone else or the orders are cancelled the instant I send my buy order. What just happened?

I can also send the order directly via an ecn like ARCA with instructions to sweep other venues, but the same result occurs.
 
What is the ratio of winning trades to losing trades and if this can be further split into ones attributed to algo entry/exit only and attributed to trade management.

Second question: what are the most common general groups of algorithmic trading being used in practice and how they evolve over time.
 
Here is a nother front running mechanism - newly discovered at the CME. The CME does not sell the data feed first, BUT.... it distributes FILL first.

* Put a pricece to buy, sell into the book. Try to be in ront (il.e. start out).

* If the market moves, prices are distribued to everyone at the same time - those that colocate at the exchange get them first.

* you can STILL get info that a price swithcd to a new level ealier if you are in front, because executions get distributed BEFORE price updates.

Was on zerohedge.
 
Quote from vicirek:

What is the ratio of winning trades to losing trades and if this can be further split into ones attributed to algo entry/exit only and attributed to trade management.
More...
A typical ratio for non-equities trading (equities trading is quite different due to rebates) might be something like 90% scratches, 8% winners, 2% losers. Expected profit on the order of 1/10 tick per contract traded overall.

Quote from vicirek:
Second question: what are the most common general groups of algorithmic trading being used in practice and how they evolve over time.
More...
"Algorithmic trading" is beyond the scope of my expertise, but high-frequency trading can be very generally categorized between providity-providing and providity-removing strategies. As for their evolution, that's too broad to really describe but in general they've gotten faster and margins have gotten smaller all-around over the past decade.
 
Quote from NetTecture:

Here is a nother front running mechanism - newly discovered at the CME. The CME does not sell the data feed first, BUT.... it distributes FILL first.

* Put a pricece to buy, sell into the book. Try to be in ront (il.e. start out).

* If the market moves, prices are distribued to everyone at the same time - those that colocate at the exchange get them first.

* you can STILL get info that a price swithcd to a new level ealier if you are in front, because executions get distributed BEFORE price updates.

Was on zerohedge.
More...
This kind of classification aggravates me. This is not front-running. If you're willing to put out risk by resting an order in the market, you get paid for that risk by receiving your fill before the trade is reported in the public pricefeed. Makes perfect sense to me, and this is standard procedure on most exchanges (order-level exchanges are a rare exception). If you're willing to put out an order and accept that risk, you'd receive the fill information before the rest of the world too. On top of that, there's a lag between fills disseminated for orders at the front of the queue at a price level versus those behind as well, which is another form of benefit for putting out that order/risk earlier and/or smarter and/or faster than the guys putting in orders behind you.

Front-running is generally classified as submitting orders for prices that you know will disappear BEFORE someone else submits his order, due to unfair advantage (like if someone at a BB gets a big order from a big client and proceeds to buy the stock before that big order hits the market). Apples and oranges to me.
 
Quote from western:

How do HFT algos front run SMART routing orders?

Here's an example.

There's 500 shares available at the ask price. 100 shares each at ARCA, NSDQ, BATS, NSX, and, EDGE. I want to buy all 500 so i send a 500 buy order via my broker's smart routing system at the ask price. But I only get filled on 100 shares. The remaining 400 shares either get bought by someone else or the orders are cancelled the instant I send my buy order. What just happened?

I can also send the order directly via an ecn like ARCA with instructions to sweep other venues, but the same result occurs.
More...

Lots of possible scenarios here. Depending on the details you can eliminate one or another.

Whether you submit to your broker or ARCA, you're at the mercy of their "Smart" order routing system. One easy scenario, some crappy exchange like CHX has an offer hanging at an ask price that's equal to or lower than the price you want, so your broker is obligated to submit your order there first, wait to hear back that is missed, then proceed to send to the other exchanges you listed with the price available. By that time the price is long gone. HFT's and other smart firms know that CHX aren't likely to have that price available (due to subjective experience, historical data/statistical analysis, etc) and will route and ISO order to CHX just to meet Reg NMS regulations, then immediately send orders to the other exchanges you listed to actually try to get the fill. This results in overexposure (there's a chance that we'll get filled on both CHX and the others), but we take that risk knowing that the likelihood of that is very small and it's worth it to get a better chance at getting the fill at the other 'good' exchanges.

Now in the unlikely chance that you actually looked at all protected endpoints and the price isn't available any other than the 5 you listed(not saying you didn't do your hw, but I didn't think it likely that most retail traders look at all the lesser-known exchanges), maybe your broker's 'smart' system is set up to try 500 at the first available exchange, then wait to hear back on whether the whole thing is filled before trying others.

Another very likely possibility is that your broker sends a non-ISO order to one of the exchanges, then relies on that exchange's 'smart' system to route accordingly. Problem is, by the time the order gets to the first exchange, it's already too late to hit the others.

An HFT firm would have the ability to send ISO orders to each of the exchanges simultaneously, but that firm is also required to log its known quotes of all protected endpoints and have them on file for future investigation that it doesn't violate Reg NMS rules. You at home don't really have that option.

I must admit I'm making a few assumptions here, particularly about how your broker operates. What I can guarantee though, is that there is no way another firm can see your first order hit one exchange and unfairly hit the other exchanges before the rest of your orders hit. Then again, I suppose if your broker's line is slow enough it is a slight possibility. For example, if the exchanges are say 1, 2, and 3 miles from your broker, and your broker's line from 1 to 2 is slow enough, the HFT firm can beat your order getting to 2 after reacting to the hit at 1. I think this is pretty unlikely, but not out of the realm of possibility.
 
What is the difference aside technology/speed between Market Maker or Specialist and HFT.

This question is related to allegations of front running because MM would see buy and sell order first and then he would become intermediary in matching them pocketing the spread. People can accept that but cannot stand HFT.

If independent HFT firms would be removed from the market are MM's and Specialist capitalized enough to provide liquidity for providing orderly markets?
 
HFT dude, here's my question for you: If all you're doing is taking gains measuring fractions of a tick out of individual stocks in split-second trades, that alone should <b>not</b> have been enough to devastate the entire game for manual equities daytraders. I mean, decent fills might have been harder to come by... but we would have survived.

However, the entire leveraged prop equities trading industry has indeed fallen off a cliff in recent years, and HFT's are commonly blamed for this. Patterns that used to repeat themselves over and over again, providing us trade setups with real edge have been wiped off the face of the earth. Intra-day movement in individual equities has moved away from the repeating patterns of the past, and towards more random, unpredictable and un-exploitable random crap.

Here's the explanation I came up with, which may or may not be accurate:

Pre-HFT: There were lucrative, constantly repeating short-term patterns with +EV edge in individual equities. Once a trader recognized any specific pattern, he just had to keep trading that specific setup over and over until that pattern finally cycled out into obsolescence. This could be months or even years.

Post-HFT: The algos can recognize repeating short term patterns as well as any human. 60%+ edges are reduced to untradable 51% edges, as the patterns are over-exploited with size by the machines, who are also seeing the quotes/getting filled milliseconds before everyone else. Let alone the way they get to jump a micro-penny in front of the human's limit order, and can cancel/replace to jump to the front a lot faster than he can.
Unable to compete, manual equities daytraders have (nearly) all been forced to pursue other ways of making a living.

<b>So if the HFT's did not cause the death of all the old, predictable intra-day repeating patterns, what did?</b>
 
Quote from Rearden Metal:

Let alone the way they get to jump a micro-penny in front of the human's limit order, and can cancel/replace to jump to the front a lot faster than he can.
More...

Have you tried to use the Dark Pools (like Barclays LX, Cross+, etc..)?
 
You must log in or register to reply here.
Back
Top