There are a few changes to this post in italics. These come from the comments.
Goldman Sachs made 5.7 billion dollars of trading revenue in the last quarter. That run rate (over 22 billion per annum) is almost as much as the pre-crisis peak.
$22 billion per annum is roughly $200 per year per household in the United States.
If it is someone’s trading revenue it presumably comes out of someone else’s pocket so measuring it per household is appropriate.
The trading revenue of “Wall Street” major investment banks (including Barclays, the trading parts of Citibank and similar entities) peaked at over $500 per household in the Western world.
Revenue like this is usually paid for a service. Ultimately I thought the service was intermediation between savers in China, Japan and the Middle East (who want Treasuries) and dis-savers in the Anglo countries (who want to fund exotic credit card debt and mortgages). That remains the only service that looks large enough to justify that sort of revenue. [The real service having been finding suckers such as municipalities and insurance companies to hold the toxic waste such as CDO squared resecuritisation paper.]
That said, given almost nobody knows how to make $22 billion per annum trading and jealousy is a common trait, conspiracy theories abound. The current conspiracy theory is that this money comes from front-running clients in the market with very rapid trading. The New York Times recently promoted this view.
The idea is that by knowing client orders you can extract profits. Computers fleece clients by forcing clients to pay more when they buy and to receive less when they sell.
And it is clear this happens. We trade electronically at our fund. We were recently trading in a stock with a large spread. I have changed the numbers so as not to identify the stock – but the ratios are about right. The bid was about 129.50, offer was about 131.50. We did not want to cross the spread – so when we bid for the stock we bid $129.55. Within a second a computer (possibly at our own broker but it makes no difference which broker) bid $129.60 for a few hundred shares. We fiddled for a while changing our bid and watching the bot change theirs. We would have loved to think we were frustrating the computer – but alas it was just a machine – and we were people up late at night
Actually obtaining the stock required that we paid up – and when we did so it was probably a computer that sold the stock to us.
Inevitably we cross spreads and the computer earns spreads.
The computers make even more consistent profits with high volume low spread stocks. If you are buying or selling Citigroup it is almost certain that when you buy you will pay the offer price and when you sell you will receive the bid price. They are only 1 cent different – but in almost all cases it will be a computer that traded with you – and the computer will – through owning the “order flow” be getting the better end of that deal.
That said – these profits can’t add up to sufficient to explain Goldman’s trading profit.Interactive Brokers is (by far) the most electronic and lowest cost broking platform in the world.We use it extensively as do many others. Interactive Brokers has a 12 percent market share in option market making globally and probably a 10 percent share in all market making. Trading revenue was about 220 million. Moreover in the conference call the CEO/Founder (Thomas Peterffy) thought the influx of competition in the area had reduced market maker margins very substantially.
Anyway if 10 percent of global stock volume provides 220 million dollars revenue per quarter then there is no way that a substantial proportion of Goldman’s trading profit can come from high frequency trading. The numbers do not work.
When the New York Times quotes William Donaldson (a former CEO of the New York Stock Exchange) as that high frequency trading “is where all the money is getting made” they are quoting bunk – and they should know it.
This is a plea. Can we have a dispassionate and accurate view of where the (vast) trading profits of Wall Street in general (and Goldman Sachs in particular) come from? The last big boom in trading profits was followed by a bust which came at huge social costs. [Look what happened to Lehman.]
We cannot understand the risks “Wall Street” is taking and hence the economic downside if it all turns pear shaped, and the appropriate regulatory structure, unless we know what is happening.
Mindless articles such as the recent New York Times one – grossly inconsistent with facts are less than helpful. They are distracting.
One comment thought that this was algorithmic trading - someone really wanting to buy the stock - and bidding above our bid when we showed our bid. I wish that were true. If it were then if we were buying very illiquid wide spread things the bot would still be there. It is always there - even when buying defaulted debt that trades once per month. We simply ALWAYS find the bot.
48 comments:
If you wanted to buy, why not post a small offer below the best offer, watch the bot come in with a lower offer, repeat until the bot offer is is around mid and then lift it ?
you trade with a broker that you assume is front running your orders? really?
Most orders are executed by algos these days, but that does not mean they represent 'high frequency' trading. When portfolio managers place an order in the market it is likely to be worked by an algo, but they are clearly not trying to make the spread. Ditto, hedge funds, prop desk, etc. The algo that was pipping your bid on that stock was very likely representing someone who genuinely wanted to own the stock and was just trying to avoid paying the spread, same as you were.
I have tried crossing an option at an internet account from one broker to another. (Both mine - one in the family trust one personal.)
Guess what - it was crossed - but the broker took one side.
I could cross bang in the middle of the spread.
The bots are PERVASIVE. I have done the experiment with the bots and electronic trading over a dozen times. Every time there is a bot.
The bots are also there in penny stocks. Defaulted stuff.
For instance I was buying the defaulted debt of a recent bankruptcy at less than 10c in the dollar but with a 20% spread.
A bot was there.
Yes I assume that the bot will ALWAYS be there.
I do not care if it is my broker's bot or someone else's broker bot. Makes no difference to me.
John
I think your article is correct, except for the inference about Interactive Brokers having 10% market share and therefore 10% of high frequency revenues. Interactive Brokers has a fundamentally different business model: a large portion of their revenue is from commissions on your trades, not from being the person trading with you on the other side of the market.
They do take the other side of your trades in the electronic equity options market (though they aren't the largest player in that, my understanding is that Citadel is bigger). But in the cash equity market, the bots that you are trading with are more likely to be from Citadel, DE Shaw, Renaissance, Susquehanna, perhaps Goldman, and others, instead of IB. So I don't think you can assume IB is 10% of the market and makes $220mm. I worked at one of these funds before and the high frequency profits are an order of magnitude larger than that -- I'm not sure precisely because that was not my group, but $1b annually is conservative, I think. Last year during volatile markets they made substantially more than that.
Even so, it doesn't get you to Goldman's $22b annual number...
I know interactive take the opposite side of the options. I have experimented at length.
--
The 12% market share in electronic options globally is Interactive's own estimate. They also estimate that their share is falling.
Given they estimate a good but falling share I think the stuff is a good estimate.
J
John; Another great post, totally agree with you. The Machines are always there, I see it every time I put in an order.
I hate the fact that somone gets to see the market before I do, there is something fundamentally wrong with this. But I think a question that needs asking is does this front running narrow spreads in the market in general (a net benefit)?
S.B.H.S
First of all, is there some sort of great moral justification that makes it OK for you to make the spread but wrong for a broker's algo desk to make the spread?
I think that you are starting off taking umbrage at this. I don't understand your moral frame of reference on this. You're going to hold the stock longer than the algo system? You use a funamental analysis vs the computer using technical analysis?
This just makes it sound like your execution is sloppy. Even if your market is not so liquid, some brokerages will work with you to tune their engines to handle niche's like this. These system compete very hard for market share and will work it.
You have to also remember that someone else is on offer and is providing that liquidity that you wish to take. Remember how spreads widened after the SS restrictions. That took many of the algo systems out of the market for the first week and everyone rushed to sort out the new rules.
As far as overall bank profits, you have to remember that these banks operate in so many environments. They are providing managed swaped funding to many firms like yours, they are borrowing, servicing, settling, etc. And a lot of this goes under the "trading" category, even though one might argue that the revenues should live in another place, like the client services side. (Those fights in reality do happen a lot).
There was no moral umbrage here - and I assure you our execution (except for one notably fat-thumb exercise) has been good for the entire history of the firm. The fat thumb cost about 70bps of return.
I am not trying to moralise about the bots. They annoy me. They are there. They are part of the air I breathe.
Do they make as much profit as zero hedge and other think? No way. They are not the explanation for anything.
J
Josh. I am also a little offended that you would conclude from my careful experimentation that my execution is sloppy.
It is not.
John
I'm sure it's annoying, but it doesn't preclude you from working a good old limit order. You have a price you want to buy at, and that's that. Price takers pay liquidity premiums, price makers earn liquidity premiums. If you want the liquidity spread, setup an algo that will buy / sell in these securities and beat the bots at their own game
John, I'm sorry, maybe I didn't pick up the tone right. There have been a lot of articles recently accusing hi-frequency trading as, essentially, being a negative force that should probably be banned, regulated, etc. And that any profit that they make is untoward. In many of the screeds that you read it implies that my execution is fine.
If not for the bot I could put out there my order to buy 100k of whatever illiquid name and the market wouldn't move away, but instead a seller (who of course had held the stock for many years and is only selling to pay his child's college expenses) would take the other side.
I'm just trying to make the point that manual execution isn't really ideal for these markets if you are doing size. I think you will see with the broker systems that they smooth out your execution flow quite a bit and interact with all of the other market centers relatively efficiently.
I was having a conversation with a block trader at a brokerage recently and I asked him what they do for the customer. He told me that he was "really good" at setting the parameters on the algo system and then sending the customer orders down there. I'm not sure if that's the most viable job two years down the road....
The market changes and I'm sure will change again...
When I see people complaining about automated trading they are just really avoiding the fact that their execution is soft.
As far as where trading profits come from, my understanding is that you are correct and that only a small portion of them is from hi-frequency trading. Also, you need to remember that the term itself has different connotations. A lot of people use it to mean a rotating start arb book that trades 10 times a day.... Others mean true market making systems.
Thanks Josh. I was of course just pointing out that high frequency trading could not explain anything like all the profits at Goldies - or even a small fraction.
Indeed it is just not a starter.
It annoys me (except maybe that it ensures options are available to me). But it is not a big problem.
J
When I was trading as locals on the Sydney SFE floor, BUYing at BID and SELLing at OFFER is a dead sure way of making small but regular money which was main money we locals earned.
We called the practice chipping away gradually. You need to be on the trading floor though. The prop traders working for all institutional brokers did it then and still doing it with their high speed computers, I reckoned.
Great post. I think the idea that high-frequency trading is a major profit center for the likes of Goldman is just an artifact of the coincidence of the HFT-software-theft story coming out around the same time as the return-to-profit story. People automatically connected the two. It's hard to shake these stories, since no one really seems to know how Goldman makes its money, and the story of HFT is shadowy enough to serve in the imagination as an explanation for lots of things.
John,
I think you should learn to love the robots. If it were not for them spreads many liquid stocks would not be at 1 cent, but rather 3-4 cents. The depth would also be rather bad. I can delta hedge 2+ million shares of a large basket of stocks in 15 minutes largely because these guys are around to provide liquidity. It costs me on average about 0.5 to 1 bps of slippage (benchmarked on arrival price). If spreads were 3-4 cents and the market depth the same as before these guys arrived, I don't think it'd be possible for me to achieve that. True, I can probably do better by being smarter about my delta hedging, but I don't think I have the scale necessary to make that costly investment in research and development worth while.
John,
You might have hit a hidden order in your options crossing experience. However that said, brokers like IBKR do have right of first refusal on your option flow. In fact many prop desks pay retail brokers to gain access to flow. They are only obligated to fill you at the best price or better.
If you wanted to buy, why not post a small offer below the best offer, watch the bot come in with a lower offer, repeat until the bot offer is is around mid and then lift it?
I have done that in a futures market --- I bid, the bot bid higher, and I forced it to move it's bid up (which was much larger volume than mine) by the equivalent of about 3.5bp of yield. Not bad! Wish I'd been able to repeat it, though...
Finally someone with a sensible look at an issue, rather than the screaming and kicking Zero Hedge does all the time.
As for the 22 billion, a lot comes from credit and interest rate trading. Bid Offer spreads in both have widened dramatically, there has been a record setting amount of corporate issuing, next to record setting issuing of treasuries and lots of MBS hedging required due to volatile markets. Don`t forget the pension funds scrambling to regain some structure. All of that require swap execution and GS is a swap powerhouse. There is lots and lots of money to earn in those two markets if you have the system worked out right. Add to that the steep yield curve and the cheap overnight funding and you have a successful mix of income creating trading desks.
You are correct. High frequency trading is probably a small portion of Goldman's profit.
I think most of their trading profit is generated by large, fixed income and equity trades done with large institutions over the phone.
IBKR does 11% of the listed option trades, not equity trades.
US option exchange rules allow IBKR to participate in trading with your order as long as that is done at or better than the best displayed price. We frequently improve upon the best price as that may give us the trade and put money in our customers' pockets.
We do NOT trade with your stock orders.
IBKR has an algorithm called accumulate/distribute that will win over HFTs, please try it.
Aren't you assuming that the high-frequency guys aren't using leverage? While IB may not be, as they are merely collecting spreads, a large HFT like Goldman or Citadel is almost certainly using leverage to boost the revenues/returns on their very short-term positions.
That alone wouldn't account for the full profit amount, but it would explain why they might earn more per point of market share than IB.
John, Great Post.
For some time at least 2 categories of robots have existed:
1. Automated Trading Systems which work large orders and attempt an optimal tradeoff between price and execution speed
2. Market Making algorithms run by Renaissance and others which earn a part of the spread with the benefit of reducing spreads.
I wonder how different these "new" HFT robots really are. They seem more like MM algos in that they don't have even a medium-term economic interest in owning the shares. HFT algos are described as having more of a focus on predicting short-term price moves but that's something that MM algos and some better ATS's have also been doing.
So is something truly new happening here or not? Unfortunately neither the NYT or ZeroHedge are helping answer that question.
Also: flash orders are just an optional order type a trader can choose. Normal orders don't behave this way. And is this really new either? Sounds like fillOrKill or ImmediateOrCancel orders which are definitely not new.
My guess is that this whole issue is really just the next step by those who are gaming dark pools and only affects dark pool participants. If it becomes a problem dark pool operators will have to fix it or the traders will move elsewhere.
For reference - I am fairly sure that the Thomas Peterffy who commented is the real pioneer of electronic trading globally.
J
can we clarify a key point: if you bid for stock at 130.50, or whatever, and someone (a bot, for example) improves that bid to 130.60 that IS NOT FRONTRUNNING. It's called trading. It's called making use of publicly available information. The bots these days do it faster and better than ever - that's not a crime. No bot is forcing you to buy stock higher.
The first thing we need to do is educate people about the fact that this is not frontrunning - which is blatantly illegal, indefensible, and has a horrible connotation, thus stirring up rage with the public for no reason.
Joe Saluzzi from Themis - the architect of this whole recent anti HFT rage is the ultimate hypocrite. His JOB is to make use of his skills and experience trading to try to determine where stocks are going. Now, computers do it faster and better than him so he thinks it's unfair. See: IRONY.
If the volume created is phony, then the market is literally being held up by churn. And they can dechurn any time they want: http://hubpages.com/hub/Why-Goldman-Sachs-Is-Committing-Treason
Maybe this is naive, but didn't the specialists get information first once upon a time? I'm sure they did some front-running from time to time. Is the issue more a matter of liquidity--that the bids will get pulled when something goes wrong. . .
Thanks for the reasonable discussion of this topic.
GS are the market in many of the exotics. To paraphrase Lewis in 'Liars Poker': if you are the market and there's sellers "oozing desperation" then you as the market 'rip their eyes out'.
Have no firm numbers on any of this but if the OTC markets that I traded back in the eighties are any indication, vast profits can be made on many of these illiquid instruments.
Was also lead to understand, from the (UK) horses mouth, that they were 'raking it in' and that was nearly a year ago. No update since then but their recent numbers do not come as such a huge surprise.
Simple profit explanation: maturity of short term commercial debt purchased when under distress in Q4 2008?
"Phony" volume is not equal to phony prices. If the algo volume were to go away prices would stay fairly constant but spreads would widen and volatility would increase. It's no coincidence that pricing in eighths and 25 cent spreads have disappeared as algo volume increased.
John, I think you are completely missing zerohedge's point. They've never said that HFT is an exlanation for obscene profits. What they say is that:
- HFT is a stealth tax on investors as they have to either enter the arms race, or accept to be ripped off.
- HFT distorts the market by creating an upward pressure on prices on the way up, which can turn into a nasty downward pressure on the way down
- HFT hides and drives out real liquidity, replacing it with a shallow version, which can end in a nasty liquidity hole, a drastic downward price movement.
For HFS being a stealth tax on investors - I am inclined to agree. The proportion of time at which I am forced to cross the spread has increased.
The BOT forces me to cross the spread. That is a stealth tax on me.
It would also apply to the selling.
Zero hedge is right that it is a stealth tax on investors. My point is that it is not large - and it is in no sense explanatory of Goldman's profits or at the root of any evil in Goldman Sachs.
HFT is a bogey issue - to be rated along with garbage men who spill a little rubbish as they collect the garbage - and not to be central in any financial reform.
As for real liquidity and phoney liquidity - I have never noticed that real liquidity exists in any panic. Pockets of lack of liquidity have been with us much longer than HFT.
My guess is that HFT will provide a small amount of continuous liquidity. But the 1000 shares that get in front of any of my orders - they are just an annoyance.
Part of the air that I breathe.
John
And where is the benefit of such computerized trading?
It seems all it does is to make the prices go up, even though there is no justification in reality for prices going up.
Computers feed from data from within the system and are unable to account for externalities they have not been programmed for, so where is the benefit here?
If I did figure that out right than adding liquidity (i.e. buying) is favored by exchanges by giving you a rebate, whereas removing liquidity (i.e. selling) is disfavored by exchanges by not making it more expensive then adding liquidity.
So the only thing this does is to bias the market go up, because buying is more attractive than selling.
How can this work if it is not a pyramid scheme? At one point someone will have to inject massive amounts of money because the prices paid at the exchange have no relation to the actual value of the item that is traded.
If no money is injected, the thing crashes and burns.
This has all the hallmarks of a pyramid scheme. Works only as long as new capital is coming in.
Who cares!
If GE stock is trading at $25 and you think it's worth $50 then you buy it.
If a penny or two from the tradw with whatever market maker computer is lost, and that penny is so important, then you shouldn't be in this business.
John, regarding your original trading example, I think it's equally likely that the bot represented another buyer who (like you) wanted to be a bit more aggressive than just joining the bid and waiting for other participants to cross. This is bound to raise the bid until one of you cries uncle or gets filled. If you had been less aggressive and just joined the bid chances are the bid wouldn't have changed.
On flash orders: people seem to believe that every order they place is flashed to the HFT players. Not true. It is only flashed if you want it to be flashed. Those who choose the flash do so because they believe it will aid in their search for liquidity. They have to balance that against the possibility that it might tip their hand to HFT or human traders who would bid the price up. But your normal (ie non-flash) order is seen by the whole marketplace at the same time.
I didn't state my point clearly enough: the people who are allegedly harmed are those placing flash orders that are being "front-run" by HFT. But if they are really harmed why would they continue to use flash orders?
Another ironic twist to these arguments is that a trader who is not privy to the flash orders is simply deprived of particpating in the alleged raping of the original institutional order.
I would think that the forced selling in the credit markets by AIG found that there were few market makers and that GS et al took tem to the proverbial cleaners...
I haven't read the filings, but my guess would be that GS buckets earnings from SSG and other desks that are made by risking the firm's capital in a large, long-term sense with 'trading' profits.
If SSG buys a string of Japanese golf courses or a book of South Korean car loans and sells them a year later for 5x, GS probably calls this a 'trading' profit, as if risking the firm's capital in this way is the same as using the firm's capital to buy a liquid name from a client before selling it to another client/firm.
Similarly, FICC desk MDs/partners can take large proprietary risks and generate profits that are not strictly linked to trading on behalf of clients. They are just straight market bets. I think if you manage a book of say IR derivatives, its a little tough to separate hedging activity ('we are too long these vol surface points') with proprietary trading activity ('lets get a little shorter these vol surface points').
Since desks like SSG essentially run like asset management firms that invest GS's considerable balance sheet opportunistically, they can generate 'trading' revenues that are in scale with an asset manager running a fund that large.
Again-- this is an uninformed guess, but hey, what would blog comments be without uninformed guessing from the peanut gallery?
One explanation for Goldman's profit was also mentioned by Zero Hedge (and Felix Salmon): VaR. In Q2, Goldman's VaR hit an all-time record of $245m. On top of this, Goldman received a special exemption from the SEC allowing it to calculate VaR in a proprietary fashion that likely understates the figure relative to the old method.
Some posts on the subject are here:
http://zerohedge.blogspot.com/2009/07/as-if-anyone-expected-less-than-one-of.html
http://feeds.felixsalmon.com/~r/felix-all/~3/htes607hyjI/
http://www.zerohedge.com/article/why-does-goldman-need-fed-exemption-var-calculations
I traded illiquid stuff before everything was electronic. If I was the one standing there all day long giving a bid and offer and somone asked for a market and then jumped in ahead of me then my response was usually to jump back ahead of them. This was especially the case if I was short.
A lot of this complaining about hft is by people that really do't understand the how much of a poker game it was before computers dominated. They've made the game faster and more complicated but it it is not that different from the slower manual contest that existed before.
Hi John,
Interesting thread. The big profits this quarter at the remaining IB/brokers came from a massive widening of spreads on interest rate derivatives, enhanced by an increase in trading volume owing to reduced competition.
The GS algo trading desks probably are obscenely profitable, but nowhere near fixed income and currencies. The code theft furore was about 'event driven' trading software, ie, their computers parse reuters' feed and trade on it before the rest of the human market can read the first sentence. GS also do big stat arb, which is likely behind the volume spikes zero hedge has commented on. These strategies can also be extremely profitable.
I do wonder about the net benefit to anyone who isn't RenTech Citadel DE Shaw or Goldmans of having microsend trading. The liquidity argument is clearly a furphy, so I think all your left with is a scale premium: if you're too small to play high frequency games you pay more. Tax is the wrong word, market asymmetry is closer to the truth.
Keiran
There's competition among the HFTs, and they only make money when they are right. The prices they cause, via their actions, are more accurate with respect to what's actually knowable about order flow. Most of what they process is noise, and picking out the signal from this noise is non-trivial, requires quite a bit of capital and know how to achieve. How does that hurt anyone? Don't you prefer to trade at prices that are more accurate, with smaller spreads? As someone who's traded since 1965, I can assure you that liquidity and costs are massively cheaper for the individual investor, and continue to improve. Tamper ignorantly with this ecosystem at your (and my) peril. Thank you John for a sane analysis of what's really going on. And thank you for the best banking analysis in the blogosphere.
Wow, I'm speechless. In your example, someone else had entered a pegged order. I don't know who you are or what role you have at your firm, but the fact that you encountered an order pegged to the bid (these have been around for many many years) and have concluded that you're being front-run is seriously scary.
Well - front run - who cares... the pegged order is annoying.
Its part of the air I breath.
And - yes - in some colloquial sense I was being front run. People had my bid - and were getting in front of it.
What is scary about that?
Its profitable for the computer that does it. But not very.
J
Apparently two research groups, the TABB Group and FIXProtocol, have pegged total HFT profits as in the $20B neighborhood. So if GS has 20% of that - we're looking at $4B, which is no longer such small beer. I wonder how they came to such different conclusions from your estimate.
Great point about the pegged orders.
John, you call the pegged order annoying - so are scalpers, specialists and anyone else who wants to buy the stock at the same time I do. In fact the only people who aren't annoying are those who are selling the stock to me (unless they are right that the stock will go down - then they are annoying too :)
I loved your post because it called out that the conspiracy theorists have neither facts nor logic on their side. But I'm concerned about your loosening of the def'n of front-running. If I tell my broker my size intention and they abuse that trust by trading in front of me or tipping off friends, *that's* front-running and thankfully it's illegal.
But if I place an order to buy 500K sh AAPL directly with the exchange everyone in the world can see that order. And I wouldn't be surprised if suddenly a whole lot of liquidity was withdrawn at the ask as they feared my order would drive up prices, a pretty rational fear. If it's bad behavior for buyers to step in front of my order wouldn't the sellers also be acting badly by withdrawing their orders? Or should they be forced to let their orders fill and be instantly underwater by 20 cents or more?
It's all just investors responding to publicly available information which is clearly different than front-running.
I don't think IB is the cheapest broker out there. I use thinkorswim for trading options and their commission rates are cheaper (if only marginally). I also know IB takes the opposite side of your trades from experimenting. They will not fill your order at the best possible bid/offer if your instrument is traded across different exchanges. Their trading program will ALWAYS earn at least a spread from your order.
everyone forgot that goldman sachs also trades international futures and currencities and other things... with the volume in those markets its no suprise they made that much money
Check out the new book "High-Frequency Trading" by Irene Aldridge (here is the link to teh book's Amazon.com page: http://www.amazon.com/exec/obidos/ASIN/0470563761). This is an excellent resource that will help even the smallest investors make money in the HFT space.
At the core, Sunrise Capital is a team of highly accomplished financial professionals with a range of skills and qualifications which enable us to advise confidently and competently on most aspects of portfolio management and associated financial planning issues including specialist areas. kse pk
Post a Comment