You're now on the archive server. Commenting has been disabled.

Goldman's $4 Billion High Frequency Trading Wildcard

Tyler Durden's picture




A recent story in Advanced Trading goes after some of the minutae of High Frequency Trading and provides a glimpse of the total value that HFT may provide to behemoth PT powerhouses such as Goldman Sachs. The article presents a very valuable perspective on just why HFT is so critical these days, especially when cash traders go for 6 hour Starbucks breaks between 10 am and 3:30 pm: "high frequency trading firms, which represent approximately 2% of the 20,000 or so trading firms operating in the US markets today, account for 73% of all US equity trading volume. These companies include proprietary trading desks for a small number of major investment banks, less than 100 of the most sophisticated hedge funds and hundreds of the most secretive prop shops, all of which operate with one thing in mind—capture profit opportunities by being smarter and faster than the closest competition." And as the market keeps going up day in and day out, regardless of the deteriorating economic conditions, it is just these HFT's that determine the overall market direction, usually without fundamental or technical reason. And based on a few lines of code, retail investors get suckered into a rising market that has nothing to do with green shoots or some Chinese firms buying a few hundred extra Intel servers: HFTs are merely perpetuating the same ponzi market mythology last seen in the Madoff case, but on a massively larger scale. When it all blows up, the question is whether the SEC will go after the perpetrators of this pyramid with the same zeal that it pursued Madoff himself. We think not.

The reason for this, as the AT article points out, is that HFT has become the biggest cash cow for Wall Street: "The incredible capabilities offered by technology have given meteoric rise to a relatively few high frequency proprietary trading firms that now wield far greater influence on the markets today than most people recognize." How big of a cash cow:

"Proprietary trading takes in a number of unique strategies, including market making, arbitrage (ETFs, futures, options), pairs trading and others based on the linked trading of more than one asset class, e.g., futures index and cash equities. In fact, TABB Group estimates that annual aggregate profits of low latency arbitrage strategies exceed $21 billion, spread out among the few hundred firms that deploy them."

The $21 billion estimate is smack in the middle of the FIXProtocol estimated $15-$25 billion in revenue that HFT generates. So let's do a back of the envelope calculation: Goldman controls roughly 50-60% of principal program trading on the NYSE, which in turn accounts for 30% of all global program trading. Throw in Goldman's domination of dark pool trading through Sigma X, and one can come up to quite a sizable number - It would not be a stretch to conclude that, through various conduits, Goldman is directly responsible for over 20% of global HFT trading. 20% of $21 billion is over $4 billion a year. As margins on HFT are sky high (it doesn't cost all that much to tweak a few hundred lines of code - and if Sergey Aleynikov is any indication, $400,000/year for VPs in the program is peanuts for a firm like Goldman), this $4 billion likely drops to the bottom line almost dollar for dollar. Let's recall that Goldman's Q2 earnings were $3.44 billion. Does this mean that HFT/PT accounts for roughly 25% of earnings for the firm that is a hedge fund in all but FDIC backing? Zero Hedge would in fact take the over, especially in this environment where M&A fees are a distant memory. We leave this question open, but even if we are off, it would not be by order of magnitude, and would explain why Goldman has thrown the kitchen sink into dominating such NYSE programs as the SLP, and is expending so much energy to dominate dark pools as well.

Going back to the AT article, which provides some additional critical observations, especially with regard to the Aleynikov arrest and his ludicrous $750,000 bail which surpasses that of indicted Ponzier Sir Allen Stanford:

First, strategies that optimize the value of high frequency algorithmic trading are highly dependent on ultra-low latency. The right decisions are based on flowing information into your algorithm microseconds sooner than your competitors. To realize any real benefit from implementing these strategies, a trading firm must have a real-time, colocated, high-frequency trading platform—one where data is collected, and orders are created and routed to execution venues in sub-millisecond times.

Next, since many of these strategies require transacting in more than one asset class and across multiple exchanges often located hundreds of miles apart, i.e., NY to Chicago, that infrastructure will often require roundtrip long haul connectivity between the data centers. [TD:Any real estate professionals out there who can determine just how easy it is to set up a colocated station within millisecond distance of the NYSE, and whether or not Goldman has any rights of first refusal on this real estate optionality? Nothing like a little derivative monopoly to keep potential SLP vendors at bay]

Lastly and most importantly, this code has a limited shelf life, whose competitive advantage is diluted with each second it is outstanding. While a prop desk's high level trading strategy may be consistent over time, the micro-level strategies are constantly altered—growing stale after a few days if not sooner—for two important reasons. Firstly, because high frequency trading depends on ridiculously precise interaction of markets and mathematical correlations between securities, traders need to regularly adjust code—sometimes slightly, sometimes more—to reflect the subtle changes in the dynamic market. The speed and volatility of today's markets is such that the relationships forming the core of our algorithm strategies often change within seconds of our ability to implement the very strategies that exploit them. Secondly, competitive intelligence is so good across all rival trading firms that each is exposed to the increasing susceptibility of their strategies being reverse engineered, turning their most profitable ideas into their most risky. As a result, any firm acquiring the "stolen" code would gain benefit from it for no more than a few days before that firm would need to adjust the code to the dynamic conditions. Since these changes build on themselves, in a matter of weeks that code would look quite different from that which was originally "stolen."

And the conclusion:

There's no doubt that Goldman Sachs, or any other proprietary trading firm, could indeed lose tens of millions of dollars from its proprietary trading if their strategies are stolen—and that is very serious. The competitors that obtain access to these trading secrets could (and would) use it to front run or trade against it, ruining even the most well-planned tactics. This news story contains many very important sub-plots: trading espionage, the necessity for a trading firm to have sophisticated security systems built around its technology, the requirements for risk management, and even the potential for proprietary trading software to be targeted on a wider scale for terrorist activity; but more than anything else it highlights the critical role played by high frequency prop trading in this new market.

This is indeed, a conclusion that Zero Hedge has been pounding the table on for months. It is imperative that Wall Street firms shed much more light into this astronomically profitable yet highly misunderstood and under the radar concept. In the absence of more information, the likelihood that Wall Street firms who dominate order flow and have material unfair advantages over virtually everyone else, should be isolated from trading up to the point where they provide sufficient information to make the market a fair and equal playing field for all investors. Until that moment, investing, trading and speculating is doomed to have the same outcome for the majority of market participants as playing roulette with 35 instances of 00, a much lower fun coefficient and no ability to be comped for your room in light of significant trading losses.




Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.
Fri, 07/17/2009 - 14:21 | Link to Comment Anonymous
Fri, 07/17/2009 - 15:56 | Link to Comment aldousd
aldousd's picture

Whatever this is called, it is not Capitalism. Its a shame that capitalism will take the rap for it though. It already is, even in comments I've seen on this site, where people usually know better.

Fri, 07/17/2009 - 14:23 | Link to Comment Anonymous
Fri, 07/17/2009 - 14:29 | Link to Comment Anonymous
Fri, 07/17/2009 - 14:54 | Link to Comment Anonymous
Fri, 07/17/2009 - 15:12 | Link to Comment lizzy36
lizzy36's picture

FTW

Fri, 07/17/2009 - 14:30 | Link to Comment Anonymous
Fri, 07/17/2009 - 18:26 | Link to Comment Anonymous
Fri, 07/17/2009 - 19:38 | Link to Comment Anonymous
Fri, 07/17/2009 - 20:18 | Link to Comment Anonymous
Fri, 07/17/2009 - 14:36 | Link to Comment Anonymous
Fri, 07/17/2009 - 14:36 | Link to Comment Anonymous
Fri, 07/17/2009 - 14:40 | Link to Comment Anonymous
Fri, 07/17/2009 - 14:46 | Link to Comment Anonymous
Fri, 07/17/2009 - 15:14 | Link to Comment lizzy36
lizzy36's picture

conclude that they ARE front running clients until proved differently.  hence the lack of GS agency numbers every week when NYSE comes out with volume #'s.

Fri, 07/17/2009 - 15:21 | Link to Comment ptoemmes
ptoemmes's picture

Well, if there is, let's hope he/she is not an aging half-blind nearly broke former Cobol programmer on water-cooled 360 platforms like m....

Fri, 07/17/2009 - 15:58 | Link to Comment aldousd
aldousd's picture

fwiw: you can't sniff packets unless someone's given you a span port. now THAT would be a conspiracy.

Fri, 07/17/2009 - 16:09 | Link to Comment Anonymous
Fri, 07/17/2009 - 16:14 | Link to Comment aldousd
aldousd's picture

You're right, I was just responding to the implication above.  I'm not implying that everything is therefore kosher. 

Fri, 07/17/2009 - 19:33 | Link to Comment Anonymous
Fri, 07/17/2009 - 14:47 | Link to Comment Woodshedder
Woodshedder's picture

Anon, HFT firms get paid by the share from the exchanges to churn. Literally, the NYSE is paying Sachs per share to buy/sell.

Part of the problem with this is that GS is literally able to look and see the orders coming in, due to their servers being co-located at the exchanges, and then places their bid/ask to meet the order demand. They earn their fraction of a penny per share for "providing liquidity" and they do not have to worry to much about getting caught holding the bag due to their ability to beat their competitors to the order flow.

Someone feel free to correct me if my interpretation is incorrect.

Fri, 07/17/2009 - 14:54 | Link to Comment Anonymous
Fri, 07/17/2009 - 14:56 | Link to Comment Tyler Durden
Tyler Durden's picture

Do a search for SLP on Zero Hedge. All the answers are here... (but FYI, Goldman is the only SLP)

Fri, 07/17/2009 - 15:21 | Link to Comment Anonymous
Fri, 07/17/2009 - 15:06 | Link to Comment Anonymous
Fri, 07/17/2009 - 15:13 | Link to Comment Anonymous
Fri, 07/17/2009 - 17:58 | Link to Comment JohnKing
JohnKing's picture

Yah, nice relationship:

 

 

But the man promoted from within to replace John Thain as the new CEO of NYSE Euronext (NYX) suddenly finds himself among the elite Wall Street powerbrokers.

 

Niederauer, 48, whose Wall Street pedigree includes 22 years at investment bank Goldman Sachs, has broad experience trading stocks and is a proponent of electronic trading. The ex-chief operating officer, who joined the exchange in April and was viewed as a likely successor, is now responsible for building on the massive modernization and foreign expansion the exchange has undergone under Thain, who was named CEO of Merrill Lynch Wednesday.
Fri, 07/17/2009 - 15:44 | Link to Comment Anonymous
Fri, 07/17/2009 - 16:09 | Link to Comment channel_zero
channel_zero's picture

Part of the problem with this is that GS is literally able to look and see the orders coming in,

This, practically speaking, would be extremely difficult.  Most trading platforms are exceedingly complicated silos.  The notion they could hook into them is a good evil plot twist for a movie, but not a sub-second algo trading application.

 

Algo trading is not illegal.  Even if GS did front-run clients, what's the penalty?  I'm not saying GS isn't evil, because they are.  The heat in this story is the way GS is "paid" to make trades.

Fri, 07/17/2009 - 14:51 | Link to Comment evanesce
evanesce's picture

The speed of light is roughly 186,000 miles per second. That works out to 186 miles per millisecond, or 0.186 miles (982 feet) per microsecond. That's not a great challenge; even allowing for optoelectronic conversions cutting speeds in half, getting several microseconds ahead of the competition strikes me as quite feasible.

Fri, 07/17/2009 - 15:01 | Link to Comment Anonymous
Fri, 07/17/2009 - 15:30 | Link to Comment channel_zero
channel_zero's picture

drop disk i/o from your back of the envelope calculation.  The trading algo would run entirely in RAM.  If there's some DB queries the tables should be entirely in RAM too. 

Network IO and middleware would be the biggest bottleneck.  I just did a cross-country ping over a private point-to-point and got 176ms latency.  A ping inside my plain-vanilla LAN is < 1 ms.

So, it's very probable you could initiate a minimum of 5 trades per 100ms.  I don't know how fast the message would get passed to the trading platform via messaging middleware then get confirmation back.

Fri, 07/17/2009 - 15:41 | Link to Comment Anonymous
Fri, 07/17/2009 - 14:56 | Link to Comment channel_zero
channel_zero's picture

Interesting.  Now it should be clear why erlang is a strong contender for trading algos. 

-Excellent multi-way processor support.  C, C# just don't work like this.

-Ability to update code while the application is running.

As far as connecting hosts via private circuits, this is nothing new or special.  A couple of routers and a t1+LEC you are good to go.  The messaging isn't a whole lot of bandwidth, so a t1 should be sufficient.

The article borders on hyperbole in some instances.  For example, you could easily do nano-second trades based on an algo if the box is approximate to the exchange and is collecting input from that exchange.  But trying to do nano-second trades from NY to Chicago is not probable.  Some hybrids are totally possible.

Again, by itself there's nothing illegal going on. 

Sticking with the systemic risk angle will only get you a "Hey, that zerohedge was right."  AFTER it blows up.  I mean, how long were people warning about leverage?  And what was the response at the time?  "You crazy!!"

 

Fri, 07/17/2009 - 14:59 | Link to Comment kaiserwongze
kaiserwongze's picture

Any traders have this issue recently?  I usually trade in hidden orders with arca and NSDQ, but even if I put out such an order, it seems as if the algorithms, or whoever is market making, can actually see the size.  Doesn't that defeat the entire purpose of hidden orders? 

Fri, 07/17/2009 - 15:00 | Link to Comment Anonymous
Fri, 07/17/2009 - 15:04 | Link to Comment rigger mortice
rigger mortice's picture

would be interested to know how long HFT has been more than half of volume on the NYSE.Anyone know?

Fri, 07/17/2009 - 15:26 | Link to Comment Anonymous
Fri, 07/17/2009 - 15:29 | Link to Comment quant-this
quant-this's picture

Not only is this kind of business so profitable but everyone trading is paying for their transaction costs. These guys pay no commissions and get their exchange fees rebated to them for creating volume which then tricks other suckers into buying or selling based on false movement. 

I guarantee you that if I let any of you trade for me and you had no cost (including cost of capital), it would be very hard for you not to make money. First you would try to make it $0.10 at a time, then $0.1 at a time. If I then told you I would pay you to the closest $0.01, then you would try to make $0.001 at a time and I would pay you $0.009. If the exchanges or SEC wanted, they could kill this business overnight by stopping rebates and forcing anyone with a low latency connection (exchange backbone connected) to be a true market maker like in the early NASDAQ days.

 

I

Fri, 07/17/2009 - 16:51 | Link to Comment Anonymous
Fri, 07/17/2009 - 15:29 | Link to Comment Anonymous
Fri, 07/17/2009 - 15:47 | Link to Comment Anonymous
Fri, 07/17/2009 - 15:48 | Link to Comment Anonymous
Fri, 07/17/2009 - 15:54 | Link to Comment Anonymous
Fri, 07/17/2009 - 16:07 | Link to Comment Anonymous
Fri, 07/17/2009 - 19:47 | Link to Comment Anonymous
Sat, 07/18/2009 - 02:26 | Link to Comment Anonymous
Fri, 07/17/2009 - 15:53 | Link to Comment Anonymous
Fri, 07/17/2009 - 16:10 | Link to Comment Anonymous
Fri, 07/17/2009 - 16:12 | Link to Comment Anonymous
Fri, 07/17/2009 - 16:58 | Link to Comment Anonymous
Fri, 07/17/2009 - 16:11 | Link to Comment Anonymous
Fri, 07/17/2009 - 16:50 | Link to Comment 3Gonads
3Gonads's picture

 Seriously, are you folks just catching on that the bots are skimming the market?

 

 This has been painfully obvious for at least 3 years, and mildly obvious well before that,

 

 Nice of you to finally get religion.

 

 There's nothing you can do , unless you BAN co location and force some minimum display time on ALL quotes.

 

 When it suits them , they can do it. SELECTNET instituted a rule years ago that said you could not cancel a posted bid/offer for at least 30 seconds.   I think a one or 2 second rule for ALL market participants would level the field a bit.  If you are not willing to keep a quote up for 2 seconds, then you are not providing a damn thing in terms of liquidity.  You're an arb or a rebate humper.

Fri, 07/17/2009 - 18:18 | Link to Comment poydras
poydras's picture

I highly doubt the NYSE allows access to order flow.  I suspect they have models based on their client books and orders.  I further suspect they have models that involve the influence of trade behavior of other participants.  That seems a natural progression of the quant world.

Fri, 07/17/2009 - 18:56 | Link to Comment poydras
poydras's picture

One last point...A large instituional fund has to appear like one of these jumbo ships.  They can easily spot their activity and profit accordingly.

The fundamental problem is that most accept the idea of a low value-add intermediary garnering exceptional profits.  It effectively amounts to a tax.  The big problem we have is continually enabling financial institutions to game a simple process.  They go out and create a massive pile of loans based on unrealistic, long-term assumptions and sloppy underwriting while collecting profits on the front end.  The ultimate financial scam is creating such a mess so that you require further compensation to help clean it up.

 

Sat, 07/18/2009 - 02:19 | Link to Comment Anonymous
Fri, 07/17/2009 - 19:29 | Link to Comment Anonymous
Fri, 07/17/2009 - 19:53 | Link to Comment rahbii
rahbii's picture

PBS's Jim Leher is now questioning Goldman Sachs.  Any way to get that soundbite? 

Fri, 07/17/2009 - 20:22 | Link to Comment Anonymous
Sat, 07/18/2009 - 09:01 | Link to Comment Anonymous
Sat, 07/18/2009 - 09:57 | Link to Comment Anonymous
Sat, 07/18/2009 - 13:17 | Link to Comment Anonymous
Sun, 07/19/2009 - 11:40 | Link to Comment Anonymous
Sun, 07/19/2009 - 21:16 | Link to Comment Anonymous
Thu, 09/03/2009 - 13:53 | Link to Comment Anonymous
Do NOT follow this link or you will be banned from the site!