What is the state of the art of algorithmic trading?

Ask your investment, budget, and other money related questions here
Post Reply
Lucky C
Posts: 268
Joined: Sat Apr 16, 2016 6:09 am

What is the state of the art of algorithmic trading?

Post by Lucky C » Thu Sep 06, 2018 11:23 am

What resources are out there to learn a bit about what is currently implementing in the world of algorithmic trading? Obviously if there are some wildly successful algorithms out there, all the details will be kept secret, but I'm interested in learning more about the state of algos in general.

Are algos taking over trading or are they still unlikely to have a significant impact on the performance of regular human investors? I want to know if I am seriously competing with algos or if they're just a small percent of trades. This leads to the question of whether prices should continue to move (in the foreseeable future) as if they are still set by emotional humans (greedy or fearful)?

What has been implemented with a significant amount of assets under management? Just small-scale rules-based trading operations, or are there supercomputers implementing sophisticated AIs that are trading billions?

How much do we actually know about real-world use of algorithmic trading and how much are we in the dark?

Dream of Freedom
Posts: 218
Joined: Wed Aug 29, 2012 5:58 pm
Location: Nebraska, US

Re: What is the state of the art of algorithmic trading?

Post by Dream of Freedom » Thu Sep 06, 2018 12:03 pm

They are funneling money to build skynet and develop terminators.

jacob
Site Admin
Posts: 10442
Joined: Fri Jun 28, 2013 8:38 pm
Location: USA, Zone 5b, Koppen Dfa, Elev. 620ft, Walkscore 73
Contact:

Re: What is the state of the art of algorithmic trading?

Post by jacob » Thu Sep 06, 2018 12:06 pm

You're basically going to find nothing published in book form that's not older than 10-15 years. That's actually not that different from any other kind of research: What you see in the newest textbooks is what what you heard at conferences 30 years ago and what people published in journals 25 years ago. All this stuff is IP and subject to non-disclosures and non-competes. People go to prison for violating this stuff. The research/IP can be worth [many] millions.

When I quit back in 2015, algos were responsible for more than 90% of all posted orders. That's a bit misleading though. Very many algos really do nothing much beyond providing liquidity. E.g. if you sell 100GE on NYSE, an algo might snap that up and then immediately (literally at the speed of light) turn around and sell it on AMEX for a 0.5 cent profit.

So here's how to think about it. Imagine a diagram that classifies agents in the market according to speed and size. A private investor would operate on a time scale of hours to years and a size of $500 to $50,000 tops. A professional clicker-trader (that's a human clicking on a mouse) would operate on a time scale of seconds to hours and a size of $50,000 to $5,000,000. A pension/mutual fund operates on a timescale of days to decades and is enormously large (block traders). A HFT algorithm operates on a timecale of 0.005s to minutes and sizes down to $50.

You can never tell who is taking the other side of the trade .. however odds are that if you're a scalping daytrader, you're competing with HFT algos. If you're a generic daytrader you're competing with clicker traders. And so on...

In terms of implementations, things were/are moving towards sophisticated order types. Most people are familiar with market orders and limit orders, but there are now more than 150 different order types that makes it possible to circumvent the 30 millisecond latency by putting part of the behavior directly on the matching engine (<- that's the server that figures out who trades with who) ... e.g. "fill or kill at midprice" or "sliding ourside".

When I quit the game back in 2015 headhunters were actively trying to recruit me for AI stuff. I suppose that's where things are heading. The reason is that the hardware/speed game is maxed out and "owned" by a few massive players which invest in microwave antennas and specialized chips. It's really similar to bitcoin that way ... you're not gonna mine anything with your personal laptop anymore. There's also no way that a human can be faster than humans already are and maintain concentration. Therefore the inefficiency is to try to make algos slightly smarter with AI, etc.

What I can tell you is that people who are not professionals know very little about real-world use.
The popularizations I read back then were eye-rollingly bad/inaccurate.

You should also keep in mind that algos are rarely if ever autonomous. An algo is just a tool managed by a real human who answers to a human boss and which at the very least has an off button. Flash crashes tend to happen when the humans behind the algos panic and leave the market. In that sense, fear and greed are an inherent feature of the market.

Lucky C
Posts: 268
Joined: Sat Apr 16, 2016 6:09 am

Re: What is the state of the art of algorithmic trading?

Post by Lucky C » Thu Sep 06, 2018 1:44 pm

Thank you, Jacob. Very interesting and reassuring. As impressive as some AI developments have been in other fields, it sounds like there is no chance that algos would have any real impact on my investment strategy any time soon.

User avatar
Bankai
Posts: 262
Joined: Fri Jul 25, 2014 5:28 am

Re: What is the state of the art of algorithmic trading?

Post by Bankai » Thu Sep 06, 2018 5:12 pm

If most are after multiple tiny profits, would tax similar to stamp duty in the UK (0.5% of transaction value on each purchase) not be quite effective in curbing this?

Also, considering the 0.5% tax in the UK, I imagine are way less common than in the US? The same way day trading must be much more difficult with the tax.

suomalainen
Posts: 473
Joined: Sat Oct 18, 2014 12:49 pm

Re: What is the state of the art of algorithmic trading?

Post by suomalainen » Thu Sep 06, 2018 5:34 pm

My impression is that the impact on a human's trading is that it is actually helpful in that it narrows spreads. So algorithmic trading is not some scary front-running thing which is sometimes how it's portrayed (anyone read Flashboys? That was my impression from reviews, but I never read it). I take it from @Jacob's description that my impressions is reasonably accurate. I.e.,
jacob wrote:
Thu Sep 06, 2018 12:06 pm
When I quit back in 2015, algos were responsible for more than 90% of all posted orders. That's a bit misleading though. Very many algos really do nothing much beyond providing liquidity. E.g. if you sell 100GE on NYSE, an algo might snap that up and then immediately (literally at the speed of light) turn around and sell it on AMEX for a 0.5 cent profit.
...
A HFT algorithm operates on a timecale of 0.005s to minutes and sizes down to $50.
...
You can never tell who is taking the other side of the trade .. however odds are that if you're a scalping daytrader, you're competing with HFT algos. If you're a generic daytrader you're competing with clicker traders. And so on...
In my experience, if I'm worried about paying 74.07 vs 74.06 (and losing that penny to an algo), I'm worrying about the wrong thing.

jacob
Site Admin
Posts: 10442
Joined: Fri Jun 28, 2013 8:38 pm
Location: USA, Zone 5b, Koppen Dfa, Elev. 620ft, Walkscore 73
Contact:

Re: What is the state of the art of algorithmic trading?

Post by jacob » Thu Sep 06, 2018 5:38 pm

Something like Tobin tax for trading. What this would do is to increase the spread between the bid and the ask or perhaps translate it into commissions that hide/insure the holding risk. A lot of financial innovation in the algo-space has been to reduce the spread (now down from 25cents a couple of decades ago to 0.5cents) and transaction costs (near zero and actually a profit source in some cases). Effectively, the tax would be pushed onto retailers/consumers. You can wave goodbye to those low-fee/no-fee instruments people have growth accustomed to. IMHO, Tobin tax is an attempt to eat a free lunch that isn't really there. It's a bad idea.

Multiple tiny profits is not nearly as bad as occasional large profits/losses due to lacking liquidity. At least that's the theory.

Another way of seeing algo-trading is as the robotization of human traders. If one computer can replace [the mostly Gaussian aggregate behavior of] 50 traders that saves people (the company and its customers) money in the form of salary. To a large degree this is no different than things like Black Scholes and the EMH and the humans those inventions eliminated.

Post Reply