Insights into High Frequency Trading from the Virtu IPO [pdf] • Hacker News Board

Re: Insights into High Frequency Trading from the Virtu IPO [pdf]

#121

Post by bovine3dom » Fri, May 05, 2017, 7:51 AM UTC

anigbrowl wrote: post #92

tptacek wrote: post #80

Earlier quoted context omitted.

I apologize for repeating a point I just made, but I'll make it here too: You can mentally substitute "cost of trading" for "liquidity". It's true that you can get costs down to a point where they don't matter. But I think you'll have an easier and more intellectually honest time engaging with the issue if you avoid the jargon and focus on the impact of the jargon. Cost is why we care about liquidity. As markets get…

I get this. I just have an intuition that some friction in a dynamic system is actually a good thing, and also wonder if the growth of HFT might not lead to an unsustainable deviation from an underlying power law distribution of trading activity.

You have reminded me of an excellent article on frictional costs in economics from The Economist [1]. There is a slight snag in that it is an April Fools joke.

[1] http://www.economist.com/news/leaders/21719799-it-time-recog...

Re: Insights into High Frequency Trading from the Virtu IPO [pdf]

#122

Post by ForHackernews » Fri, May 05, 2017, 9:07 AM UTC

consz wrote: post #100

ForHackernews wrote: post #46

Earlier quoted context omitted.

If nothing else, it means that firms that can afford to collocate their servers in the cage next to the exchange's servers have an advantage over those that can't. It seems like something like the roundtrip time for a packet from NYC to Tokyo would be a fair "speed limit" for exchanges, and ensure that all firms, globally, are on an even footing.

The round trip time between NYC to Tokyo is very short; Any computer system could beat any human, even at that scale. I'm not sure what you think you'd accomplish here; you'd just have HFT with 150ms response times instead of 1us. But you'd still have HFTs.

Yes, obviously.

The point is that there would be no incentive for HFT firms to invest endless money and effort trying to eek out a few more microseconds because that would no longer be a viable trading "strategy" (if you can even call it that).

The problem with HFT firms isn't that they're fast, it's that their only focus is being fast. There are all kinds of interesting algorithmic trading strategies that use data analysis or complex models to make rapid trades: those systems are adding information to the market and making it more efficient. In contrast, pure HFT only adds dubiously-necessary "liquidity" by being quicker on the draw than other firms.

Re: Insights into High Frequency Trading from the Virtu IPO [pdf]

#123

Post by ForHackernews » Fri, May 05, 2017, 9:07 AM UTC

MichaelGG wrote: post #59

ForHackernews wrote: post #46

Earlier quoted context omitted.

If nothing else, it means that firms that can afford to collocate their servers in the cage next to the exchange's servers have an advantage over those that can't. It seems like something like the roundtrip time for a packet from NYC to Tokyo would be a fair "speed limit" for exchanges, and ensure that all firms, globally, are on an even footing.

I think you'll find, when you try to get down into it, there's no easy way to enforce a speed limit. It's a naturally arising behaviour. You could try making a market that crosses orders every X seconds, but then you move the speed to the edge of those periods. If you wanna slow down HFT, then kill the stupid restriction that stocks be priced in pennies. With 8 digits of resolution, HFTs would then be forced to compe…

That's an interesting idea. Is that rule just a holdover from the days of manual accounting?

Re: Insights into High Frequency Trading from the Virtu IPO [pdf]

#124

Post by consz » Fri, May 05, 2017, 10:37 AM UTC

ForHackernews wrote: post #122

consz wrote: post #100

Earlier quoted context omitted.

The round trip time between NYC to Tokyo is very short; Any computer system could beat any human, even at that scale. I'm not sure what you think you'd accomplish here; you'd just have HFT with 150ms response times instead of 1us. But you'd still have HFTs.

Yes, obviously. The point is that there would be no incentive for HFT firms to invest endless money and effort trying to eek out a few more microseconds because that would no longer be a viable trading "strategy" (if you can even call it that). The problem with HFT firms isn't that they're fast, it's that their only focus is being fast. There are all kinds of interesting algorithmic trading strategies that use data a…

I'm not sure where you get the impression HFT firms only focus on speed. They focus both on being smarter (having better signals) and being faster; it's only very niche HFT firms which rely entirely on speed as their edge.

Also, what's your goal here? It doesn't sound like it'll necessarily impact the profits of the HFT industry as a whole if you do this, merely lower the cost, so on net making the average HFT firm richer.

Re: Insights into High Frequency Trading from the Virtu IPO [pdf]

#125

Post by consz » Fri, May 05, 2017, 10:49 AM UTC

anigbrowl wrote: post #68

21 wrote: post #7

Earlier quoted context omitted.

No they would not. Even if you introduce a 10 sec delay, there will still be someone which will be first after the 10 sec pass, and you can still create models which will be tuned to predict what the price will do 10 sec into the future. And what do you think about a law which forces all news agencies to wait 1 hour before reporting major news, so that everybody gets a chance to report it, and not only the huge agenc…

GP specifically said "a bit of random latency". Don't change the terms of his argument to make it easier for you to refute, that's dishonest.

There's no reason HFT can't or wouldn't exist even if you add random latency.

Re: Insights into High Frequency Trading from the Virtu IPO [pdf]

#126

Post by consz » Fri, May 05, 2017, 10:53 AM UTC

ForHackernews wrote: post #123

MichaelGG wrote: post #59

Earlier quoted context omitted.

I think you'll find, when you try to get down into it, there's no easy way to enforce a speed limit. It's a naturally arising behaviour. You could try making a market that crosses orders every X seconds, but then you move the speed to the edge of those periods. If you wanna slow down HFT, then kill the stupid restriction that stocks be priced in pennies. With 8 digits of resolution, HFTs would then be forced to compe…

That's an interesting idea. Is that rule just a holdover from the days of manual accounting?

From my own conversations with exchanges -- I think this is just the exchange appealing to layman investors. The exchange wants to appear liquid, and they like to do this by having a lot of shares on the best bid or offer. If you reduce the tick sizes, you'll definitely see fewer shares on the inside bid or offer (even though you'll see more shares on the inside x% for any x, aka. you'll see more liquidity), but the layman investor (e.g. mom and pop, retail traders, people like brad katsuyama, etc.) will suddenly think you're less liquid because the best price now has fewer shares.

That's, at least, the strongest argument I've heard for why tick sizes aren't reduced more rapidly.

Re: Insights into High Frequency Trading from the Virtu IPO [pdf]

#127

Post by consz » Fri, May 05, 2017, 10:55 AM UTC

jasonmunro wrote: post #115

jandrese wrote: post #30

Earlier quoted context omitted.

HFT aren't the firms putting the markets on computers. HFTs sit between the computers and the retail traders and use that information advantage to front run. There's no reason there needs to be a middleman here. Buyers can buy from sellers directly. In fact that's what they think they are doing, except that the HFT firms are basically adjusting the price on them underneath the sheets. The real losers are the sellers.…

The liquidity comes from the HFT market maker's own money. Market makers are required to post bid and offers of a reasonable level of liquidity in all symbols they are registered to "make markets" in. It may seem like a simple skim off the top operation, but it's not that simple. If they do not maintain their bids/offers, they are disqualified from market making. If they do maintain them, they reap a number of benefi…

> which IMO should be off-limits for all trading firms

Out of curiosity -- why?

Re: Insights into High Frequency Trading from the Virtu IPO [pdf]

#128

Post by ForHackernews » Fri, May 05, 2017, 11:34 AM UTC

consz wrote: post #124

ForHackernews wrote: post #122

Earlier quoted context omitted.

Yes, obviously. The point is that there would be no incentive for HFT firms to invest endless money and effort trying to eek out a few more microseconds because that would no longer be a viable trading "strategy" (if you can even call it that). The problem with HFT firms isn't that they're fast, it's that their only focus is being fast. There are all kinds of interesting algorithmic trading strategies that use data a…

I'm not sure where you get the impression HFT firms only focus on speed. They focus both on being smarter (having better signals) and being faster; it's only very niche HFT firms which rely entirely on speed as their edge. Also, what's your goal here? It doesn't sound like it'll necessarily impact the profits of the HFT industry as a whole if you do this, merely lower the cost, so on net making the average HFT firm r…

Lowering costs should allow more firms to enter the market, not just pad the profits of existing participants.

If a HFT firm is richer because they're smarter, I don't have any problem with that. That's how investing works. Developing smarter ways to analyze data, setting prices accurately, having a market that responds to new information, these are all social goods.

I am less convinced that everyone trying to be in the same datacenter as the NYSE (https://www.bloomberg.com/news/articles/2016-04-13/inside-eq...) adds any value for society.

Re: Insights into High Frequency Trading from the Virtu IPO [pdf]

#129

Post by consz » Fri, May 05, 2017, 12:07 PM UTC

ForHackernews wrote: post #128

consz wrote: post #124

Earlier quoted context omitted.

I'm not sure where you get the impression HFT firms only focus on speed. They focus both on being smarter (having better signals) and being faster; it's only very niche HFT firms which rely entirely on speed as their edge. Also, what's your goal here? It doesn't sound like it'll necessarily impact the profits of the HFT industry as a whole if you do this, merely lower the cost, so on net making the average HFT firm r…

Lowering costs should allow more firms to enter the market, not just pad the profits of existing participants. If a HFT firm is richer because they're smarter, I don't have any problem with that. That's how investing works. Developing smarter ways to analyze data, setting prices accurately, having a market that responds to new information, these are all social goods. I am less convinced that everyone trying to be in…

I mean, I'm all for lowering costs in the HFT industry, it'll mean the company I work at makes more money -- I just didn't realize you were arguing for the industry, not against it. I don't think any HFT firm would disagree with lowering costs to themselves, the same way they'd prefer that CPUs were never upgraded (so they didn't have to waste money buying the latest ones).

Co-location is good because it commoditizes the closest location to the exchange. Speed always matters, even in randomized auctions, even if you have artificial latencies added to your order, even if... Without co-location done the way it is now, you get into situations where a single firm buys the actual nearest location, and has a monopoly over being the fastest competitor. Now, anyone can access it.

Re: Insights into High Frequency Trading from the Virtu IPO [pdf]

#130

Post by pdovy » Fri, May 05, 2017, 2:40 PM UTC

the_cat_kittles wrote: post #101

vzcx wrote: post #77

Earlier quoted context omitted.

Beyond the "providing liquidity" argument, another side effect of HFT and arbitrage strategies that one might see as valuable is the promotion of global invariants across the financial network, which is rather complicated distributed system. Would it make sense if there were two very different prices for something at two nearly identical venues? If large block of a stock is dropped on one exchange, would it make sens…

>> Beyond the "providing liquidity" argument, another side effect of HFT and arbitrage strategies that one might see as valuable is the promotion of global invariants across the financial network, which is rather complicated distributed system. Would it make sense if there were two very different prices for something at two nearly identical venues? If large block of a stock is dropped on one exchange, would it make s…

>> i dont disagree with arbitrage and liquidity, and a globally invariant price for goods. i just think HFT isnt really improving that in a meaningful way. i ask you- at what timescale does it stop mattering? thats really what im saying. not that liquidity and invariant prices are suspect, but that things are happening on such a minute time scale that it doesn't matter. hey, i could be wrong, but no one has argued why going from milliseconds to picoseconds is improving our lives. thats the thing i would like explained.

This is a little more complicated than it would initially appear, I'll try to explain from a market makers perspective why the speed race exists and why being faster (as a liquidity provider) is better for the market, at least the way it's currently structured.

Market making 101 is basically that you want to come up with a fair value for the product you're trading, and then put out orders to buy for a little less than FV, and sell for a little more than FV. If you buy and sell at those prices you're providing liquidity to the market and capturing a small spread for your effort, great. Do that repeatedly and you have a business. But how much should your "a little less" and "a little more" than FV actually be? The smaller the better for the market, and this ideally should be the primary vector on which market makers compete with one another.

Okay, so in our optimal scenario you'd always quote as tight as possible (limited by the granularity of pricing on the exchange) around the fair value. (I'm glossing over a lot here, calculating the fair value is non-trivial and your level of uncertainty about it will also determine the spread you can quote, but ignoring that for the moment). The problem with this is that if the fair value moves then some of your orders become mispriced, as they represent an opportunity to buy below or sell above fair value and if they execute will be a loser. Smart participants will recognize this and race to pick those off before you can reprice them. If this happens too often, you're not making money anymore, crap. Really only two options here, #1 is to widen your quote so that you are less sensitive to such movements and you are capturing a fatter average spread which compensates you for the losing trades. #2 is to get faster than those other guys.

#2 yields a better outcome for the market, but necessitates a speed race as an additional vector of competition. Exchanges recognize this as well and have long played around with various schemes to give liquidity providers a systematic advantage, e.g., via rebates, or otherwise. It's a tough problem.