PDA

View Full Version : How does high-frequency stock trading work (as in make money)?


Drum God
03-23-2011, 11:05 PM
This article (http://io9.com/#%215784758/the-speed-of-light-could-turn-the-middle-of-the-ocean-into-a-stock+trading-center)at Gizmodo's iO9 website talks about high frequency stock trading. The issue it specifically addresses is the idea that the speed of light is becoming a factor in some stock trades. It says that some transactions are profitable for only a fraction of a second.

So, how does this work? I don't mean the infrastructure of how they get it done, but how can a transaction be valuable for a tiny fraction of a second? Who is doing this? Is this sort of thing a good idea?

Exapno Mapcase
03-25-2011, 04:56 PM
Your link should work but all I get is io9's front page. And I get stack overflow errors all over the place. I found the article after a while, though.

Anyway, here's the underlying idea. Stock prices change from moment to moment. With modern technology they change faster than ever. But why do they change? Well, theory says that people with better information about the value of the stock take advantage of that knowledge by exploiting the price through higher or lower bids.

Again, theoretically, if you have a piece of information first you can exploit it best. First can literally mean milliseconds. If a stock is being asked at 1.65 and you think it's worth 1.66 then by placing a bid of 1.66 first you can shut out the others. Those others may now bid 1.67 for it and you sell, making your penny profit times however many shares you bought. For 1,000,000 shares that's $10,000. And the more times you can do that in a day, the more money you can make. If milliseconds count, then the fastest software and the fastest connections and the best location can make you more money than the slower pokes.

The article also talks about prices being different in different markets and making money off buying at the cheap price and selling at the high prices. That's called arbitrage. It's another way to value better information. Prices will tend to reach an equilibrium but the first ones in can make money.

Personally, I don't see this as anything much more than a technical variation of what day-trading was during the stock boom. It sounds all futuristic, but the principal of having better information sooner goes back to the days of fast ships and carrier pigeons. Telegraph systems started in several countries at the behest of traders, and even where they didn't start them they soon became the biggest users. The exploitation of arbitrage is also a classic technique of the math gurus. Today everybody looks to exploit the same set of differences, which dampens the amount that can be made. Speed is still helpful but there's less value there to exploit.

The down side of this is that while you can make money faster, you can also lose money faster. Your information has to be correct and your interpretation of that information has to be correct. Day-traders got killed by bets that the market would keep going up. The expenses inherent in multiple trades don't go away by speed. If you can buy a stock and have it appreciate 10% over a year with one purchase and sale commission expense, then you're far ahead of someone who got a 10% return with a thousand purchase and sale expenses. You'd have to consistently return, say, 20% to make up for those expenses. And nobody has ever done that without cheating.

That's why all this is so controversial. Perfect information is self-canceling. There are no differentials to exploit. The same advances that allow for speedier exploitation are also used to spread the knowledge everywhere equally. If the value of the advantages has to be more than the cost of the expenses then there are fewer ways to make money.

It's a fascinating field to read about because it's a real world example of the value of pure information. I wouldn't get anywhere near it with my money, though.

phreesh
03-25-2011, 05:12 PM
Yeah, I read a super-cool article in wired or somewhere that visited the very mundane-looking building that housed some of these trading computers. I think it was in response to that weird massive trade that dropped the NYSE several hundred points for five minutes or whatever.

At any rate, these companies literally place their computers closest to the incoming wire into the building in order to save that fraction of a fraction of a second so as to get their trades in first. Weird to think that those sorts of distances - a few feet - could make any difference when dealing with information travelling at the speed of light, but it does.

I have nothing informative to add to Expano Mapcase's post, which actually, you know, answers the OP.

Shagnasty
03-25-2011, 06:05 PM
You have to be super-careful about assumptions that this actually works consistently to begin with for reasons that Exapno Mapcase gave and others. There are some very smart people who are not only world-class crackpots but crooked or a combination of the two especially in the finance world. These types of operations are secretive by design so it is hard to evaluate what they are actually doing and accomplishing. They likely aren't all losing money but that doesn't mean much. A rising tide lifts all boats so it isn't hard to show gains doing most things when the market is going up.

That isn't proof that a specific strategy actually works. The control group is a buy and hold index fund and even the vast majority of mutual funds lose out to those yet it is a huge industry with lots of highly paid and educated people who do nothing but make less money for people than they could have by essentially parking their money in diversified investments and doing nothing else which is essentially the same thing as losing money for a lot of additional work.

I don't know if the speed of light trades actually work better than doing essentially nothing but that is the beauty of it. The people that build such things don't do it so that they can make money only off of their own money. They do it to impress new investors with super-technical talk in a hush, hush kind of way, invest their money, hopefully show some gains, and skim off the top of the investments. If it really worked that well, they wouldn't want or need additional investors. A small group could invest their own money and watch it grow incredibly quickly through trades happening that fast. A computer trading that quickly should make a person a billionaire in short order if the algorithms were right even 51% of the time and that hasn't happened so there is something else to it.

Derleth
03-25-2011, 06:17 PM
All this puts me in mind of Admiral Grace Hopper's famous nanosecond (http://www.flickr.com/photos/shinythings/154816771/), which is a 30 cm (11.8 in) length of copper wire. The point is that it takes an electromagnetic signal (which goes at the speed of light in the medium) a nanosecond, or one billionth of one second, to travel 30 cm (11.8 in) in a vacuum (it goes slower along copper wire, or through glass, or through air).

China Guy
03-25-2011, 08:01 PM
It's not about acting on market information. It is purely arbitrage. A mis pricing allows one to buy and sell simultaneously and lock in the difference minus trading costs.

In the old days, traders used to do this in the trading pit. Now it's computers closest to the exchange feed.

Tied in with this is the automatic trading. In the case of that big intra day fall, a wrong trade was entered. I forget the details but it was big enough to push down the market xx amount, which triggered automatic sell programs from non-arbitrage automated computer selling, which triggered a market sell off, which in turn triggered more selling until the market circuit breakers kicked in.

During the mandatory no trading period, the original bad trade was discovered and reversed. This IIRC also triggered automated buy programs and the whole thing went in reverse. The bad thing is that the market whipsaw really hammered some real end investor trades as collateral damage.

I remember watching the Hang Seng Index the day that Soc Gen announced Jerome Kerviel's fraud and liquidated the positions. It was a full trading day of massive market swings for big losses to big gains several times throughout the day. It was almost all computer generated programmatic trading.

Chronos
03-25-2011, 10:31 PM
The way it makes money is by being a broker and charging a commission from the people attempting it.

China Guy
03-26-2011, 09:59 AM
Chronos, you're mixing up different types of trading.

High frequency trading is done for a proprietary account and usually by one of the big banks. It really is just a fast sophisticated computer program to capture momentary mispricings in the market. The act of making that trade usually removes the mispricing.

A real simple example. if Apple is trading for $200 on Nasdaq but for $199.95 on the DAX. Buy low sell high. Buy Apple on Dax, sell it simultaneously on Nasdaq and make a nickle minus costs.

More complicated example. A long call and a short put with the same expiration are the equivalent of a long future. If the price of buying a long call, selling a short put is higher than selling a short future and holding it until expiration, then put on this trade and it is a risk free arbitrage until you sell it at expiration and take the profit minus costs.

High frequency trading does this type of thing all day long.

Crusoe
03-26-2011, 10:13 AM
I used to work for a company that did algo trading, but hadn't come across the term HFT before. Is HFT a subset/superset of algo trading?

hogarth
03-26-2011, 10:40 AM
I used to work for a company that did algo trading, but hadn't come across the term HFT before. Is HFT a subset/superset of algo trading?
Algorithmic trading is a broader term. For instance, breaking a large block of shares into smaller trades (to avoid having a big jump in the price from a large sale) can be done algorithmically, but it wouldn't count as high-frequency trading.

septimus
03-26-2011, 10:51 AM
High frequency trading is done for a proprietary account and usually by one of the big banks. It really is just a fast sophisticated computer program to capture momentary mispricings in the market. The act of making that trade usually removes the mispricing.

This is essentially correct. (Surprised that the earlier responses were way off-base, as we've discussed this before here.)

Worth adding is that the "momentary mispricing" is usually not directly due to some real-world event. (After all, we're talking milliseconds, too short for a CNN news reporter to utter a word, let alone a sentence!) The momentary mispricing is due to recent uncleared orders. In fact, just as in other algorithm-driven games, a successful program will often be one that can predict the strategies of other algorithmic traders.

Several months ago I bookmarked an interesting Youtube item, in which someone claimed to observe illegal manipulations by a high-speed trader. Checking now, I see that Youtube item has become password-protected, but give the link in case anyone can "hack" it:

http://www.youtube.com/watch?v=xOr5suFJ6-k&feature=player_embedded

(If you Google that URL, you'll find the SDMB thread that mentioned it. I won't link to that thread as OP got mostly incorrect responses. :D )

If this is the same illegal manipulation I read of, a high-speed trader sent many limit orders to influence the competition's algorithms, then canceled them before they executed. It is supposedly illegal to place an order with no intention of execution.

Chronos
03-26-2011, 12:39 PM
Worth adding is that the "momentary mispricing" is usually not directly due to some real-world event. (After all, we're talking milliseconds, too short for a CNN news reporter to utter a word, let alone a sentence!) The momentary mispricing is due to recent uncleared orders. In fact, just as in other algorithm-driven games, a successful program will often be one that can predict the strategies of other algorithmic traders.
In other words, it's not a way to make money, it's a way to play paper-rock-scissors for high stakes.

LSLGuy
03-26-2011, 01:05 PM
If by "make money" you mean create value for the world at large, then no.

But if by "make money" you mean create real pocketable spendable revenue for the I-bank(s) playing the game, then yes, it does make money. Rather a lot of it.

septimus
03-26-2011, 01:58 PM
In other words, it's not a way to make money, it's a way to play paper-rock-scissors for high stakes.

No. :D These are not misguided individuals who gave up their day jobs after watching too much Cable TV. These are sophisticated Wall Street firms with sophisticated software and big money, making more money. The profits they make come in part from other big traders operating at slower speed. Those slower traders may also be profitable: Ever wonder why when you buy 100 shares of XYZ for $66.15, your friend who sold his shares at the same moment only got $66.13 ?

Chronos
03-26-2011, 03:41 PM
Ever wonder why when you buy 100 shares of XYZ for $66.15, your friend who sold his shares at the same moment only got $66.13 ? Because some middleman is taking a commission, of course.

China Guy
03-26-2011, 06:36 PM
Because some middleman is taking a commission, of course.Chronos - you're missing something but I'm not sure what.

High frequency trading is done solely by computers to capture mispricings. The mispricings are on the same instrument (Eg Apple) traded on different exchanges. Or the mispricings are one equity derivative combinations that create synthetic stock. I gave an example earlier of the most basic synthetic stock and that is long call/short put. This used to be done by pit traders (eg people) certainly through the 1990's but I've been out of the biz for a while and not sure when this was automated. I'm guessing, and someone still in the biz can chime in, but I'm almost certain that the really basic synthetic options have been taken over by the high frequency trading. This is done in when hundreths of a second mean the difference between making money or not doing the trade. This is purely a speed game.

Algorhymic trading is not high frequency but is computer trading the layer after the high frequency is done. These are sophisticated models that do the trading but it's not necessarily a risk free arbitrage. The is probability trading based on complex models. Eg, 80% of the time the trade should make money and 20% lose money. For example, maybe the model sees that the Apple and Google shares trade in a correlated fashion, and when the correlation gets too wide, then it buys the "cheaper" one and sells the "expensive" one.

I banks also do cross trades. As in, you Mr retail place an order to buy 100 share of Apple. The ibank is trying to sell a million shares of Apple. Before they sell the million shares in the exchange, the ibank also fills your order (takes your commission and does not pay a commission to the exchange).

There are still day traders that make money but they take on much higher risk and pay higher commissions. What they do is not a risk free arbitrage.

Note: often you see the term hedge traders or arbitrage traders, but usually these are not 100% (risk free) hedged or arbitraged.

IMO, high frequency trading is not a good thing and there are not enough circuit breakers in the system.

So, to reiterate, high frequency trading is designed to be a 100% risk free arbitrage. It is not a gamble or based on a model, it is based on being able to put on a trade this mili-second and know exactly what the profit is regardless of what the share price or market does after the trade is executed.

Hope the above kinda makes sense.

IAmNotSpartacus
03-27-2011, 12:15 AM
Arbitrage trading across markets is only one part of high frequency trading. "High frequency trading" is a bit of a generalized term for a number of "strategies" that involve buying and selling stock rapidly.

Other ways of making money by high frequency trading include playing both sides of an ask/bid, especially on high volume stocks with some volatility in their spread; predicting big money movement by analyzing pending and incoming limit orders; earning rebates from exchanges by offering liquidity to trades on that same exchange, avoiding cross-exchange fees for the trade.

There are other ways to do it, too, with HFT being such a vague and generalized term.

septimus
08-02-2012, 10:36 AM
IMO, high frequency trading is not a good thing and there are not enough circuit breakers in the system.

With the latest snafu (http://www.reuters.com/article/2012/08/01/us-usa-nyse-tradinghalts-idUSBRE8701BN20120801), more major investors start to agree with that opinion (http://www.cnbc.com/id/48443271):

This algorithmic trading is kind of out of control. It seriously hurts investor confidence.

Chimera
08-02-2012, 10:45 AM
The people that build such things don't do it so that they can make money only off of their own money. They do it to impress new investors with super-technical talk in a hush, hush kind of way, invest their money, hopefully show some gains, and skim off the top of the investments. If it really worked that well, they wouldn't want or need additional investors. A small group could invest their own money and watch it grow incredibly quickly through trades happening that fast. A computer trading that quickly should make a person a billionaire in short order if the algorithms were right even 51% of the time and that hasn't happened so there is something else to it.

This.

NO ONE develops an incredibly foolproof of turning thousands into billions and then says "You know, I'm help other people use this too". No, people with a real method keep it to themselves.

Anyone claiming to have such methods and selling them, via brokers or Seminars or books, is a scam artist whose "method" of making money is to sell you the phony method.

Remember all those "How to make millions on mail order" ads you used to see? (if you're old enough) Send us $5.99 and we'll tell you how? Well, you send me $5.99 and I tell you to find a couple of thousand suckers willing to send YOU $5.99.

hogarth
08-02-2012, 10:54 AM
This.
I can assure you that large investment banks that develop proprietary HFT software do, in fact, build those system to make money on their own behalf.

NO ONE develops an incredibly foolproof of turning thousands into billions and then says "You know, I'm help other people use this too".
What examples do you have of someone doing this?

allotrope
08-02-2012, 11:08 AM
You're all wrong. 99% of all HFT trades are canceled. THAT's why it's called high frequency trading.

Now since you're all so smart, why don't you take a stab at why that might be.

This should be amusing.

Derleth
08-03-2012, 12:41 AM
99% of all HFT trades are canceled.Who says?

grude
08-03-2012, 12:54 AM
How do these trades make financial sense? It seems there is an intrinsic cost in a trade with fees, so how does that not cancel out any profit when we're talking about fractions of cents?

septimus
08-03-2012, 01:09 AM
How do these trades make financial sense? It seems there is an intrinsic cost in a trade with fees, so how does that not cancel out any profit when we're talking about fractions of cents?

What fees? Are you referring to the 0.0034% transaction tax currently imposed by the U.S. government? That works out to $1 for every $30,000 trade -- almost insignificant (though good enough to add billions of dollars annually to U.S. Treasury coffers).

Some Congressmen want to impose a tax of 0.03% (http://www.huffingtonpost.com/2011/11/07/wall-street-transaction-tax-revenue_n_1080493.html). Anyone who claims such a tiny tax would have a severe adverse effect on small investors has been drinking the wrong Kool-Aid.

A minuscule tax on financial transactions proposed by congressional Democrats would raise more than $350 billion over the next nine years, according to an analysis by the Joint Tax Committee, a nonpartisan congressional scorekeeping panel.

The analysis was sent Monday to the offices of Sen. Tom Harkin (D-Iowa) and Rep. Peter DeFazio (D-Ore.), the lawmakers who proposed the tax, and provided to The Huffington Post.

The Wall Street Trading and Speculators Tax Act would impose a tax of 0.03 percent on financial transactions, meaning that longterm investors would barely notice it, but traders who move rapidly in and out of positions would feel its sting and, the authors hope, reduce the volume of their speculation in response.

Of course the tax would not apply to canceled orders. While it may be true that most HFT orders are not executed, it is illegal (but hard to prosecute) for a HFT order to make orders he intends to cancel quickly.

allotrope
08-03-2012, 01:16 AM
Septimus: :golf clap:

septimus
08-03-2012, 01:21 AM
... it is illegal (but hard to prosecute) for a HFT order to make orders he intends to cancel quickly.

Evidence has been presented that such orders have been used to manipulate prices. Since a typical small investor will lose only a few pennies on such a manipulation, these are often presented as "victimless crimes" but I disagree.

I read that Wednesday's trades will be unwound if the price differed by at least 30% from the opening. I'd be unhappy if I bought 29% too high. Those claiming these shenanigans do not affect small investors are, again, drinking Kool-Aid.

septimus
08-03-2012, 01:33 AM
Septimus: :golf clap:

Do you have an intelligible contribution to make? :stifles a fart:

allotrope
08-03-2012, 01:37 AM
Do you have an intelligible contribution to make? :stifles a fart:
Nope, you're doing fine - in sharp contrast to your predecessors I might add.

grude
08-03-2012, 01:40 AM
What fees? Are you referring to the 0.0034% transaction tax currently imposed by the U.S. government? That works out to $1 for every $30,000 trade -- almost insignificant (though good enough to add billions of dollars annually to U.S. Treasury coffers)..

Aren't there fees for even using the trading systems? Or do these big traders operate in a different arena?

si_blakely
08-03-2012, 02:13 AM
Aren't there fees for even using the trading systems? Or do these big traders operate in a different arena?Yes, there are fees. While it may be propriety trading (thus no brokerage fees), the companies engaged in HFT are paying for access to the exchanges, and for the high speed links between systems (including massively expensive low-latency international fibre). They pay extra for the server racks in the optimal locations and preferential access to trading data systems. They also pay big money to the best maths and engineering graduates they can buy to write these systems. And (sometimes) they pay massive compensation when it all goes wrong (e.g Knight Capital).

They make fractions per share, and go for volume (both number of shares and number of trades). And the battle is to be faster or to game the system - Core Wars played out on the world financial markets. Risky business, indeed.

Si