Spread Networks recently spent $300 million to build a fiberoptic cable that will let Wall Street traders shave .003 seconds off their execution times.
What’s the social value of that cable? If you can shave .003 seconds off the time it takes to execute a trade, how much good have you done the world?
Clearly, the full value of the cable resides in its ability to get things done faster. So start with a vast overestimate: Suppose the entire economy is on hold waiting for that trade to be completed. Then, thanks to the cable, we can all get on with our lives .003 seconds sooner and produce an extra .003 seconds worth of output.
In a $15-trillion-a-year economy, that comes to about $1500.
If we assume, more realistically, that just 1/1000 of the economy is hanging fire waiting for this one trade, the social contribution of a .003-second speedup is roughly $1.50. I’m confident it’s even more realistic to replace that 1/1000 with 1/1,000,000 . That gets us down to about an eighth of a cent.
But chances are you’d be willing to pay a hell of a lot more than an eighth of a cent for that extra speed, which is why Spread Networks is willing to pour $300 million into this thing, and why, quite generally, we should expect there to be more invested in such projects than they return in social value.
Economic theory tells us that under quite general hypotheses, the private value of an activity is in synch with its social value. If growing an orange makes you a dollar richer, that’s because growing that orange makes the world a dollar richer. And that’s good, because it encourages people to grow all and only those oranges that are (socially) worth growing.
But while those general hypotheses apply to the markets for oranges, locomotives, haircuts and chocolate chip cookies, they do not apply to high frequency trading. Here’s the key difference: If I start growing oranges, I slightly diminish the value of existing orange groves — but that cost to my competitors is exactly offset by benefits to my customers. By contrast, if I build a fiberoptic cable that’s just a hair faster than your fiberoptic cable, I do not slightly diminish the value of your cable; instead, I (at least potentially) devastate it. The reason my cable can be worth more to me than it is to the world is that it effectively transfers wealth from you to me. That’s unfortunate, because the world works better when people concentrate on creating wealth, not taking it from others.
That creates at least the possibility that projects like Spread Networks’ are socially wasteful. To determine whether they are socially wasteful, we need to compare their external benefits with their external costs. (“External” here means “felt by someone other than the decisionmaker”.) We know that high frequency trading has big external costs. For it to be socially desirable, it would have to have even bigger external benefits. Over on another thread, we’ve been debating whether that’s plausible. As far as I can see, the pro-HFT arguments offered there have amounted to nothing more than a) denying the obvious external costs, with no reasons offered, b) mistaking private benefits for external benefits, and c) asserting that there are external benefits that must surely outweigh the external costs, with no reasons given.
My own position has been that the external costs are obvious, and that there might or might not be external benefits that outweigh them. That was before I did the back-of-the-envelope calculations you see at the beginning of this post. Now that I I’ve got an upper bound on the benefits, I’m pretty well convinced that the matter is settled.
There remains the question: If HFT is socially destructive (as I now think it clearly is), what should be done about it? Outlawing, directly regulating, or taxing it seems like a nightmare to me for multiple reasons, including the implausibility of effective enforcement and my general reluctance to cede power to regulators who all too often wield that power in unanticipated and obnoxious ways.
But here, perhaps, is an easy fix: What if trades were executed anywhere between 0 and 1 second after they were submitted, with the delay chosen randomly? On average, we’d slow everything down by half a second. At the same time, we’d pretty effectively destroy the value of a .003 second head start, and Spread Networks might start looking for an investment that could actually enrich the world — perhaps an orange grove.
Would it destroy the benefit? The HFTers would still have their .003 second advantage on average.
If that much regulation doesn’t offend you, then how ’bout this: move from continuous trade times to quantum trade times. Have the exchange establish a rule that all (comparable) trades for a given stock clear simultaneously every [hour? minute? second? tenth of a second?]. Traders can put conditions on their orders [E.g., Don’t sell/buy if the price goes below/above some specified price], and algorithms set the market clearing price for all comparable transactions. Thus, you’re free to submit your trade .003 seconds earlier than everyone else — but unless it’s the .003 seconds that straddles the deadline for the next transaction period, it won’t provide you with any advantage.
You posted about Rosen 81 and the tournament problem 4 years ago. http://www.thebigquestions.com/2010/02/23/the-olympics-bernie-madoff-and-me/ and also http://www.thebigquestions.com/2010/02/26/arsenic-and-gold-medals/. It seems like the issue was as contentious then as it is now.
Steve,
You write above:
“But while those general hypotheses apply to the markets for oranges, locomotives, haircuts and chocolate chip cookies, they do not apply to high frequency trading. Here’s the key difference: If I start growing oranges, I slightly diminish the value of existing orange groves — but that cost to my competitors is exactly offset by benefits to my customers. By contrast, if I build a fiberoptic cable that’s just a hair faster than your fiberoptic cable, I do not slightly diminish the value of your cable; instead, I (at least potentially) devastate it. The reason my cable can be worth more to me than it is to the world is that it effectively transfers wealth from you to me. That’s unfortunate, because the world works better when people concentrate on creating wealth, not taking it from others.”
So would you say that Amazon.com generates net social waste? The advent of on-line retailing has destroyed a great deal of the capital value of brick-and-morter retailers. Borders, in fact, launched a major expansion – building lots of physical stores – in the 1990s only to find itself struggling and then bankrupt within a few years. Was benefit of the the added convenience of buying books and music on-line sufficiently high to justify this destruction of capital value?
You know I respect you greatly (and that’s not just hyperbole or a string of empty words), but I believe that you are fundamentally wrong to equate competition mediated through consumer choices with robbery, or wealth transfers akin to robbery.
My schedule this week (through Sunday) is a bear, but I hope to soon on my own blog (Cafe Hayek) return to this debate.
With respect,
Don
Multiply your number by the billions of trades executed every year and you’ll see you have the right orders of magnitude.
You have social cost and social benefit wrong.
The speed of execution is not important because nobody can get on with their lives until the trade goes through. It is important because the slower the execution, the more risk ALL market participants are forced to bear.
Investors demand a higher required return than they otherwise would to compensate for that risk. The external effect of that is a higher cost of capital for firms and less investment. I think we agree that less investment is not beneficial to society.
The benefit to market participants of any incremental reduction in speed of execution is an incremental reduction in risk. That incremental reduction in risk borne by investors translates into a lower required return. The external benefit of that reduction in the investor’s risk is a lower cost of capital for firms who are not party to the original transaction. A lower cost of capital means more projects meet the hurdle rate and I think we would agree that this is beneficial to society.
I’m not sure how you calculate this benefit, but I think you’d have to look at the changes in how stocks behave after a trade is executed and then correlate that to changes in cost of capital over the same time period. You’d have to account for lots of other variables that influence equity cost of capital, though. There might be a better way, but that’s off the top of my head and a “back of the envelope” calculation isn’t going to cut it.
However, what you have calculated is utterly meaningless because you’re calculating the wrong thing.
Don: I appreciate your kind words. But:
1) If we reach a point where the marginal cost of supplying books falls to essentially zero (because they are all delivered electronically), so that they are destined to be supplied by a monopolist (because under competition the price would drop to zero, killing off the industry), then I certainly believe that competition to acquire that monopoly has the potential to become socially wasteful. If Amazon makes a trillion dollars a year, and if I can, for an investment of half a trillion, provide an extra 10 cents worth of gain to the consumer, then I can, for an investment of half a trillion, redirect a trillion dollars worth of profits to myself while adding only 10 cents to social value. That’s a recipe for overinvestment.
2) At the moment, the marginal cost of supplying books is not zero. Nevertheless there are a variety of reasons why it seems to be efficient to have one large monopoly online bookseller. So yes, I think there can potentially be inefficient competition for that monopoly power.
3) That’s not to say the competition is definitely inefficient — some of it leads to huge improvements for the consumer (as was, I think, the case when the web drove out the brick-and-mortar stores).
4) Conclusion: We have no theory that can tell us when this scrambling for market position is or is not inefficient in general. (Fortunately we do have a theory that says that under reasonable assumptions about cost curves, competition is sure to be efficient. That theory does not apply when the marginal cost curves are flat at zero.)
5) Because we have no general theory, we have to look at the specifics of each case. The first few paragraphs of this post are intended to look at the specifics of the HFT case, via a back-of-the-envelope calculation of an upper bound on the benefits. I think it would be much harder to do such a back-of-the-envelope calculation for competition in the book business.
6) We all know that competition can sometimes be unhealthy. For example, competition for government subsidies can be unhealthy. We also know that competition is often healthy, and we understand the reasons for that. Sometimes those reasons don’t apply.
7) Even when competition is unhealthy (in the precise sense of being welfare-disimproving) it does not follow that regulation can improve matters.
Doesn’t there need to be some sort of present-value adjusted stream of future benefits? In your example, it seems like a one-off calculation, where in reality the benefits repeat.
Would I pay to have one page on the internet load a second faster? Maybe a little. Would I pay to have every page load a second faster? For sure. It’s a different calculation.
I do understand that this calculation is not your main point.
@Don
Some people enjoy walking around borders and buying books. Given amazon’s success, it seems more people enjoy sitting in their pajamas buying books online. Similarly, to use Steve’s example, some people prefer buying oranges for $1, but given other circumstances, people might be willing to buy oranges for $1.05 (for example, if the more expensive orange are sold at a store more conveniently located).
However, there aren’t ANY conceivable circumstances where, all other things being equal, people would prefer investing on a slower fiberoptic line. Steve’s point is that the amazon does not completely strictly dominate competitor firms in consumer preferences, but faster fiberoptic lines do allow such domination, which is where social loss might be positive.
Steve – Budish et al recently wrote a paper with an idea similar to yours: move trading to discrete time and conduct auctions at fixed intervals (say, once every second). They also provide an thorough analysis of the benefits and costs of HFT, and point to fundamental flaws in current market structure:
http://faculty.chicagobooth.edu/eric.budish/research/HFT-FrequentBatchAuctions.pdf
Why wouldn’t you multiply your $1,500 benefit estimate by the number of 0.003 trade seconds in a year?
It seems like you get that $1,500 for each 0.003s period and so you would get that benefit for each one, not just one — or am I missing something?
Brian Albrecht: I stand by the calculation re the social benefit of speeding up a single trade. We need to compare that to the private benefit of speeding up a single trade. Alternatively, we could compare the social benefit of speeding up a billion trades with the private benefit of speeding up a billion trades, but that just multiplies each number by a billion and does not affect the question of which is larger.
What if trades were executed anywhere between 0 and 1 second after they were submitted, with the delay chosen randomly?
Well, um, they are now executed at between 0 and 1 second.
Of course, market participants are really going to be stoked about your idea to randomly impose the threat of adverse selection on them. There’s nothing investors love more than increased randomness and the probability of being picked off. Artificially increasing the cost of trading would certainly be nothing but beneficial for everyone! You CANNOT be serious, Steve Landsburg.
Mike Nigro: Thanks for this reference. I was unaware of it, but it looks highly relevant to this discussion.
suckmydictum: Well put. Thank you.
“But while those general hypotheses apply to the markets for oranges, locomotives, haircuts and chocolate chip cookies, they do not apply to high frequency trading. ”
I’m missing the logic behind this statement.
If my business is providing financial services then I may decide that an investment in HFT technology of $300m is worth it as it will allow me to offer my customers better returns and lower risks. If my decision is good I will increase my profits and if I am wrong they will fall. If I take market share from my competitors this may cause some of their previous investments to fall in value or even become worthless.
I do not see how this differs between HFT technology and any other investments in any other industry.
In all cases it seems possible that the short term loss in terms of capital value destroyed as a result of investment in a new technology may exceed its specific utility benefits to consumers. In theory a central planner with perfect knowledge could do a better job of allocating resources to consumer need than the the free market.
High speed rent seeking vs. Low speed rent seeking:
http://streetwiseprofessor.com/?p=8374
and
http://streetwiseprofessor.com/?p=8367
The social benefit to quicker trades is not getting money for your sales sooner. The benefit is that the market makers can operate more efficiently, thereby reducing the bid-ask spread. This means that you get a better price when you buy or sell a stock.
‘However, there aren’t ANY conceivable circumstances where, all other things being equal, people would prefer investing on a slower fiberoptic line’
Why ? Suppose you could get 99% of the speed at 50% the cost and make up for the speed difference by investing some of the savings in writing algorithms that are faster / better. Or just offer a slightly less efficient investment service to your customers at a lower cost?
Is HFT really a winner-take-all thing ?
@20
When I said “all other things being equal”, I included cost as one of things being held equal.
“Is HFT really a winner take all thing ?”
Not necessarily, but a faster fiber optic line is, assuming costs are equal.
Roger:
The social benefit to quicker trades is not getting money for your sales sooner. The benefit is that the market makers can operate more efficiently,
If all of history were delayed by .003 seconds, we’d have exactly the same bid-ask spreads, just .003 seconds later.
Rob Rawlings:
I do not see how this differs between HFT technology and any other investments in any other industry.
The difference is in the winner-take-all nature, and you’re never going to see it if you continue to ignore it.
“That’s not to say the competition is definitely inefficient”
I loled on that.
Don cited Amazon, I will cite Intel. Whenever Intel puts a new processor on the market, they make all their competitors (and previous chips) obsolete. What is the “social” value of a few GHz more ? Nothing. Yet thanks to that we have great computers know.
Are you poo-pooing the importance of that .003 seconds?
My college paper included a quote from a sprinter who missed setting a new school record by only a fraction of a second: “I can’t believe I blew it! I was counting on having the extra time for studying.”
I think that taxation can be a way to discourage activity. I.e. if you want to help discourage people from making income, then you would tax income.
You could tax HFT:
– Stocks held for less than 1 second have a 70% tax.
– Stocks held for less than 1 minute have a 60% tax.
– Stocks held for less than 1 hour have a 50% tax.
– Stocks held less than than 1 day have a 40% tax.
Use the revenue gained to reduce the long term capital gains tax on stocks held for more than 1 year.
Methinks, thanks for the huge amount of interesting input on these posts.
Sorry on the above I should have mentioned capital gains tax. I wouldn’t think you would need to tax trades done at a loss.
So what about the argument that HFT is good because it reduces other types of rent seeking?
” By contrast, if I build a fiberoptic cable that’s just a hair faster than your fiberoptic cable, I do not slightly diminish the value of your cable; instead, I (at least potentially) devastate it”
I think this is not quite right. Keep in mind the HFT grabs an offer in front of me, but then it has to get *rid* of it at a higher price (assume back to me). Note I can always NOT take that second fill. Then the HFT gets stuck with the stock, and if (e.g. due to limited capital) they have to be flat end-of day, well, then I will certainly extract my pound of flesh. And I will be willing to trade on a slower line for that second trade as well.
It is the capacity and risk-carrying issue that limit HFTs to a relatively trivial market pnl share (8B overall per year if that?)
I am not an aconomist so unqualified to talk about social benefits etc. But is as people believe the HFTs have crushed bid-asks to a fraction of what they were before they came in, and they take out less than 10B per year out of the market for that, seems net we are somehow better off – even aside from the Keynesian boost from burying bits in mountains (sorry couldn’t resist).
#21
I’m trying to understand and not ignore.
I’m not convinced HFT is winner take all but assume it is for the sake of argument.
Won’t investors take that into account when investing ? If you have a 20% chance of $1T profits and 80% of $0 profits you will limit your exposure and balance your portfolio so that the the risk matches the reward. As long as investors correctly calculate the risks of “winner-take-all” scenarios then won’t this minimize or eliminate the external costs?
Simon:
What is the “social” value of a few GHz more ? Nothing. Yet thanks to that we have great computers know.
In other words, you didn’t think for an instant before posting this — either that or you attribute zero social value to a great computer. Either way, you’re welcome to contribute to the discussion, but if you’re just here to say intentionally stupid and self-contradictory things, then please take it elsewhere.
Have I missed this point? But it seems to me that there may (may) be significant external benefits to this which no-one seems to have mentioned. It’s like the lunar landings and porn. The value to society of the lunar landings is debateable (I think they were great, many think they were a waste of time) but didn’t the programme speed up development of microchips? Likewise porn; it’s social value is pretty dubious but the porn industry demanded high quality video streaming systems long before broadcast TV and Hollywood (weren’t they a bit snooty about the internet?). So, in this case, the fact that engineers and software designers are perfecting and developing such sophisticated products could bring tremendous future gains in unexpected ways. Whether this compensated for Steven’s argument about external costs, I am not able to judge, but I suggest these are real or at least potential benefits which might not arise without HFT.
I agree with Simon. Quicker is better. Denying that those 0.003 seconds make a difference is about like denying that GHz on an Intel processor makes a difference. What if all of history were delayed a few nanoseconds? Humans cannot notice a few nanoseconds? These sorts of arguments could be made against almost anything being quicker. And yet quicker is better. If this HFT is somehow an exception to this general rule, it would take a whole lot better argument to show it.
Steve, I think you miss Roger’s point. The point is not that you are fast in any objective sense but instead that you are faster than the competition in reducing the bid-ask spread. I’m concerned that the social value of improving speeds can be objectively assessed because of unintended consequences. It’s assumed that the only social benefit is to high speed traders but couldn’t there be benefits beyond the direct ones? The technology developed in this “wasteful exercise” could have benefit in other areas and it’s not always clear what those are or might be from the beginning. I’m skeptical of dismissing these benefits cart blanche. Imagine if academic research were restricted based on the expected immediate economic value. Would that be a useful exercise?
Time is continuous. Prices are discrete. Why focus on time and not marketing pricing mechanisms. If prices have to move (quantum jump) by easily discernible amount what effect would that have on time differences. There are plenty of levels of infinity, integer are a little different.
Re #20. Steve, I think you’re way off here. If a technology allowed airline pilots to react 0.25 sec faster, the benefit would not be simply that passengers arrive at their destinations 0.25 sec earlier. The benefits would arise from pilots being able to avoid accidents.
Similarly, the benefit of high-speed data lines is that HFT market makers can reduce their (non-zero) marginal cost of providing each additional share of liquidity. That marginal cost includes transactions costs of offsetting trades and cost of risk. For example, if someone is providing liquidity in SPY (the S&P 500 index ETF), the high-speed line allows them to monitor the 500 component stocks and adjust their bid-offer prices for SPY more quickly in response to changes. That, in turn, allows them to narrow their bid-offer spreads.
Scratch my previous benefit estimate. But, still not sure I buy the $1,500 high-end estimate.
It seems like a % time improvement would be a better estimate. If 0.003 seconds is the improvement on a 0.5s transaction average, then that’s a 0.6% improvement.
0.6% improvement on $15T is $90b.
Now we’re talking rent seeking;
http://www.reuters.com/article/2014/04/18/ca-robbins-geller-idUSnBw185295a+100+BSW20140418
‘Plaintiff seeks to recover damages on behalf of all public investors who purchased and/or sold shares of stock in the United States during the Class Period on a registered public stock exchange or a United States-based alternate trading venue (the “Class”). The plaintiff is represented by Robbins Geller, which has expertise in prosecuting investor class actions and extensive experience in actions involving financial fraud.
‘Robbins Geller represents U.S. and international institutional investors in contingency-based securities and corporate litigation. With nearly 200 lawyers in ten offices, the firm represents hundreds of public and multi-employer pension funds with combined assets under management in excess of $2 trillion. The firm has obtained many of the largest recoveries in history and has been ranked number one in the number of shareholder class action recoveries in MSCI’s Top SCAS 50 every year since 2003.’
“By contrast, if I build a fiberoptic cable that’s just a hair faster than your fiberoptic cable, I do not slightly diminish the value of your cable; instead, I (at least potentially) devastate it.”
Steve, could there be alternative uses for fiberoptic cable that is no longer competitive as a HFT conduit?
I’m seeing the general case to be something like this :
Assume there is a winner-take-all industry (industryA) that has some % (say 10%) of total capital invested in it.
An entrepreneur persuades investors to take capital from the other 90% of the economy and build an business that is 1% more efficient than the Winner-take-all industry. It succeeds and renders all the original investment in industry totally obsolete.
As a result the productive capacity of the economy has been reduced by close to 10% (and the effectiveness of industry A has only increased by a small amount).
In this circumstances would it make sense to introduce rules that prevent IndustryA being winner-take-all ?
I would argue not:
1. Even if the new firm in industry is only marginally more efficient that the old then in the long term the benefits will still outweigh the costs of short term capital destruction.
2. The original incumbents in industryA should have known their industry was winner-take-all and built this into their business model. If they got this right then by the time their capital is rendered obsolete they will already have made sufficient profits (and generated sufficient social benefits) to justify their original investment.
If they got it wrong then they will likely be eliminated from the entrepreneurial pool.
It seems to me that Landsburg is falling victim to the spherical cow model problem. His model of a complex system is far too simplistic to make any useful conclusions.
@Seth,
“Scratch my previous benefit estimate. But, still not sure I buy the $1,500 high-end estimate.
It seems like a % time improvement would be a better estimate. If 0.003 seconds is the improvement on a 0.5s transaction average, then that’s a 0.6% improvement.
0.6% improvement on $15T is $90b.”
So if the transaction time was 0.5s and we shaved off 0.25s, we could increase our economic output by 50%!!! Lol.
Some more thoughts regarding your contention the second line is basically worthless. Another reason to disagree, is the simple reason that if I cross the bid-ask at one exchange, no HFT can execute ahead of me (at least in the basic scenario Lewis describes in his book). So that older line can be offered to investors for orders who are not intending to sweep the market across the exchanges, for cheaper. However, I argue even these slow coaches get benefit from the new line, for the following reason:
Case A: We live in a world with one exchange. We have one price. But the exchange controls and enforces rules, collects rent, speed and competition is low, bid-ask high (see NYSE a few years ago).
Case B: Multiple exchanges. Less direct exchange rent extraction, faster technology, more innovation. Bid-ask can go down on average for all investors, including the retail investors, but *only* if you can make sure an order get the same/best price at various exchanges. If you sweep large orders across the markets, the HFTs in effect rent you this high-speed interconnection – they bring the offers from various exchanges to you; their competition eventually drives the inter-exchange spread to zero. However, even if you trade so your orders fit on one exchange, you still get benefit from the HFTs being there, as you are guaranteed to get a fill close to the best at all other exchanges.
I have no numbers or time to model this, but I can see that such a systems analysis could show some overall beneficial net to everyone.
Apparently I missed the first part of this argument so forgive me a bit, but at last you’re writing about something I know intimately. I worked in financial services for 8+ years, including 5+ at a high-frequency firm (arguably one of the first). I no longer work there, and have no financial stake in any trading company. I also don’t give financial advice, especially not in comments on blog posts. With that out of the way…
It appears that you’re conflating the cost of a particular piece of equipment (a high-speed fiber optic line) with the use to which it is being put. That is, you’re making judgment about high-frequency trading based on its infrastructure. That’s roughly like making judgments about commuting to work based on the cost of a car from Tesla Motors. That fiber-optic line could just as easily be used for things like real-time streaming of live events, tele-medicine, or pick your other purpose. Surely you’re not arguing against a piece of infrastructure simply because of the use to which it is put?
Furthermore, isn’t the purpose of free markets to enable the creation of products that meet demands? If Spread Networks perceives a demand, what’s wrong with them bringing a product that meets that demand? Are you arguing that the presence of this line creates undesirable forces?
There’s also a detail flaw in your argument in that Spread Networks’ higher-speed line doesn’t wipe out the value of another fiber-optic line because it’s a finite resource with a cost. You might say that an EZ-Pass lane on a highway wipes out the value of a cash toll lane but in fact it doesn’t. They co-exist and people pay the costs of each based on their ability to pay and the perceived values of each.
Next, it appears you’re singling out HFT as rent-seeking, but in fact all trading is rent-seeking. If you want to critique HFT you need to distinguish how it’s worse than other models of trading. For example, automated trading largely replaced human trading. This had the effect of greatly speeding up the pace of trading (leading to HFT) and also had the effect of narrowing spreads, as well as driving down the cost of trades. I would argue that both of these things brought benefits in terms of being able to get business done more quickly and at lower cost. In addition, the rise of HFT led to the rise of rebate strategies where traders now get paid to bring liquidity to markets; previously, both sides of a trade paid a cost to the trader.
So if you want to abolish all trading including HFT on the basis of externalities and rent-seeking well that’s one argument. But if you’re claiming HFT is uniquely bad then you need to account for the benefits it brings and balance those against its costs, no?
The purpose of financial markets is in fact to transfer wealth, with associated costs and risks. That’s generally considered a necessity, both in equity and commodity arenas. Again, if you want to argue that capital markets as a whole bring only private benefits and not external benefits I’ll grant that argument only insofar as that’s an argument against capital markets in general and not uniquely against HFT.
For anyone looking for what, exactly, is wrong with HFT, here is a pretty good overview.
http://blogmaverick.com/2014/04/03/the-idiots-guide-to-high-frequency-trading/
@Kevin Erdman,
Thank you.
@BC
Similarly, the benefit of high-speed data lines is that HFT market makers can reduce their (non-zero) marginal cost of providing each additional share of liquidity.
That’s one benefit. Another benefit of faster execution is that it lowers the probability of being picked off for liquidity takers.
Which, btw, is why Steve’s proposal will never see the light of day. Neither liquidity takers nor liquidity providers benefit from additional randomness and exposure to the threat of adverse selection. The only place such proposals will get traction is in academic circles located at safe distances from reality.
I have read the arguments by Michael Lewis, Steve, the above iodiot’s guide, etc, against HFT, and they all seem like Luddites to me. Traders make markets, and all other things being equal, more and quicker trades make a better market. And yet they are somehow upset that money is involved in the pursuit of HFT.
If you have an idea for making the market more efficiently, let’s hear it. But just to complain that money is spent or made? This is like complaining that General Motors spent some money to replace some assembly line equipment, while you were happy with the cars from the old assembly line.
Kevin Erdmann:
Steve, could there be alternative uses for fiberoptic cable that is no longer competitive as a HFT conduit?
Sure. But I still don’t see (and don’t anticipate seeing) any argument that says these social benefits will equal the social costs at the margin.
Rob Rawlings:
An entrepreneur persuades investors to take capital from the other 90% of the economy and build an business that is 1% more efficient than the Winner-take-all industry. It succeeds and renders all the original investment in industry totally obsolete.
As a result the productive capacity of the economy has been reduced by close to 10% (and the effectiveness of industry A has only increased by a small amount).
Yes, you’ve essentially understood the problem.
Even if the new firm in industry is only marginally more efficient that the old then in the long term the benefits will still outweigh the costs of short term capital destruction.
As far as I can tell, you’re just making this up. If there’s a reason to expect this, I’d like to know what it is.
The original incumbents in industryA should have known their industry was winner-take-all and built this into their business model. If they got this right then by the time their capital is rendered obsolete they will already have made sufficient profits (and generated sufficient social benefits) to justify their original investment.
That’s absolutely true. But it does not address the divergence between private and social benefits, so it does not address the issue of overinvestment.
JohnW:
It seems to me that Landsburg is falling victim to the spherical cow model problem. His model of a complex system is far too simplistic to make any useful conclusions.
Where in the argument did you perceive a significant gap?
Alan Wexelblat:
So if you want to abolish all trading including HFT on the basis of externalities and rent-seeking well that’s one argument. But if you’re claiming HFT is uniquely bad then you need to account for the benefits it brings and balance those against its costs, no?
Ummm…that’s exactly what I tried to do in this post. Did you not read it?
But just to complain that money is spent or made?
Roger, you’re the second person I’ve had to say this too on this thread: Feel free to express your opinions forcefully but don’t feel free to make up stupid opinions and attribute them to other people, particularly me.
@48 (and 39) Spread Network must believe that their advantage will persist long enough to recoup this massive investment. Surely they know that their is a reasonable probability someone else will render their cable obsolete and also some probability that laws will change and remove the advantage of speed.
Admittedly, I don’t see a massive social benefit here, but we have to be missing some genuine value creation. The profit opportunity is simply too ephemeral to justify the expense on short-term rent seeking groups.
blink: Or, maybe, they realize that in the event that their HFT advantage isn’t persistent, the broader value of a fiber optic line for many other uses will provide positive returns. In their NPV analysis for this investment, the worst scenarios are probably pretty good.
Never before has so much digital ink been spilled in the effort to imagine how a state of the art fiber optic line could possibly add any utility to a country’s infrastructure.
I think that I understand better now where the confusion is arising. If one thinks of HFT as some sort of race to pick up some $20 bills that happen to be lying around, that the one with the fastest transmission line wins the race, and that the marginal cost of picking up each bill is basically zero then, yes, it would be socially wasteful to build transmission lines to win the race. After all, there is no wealth generated from picking up the bills. The race only determines who gets the bills. If someone can already pick them up with zero marginal cost, then any additional cost incurred to pick them up would be wasteful.
I think, though, that we should pause to consider why someone is leaving $20 bills all over the place. In the case of HFT, the “$20 bills” (more like fractions of a cent) are there to compensate liquidity providers. (It’s as though the original owner of the $20 bill gets a benefit each time someone physically leans down. He leaves $20 bills to get people to lean down.). Also, the marginal cost of picking up a bill is *NOT* zero. One must expend energy to bend down and pick up a bill. The “energy expended” by an HFT liquidity provider is that, each time he provides liquidity, he takes a share of stock into his portfolio. That either adds risk or requires him to incur transactions costs to make an offsetting trade to remove that risk. (Think of a liquidity provider selling a share of SPY and offsetting the risk by buying the 500 stocks in the S&P 500 index. The transaction costs in buying the 500 stocks is part of his marginal cost of providing liquidity in SPY. This is conceptual: most likely, this would be too expensive to be effective in hedging risk.) The marginal cost of leaning down will be $20. If it was higher, then no one will lean down to pick up the $20. If it was lower, then people will keep picking up bills until they get tired enough that their marginal cost increases. So social marginal benefit is $20 and social marginal cost is $20. All is good.
Along comes someone who installs a new transmission line, who is able to lean down more cheaply. Both I (#35) and Methinks (#45) have explained why. The transmission line doesn’t just allow one to execute a trade sooner so that “we can all get on with our lives”. In my 500 stocks vs. SPY example, the transmission line allows a liquidity provider of SPY to adjust his prices more quickly in response to price changes in the 500 stocks. It reduces his risk of getting “picked off” as Methinks says. An example of getting “picked off” would be if the liquidity provider believes that SPY is worth 100.00 so he offers to sell at 100.05 or buy at 99.95. Some subset of the 500 stocks increase in price pushing fair value of SPY to 100.10. Before the liquidity provider is able to update his prices to 100.05 and 100.15, someone takes him up on his not-yet-updated offer to sell at 100.05. Part of the 0.10 bid-offer spread compensates him for this risk of getting picked off. The faster his transmission line, the less risk of getting picked off, the narrower bid-offer spread he needs. With a faster line, he might be able to narrow is spread to 99.98-100.02, for example.
Because this newcomer has lower marginal cost of leaning down, some people don’t need to leave $20 bills; they can leave $10, the newcomer’s marginal cost. Each bill newcomer picks up produces $10 of leaning down benefit at a marginal cost of $10. At some point, newcomer may become tired (he acquires too much risk in his own portfolio to want to keep providing liquidity), so eventually people may resume leaving $20 bills to get the old incumbent leaning down. For each bill picked up (whether $10 or $20), marginal social benefit of leaning down exceeds or equals marginal cost. So, it’s welfare maximizing.
I have deliberately left aside the issue of the sunk costs or asset depreciation incurred by the old incumbent. What matters is *marginal* cost and benefit of leaning down to pick up each additional bill. When newcomer leans down, it does not impose any additional cost (use of resources) on incumbent. Any depreciation of incumbent’s asset does not represent an increased use of economic resources by incumbent. There could, of course, be an accounting cost. The cost of newcomer’s transmission line *is* a marginal cost of course (at least for the first bill). But, newcomer is already incorporating such cost into his decision to pick up bills. It is part of his marginal cost calculation. Also, incumbent can continue to use his line to pick up $20 bills when newcomer gets tired.
Hopefully, this will help people understand HFT from liquidity-provision perspective. I suppose it’s true that if people *are* leaving $20 bills around for no apparent reason and no benefit to themselves, then we might need policies to discourage other people from wasting resources to pick them up. I don’t see many $20 bills lying around though.
‘ Also, incumbent can continue to use his line to pick up $20 bills when newcomer gets tired.’
Or, more likely, he can use his old line for its next most highly valued use. So society has BOTH high speed trading of stocks and whatever the old line is used for.
BC does a nice job of explaining how HFT enhances liquidity provision. I would also advise people to go to Craig Pirrong’s Streetwise Professor site and read his detailed explanation of why HFT is simply the latest iteration on the tradeoff between market-making (providing liquidity to buyers and sellers) and having asset prices reflect private information. e.g. http://streetwiseprofessor.com/?p=8340
Anyone providing liquidity is subject to the risk that when a big order comes in the party making the order has superior information and so the market maker will lose money on the deal. All exchanges that have market-maker functions have ended up with those liquidity providers trying to use every bit of information they have on trading patterns to try to suss out these informed traders (“liquidity abusers” perhaps from the standpoint of the market makers) and protect themselves by marking up or down the purchase or sale price to the informed party.
When an informed party tries to spread its orders around to different exchanges to hide them, it is engaged in a rent-extraction game from the market maker, which if it wins in equilibrium will drive up the costs of liquidity for everyone. So HFT, and all the other older ways in which market makers try to protect themselves, has a bad effect on the markets by discouraging traders from developing information about securities (because it becomes harder to cash in on this information) but HFT also makes it easier for all the traders who just want to rebalance their portfolios or cash out to execute their trades at lower spreads.
You can’t use Steve’s back of the envelope calculation to weigh these tradeoffs.
BC:
I think that I understand better now where the confusion is arising. If one thinks of HFT as some sort of race to pick up some $20 bills that happen to be lying around, that the one with the fastest transmission line wins the race, and that the marginal cost of picking up each bill is basically zero then, yes, it would be socially wasteful to build transmission lines to win the race. After all, there is no wealth generated from picking up the bills. The race only determines who gets the bills. If someone can already pick them up with zero marginal cost, then any additional cost incurred to pick them up would be wasteful.
I haven’t yet read the rest of your long post, but this part is exactly right. I understand that in the part I haven’t read, you are going to argue that there are important differences between $20 bills and HFT. I look forward to reading it.
Steve, you said: “Spread Networks recently spent $300 million … What’s the social value of that cable? … If HFT is socially destructive (as I now think it clearly is), what should be done about it?”
This seems to me to be a complaint about money being spent pursuing HFT. I did lump you in with Michael Lewis, but only because he has launched a more detailed and highly publicized attack on the supposed evils of HFT. But it is your blog, and go ahead and clarify however you please.
BC:
I had briefly posted a reply to the bulk of your comment but am reconsidering whether it’s correct. More later…..
#48
on “Even if the new firm in industryA is only marginally more efficient that the old then in the long term the benefits will still outweigh the costs of short term capital destruction.”
Take the extreme case where the new firm renders all the existing capital invested in industryA obsolete. At that point the productive capacity of the economy has been reduced (which is a social cost) but industryA is now slightly more efficient as a result of the investment in the new technology.
If nothing else changes then after a few more years the economy will replace the obsolete capital from IndustryA with increased investments in the rest of the economy, and the new steady state will be at higher level of productive capacity (all other industries will be the same as before but Industry A will be a bit more efficient).
on “it does not address the divergence between private and social benefits, so it does not address the issue of overinvestment.”
The way I’m looking at that this is: If investors in winner-takes-all industries invest knowing that their gains will be short lived then it is an optimal and not an over-investment. This assumes that investment in HTC is of benefit to both the investors and society as long as it doesn’t result in capital being rendered obsolete (as is the case when the incumbents had invested expecting their investment to be long -term).
That was great, BC.
Steve:
If all of history were delayed by .003 seconds, we’d have exactly the same bid-ask spreads, just .003 seconds later.
The new link doesn’t mean history plays itself out the same way, just advanced by 3ms. Compared to the previous link, it means the market trades ~20% faster (13.3 vs 16.3ms according to the linked article). The benefits of that are debatable, but they don’t have to be all that big to add up.
$300mm just isn’t that much money compared to equity market trading volumes. The SP500 emini contract for example trades $150bn+ daily. It doesn’t take much of a tightening of bid/offer spreads to add up to $300mm — 0.015 index points (on a mid of ~1865) would be enough to pay back the investment in just a year.
“Spread Networks might start looking for an investment that could actually enrich the world”.
Nah, I’m thinking the world will be better off if the owners of companies like Spread Networks continue to pursue their own objectives by looking for investments that they believe will enrich themselves. They dug the $300 million tunnel because they believed that certain others would be willing to pay for access to their line to a degree that would return to them a very VERY nice profit on their investment.
Do I like the idea that HFT algos can accurately anticipate someone else’s remaining buy order after the first trade executes, then go buy up the inventory and sell it to the unsuspecting buyer for a penny or two more per share? Honestly, no. But, Steve, I’m more concerned with your reaction to it…
Getting into these weeds in a manner only really bright folks like yourself can do is not only not useful in my estimation, it’s inherently dangerous. While I know you’re sincere in your concerns over government grabbing control of the situation, when reputable economists, such as yourself, begin labeling legal private acts as lacking social value (which, frankly you can’t, in the grand scheme of things, know), you invite, add credibility even, to the very thing I believe you’d agree can only make matters much MUCH worse (government intervention)…
As for the fix, didn’t Lewis’s protagonist already figure that one out? I believe he’s investing in his own exchange (IEX) that will circumvent the front runners by making sure orders hit every exchange at precisely the same time. Should we now explore the investment dollars poured into IEX and measure its social value? Or can we conclude that, in the end, or, for now, Spread Networks inspired yet more innovation that, who knows, may someday benefit society in ways we’ve yet to imagine?
Lastly, in your earlier post you, as did I in a post of my own (http://www.betweenthelines.us/krugman-shouldve-loved-this-one/), point to Krugman’s, on this topic, abandonment of Keynesianism (wasteful spending sometimes good). Please be careful not to contradict your own ideology going forward. For yours, unlike Krugman’s, is indeed socially useful…
Steve, one major problem is competition between exchanges.
We could outlaw non-random continuous time trading, but what’s to stop HFT firms from trading US companies on a Canadian exchange?
Take Italy, where there is excessive regulation through a transactions tax. No one use the Italian exchanges.
I think you have the right conclusion, and sometimes a market failure isn’t as costly as a government failure. The only cost of this failure is over-priced cables, but a government failure could cost a lot more.
Spreading the 0.003s across the entire economy did not make sense to me either.
The US equities market is about $10T. Assuming 9% growth a year, that comes out to $900B. If HFT is 1% of that, a $300M investment to improve it does not seem so outlandish.
I also note that the commenters who have real-world experience with HFT, such as Methinks, do not believe current trading infrastructure will suddenly become worthless.
That said, Landsburg has convinced me there is a good chance of overinvestment. I just don’t believe it’s possible to calculate that out.
No, I don’t think it’s going to become worthless, nor do I believe that investment in trading infrastructure is any more immune to over-investment than anything else. However, this back of the envelope calculation isn’t even measuring the appropriate benefit and measuring the appropriate benefit is much more involved than this basic arithmetic.
I also fail to see how one firm can possibly take all. Spread leases its lines and it already has a competitors digging more. They all lease their lines to us (and I re-iterate, I’m not technically an HFT trader and I do find it bizarre that I’m defending my competitors against people with absolutely no skin in the game). Multiple HFT shops have leased those lines already and eventually all HFT – followed later by the rest of us – will move to the faster lines as supply increases and prices drop. The ideal in trading for both liquidity providers and liquidity takers is zero latency and innovation will continue to move in that direction with or without crazy regulations like those proposed here. If regulation gets too onerous in the United States, trading will move to another venue out of the SEC’s jurisdiction.
I am reminded of a similar discussion about playing online poker professionally (which I myself did). Some people claimed that professional poker players added “nothing to society”, whereas some others would claim that by ensuring games were running, they provided entertainment for the recreational players.
Now, that may be true in aggregate. Without any pros at all, some recreational players would find it hard to play some higher stakes games whenever they wanted. So there is some benefit to having pros. However, the marginal benefit is much lower, especially for pros playing popular games. For instance, no-one should have any trouble finding a seat in a small stakes No Limit Texas Hold’em ring game. Thus, the social benefit is virtually zero – all the pro does is lower the earnings of other pros and make the recreational players lose money faster.
This is similar to HFT and other speculation. There might be some modest benefits in increasing liquidity and what-not. But what is the marginal benefit? Liquidity is already very high for many financial instruments, I struggle to see how a slight decrease in liquidity would have significant social costs.
@methinks 65
My reading has definitely been enriched from your insight in these threads (seriously), but I posit the most outrageous pronouncement you’ve made (so far) seems to be this:
“I do find it bizarre that I’m defending my competitors against people with absolutely no skin in the game”
This misses Steve’s point obviously and fundamentally. ANYONE who invests in a shop trading on the slower line “has skin in the game”. You don’t need to be running that shop to be committing skin to this game. When my investment in quant firm X yields crap results because its lined has been replaced by this slightly faster one, I am a non-consenting stakeholder. Please don’t pretend there are no stakeholders in this outside of the financial sector.
Keep in mind that the cable that is being installed has alternative uses, like the cable it replaced. So the “social cost” of the cable has to be balanced against the “social gain” of seeing the old cable put to some other use.
Now as it turns out, most Internet applications are not sensitive to 2-3 ms of latency – no user of an typical application will notice it. However, I can envision a scenario where a general purpose pipe with high-bandwidth / medium-latency getting replaced with a low-bandwidth / low-latency pipe frees up lots of bandwidth for other applications, creating a huge benefit. Or, the old pipe find new uses, for example playing live music with people in remote areas where 2-3 ms actually matters. I don’t know for sure – but generally given our increasing reliance on technology, I find it very hard to believe that building a new information pipeline is socially wasteful.
dictum, please change your handle. I don’t mind juvenile humor but it doesn’t seem appropriate on the smartest comment section I’ve seen. I won’t pester you again about it, I promise.
Back to the subject. I hesitate to bring it up because so many people redefine “public good”. Would prevention of such an inefficient arms race through regulation be a public good? I’m naturally hesitant about being too loose about defining a public good.
Most of the commenters seem to be at least intuitively aware of Coase’s dictum that ‘The Problem of Social Cost’ is a reciprocal problem. Odd that our host doesn’t seem to get that.
Steve, I did read your post, where you say things like “If HFT is socially destructive, which I think it is.” Throughout the post you call out HFT and assign it evils that I think could equally be argued around all forms of trading; however, you then have to engage not only with the benefits of HFT but with the benefits of capital markets in general.
For example, how would businesses get liquid capability to expand in the absence of capital markets? How would financial institutions find ways to lay off risks in the absence of capital markets? Et cetera.
It also looks like @BC and @Methinks are attempting to explain markets and liquidity so I won’t belabor their points and will simply stick to my contention that your argument is an argument against trading and capital markets in general and therefore must account for those larger benefits. You are not making arguments against HFT qua HFT.
I do not understand the logic of the calculation about the benefits of high speed trading. It seems to be resting on the idea that speeding up the trades by 0.003 s is equivalent to time shifting the universe by 0.003 s.
This seems wrong to me. The information passing on the fiber is moving faster *relative* to the rate other information is moving around.
To illustrate, say we were knew of aliens that were able to coordinate 333 times faster than us. I think this would seem very valuable to us. They would be able to get things done more quickly relative to us.
I’m not sure if this changes the conclusion, but it makes me suspicious of your calculation of the benefits.
Alan Wexelblat:
Steve, I did read your post, where you say things like “If HFT is socially destructive, which I think it is.”
Okay, you read that part. But apparently you skipped the actual content of the post, namely the calculations that led to that conclusion.
Pat says:
dictum, please change your handle. I don’t mind juvenile humor but it doesn’t seem appropriate on the smartest comment section I’ve seen. I won’t pester you again about it, I promise.
Seconded.
When my investment in quant firm X yields crap results because its lined has been replaced by this slightly faster one, I am a non-consenting stakeholder. Please don’t pretend there are no stakeholders in this outside of the financial sector.
The only reason your investment in X would yield crap results is that X refuses to upgrade its technology but insists on continuing to attempt to compete in a space in which it is no longer competitive.
Your crap results don’t stem from firm Y’s decision to improve its equipment in order to gain a competitive edge. Your crap results are rooted in the crap management decisions of the (obviously) crap firm you invested in. By remaining invested in X, you have consented to their crap decisions. Divest!
Would prevention of such an inefficient arms race through regulation be a public good?
Is there anything less efficient way to accomplish something than regulation? I think you’ll find that you introduce much more inefficiency by trying to solve problems via regulation than the inefficiencies you’re trying to solve. Remember: insiders own the regulator.
@pat and Steve
Just remember that juvenial humor ably sustains us all. I will henceforth be known as “dictum” only.
@methinks
Thank you for graciously conceding my point.
Dictum:
I will henceforth be known as “dictum” only.
Thank you.
Dictum,
Here’s the thing: the overwhelming majority of liquidity providers (including really big ones) don’t have outside investors.
We’re not hedge funds seeking to slap a 2% fee on the largest amount of capital we can convince people to give us. We use as little capital as possible and lever as much as we can. So, there’s not a lot of other people’s skin in this game and if there is, they should know that they can lose to a competitor.
BTW, Dictum, I didn’t miss Steve’s point. I contend that Steve hasn’t made a point.
I’ve got a question (still waiting for SL to respond to BC though).
What’s the difference between HFT and ticket scalpers?
Economics 101 teaches that ticket scalping provides a great service to society, it distributes them to the buyers willing to pay the most for the enjoyment of watching the event. This is possible because the quantity demanded of tickets exceeds the quantity supplied, usually because the performance is at a smaller venue or doesn’t show enough shows (this is why you never see ticket scalpers for movies). Ticket scalping is rent-seeking, but it still increases total surplus, so we can conclude the best public policy would be to repeal anti-scalping laws.
So I finally caved and bough Lewis’ book and am in the midst of reading it to understand HFT more accurately. Lo-and-behold, he sets up a hypothetical to explain the argument against HFT, and even nicknames the hypothetical HFT firm “Scalpers, Inc.”
I’m going to create another post with the relevant quote because I hate long posts, but what is the difference? Ticket scalpers “Buy low and sell high!” and don’t HFT find a low pricing seller and transfer to a high bidder?
Steve, as you’ve no doubt read in the comments and other places, the claimed external benefit of HFT is that it results in lower bid/ask spreads, which makes transaction costs cheaper for non-HFT market participants. Your calculation in the original post doesn’t address lowered spreads or trading costs. It seems as if you are explicitly rejecting the idea that improving marginal efficiency in HFT can lower spreads further and reduce trading costs even more. I think a convincing argument (to me) would have to show that 1) HFT doesn’t actually lower trading costs or 2) the Spread line wouldn’t lower trading costs by $300M, and that these arguments would have to involve a story about market liquidity and risk and historical trading costs.
Also, I am swayed by the argument made yesterday on Cafe Hayek that a voluntary private transaction does not need to justify its social benefit (which in this case is likely diffuse and hard measure) to avoid being regulated.
I apologize for the length, but I’m trying to avoid a misunderstanding of the author’s content.
Imagine, for instance, that someone passed a rule, in the U.S. stock market as it is currently configured, that required every stock market trade to be front-run by a firm called Scalpers Inc. Under this rule, each time you went to buy 1,000 shares of Microsoft, Scalpers Inc. would be informed, whereupon it would set off to buy 1,000 shares of Microsoft offered in the market and, without taking the risk of owning the stock for even an instant, sell it to you at a higher price. Scalpers Inc. is prohibited from taking the slightest market risk; when it buys, it has the seller firmly in hand; when it sells, it has the buyer in hand; and at the end of the trading day, it will have no position at all in the stock market. Scalpers Inc. trades for the sole purpose of interfering with trading that would have happened without it. In buying from every seller and selling to every buyer, it winds up: a) doubling the trades in the marketplace and b) being exactly 50 percent of that booming volume. It adds nothing to the market but at the same time might be mistaken for the central player in that market.
…In spirit Scalpers Inc. was less a market enabler than a weird sort of market burden. Financial intermediation is a tax on capital; it’s the toll paid by both the people who have it and the people who put it to productive use. Reduce the tax and the rest of the economy benefits. Technology should have led to a reduction in this tax; the ability of investors to find each other without the help of some human broker might have eliminated the tax altogether. Instead this new best rose up in the middle of the market and the tax increased-by billions of dollars.
Pages 107-109.
Curses for triple post…
Upon further reflection as I typed the above passage, some differences occurred to me.
1) Ticket scalpers assume risk by not necessarily having buyers on hand, HFT doesn’t.
2) Throughout the book, HFT firms seem to have unfair access to stock market information. For example, there’s a passage that describes how certain HFT companies set up their own stock market exchanges and incentivized people to partake in it by paying buyers to “take” while charging creators of a market for stock to “make”. However, it was in REALLY small quantities, and this information would travel to these HFT companies before the larger shares would hit the market and they could use the information to see who wanted what stock and who wanted to sell what. That seems shady to me.
Still, aren’t HFTs once you eliminate #2 just like ticket scalpers, or is there something more fundamental between them?
Clyde:
Steve, as you’ve no doubt read in the comments and other places, the claimed external benefit of HFT is that it results in lower bid/ask spreads, which makes transaction costs cheaper for non-HFT market participants.
Why is this not pecuniary?
Under this rule, each time you went to buy 1,000 shares of Microsoft, Scalpers Inc. would be informed, whereupon it would set off to buy 1,000 shares of Microsoft offered in the market and, without taking the risk of owning the stock for even an instant, sell it to you at a higher price.
That’s the adorable fiction for which Mike Lewis has become known. Is this analogy meant to illustrate a market here on earth in America or is this meant to be some imaginary world?
“Front-running” is when your broker, who has a fiduciary duty to you, receives your order and decides to trade ahead of you BEFORE sending your order to the open market. No HFT is in that position (indeed, no market maker). No HFT CAN front-run an order.
Once your order is sent to the open market, it become public information. Available to all. That’s the first opportunity an HFT algo has to see your order.
IF you are an “informed’ trader, you are seeking to extract rent from your counterparty by trading on your counterparty’s stale market. It’s stale because the information you are trying to profit from exists, but is not yet priced into the security. Your counterparty (maybe HFT, unless you are NOT an institution and then NOT an HFT) will understandably try his best to avoid being picked off by you.
To avoid being picked off, he hopes to “read” your PUBLICLY AVAILABLE order in an attempt to decide whether you’re about to pick him off or not. If he correctly identifies your order as having no informational edge, the market maker (HFT or not HFT) will happily fill your order in size.
If, however, he correctly assesses your order as informed, he knows you’re here to get paid for your not yet public information (note: that this could be different from prohibited insider trading as defined by the SEC)and the party he intends do the paying is you – and everyone else who takes the other side of his trade (which could include your grandma)
Obviously, anyone with two firing brain cells will move out of the way by moving their market – and, mind, closer to the new fair market value that the order implies. Why, a smart market maker might even join the informed order in the hope that he can get a few bites in before the price moves all the way to the new fair. This activity informs everyone else in the market that the fair value has changed, everyone else responds my moving their market to the new fair value and so the rent seeker is denied at least some of his rent (the more market makers competing in that security, the faster this happens and the less likely rent seeker will be able to profit from his information).
That’s what really actually happens. And, btw, this is what happened long before algorithmic trading was even a concept. Now our algos do this. Before algos we stood on the floor of the exchange and did this in person.
BTW, if market makers take no risk, then please explain to me why we typically win on only 51% of our trades (this is also true for HFT market makers)? Also, why do so dang many market making firms blow up (including HFT)?
1) Ticket scalpers assume risk by not necessarily having buyers on hand, HFT doesn’t.
That is correct about ticket scalpers and INCORRECT about HFT. HFT assumes risk. Just one glaring red flag that should make anyone doubt such a claim is that successful HFT’s profit only on just a little over half their trades (typical for any successful market maker). If they truly weren’t taking risk, the number of trades they would be expected to profit from approaches 100%.
Throughout the book, HFT firms seem to have unfair access to stock market information. For example, there’s a passage that describes how certain HFT companies set up their own stock market exchanges and incentivized people to partake in it by paying buyers to “take” while charging creators of a market for stock to “make”.
That is a common method of limiting adverse selection by incentivizing orders not looking to pick off liquidity providers. Lit exchanges use this method. Just off the top of my head: the Chicago Board Options Exchange’s stock exchange, CBSX, used this exact fee structure to attract volume to its exchange. It’s one method exchanges – particularly small exchanges disadvantaged by Reg NMS – employ to attract order flow. There’s nothing evil or sinister about it and, as I just pointed out, it’s not an uncommon practice in the world of your friendly neighbourhood lit exchanges. Also, people CHOOSE to send their orders there. Nobody is compelled to send orders to any exchange, so you have to wonder if maybe people directing their orders to an exchange with such a fee structure believe doing so is to their advantage.
You have to remember that Michael Lewis is best known for writing sensationalist fiction. While I enjoyed “Liar’s Poker”, which was published just before I came to Wall Street, I found a lot of his stories to be…um…exaggerated (except that bit about you starting out in a position that is lower than whale shit on the bottom of the ocean floor and sales guys like him having to get permission from a trader before they so much as burped. Sales guys are not the cream of the intellectual crop. That was true). And, btw, I was employed by Solomon Brothers very early in my career.
Also, I am swayed by the argument made yesterday on Cafe Hayek that a voluntary private transaction does not need to justify its social benefit (which in this case is likely diffuse and hard measure) to avoid being regulated.
I just hired a private burglar to steal your valuables.
Steve, I’m not sure why you think you’ve addressed the issues I raised, or why your post addresses trading in general. You certainly haven’t answered any of the questions I raised, and I suspect repeating myself wouldn’t be productive.
I’ll just close by pointing to this – http://www.frbatlanta.org/documents/news/conferences/14fmc/Stiglitz.pdf – a talk by Joseph Stiglitz at the Atlanta Fed recently that talks about markets and activity. I haven’t read it myself, yet.
Steve,
I think your back of the envelope calculation needs a better starting point. You need the FRACTIONAL improvement that the 0.003 seconds gives you.
Under this framing, the question to ask is how much faster, relatively speaking, is the 0.003 seconds relative to the previous travel time. This speed up would be a ballpark estimate of how many more executions/trades one can perform.
So for example, the time = t = d/(c/n). c is the speed of light, n the index of refraction of glass (with fiber optics this is equal to approximately 1.6), d the distance.
Over say 2000 km, this would be 0.011 seconds. The ratio of 0.003 to 0.01 is ~0.3. This 30% improvement is way too high (the servers have to do a price lookup, adding to the time in the denominator), but you get the idea. Whatever dollar value you attached before to the old technology, one gets this % improvement.
Let’s say the servers are slow, on the order of 200 ms. You are still looking at an improvement of 1+ percent:
.003/(0.2 + 2*0.011) = 0.014
(The extra factor of two is there for round-trip.)
The social value of $15 billion (which I agree is too high a starting point) at 1.4% would be $200 billion.
Trey
@Methinks,
“IF you are an “informed’ trader, you are seeking to extract rent from your counterparty by trading on your counterparty’s stale market. It’s stale because the information you are trying to profit from exists, but is not yet priced into the security. Your counterparty (maybe HFT, unless you are NOT an institution and then NOT an HFT) will understandably try his best to avoid being picked off by you.
To avoid being picked off, he hopes to “read” your PUBLICLY AVAILABLE order in an attempt to decide whether you’re about to pick him off or not. If he correctly identifies your order as having no informational edge, the market maker (HFT or not HFT) will happily fill your order in size.
If, however, he correctly assesses your order as informed, he knows you’re here to get paid for your not yet public information (note: that this could be different from prohibited insider trading as defined by the SEC)and the party he intends do the paying is you – and everyone else who takes the other side of his trade (which could include your grandma)
Obviously, anyone with two firing brain cells will move out of the way by moving their market – and, mind, closer to the new fair market value that the order implies. Why, a smart market maker might even join the informed order in the hope that he can get a few bites in before the price moves all the way to the new fair. This activity informs everyone else in the market that the fair value has changed, everyone else responds my moving their market to the new fair value and so the rent seeker is denied at least some of his rent (the more market makers competing in that security, the faster this happens and the less likely rent seeker will be able to profit from his information).”
But doesn’t this make it less likely that a rent seeker will seek information in the first place? My very limited understanding of the stock market tells me that the stock market’s function is to allocate capital efficiently. If the rent seeker no longer receives the value that he would have received without the HFT firms, doesn’t he have less incentive to seek information in the first place? From my passing reading of your comment, this makes me believe that HFT makes the stock market less efficient since the benefit of seeking out information about whether a companies true value is it’s current going rate is reduced and some of his value is transferred to those that have faster algo lines rather than to those that are actually making the allocation process more efficient (i.e. those seeking information)? Please tell me where I’ve erred, as I’m sure I have.
Steve:
Clive
Steve, as you’ve no doubt read in the comments and other places, the claimed external benefit of HFT is that it results in lower bid/ask spreads, which makes transaction costs cheaper for non-HFT market participants.
Why is this not pecuniary?
Why would it be a pecuniary benefit? The point of the new data link is that it is a real investment that reduces the economic cost of transactions. The bid/ask spread isn’t pure profit, it is set by the cost of the marginal provider of liquidity. When new technology (whether improved algorithms, faster or cheaper computers, or a higher speed data link) reduces that cost, that is a real economic benefit.
Steve, I must confess I’m a little baffled by your question. If the local bookstore was charging a markup of 10% on books, and Amazon comes along and offers books at a markup of 5%, we don’t normally question why this is not a purely pecuniary benefit transferring wealth from bookstores to book readers. The default assumption, I would think, absent any concrete evidence to the contrary, would be that the bookstore had economic costs of 10%, which is why they were charging their markup, and Amazon has found a way to reduce those costs to 5%.
Do you have any concrete evidence to support your contention that the new data link cannot reduce the true trading costs of just the S&P e-mini contract by a 10th of a tick for at least one year?
@Daniel,
The only way to profit from information is to reveal it to the market by trading. Given this, nobody ever expects to realize all of the calculated informational edge. Given the choice of not earning any excess return or earning part of the excess return, you will choose the latter. To maximize the edge they capture, informed traders must hide their hand as much as they can. They work orders in small increments over time, break up the order over several venues, go to dark pools, implement algos designed to fool their counterparties, etc.
Also, I should note that what I described in the passage you quote is not unique to HFT and has nothing to do with transferring info to those whose algos necessarily operate over faster lines but rather to those who have smarter algorithms that are better at differentiating order flow. We now do this electronically, but we used to do exactly this standing on the floors of exchanges. This is the behavior of any market participant because no market participant wants to get picked off. So, this is not an issue specific to HFT at all, even though Michael Lewis is presenting it as such.
@Methinks,
Okay, so the value that the rent seeker receives is not altered by the speed of the algo? In other words in a regime with faster algos the rent seeker will gain as much from his knowledge as if slower algos were in place, all else equal?
Trey: I have no idea why you think these calculations are relevant. The question at hand is: Does the social benefit of a particular fiberoptic cable exceed its social cost? If you’re going to express the benefit as a percentage of something, then you’ve got to express the cost as a percentage of exactly the same thing — in which case all you’ve done is multiply each side of the inequality by the same irrelevant constant, which does no harm, but does no good either.
Methinks and BC: Perhaps one of you has answered this in one of your many long posts, in which case I apologize, but it would help a lot if you could explain this:
Why do high-speed transmission lines lower my risk of getting “picked off” (as BC puts it)? If I’m at risk for, say, 10% less time, but the picker-offers are 10% faster, why am I not facing exactly the same risk as before?
nivedita:
Steve, I must confess I’m a little baffled by your question. If the local bookstore was charging a markup of 10% on books, and Amazon comes along and offers books at a markup of 5%, we don’t normally question why this is not a purely pecuniary benefit transferring wealth from bookstores to book readers. The default assumption, I would think, absent any concrete evidence to the contrary, would be that the bookstore had economic costs of 10%, which is why they were charging their markup, and Amazon has found a way to reduce those costs to 5%.
Okay, agreed.
The right question, though, is: What reason do we have to believe that the social benefits of HFT exceed its social costs? So far, I do not believe that *any* of the commenters have offered anything like an answer to this question.
Now if HFT does in fact reduce bid-ask spreads, I agree that we should infer this is a measure of social value. But I am skeptical of this, because I don’t see how it reduces risk. If someone can explain that to me, then I’m willing to revisit my calculation.
I think I can add something to this conversation.
High frequency traders are market makers, aka liquidity providers. In a competitive market, the “price” that they compete over is the bid/ask spread. To date, the high frequency traders have been so much more efficient than the old market making firms that they have driven most spreads to one penny.
Herein lies the problem. Due to the way current regulations are written, they can not drive the bid/ask spread any lower. If regulations would allow it, most spreads would be sub-penny by now.
Because spreads can not get any thinner, HFTs compete by getting their orders in the book first. He, who is first wins. The liquidity taker sees no benefit to this race to match his order first, so Stephen is most likely correct, that there is no NET social benefit to this HFT arms race.
What we need to do is rework the regulations so sub-penny spreads are possible. Then liquidity takers will, again gain from the competition between HFT’ers.
Clyde:
Also, I am swayed by the argument made yesterday on Cafe Hayek that a voluntary private transaction does not need to justify its social benefit (which in this case is likely diffuse and hard measure) to avoid being regulated.
You should probably keep that discussion on Cafe Hayek. In general, I like to let threads drift around here, but not when they got this long and complicated. The question at hand is: Does the social benefit of HFT (and in particular of fiberoptic cables devoted to HFT) exceed the social cost? The question of when and whether regulations are justified is quite separate and I think it’s a bad idea to clutter this discussion with that one.
Follow this link:
http://www.chrisstucchio.com/blog/2012/hft_whats_broken.html
This guy explains the details better than I could.
Clyde:
I think a convincing argument (to me) would have to show that 1) HFT doesn’t actually lower trading costs or 2) the Spread line wouldn’t lower trading costs by $300M, and that these arguments would have to involve a story about market liquidity and risk and historical trading costs.
These are entirely reasonable criteria. I have asked (in the couple of comments preceding this) for clarification of how HFT is supposed to lower trading costs, beyond the value of the speedups that I’ve already accounted for in this post. The answers will presumably address at least your question 1).
@Daniel,
To the question of whether diminishing the value of that information to the rent seeker will discourage the rent seeker from obtaining it the answer is “no” because informational edge is always alpha. The rent seeker will only stop seeking rent when he is sure there will be no rent.
Speed is always important. If a floor trader can react faster than the rest of the crowd then he has edge. However, if he is faster but dumber than the crowd, then the speed edge is eroded and may actually work to his detriment. Same goes for algos.
Why do high-speed transmission lines lower my risk of getting “picked off” (as BC puts it)? If I’m at risk for, say, 10% less time, but the picker-offers are 10% faster, why am I not facing exactly the same risk as before?
It’s a race between all market participants. Events occur in continuous time, not in discreet time increments. Zero latency is the ideal in trading and we have been moving toward that ideal since the dawn of time. IF I have informational edge (which most likely comes from my continuously updating pricing model for which speed is also obviously important) and I see a market maker’s quote and realize that the information I have is not priced in, I can trade with him before he moves his market. I can realize my theoretical edge by picking him off. Obviously, it’s in the market makers interest to update his pricing models faster and move his market before I pick him off. This was the case before algorithmic trading and before HFT and it’s still the case and always will be. I always want to have better pricing models and smarter and faster algos than my competition. There’s nothing you can do, no regulation you can propose to change these facts or alter these incentives.
The question of when and whether regulations are justified is quite separate and I think it’s a bad idea to clutter this discussion with that one.
But, Steve, the proposed “fix” in your main post IS regulation.
What if trades were executed anywhere between 0 and 1 second after they were submitted, with the delay chosen randomly?
Who would create such a rule? Who would impose it on unwilling exchanges? Who would ensure compliance? The regulator. Any rule inflicted on all in the financial markets is regulation.
Unless you argue for voluntary adoption of that rule. In that case, we need to look to IEX, which is a start-up exchange that imposes a speed bump. The IEX has been able to attract only 1% of U.S. trading volume.
@Methinks,
Then what you say is contrary to economic thought. Reducing the inventive for an action will almost always make it occur less since you are reducing the marginal benefit to obtain the next piece of information? If you think it is not the case in this specific instance you need to explain why. Am I wrong here Professor Landsburg?
What we need to do is rework the regulations so sub-penny spreads are possible. Then liquidity takers will, again gain from the competition between HFT’ers.
There is no question that sub-penny pricing shouldn’t be prevented – although, my vote is to eliminate regulation entirely because it invariably hinders rather than improves markets with this being one example.
However, as usual, market participants have already figured out how to get around this regulation and drive spread pricing to sub-penny while remaining technically in compliance (just as participants got around the uptick rule and other moronic regulations).
Say the market is 0.01 at 0.02
A customer order to sell 1000 shares is submitted. The market maker pays 0.02 for 100 shares and 0.01 for 900 shares. The average execution price for the order is 0.011. Voila, sub-penny spreads.
This promise of tighter spreads is one way of paying for order flow.
Obviously, this method is way more cumbersome and spreads are still much wider they would be if the SEC didn’t pour sand in the gears of competition by mandating penny increments.
@Daniel,
Perhaps it would help if I explicitly mentioned that technology is also driving down the cost of obtaining information. So, if the marginal revenue from exploiting information at least equals the marginal cost of acquiring it, you will continue to acquire and exploit information. We probably agree that the profit maximization point is still MC=MR?
@Steve Landsburg
You probably figured out already from my response to your question about speed that if the market maker and I are both getting the information at the same time the MM moves his market at exactly the same time my order arrives and there is no trade, but the new fair value price is reflected. So, if market participants move at the same speed neither can get the better of the other on speed alone. However, price discovery still happens faster. Information flows at its own (continuous) pace regardless of the speed at which we can respond to it.
@Methinks,
Of course we agree on the profit maximization point. However, bringing in other factors seems like a way to dodge the issue. It’s like saying that orange growing is not harmed by orange thieves because we’ve invented this new fertilizer that decreases the cost of growing (sorry to use the analogy of thieves again, it’s just the easiest direct harm I could think of). The orange groves are still being harmed by the thief but the mc is being reduced at the same time that your marginal benefit is being reduced, so it’s possible to end up at the same point or higher. At the very least from what you’ve told me, we need to consider the cost to the information finder as another externality to the speed up of HFT.
@Daniel
You’ve changed the subject.
Thieves cause harm, of course, but even if you lose part of your crops to thieves, if the marginal cost of farming oranges doesn’t exceed the marginal revenue, then you’ll keep farming.
That speaks to your original question whether a reduction in rents to rent seekers translates into a loss of incentive to seek and exploit information (in this case) which then leads to a less information efficient market. The answer is clearly “no”. As information provides edge (alpha, excess return, rent) everyone has incentive to seek and exploit it. If there is no alpha, then that means that the security is priced correctly and there’s nothing to exploit.
You have to understand that HFT firms (like ll marktet makers) are ALSO seeking and trying to exploit information. Non-HFT firms also have algos that try to pick off HFT and all other market participants – including that poor old grandma and the mutual fund she invests in that always gets dragged out whenever an emotional argument is made about the markets.
Plus, it’s not really like theft. It’s more like reading your opponent’s hand by his actions at the poker table. Like poker, trading is a high volatility game. Even if you lose to me today, assuming we’re reasonably well matched, there’s a high enough probability of you winning against me tomorrow that you are willing to play again tomorrow. The key is knowing at which table to sit.
@Methinks,
“Thieves cause harm, of course, but even if you lose part of your crops to thieves, if the marginal cost of farming oranges doesn’t exceed the marginal revenue, then you’ll keep farming.”
Right, and for some farmers marginal cost will exceed marginal revenue due to the orange grove thievery (in a competitive market) so they’ll either drop out of farming or farm less until marginal revenue once again equals marginal cost. This would translate into a leftward shift of the supply curve and less oranges for everyone. I see that everyone is seeking information but if some of the value of seeking this information is leaking to competitors that have better or faster algos, that will reduce the incentive overall to seek information. Unless for some reason you’re arguing that information seeking is not a competitive market? The question is not whether information seeking will still happen or not, the question is will it happen less as a result of faster algos. From what you’ve said so far, I am persuaded that faster algos shifts the supply curve for information seeking leftward.
“Plus, it’s not really like theft. It’s more like reading your opponent’s hand by his actions at the poker table. Like poker, trading is a high volatility game. Even if you lose to me today, assuming we’re reasonably well matched, there’s a high enough probability of you winning against me tomorrow that you are willing to play again tomorrow. The key is knowing at which table to sit.”
I know it’s not exactly like stealing, I was just using it as an example of a cost to the supplier. I don’t see this exactly like poker either. In poker, nothing is being produced except for the enjoyment of playing the game. And might I say that poker is a really bad analogy for your position as well since it’s always a zero sum game. No, I think the financial sector is producing something (efficient allocation of capital) and so shifting the information seeking supply curve leftward is an external cost of faster algos.
@Daniel,
Your orange theft is a poor analogy not because you used the word “theft” but because when thieves steal the oranges from the orange grove, the thieves are the only ones who benefit. When information is read from a counterparty’s publicly available order flow it is by the reader’s response released to the entire world and the entire world benefits.
Imagine if a poker player raises. That makes you think he has a strong hand. If the next guy re-raises, you think the second player has a stronger hand than the first. You assess your hand relative to the observed behaviour of players one and two and make a decision about your next action. BTW, Poker is so analogous to trading that one very large Options market maker has always used poker to train its traders.
@Daniel,
if some of the value of seeking this information is leaking to competitors that have better or faster algos, that will reduce the incentive overall to seek information.
We’ve covered this. It’s a symbiotic relationship. Without anyone to extract rent from, the information is useless to the information seeker. To extract any rent at all, the rent seeker must necessarily lose some of the value to the parties its trading with.
that will reduce the incentive overall to seek information
No, that’ll reduce the number of participants who aren’t keeping up with the closing gap between information’s appearance in the world and their ability to respond to it.
BTW, third parties who are not in this fight benefit because they get the information for free. So, If AQR is better at pricing stock X and releases that information to the world by trading, it will realize some but not all of the theoretical edge and this activity reprices the stock to the new fair value almost immediately. Mutual fund ABC which is just looking to rebalance its portfolio will then have a higher probability of trading at the actual fair value (minus the dealer’s spread)
We are all slaves to information, not to HFT. HFT is as much slave to information anyone else. Information changes valuations and any lag between its appearance in the universe and your ability to respond to it is a time in which you are vulnerable. There’s always incentive to seek information.
Empirically, what you describe is not happening. It’s the opposite. More liquid, faster markets reflect information faster. Slower, less liquid markets are rife with mispricings (unreflected information).
Methinks:
You probably figured out already from my response to your question about speed that if the market maker and I are both getting the information at the same time the MM moves his market at exactly the same time my order arrives and there is no trade, but the new fair value price is reflected. So, if market participants move at the same speed neither can get the better of the other on speed alone. However, price discovery still happens faster. Information flows at its own (continuous) pace regardless of the speed at which we can respond to it.
So this seems to confirm the validity of my calculation — When everyone gets faster, there’s no change in risk, so the only benefit is the speed itself, the value of which I’ve put an upper bound on.
@Daniel
Given that reality is not supporting your theory, I think that you must explain why your theory doesn’t fit reality rather than me explaining why reality doesn’t fit your theory.
re Poker: If all Poker produces is enjoyment and enjoyment has value, then playing poker produces something of value. Also any single trade IS zero-sum at the instant it happens (not necessarily afterwards). But that wasn’t my point. I was using Poker as an analogy for decision making under uncertainty with asymmetric information.
I’ll check back later. Right now I have to go spread evil throughout the world with my satanic algorithms. MWAAAAAHAHAHAHAHA!
@Steve Landsburg
The risk is generated by the lag between the appearance of information and your response to it. That’s the vulnerability. The longer the lag, the more information you are potentially exposed to and so the more risk you are forced to accept. As you obviously know, this is why, ceteris parabus, interest rates on long-duration bonds are higher than on short duration bonds.
You can force everyone (at least on U.S. markets_ to take more risk by slowing down trading by any amount you want, but it’s wrong to think you’re neutralizing risk by doing this. And you’re fighting strong incentives to reduce risk by getting faster.
Also, Steve, you are slowing down price discovery by slowing down trading. What is the benefit of that?
This is where an edit function would be nice. I meant to say you’re slowing down price discovery by slowing down the lines that carry feeds.
@Methinks,
Okay, it might still be my misunderstanding but I still don’t see how IF the information seeker does not gain as much as he otherwise would have given the faster algo (as opposed to a slower algo not as opposed to nothing), we can say that exactly the same amount of information will be sought. You still haven’t answered concretely on the question, if all else equal we have faster algos, will the information seeker get less of a benefit by revealing information through trades? I understand that your job precludes you from answering right now, and appreciate all the time you’ve put in helping us to understand.
Methinks:
Also, Steve, you are slowing down price discovery by slowing down trading. What is the benefit of that?
That’s not a benefit; it’s a cost. In fact it’s exactly the cost that I (over)estimated at the top of this post.
Methinks:
The risk is generated by the lag between the appearance of information and your response to it. That’s the vulnerability. The longer the lag, the more information you are potentially exposed to and so the more risk you are forced to accept.
This continues to make absolutely no sense to me. If the lag is ten times as long and the people trying to take advantage of me are ten times as slow, how is my risk increased?
Steve,
The crux of the matter is that 0.003 seconds is a lifetime when you’re talking about transaction times on the order of 1/5 second (travel time along cable, OS context switch, going out to DRAM, etc.).
Your normalization (denominator) is the number of seconds in the year. You are basically saying that Spread Networks saving of 0.003 seconds should be measured relative to the number of seconds in a year; or that Spread Networks has shaved off 0.003 seconds in a year. Why would this have any relevance to the question at hand? Can only one transaction per year be done over this fiber-optic cable? That is the calculation you are suggesting!
I’m suggesting that the flow of information has increased substantially.
Let me illustrate our two back of the envelope calculations another way. Consider a pipe that carries 15 trillion units of something per year. Your calculation suggests that the pipe is able to increase production by 0.003 * 15*10^12 /(24*3600*365) = 1400 units. Very small.
My calculation is the improvement of 0.003 seconds relative to the old baseline. (In other words, how much the flow rate of the pipe has increased.) The baseline I was using was 0.2 seconds, the old transaction time. So the flow has increased by 0.003/0.2; i.e. 1.5%, not an insignificant amount when you’re talking about 15 trillion units. (Maybe my 0.2 seconds is an order of magnitude too low, but your argument says that the baseline was one year in seconds. Very mysterious.)
Now consider that that pipe is not carrying oil or natural gas or water, but rather photons or information or dollars. The information is traveling a good deal faster using my calculation, whereas it is an insignificant speed-up in your calculation.
I’m not saying my calculation is perfect, but it is a much better starting point.
Trey
if all else equal we have faster algos, will the information seeker get less of a benefit by revealing information through trades?”
Yes!
Another way to look at that is the market will pay less for information.
Information not yet priced in (alpha) will continue to be sought until MC=MR. At MC=MR, the security reflects all relevant information available in the universe.
If we are in a constant state where there is no alpha in seeking information it is because all information is reflected in the price INSTANTLY. We don’t have to expend resources to discover it. That is perfect efficiency.
That perfect efficiency is what we’re moving toward. Market making spreads will continue to decline and upon reaching theoretical perfect efficiency there will be no market making spreads at all and I’ll be out of a job completely, but …WAAAAAIT!!!!! Efficiency renders all my assets less valuable until they are worthless!!!!!
I change my mind and join all the other ancient and slow market makers complaining they can’t compete with traders moving the market to perfect efficiency and join our lobby to demand a slow-down of orders (not my quotes, of course. Those I should be able to move as fast as I want. I only want orders slowed down)!
Of course, it’s possible that the benefits of a perfect market don’t outweigh the cost of achieving it at current prices for available technology (assuming the technology exists), but I still don’t think we’ve actually looked at the benefit of moving toward perfect efficiency because the reason we’re getting faster isn’t because anyone is holding their breath, unable to get on with their lives until the trade is executed.
@Steve Landsburg,
This continues to make absolutely no sense to me. If the lag is ten times as long and the people trying to take advantage of me are ten times as slow, how is my risk increased?
Orders are submitted continuously, not in three millisecond intervals. If your order is submitted one millisecond before an event occurs to change the fair value, the speed bump prevents you from canceling and replacing your order before it’s matched against another order that has been sitting on the book for two milliseconds when your order arrives. You trade at a price you wouldn’t have wanted to trade at. And the other party doesn’t necessarily have to have any information for that to happen. You’re going to require a higher spread to compensate for that risk.
Whether trades get executed 0.03 seconds faster or slower has no impact at all on the real economy.
The entire purpose of the investment is the creation of an information gap, which traders then exploit. From the viewpoint of the economy, the investment is a deadweight loss.
It’s a good example of the modern US financial industry.
A fairly large part of its revenues and profits derives not from capital intermediation, but from rent seeking, information destruction, breach of contract and the exploitation of information asymmetries created specifically for this purpose.
The externality only falls on users of that stock exchange and so creates an entrepreneurial opportunity for other exchanges (although it’s a fairly thin market with high switching costs).
Methinks: So you’re saying that if I’m being chased by a lion, it doesn’t matter how fast the lion is?
Methinks: So you’re saying that if I’m being chased by a lion, it doesn’t matter how fast the lion is?
You’ve lost me. I don’t understand how this relates to my last response to you.
SL:
Assume a spherical lion….
Steve:
Now if HFT does in fact reduce bid-ask spreads, I agree that we should infer this is a measure of social value. But I am skeptical of this, because I don’t see how it reduces risk. If someone can explain that to me, then I’m willing to revisit my calculation.
There are two sources of risk to a market maker — the first is informed, sophisticated traders who have access to the same high-speed technology as the market maker himself, and who may (not all of them are high-speed traders) speed up by the same degree with lower latency links. The second is random noise, from events in the real world as well as from uninformed traders who are not using high-speed links (because it isn’t worth paying for it to them).
A model for a market maker could be that he posts bids and offers around his estimate of fair value. Uninformed traders at random times hit his bids or lift his offers in random size. All this happens in continuous time (the real world is continuous time, not subject to a discrete ticking clock). The “inventory”, or net risk position of the market maker will follow a random walk, with deviations increasing with the square root of real time. To control his risk, the market maker must become a more aggressive buyer if his risk position drifts to the short side, and vice versa a more aggressive seller if his risk position drifts to the long side, purely from chance fluctuations. The quicker he can make adjustments, the lower the capital he needs to support his risk, and the tighter he can make his bid/ask spread.
This also applies to a market maker operating in both Chicago and New York. He has to communicate his updated risk positions back and forth between the two markets, and the faster he can do it, the lower the bid/ask he needs to charge in both markets.
I have another question about the notion that the first past the post nature of a high-speed link making overinvestment more likely.
Isn’t it more likely to result in underinvestment? Let’s say that there are 5 companies who all have the idea of building a high-speed link. All of them know that only the fastest one will be able to generate any revenue. They’re all wildly optimistic, so each of them believes that there is a 50% probability that they will be the winners. Even in this scenario, the investment will only happen if the social benefit (which is the most that the winner can collect in revenue) exceeds the cost by at least a factor of 2. So it seems like there’s a cliff effect, if the benefit is less than 2x the cost, no-one will invest and no link will get built. If it’s more than 2x the cost, all 5 of them will invest and we’ll end up with 5 links of which only one is useful.
What reason do we have to believe that we are in the region where overinvestment is likely as opposed to underinvestment?
Trey:
Your normalization (denominator) is the number of seconds in the year.
Not at all. I said that a 15-trillion dollar-a-year economy produces about $1500 worth of output in .003 seconds.
I could just as easily have said that a 150-trillion-dollar-a-decade economy produces about $1500 worth of output in .003 seconds.
Or that a 1.25-trillion-dollar-a-month economy produces about $1500 worth of output in .003 seconds.
It doesn’t make a bit of difference whether you quote the rate of output in years, decades, months or seconds. I quoted it in years because 15-trillion-a-year is an easily recognizable figure.
What matters is the $1500-per-.003 seconds. That’s an absolute figure. It’s not relative to anything. There is no “one year baseline”. You are extremely confused.
nivedita: Thanks for attempting to answer the question that Methinks keeps ignoring.
So let’s make sure I understand this:
I’m in the jungle, where I’m periodically chased by lions, and periodically chased by snakes. If both I and the lions double our running times, my risk from the lions is unchanged but my risk from the snakes is reduced. Is that, in essence, your story?
Methinks:
You’ve lost me. I don’t understand how this relates to my last response to you.
That’s because you’ve so thoroughly ignored my question that at this point we’re talking about two different things.
nivedita:
Isn’t it more likely to result in underinvestment? Let’s say that there are 5 companies who all have the idea of building a high-speed link. All of them know that only the fastest one will be able to generate any revenue. They’re all wildly optimistic, so each of them believes that there is a 50% probability that they will be the winners. Even in this scenario, the investment will only happen if the social benefit (which is the most that the winner can collect in revenue) exceeds the cost by at least a factor of 2. So it seems like there’s a cliff effect, if the benefit is less than 2x the cost, no-one will invest and no link will get built. If it’s more than 2x the cost, all 5 of them will invest and we’ll end up with 5 links of which only one is useful.
I don’t think you’ve described an equilibrium. If nobody builds, then everyone wants to build.
The equilibrium is that everyone builds with probability p, where the fact that everyone else is building with probability p is just enough to make me indifferent between building and not building.
To compute that probability p, I’d have to know how your expectation of success varies with the number of other entrants. If I take your model seriously, it says that my expectation of success is 50% *regardless* of the number of other entrants, but that’s not plausible. So: You give me the expectation as a realistic function of the number of other entrants and I’ll compute p for you.
Steve:
I’m in the jungle, where I’m periodically chased by lions, and periodically chased by snakes. If both I and the lions double our running times, my risk from the lions is unchanged but my risk from the snakes is reduced. Is that, in essence, your story?
Roughly speaking, yes.
Your equilibrium point is interesting as a math problem at least.. Suppose to begin with the expected probability is just 1/k if k people decide to go ahead. For optimism maybe just try min( 1, 2/k )
Steve: btw, it’s only “roughly”, because a lower latency link can also help you against the lions.
Here’s an example: you’re making markets in both New York and in Chicago. Each market on its own trades much faster than the latency of the link between them. An informed buyer tells his computers in both markets to start buying. Until your Chicago and New York computers can talk to each other, they don’t know the true extent of the new demand, and will not move their prices up aggressively enough.
The lower latency link, by making that communication faster relative to the speed of the two individual markets, lowers your risk of getting picked off by the informed trader. On the other hand, the link doesn’t help the informed trader do much better, since he’s going to be willing to keep buying in both markets as long as the price hasn’t moved too much against him (he’ll benefit from knowing how much has already been filled in the other market, but he’s likely not as much concerned about getting longer than he planned as the market maker is concerned about getting shorter than he planned).
nivedita:
I wrote:
I’m in the jungle, where I’m periodically chased by lions, and periodically chased by snakes. If both I and the lions double our running times, my risk from the lions is unchanged but my risk from the snakes is reduced. Is that, in essence, your story?
You wrote:
Roughly speaking, yes.
Okay. But if, eventually, the snakes also double their running times, then my risk is right back where it started, right? And if there was a cost to doubling everyone’s running times, then that cost is pure social waste, right?
nivedita:
Here’s an example: you’re making markets in both New York and in Chicago. Each market on its own trades much faster than the latency of the link between them. An informed buyer tells his computers in both markets to start buying. Until your Chicago and New York computers can talk to each other, they don’t know the true extent of the new demand, and will not move their prices up aggressively enough.
The lower latency link, by making that communication faster relative to the speed of the two individual markets, lowers your risk of getting picked off by the informed trader. On the other hand, the link doesn’t help the informed trader do much better, since he’s going to be willing to keep buying in both markets as long as the price hasn’t moved too much against him (he’ll benefit from knowing how much has already been filled in the other market, but he’s likely not as much concerned about getting longer than he planned as the market maker is concerned about getting shorter than he planned).
Aha! At last a story that, at least on the face of it, appears to make sense and to point to a social benefit I overlooked. (Though I want to think harder about this before completely buying in.)
Assuming this is right, though, it leaves us here: I pointed to a social cost, you pointed to a social benefit, and we have no idea which is bigger.
Snakes are volatility, but lions are pure negative expectancy. Lions don’t just run. Lions sneak up on you so they don’t have to run.
Even if speed sometimes increases an informed trader’s chances of picking off a market maker, any single trader is an informed trader only some of the time at best. So an extremely fast line is unlikely to be worth the additional expense. Without it, a trader can use stealth to at least partially accomplish his goal when he is informed.
The market maker, on the other hand, faces the threat of lions 100% of the time. To a market maker, speed is much more important because dodging lions is crucial to survival.
I have been away, and trying to catch up with this one, but probably failing completely. The thing that worries me is the nature of the information everyone is talking about. The market will exhibit noise, with almost random fluctuations. The information about these fluctuations tells us nothing about the underlying businesses. If an informed trader has information about the real value of a company affecting the share price then establishing a new price is a social benefit. Is it social benefit to predict the fluctuations a millisecond ahead which have nothing to do with actual conditions? Methinks said that if information were perfect and instananeous, then on one party gaining new information, the new “fair” price would be established without trade taking place as both seller and buyer would adjust at the same time (at least I think that is what he said). Given that there will always be fluctuations in the price that do not represent meaningful information, but are just “noise”, then does not faster trading simply increase the volume of trades without reflecting anything useful in the information. The faster the communication and the algos, the more sensitive to noise rather than “real” information the markets become. My worry is that noise could come to dominate the trades.
I suspect I am hopelessly adrift, but if it is possible to clarify this easily I would appreciate it.
Harold,
If the fair value changes from .42 to .43, this is a real event and there are only two ways to reflect it:
1.) If there are no orders on the book, market makers will move the market (bid/ask) from .41 bid at .43 to .42 at .44 with no trade printing.
or
2.) Any shares offered at .43 are lifted and those trades are printed as the market moves to .42 at .44.
But, that’s not noise. That’s real information about the fair value of the security.
Noise is generated when, say, market makers quickly post and cancel markets in an effort to glimpse lions (informed traders) in the tall grass and by lions sending and canceling orders in order to confuse market makers as they stalk (Sorry. I’m really warming to this analogy).
This noise is typical of any negotiation, though. You try to get as much information about the party you’re negotiating with and give up as little to him as you can. You would never start a real estate negotiation by gushing about how much you love the house and can’t imagine living anywhere else, for example.
Steve:
Assuming this is right, though, it leaves us here: I pointed to a social cost, you pointed to a social benefit, and we have no idea which is bigger.
I have also pointed out though, that the benefits in reduced bid/ask spread only have to be miniscule before the benefits exceed the cost.
Further, your argument is based on your premise that a faster data link lowers the value of the current best link by a large amount. This is not clear. Most traders probably do not currently subscribe to the fastest link (outside of the big market makers), but might if the price dropped a little. Once the big guys move off to the faster link, the second tier might well subscribe to the existing link as its price comes down. So the external cost on existing links is likely to be quite small. Where’s your argument for why it should be large?
Methinks (and Harold), I think there is real noise in the market as well — if an uninformed trader buys, and the price moves in response to that because the market couldn’t recognize it as an uninformed trade, that is noise. The improved technology allows the market to better distinguish informed and uninformed trading, and so in fact helps to reduce these noise moves.
Methinks and nivedita. Thanks for responding to my naive questions. One more quick one, then I will retire to read more abuut it. What is the “informed” trader actually informed about?
From here:
“The aggregate benefit to investors from the presence of HFT [through its effect on spreads] is on the order of $9-billion per year – nearly an order of magnitude greater than the entire HFT industry profit for 2013.
HFT can offer vastly narrower bid/ask spreads than traditional market makers precisely because they trade so fast. By executing a round trip buy/sell or sell/buy in seconds or even milliseconds, the HFT reduce to a bare minimum the risk that the price of the security will change adversely between the first and second leg of the round trip. The speed of HFT trading thus benefits all traders by enhancing liquidity and lowering trading costs.”
Biopolitical:
We’ve been over this ground. The story you’re quoting, on its face, makes no sense. If you double your speed and the lions who are chasing you also double their speeds, then you’re no less likely to get caught than before. So this story, by itself, cannot account for a reduction in bid-ask spreads.
At the very least, you need a more sophisticated story along the lines of what nivedita has suggested. I’m still mulling that one over.
methinks @143 is an accurate picture of the real world finance in the US.
HFT strategies are mostly market making strategies. Lions don’t currently invest in speed, and don’t consistently/frequently have enough information to make acquiring speed worth their while (sneaky is good enough).
nivedita:
Perhaps you can help me with another question:
You’re arguing that the social benefit of HFT comes through risk reduction. But risk reduction is valuable only insofar as traders are risk-averse. If these traders are so risk-averse, why are they playing in these markets in the first place? Don’t risk-averse people just buy Vanguard mutual funds?
Miko,
I believe that’s what I said. Even if speed might help on the occasions when they are lions, the lions are lions only some of the time (at best), so investing in speed isn’t worth it for them.
Other market makers are also sometimes lions and they will pick off slower or less astute market makers.
@methinks
Sorry, my wording was unclear, i was reiterating your point, and saying that i thought your post was the correct response to some of the questions raised in this thread (J^-^)J
The mistake is mine, Miko. the first few letters blurred together and I read “inaccurate”. At the screen too long. My apologies.
Steve: I’m arguing that the benefit comes from reducing the risk of liquidity providers (i.e. market makers), which means they can provide more liquidity at a lower transaction cost. This is no different from Amazon reducing the cost of intermediating between book publishers and book readers and providing a net social benefit: would you argue that this reduction in cost only has a social benefit to the extent that book readers are actually averse to paying high markups, and if they’re so averse, why don’t they just go to the public library instead? Your average Joe who doesn’t have the time or inclination to become well informed about all the various publicly listed companies in the market, and how to efficiently purchase shares in them, can simply buy the Vanguard mutual fund, and because of HFT, his mutual fund has lower management costs than it otherwise would. All this infrastructure is providing him a great product — a pre-packaged way of investing in a diversified stock portfolio — at an astonishingly low cost. What’s not to like about it?
Your question doesn’t make much sense anyway. Risk is just a cost of living and participating in society, in general, not just in the financial markets. The fact that you’re risk-averse should not raise the question of why you’re not just cowering under your bed in Mom’s basement, instead of risking your neck commuting to work.
Steve,
>> What matters is the $1500-per-.003 seconds. That’s an absolute figure. It’s not relative to anything. There is no “one year baseline”. You are extremely confused.
No, as we will see, the proper result is $1500/TRIP (or round trip — I’m not sure from the article you linked to). It IS relative to something — the number of trips. You’ve discarded an important unit. So I am only confused by how you could have missed this.
Let’s look at the units in your calculation more closely. The units would be the following:
GDP: $/year
dt = delta time (= 0.003): seconds / transaction (there HAS to be a “transaction”, or “cycle”, or “round-trip” in the denominator. THIS is what Spread Networks is offering — an improved time PER TRIP.)
SPY = “seconds per year” (= 24*3600*365): seconds / year
Given that, your calculation is
GDP * dt/ SPY
and the unit breakdown is
($/year) (seconds/transaction) / (seconds/year)
The units are then dollars per transaction. To calculate the social value you have to multiply $1500 by the number of transactions (which could be very large and would put a huge number on the social value).
You are certainly not suggesting that there is only one transaction ever used on this network???
Since I doubt this will convince you, I came up with an analogy.
Suppose in the future we have a city on the moon. In the past a reusable rocket has been able to make the round trip in 13 days. Over the course of a year, a total of 15t of something is delivered from Earth to the moon. (One can think of “t” as standing for Tons of fuel, or Tera nanoliters of water.) Along comes Spread Rocketworks: they are able to make the trip to the moon in 10 days, with the same amount of cargo, thus saving three days (which is analogous to the 0.003 seconds that Spread Networks is able to shave off).
Here is your calculation:
(15t/year) (3 days) (1/(365days/year)) = 0.12t
According to your method, the total social value to the citizens of the moon is 0.12t.
(In the above calculation I purposely left the units of decreased travel time as you had: (3 days) instead of (3 days/round trip).)
Let’s calculate the real value to the Loonies (SciFi reference). Using Spread Rocketwork’s faster technology, the new total amount delivered is the ratio of the old time to the new time:
(15t/year)*(13 days/round trip)/(10 days/round trip) = 19.5t per year.
The improvement is 19.5-15.0 = 4.5t per year. Yet by your calculation the value is a much smaller 0.12t, a factor of 36 or so lower. What gives? Once again, your value doesn’t include the number of round trips.
We can recover the 4.5t per year by taking the 0.12t per round trip (the actual units) and, thanks to the improved speed, multiply that by 365/10, the number of round trips per year with the faster rocket.
(0.12t / round trip ) * (365 days/year) / (10 days / round trip)
Viola! Out comes 4.5t /year.
I hope that you now see that $1500 is relative to round trips, transactions, cycles, … something. You have to factor those in to get meaningful answers to Spread Networks value, just as one has to factor in the number of trips to Luna to get the overall value of Spread Rocketworks to the Loonies.
Trey
Trey:
The units are then dollars per transaction. To calculate the social value you have to multiply $1500 by the number of transactions (which could be very large and would put a huge number on the social value).
Yes, of course the units are dollars per transaction.
If we assume that the entire economy is on hold awaiting a given transaction, then the social benefit of shaving off .003 seconds per transaction is $1500 per transaction. If we assume that 1/1,000,000 of the economy is on hold, then the social benefit of shaving off that .003 seconds per transaction is about an
eighth of a cent per transaction.
That social benefit has to be compared to the private benefit of shaving off the same .003 seconds per transaction.
I claim that a) assuming 1/1,000,000 of the economy is on hold awaiting the transaction is probably an overestimate.
b) Therefore the social benefit of shaving off .003 seconds per transaction is at most 1/8 of a cent per transaction.
c) The private benefit of shaving off .003 seconds per transaction is probably greater than 1/8 of a cent per transaction.
d) Therefore the private benefit of shaving off .003 seconds per transaction is greater than the social benefit of shaving off .003 seconds per transaction.
You can do everything in total dollars, or you can do everything in dollars per transaction. I chose to do the latter.
Which, if any, of the four steps a) through d) are you not following?
“c) The private benefit of shaving off .003 seconds per transaction is probably greater than 1/8 of a cent per transaction.”
Why do you assume the private benefit is greater than 1/8 of a cent per transaction?
The NYSE pays HFT liquidity providers $0.0015 per transaction.
US trading volume is roughly 7billion transactions a day. HFT account for 50% of that. The total profits for all HFT firms were about $1 billion (and some of that money comes from other markets besides equities). If you work out the profit per transaction, is comes to .07 cents. Obviously the private gain of the spread cable must be lower than that.
Steve said: “If you double your speed and the lions who are chasing you also double their speeds, then you’re no less likely to get caught than before.”
Maybe the key here is that relevant events in the world outside htf do not accelerate. If communication among potential counterparties accelerates relative to those events, there is less risk that prices become obsolete. As a result, spreads, risk and liquidity improve for all investors fast and slow.
In other words, hft would not simply be a rat race where the winner takes all but a system where several parties gain (less risk and more liquidity) by sharing private information (such as their willingness to take risks or their liquidity) faster relative to the rate at which relevant public events occur, and the improvement in risk and liquidity spills over to other investors not so interested in fast communication.
I think several commenters, including Methinks, nivedita and Steve have made similar points, so I am not sure whether I am saying something new or restating the obviously wrong.
I’m still trying to work out this dubious “winner takes all theory of market making”.
After all, that would mean that one market maker would have to not only be faster than all other market makers, but have unlimited capital, be able to bear unlimited risk and have perfect pricing ability. Since this is fantasy, the monopoly firm’s spreads will be very wide and the size it is willing to do will be relatively small (liquidity will be very low), providing enough rent to entice competitors to serve the market.
Where have I erred?
Eric Johnson:
Why do you assume the private benefit is greater than 1/8 of a cent per transaction?
Perhaps this was an error, and the numbers you quote are good evidence for that.
This is quite irrelevant to Trey’s confusion, which has nothing to do with the accuracy of the numbers and everything to do with the conceptual question of what to do with the numbers once you’ve got them. So in the mini-discussion with Trey, I’ll stick to the original numbers rather than muddy the waters by trying to deal with two issues at once.
But as far as the broader discussion goes, you could well be right that 1/8 of a cent is a vast overestimate of the private benefit. On the other hand, I expect it’s also a vast overestimate of the social benefit, and I’m not sure which overestimate to believe is bigger. Your numbers are a big contribution to resolving that, though. Thanks.
Trey:
Here is your calculation:
(15t/year) (3 days) (1/(365days/year)) = 0.12t
According to your method, the total social value to the citizens of the moon is 0.12t.
This bears not the slightest resemblance to my calculation.
The correct computation (which exactly mirrors the computation I did for HFT) is: Suppose people on the moon are able to produce $1000 worth of moonstuff per day, but production comes to a halt while waiting on this delivery from earth. Then the social value of getting the delivery there 3 days sooner is 3 days x $1000 per day = $3000 .
(Your calculation treats the total amount of stuff delivered to the moon as somehow relevant. This would be like treating the total volume of transactions on the financial markets as somehow relevant. Neither makes a lick of sense. Note that while your calculation requires knowing that total volume per year, mine does not. And note that while my calculation requires knowing something about the rate of production of real goods, yours does not. So the calculations cannot be analogous.)
I am surprised that there is still argument over whether HFT reduces bid-ask spreads. The cited article says “impressive body of academic evidence demonstrating that when HFT become active in a given financial market, bid/ask spreads substantially narrow.” I thought that this would be an obvious consequence of how markets work. Is there some error in all those academic papers?
Maybe the lion-snake analogy shows that the fast HFT traders offer no social benefit to the slower HFT traders. I can believe that. But the social benefit is to ordinary investors who simply want a good price when they decide to buy or sell. They are not trying to outrun lions.
Steve: the total volume of goods produced (or at least their dollar value) per year is the rate of production of real goods. And both of you are assuming this value is directly proportional to the amount of inputs delivered, otherwise your calculation doesn’t make sense either — you’re assuming that if they had the stuff from the earth those three days, they’d be able to produce value at the same rate as on the other days.
Trey’s calculation is really exactly the same as yours just expressed in different units. 0.1233t is the benefit per (new) trip, 4.5t is the benefit per year. 4.5 = 0.1233 * 365/10.
and of course, there’s some exchange rate depending on the dollar value of those goods that converts between “t” and “$”
Now, coming back to your calculations in the post — of course the volume of financial transactions is relevant. How else are you going to convert the $300mm into a number per transaction so you can compare it to your estimate of benefit per transaction? I think that doing it per transaction is an unnecessary exercise, because most economic data is available on an annual or quarterly basis, not some sort of per-transaction basis, but if you want to do it per-transaction, you better know what the number of transactions is.
Given that the financial sector, for better or worse, currently represents 6.6% of US GDP according to the BEA, an estimate of 0.0001% is likely to be wildly off the mark. “Securities, commodity contracts, and investments” was 1.1% of US GDP ($185bn) in 2012 (latest year on BEA website). And this is just the value-added, not the gross revenue of this industry, which would be the better measure of the value that society is putting on financial trading.
The CME alone collected revenues of almost $3bn in 2013. Assuming that 20% of this originated due to orders from New York, and that the annual volume of those orders increases by 25% (the speedup from the new line: 13.3ms vs 16.3ms), that gives us a benefit of $150mm per year.
nivedita:
Trey’s calculation is really exactly the same as yours just expressed in different units.
You are absolutely, unambiguously wrong.
Trey’s calculation is this:
(15t/year) (3 days) (1/(365days/year)) = 0.12t$
Here his “15t/year” is, by his assumption, the rate at which goods are delivered from the earth to the moon. *Nowhere* in this calculation do you see the rate at which *reals goods and services are produced* on the moon.
In my calculation you will see a “$15 trillion” per year, which represents the rate at which goods and services — heads of lettuce, locomotives, haircuts — are produced in the US.
Just because I used “15 trillion” and he used “15 t”, it does not follow that we are talking about analogous things. My $15 trillion/year is in units of “value of output per year”. His “15 t per year” is in “units of stuff delivered from the earth to the moon per year”. It has nothing to do with production.
To see this, suppose that due to a technological change, the moonfolks are suddenly able to produce twice as much per day as before, even though deliveries from earth continue at the same rate of 15t per year as before. Then by my calculation, the social benefit of a speedup in deliveries is doubled, whereas by his calculation it’s unchanged. My calculation is right and his is wrong.
nivedita:
Now, coming back to your calculations in the post — of course the volume of financial transactions is relevant. How else are you going to convert the $300mm into a number per transaction so you can compare it to your estimate of benefit per transaction?
I can’t even tell what you’re thinking here, but whatever it is I’m quite sure it makes no sense, because the volume of transactions is surely irrelevant.
If by shaving .003 seconds off a single transaction I can earn a profit of $1, then the private benefit of that shaved .003 seconds is exactly $1. It does not make a bit of difference whether that $1 represents 100% or .00000001% of all the transactions carried out today or this year.
Steve
If by shaving .003 seconds off a single transaction I can earn a profit of $1, then the private benefit of that shaved .003 seconds is exactly $1. It does not make a bit of difference whether that $1 represents 100% or .00000001% of all the transactions carried out today or this year.
Sigh. And how exactly are you going to calculate how much it cost to do that, when all you know is that you spent $300mm to speed up some unknown number of transactions?
Steve:
To see this, suppose that due to a technological change, the moonfolks are suddenly able to produce twice as much per day as before, even though deliveries from earth continue at the same rate of 15t per year as before. Then by my calculation, the social benefit of a speedup in deliveries is doubled, whereas by his calculation it’s unchanged. My calculation is right and his is wrong.
Yes, and you’re making the same assumption: how do you know the moonfolk would be just as productive without their 3 days off every 13 days? Trey is simply assuming that the value of the output of moonfolk is proportional to the amount of input. You’re assuming that the value of what they produce per unit time will remain the same even if they work every day instead of 10 days out of 13.
nivedita:
And both of you are assuming this value is directly proportional to the amount of inputs delivered, otherwise your calculation doesn’t make sense either
Huh? I made absolutely no assumptions about technology. You are attributing Trey’s assumptions to me.
Steve: you are making assumptions about the technology. You’re assuming it can actually be used every day without any dent in the production per day. He’s assuming exactly the same thing.
If you hold everything else constant (no random changes in the technology on the moon), the benefit of delivering 4.5 “t” more per year is exactly the same as the benefit of producing $3000 more every 10 days. If technology changes, that just changes the rate of conversion between “t” and $ for the purposes of this calculation.
Anyway, this is a side-show. I don’t think there’s any ambiguity in how you’d calculate the benefits of this new rocket, whether you choose to measure it in “t” delivered to the moon, $ produced by moonfolk, per trip or per year or whatever.
Let me know what you think of the GDP numbers and if I’m missing something there.
nivedita:
And how exactly are you going to calculate how much it cost to do that, when all you know is that you spent $300mm to speed up some unknown number of transactions?
I’m not estimating costs; I’m estimating benefits.
You’re assuming that the value of what they produce per unit time will remain the same even if they work every day instead of 10 days out of 13.
Yes, and this is exactly the right assumption, because I’m trying to *over*estimate the social benefit of a speedup. To do that, I assume that any production that is delayed by the *absence* of a speedup is never made up for.
You’re assuming it can actually be used every day without any dent in the production per day. He’s assuming exactly the same thing.
See previous comment.
If you delay a delivery to the moon, you are not destroying the value of that delivery; instead the social cost is given by the cost of the delay, which is the value of the delivery times the interest rate times the length of the delay. In my computation, any output that is delayed is effectively destroyed—it never takes place. These are not remotely the same computation. Trey’s makes no sense in this context. There have been several things said in this conversation where I’m not sure who’s right. This is not one of them.
For (I hope) the final time: 1) To determine whether the *total* social benefit of *all* transaction speedups exceeds the *total* private benefit of *all* transaction speedups, it suffices to ask whether the *average* social benefit of all transaction speedups exceeds the *average* private benefit of all transaction speedups. 2) This is not a hard concept; it is fourth-grade arithmetic. 3) So it’s okay for us to look at the benefits of a single transaction. 4) Therefore there is no need to know how many transactions there are. If you haven’t followed that, tell me where (at point 1), 2), 3) or 4)) you stopped following, and I’ll try my best to make it even simpler.
Steve: I’m not estimating costs; I’m estimating benefits.
I understand arithmetic very well, thank you, but I’m missing the point of stopping at estimating benefits: you have in this post estimated a per-transaction social benefit, claimed arbitrarily that it is less than the social cost, and when called out on that, your only response is that you’re not going to estimate costs?
If the social benefits exceed the social costs, this is a worthwhile project, even if the social benefits happen to be less than the private benefits.
You should note that your calculation of benefit, such as it is, has already taken into account the external costs, because it is measuring the benefit of the incremental speedup over the current state of the art: the private benefit, assuming the current cable’s value goes to zero, will be that plus whatever the existing cable was able to charge. The only remaining cost is the private cost of the $300mm investment. I’ve shown you real numbers for the benefit that are of the same order of magnitude as that investment.
Steve: If you delay a delivery to the moon, you are not destroying the value of that delivery; instead the social cost is given by the cost of the delay, which is the value of the delivery times the interest rate times the length of the delay. In my computation, any output that is delayed is effectively destroyed—it never takes place.
Actually, in this situation, a delay would destroy the value of the delivery. This is easiest to understand in the context of a one-trip delay: everything that happens if a trip is delayed by 13 days is exactly equivalent to one of the trips never happening at all, so the loss is exactly the same too. I agree that you’re overestimating the benefit.
I think I finally get what you’re saying in terms of $ (=value-add) vs “t” (=input). To put it in the terms I had in the back of my mind, if you export wheat to Japan and get back Toyotas, it seems equivalent to measure the benefit of speeding up ships in terms of tonnes of wheat shipped to Japan or number of Toyotas shipped to the USA. But really the benefit to the USA is the $ value of Toyotas less the $ value of wheat, so you should be measuring the change in that. And for Japan you should measure the yen value of wheat less the yen value of Toyotas.
nivedita:
I understand arithmetic very well, thank you, but I’m missing the point of stopping at estimating benefits: you have in this post estimated a per-transaction social benefit, claimed arbitrarily that it is less than the social cost, and when called out on that, your only response is that you’re not going to estimate costs?
If the social benefits exceed the social costs, this is a worthwhile project, even if the social benefits happen to be less than the private benefits.
It’s now crystal clear that you’ve missed the entire point.
Let me spell this out in numbered steps, using SB for social benefit, SC for social cost, PB for private benefit, PC for private cost.
1) The project is excessive if and only if, at the margin, SC > SB.
2) Nobody has suggested any reason to suppose that SC is different from PC. (The SC and the PC are both equal to the cost of digging a tunnel and building a cable.)
2a) In particular, if the cable transfers income from one trader to another, that is a net PC of zero (a positive cost to one trader and a negative cost to the other), and a net SC of zero (because a social cost has to consume real resources).
3) At the margin, profit maximization on the part of traders implies PC=PB.
4) From 2) and 3), we have, at the margin, SC=PB.
5) From 1), we want to know whether, at the margin, SC > SB. From 4), this is equivalent to asking, at the margin, whether PB > SB.
6) Therefore we are reduced to estimating PB and SB. There is no need to estimate costs.
7) We can measure PB and SB in total or per transaction, as we wish. That won’t change the truth or falsity of the inequality PB>SB. I choose to measure per transaction.
8) At the margin, PB is equal to what someone’s willing to pay for a .003 second speedup.
9) At the margin, SB is equal to what society gains from a .003 second speedup. This is at most equal to the total amount of additional production that society can achieve by gaining .003 seconds, and almost surely at most equal to 1/1,000,000 of that (because it’s hard to imagine that more than 1/1,000,000 of the economy is stopped dead by a delay in a financial transaction).
10) From 8) and 9), we need to compare the amount one trader is willing to pay to speed up one transaction with the additional amount that society as a whole can produce when that transaction is speeded up.
I am not sure at what point your confusion has set in, but when you say things like “I’m missing the point of stopping at estimating benefits” you clearly haven’t gotten as far as point 6).
I hope this is clearer now. If not, tell me which step you lost me at.
nivedita:
I think I finally get what you’re saying in terms of $ (=value-add) vs “t” (=input). To put it in the terms I had in the back of my mind, if you export wheat to Japan and get back Toyotas, it seems equivalent to measure the benefit of speeding up ships in terms of tonnes of wheat shipped to Japan or number of Toyotas shipped to the USA. But really the benefit to the USA is the $ value of Toyotas less the $ value of wheat, so you should be measuring the change in that. And for Japan you should measure the yen value of wheat less the yen value of Toyotas.
I haven’t the foggiest idea what you’re talking about.
Seve: I haven’t the foggiest idea what you’re talking about.
Really? Then I wonder what you were getting at before.
Suppose the USA and Japan are trading partners, and the USA ships some tonnes of wheat to Japan and ships back some cars. How would you calculate the benefit, to the USA, to Japan and to the world at large, of making the ships twice as fast? How would you do it in a barter economy?
Steve: @177
1) Not at the margin, either the total or interchangeably the average. In other words, the relevant criterion is not whether the last trade using the new cable has social benefits exceeding its social costs, but whether all the trades on the new cable, on average, have social benefits exceeding social costs.
2) & 2a) This is a little confusing, but I think it’s just an accounting thing and we don’t actually disagree here: I thought the point was that part of the PB of the new cable is a transfer from the existing one, and this transfer shows up in PB, and in SC, but not PC or SB — ie the fact that the cable transfers wealth from the existing one to the new investors is what makes it more likely to get built inefficiently. The way I’m thinking of this, SB = PB + EB (external benefit); SC = PC + EC. PC is $300mm, EC is whatever impact it has on the existing cable, PB is measured by what traders are willing to pay for access to the new cable (this is an underestimate of the true private benefit, because it does not account for the maximum that they would have been willing to pay, i.e. the consumer surplus). PB must be at least $300mm, including the transfer, for the cable to get built. Your calculation of the benefit of a 3ms speedup is an estimate of SB – EC (where we’re assuming that EC is the full value of the existing cable), so it can be compared to the PC of $300mm. If this transfer is not to show up in EC and hence SC, then it needs to be removed from SB somehow, is it just considered a negative contributor to EB?
3) This relation is only true in equilibrium. Clearly we’re not in equilibrium, otherwise no-one would be proposing to build a new cable. All we know is that PB > PC.
Steve: If we assume, more realistically, that just 1/1000 of the economy is hanging fire waiting for this one trade, the social contribution of a .003-second speedup is roughly $1.50. I’m confident it’s even more realistic to replace that 1/1000 with 1/1,000,000 . That gets us down to about an eighth of a cent.
Doesn’t this also imply that only about 150 people (1/1e6 of the labor force) and $15mm per year (1/1e6 of GDP) care about the speed of trades? How do we reconcile this with the fact that over 800,000 people are employed by this industry, and it produces $185bn in GDP? For HFT in particular, the estimates seem to vary between $1-2bn per year, which is still a lot bigger than $15mm.
Steve: @137
I took a stab at calculating the equilibrium, and I got some results in the case where everyone is symmetric and assuming they’re no more likely than anyone else to win.
Assume there are n investors who are thinking about investing. Suppose the private cost of investing is PC and the benefit is PB. Let p be the probability that each of them invests, and q=1-p. They each calculate that if k-1 other investors are simultaneously investing, they only have a 1/k chance of being the winner. So for each of them, given that they invest, the expected benefit is
Sum_{k >= 1} (1/k) (n-1)!/(k-1)!/(n-k)! p^(k-1) q^(n-k) = (1-q^n)/(np)
The cost is PC. At equilibrium, we want
(1-q^n)/(np) * PB = PC
For n large, p behaves like c/n for c satisfying
PB * (1-exp(-c))/c = PC
The expected social benefit (assuming that is equal to the PB of the one winner) is PB * (1-q^n), which is always equal to the expected social cost of np * PC, no matter how much bigger PB is compared to PC.
So it seems my math is forcing me to agree that if SB < PB, then a winner-takes-all situation will in fact always lead to a socially destructive outcome. The one saving grace is that (PB-PC)/PC has to be about 60% before the expected number of investors reaches even 1 (setting c = 1 in the large-n equation gives (PB-PC)/PC = 1/(e-1) ), so there’s a decent hurdle for the private investors to overcome before there’s a large investment.
Re: #98. Steve, that’s actually a good point. I was thinking of “accidental” pick-offs, where the customer just happens to trade with HFT at precisely the moment when prices move against HFT. Of course, since it’s random, the HFT could have been helped by such movements, but he is risk averse. To answer your question #152 as to why risk averse HFTs are making markets, they are less risk averse than others. Someone has to provide liquidity, and the person best able to handle the risk of doing so should do it even if he is risk averse himself. So, I am thinking of a case where HFTs as a group get faster (than other non-HFT related information movements) and thus HFTs as a group reduce their risk. That’s socially beneficial.
I agree that when HFTs are just trying to get faster relative to each other, that’s by definition zero-sum: cost incurred with no net aggregate benefit, similar to positional goods.
By the way, I think some of the earliest comments in this tread suggested that HFT was winner-take-all. I don’t think that’s the case since HFTs have finite risk appetite. HFT1 and HFT2 may both be willing to sell 200 XYZ (but not 300) at 100.00. They both enter their offering quotes and first one in line gets priority on the first 200 shares that someone wants to buy. However, if someone wants to buy more than 200 shares, then the second in line will get to sell some. So, speed matters, but there is room for more than one liquidity provider in the market. (A good question is why all investor capital doesn’t flow to the “best” HFT, which would allow that HFT to provide all the liquidity. No idea. Perhaps, it’s uncertainty about who the “best” is. It’s not just about speed. Intelligence of algorithms also matters.)
Regarding “nivedita,” Steve in #177 suggests that nivedita hasn’t “gotten as far as point 6).” In #180, nivedita then reveals that he/she hasn’t even gotten as far as point 1)!! Nivedita does not understand the basics of maximization! Back to Econ 101!
In #176 nivedita writes, “But really the benefit to the USA is the $ value of Toyotas less the $ value of wheat, so you should be measuring the change in that.” Regarding this remark, Steve (in #178) (“I haven’t the foggiest …”) is rather kind, as nivedita has revealed that he/she has no idea of economic principles of international trade that have been well understood since David Ricardo’s Principles book in 1817!
Nivedita is wise to remain anonymous.
Just to muddy up the waters in case everything was getting too clear, an related point occurred to me. The people who write the algos are very clever, highly trained people. If they were not writing algos they would be pushing back the frontiers of science, inventing wonderful things and building fabulous structures. It seems to me that the syphoning off of our best talent to writing algos is not the best use of their talents in terms of social benefit. Am I wrong?
I don’t see why trading in oranges isn’t subject to the same objection used against HFT. If I plant an orange grove and my oranges are only slightly better than your oranges, I expect that my gain will be because your customers will stop buying from you and start buying from me. My overall income will include a small portion based on wealth creation (since the oranges are slightly better) and a large portion based on wealth transference (if the customer thinks my orange is 1% better, you lose 100% of the sale and I gain 100% of the sale).
(if the customer thinks my orange is 1% better, you lose 100% of the sale and I gain 100% of the sale).
That makes absolutely no sense at all. If orange eaters think your oranges are slightly better, your competitor will have to reduce his price to make orange eaters indifferent between the two. Oh wait….
But that’s not what happens in the scary spooky world of finance (especially when *gasp* algorithms are involved!). No, economics is all myth and no logic when academic economists don’t know who the market participants are, what they’re doing there and how they interact with each other. It’s easier to write fairytales than do the hard work of figuring out what’s going on.
@ Methinks,
“No, economics is all myth and no logic when academic economists don’t know who the market participants are, what they’re doing there and how they interact with each other.”
I imagine the experts in financial economics do know these things. I don’t think you’ve found any financial economics experts on this blog. Please don’t take one dumb comment and make generalizations about the profession. It’d be equivalent to me saying that all financiers are snake oil salesmen when they sell products that they don’t themselves understand, don’t know the underlying risks, and actively bet against what they’re selling.
http://www.bloomberg.com/news/2011-04-13/goldman-sachs-cdos-bet-against-clients-misled-congress-senate-panel-says.html
David Cushman: nivedita is clearly quite confused on some very fundamental issues, but I do not think s/he ought to be embarrassed about this. Those of us who teach this stuff for a living have had a long time to digest it; our students — even very smart students sometimes — can take quite a while to master these ideas. I applaud his/her repeated attempts to get clear on them.
Harold:
If they were not writing algos they would be pushing back the frontiers of science, inventing wonderful things and building fabulous structures.
This is not a separate point; it’s exactly the market failure I’ve been pointing to.
Ken Arromdee:
If I plant an orange grove and my oranges are only slightly better than your oranges, I expect that my gain will be because your customers will stop buying from you and start buying from me.
But there is a limit to the number of oranges you can provide.
Methinks:
That makes absolutely no sense at all. If orange eaters think your oranges are slightly better, your competitor will have to reduce his price to make orange eaters indifferent between the two.
Yes, this is exactly the right response regarding oranges.
@Harold
I agree, I want talented people to go out and do other things that will make my life more fun and exciting – whether inventing cheaper ways to travel, or better-tasting strawberries that last 10x longer.
Let’s look at the problem from a different angle. We see 800,000 people going into some large building each day. On avg each leaves with $1000. During the day, anyone who is not one of the 800,000 who wishes to trade a security must do so by making an agreement with one of the 800k. The outsiders can see all trades taking place during the day (though not who did what trade), and they can see generally where the group of the 800k are willing to buy and sell. They also notice, if they are watching carefully, that many more transactions take place than just the ones interacted between the 800k and the outsiders. The 800k often trade amongst themselves.
Now, the 800k start doing something odd, they start buying a bunch of equipment, spending billions of dollars (you can tell, because you get to see all of their interactions with the outside world). Now you start to see changes in the number of trades taking place per day, and in the rate at which markets update. You have no idea what this means.
You hear some of the 800k complaining that something bad is happening, they’re aren’t making as much money. But you also hear others of the 800k saying everything is fine.
Finally, someone who is an outsider and does an enormous number of transactions per year with the 800k tells you his costs are going down.
A year later, you notice that the price to do a transaction with one of the 800k has gone down for you too. And then you start to see fewer people going into the building…
That’s what is happening right now. Things are becoming more efficient. When that happens, there will be local winners and losers, but overall society will be better off. Instead of imposing some rule to try to make these talented people do something different, the competition is driving the efficiency and soon talented people will decide of their own free will that there are more profitable ventures for them to pursue.
So far in the “debate” you are hearing from a lot of the 800k giving their views of the pros and cons of the activity. But you have also heard from one external participant telling you he has no idea what’s going on inside, but that it’s good for him (Cliff Asness, WSJ)
A reasonable question would seem to be: If after 2 years the number of people has gone from 800k to 600k, and the direct interaction costs to the outsiders has decreased (enough to account for why there are now only 600k people leaving each day with $1000), then would you agree that things were probably ok? Even if you cannot point to why from a microscopic viewpoint (you cannot see what happens inside each day)?
No, I don’t think these numbers are accurately to scale. Yes, I do think that detailing what happens inside has been shown to be too difficult to do in this forum. Methinks made a valiant effort. As far as I can tell it has resulted in a claim that the science is settled, and the answer is that HFT is bad. George Will had a great article recently about this style of analysis.
#193 – this would be a good story, but how do you explain that (from wikipedia) “The U.S. finance industry comprised only 10% of total non-farm business profits in 1947, but it grew to 50% by 2010” The same trend is happening in the UK. Far from fewer people in the building, it seems to be attracting more and more.
@Daniel
I imagine the experts in financial economics do know these things. I don’t think you’ve found any financial economics experts on this blog.
Your imagination misleads you about the profession you hope to enter. The overwhelming majority of today’s financial economists are drawn from Chinese math programs or something similar. Much more often than not they have zero (and I do mean zero) experience in finance. The academy’s bias against real world experience is astonishing. When I once entertained the idea of pursuing a Ph.D. in finance, I was confronted with naked hatred of the field itself and I was told that I would basically have to sign in blood a promise that I will remain in the academy and never soil myself by returning to industry. And don’t forget that I grew up with academics.
I read their output often (and it is often hopelessly confused drivel) and I talk to comfortably ensconced academic economists who understand the limitation and loss of granularity of the ivory tower. But, what they do is still like driving a car in theory rather than in practice. I sent you a link to the latter type of economist through email messaging. But, even he misses some things and I’m sure he’d admit as much.
And that is why academics gave us the Black-Scholes model, the all-important volatility skew was added by the market.
Still, the economists who have taken strong positions on a subject they very clearly do not understand so thoroughly that it defies imagination are not financial economists and have shown no interest in understanding that which they seek to control.
@ Steve Landsburg
Yes, this is exactly the right response regarding oranges.
That is also the right response regarding two lines of varying degrees of latency in financial trading. You would know that if you sought to understand the economics of market intermediation, who the market participants are and how they interact with each other before you jumped to calculations. Because only intermediaries, whose adverse selection risk exceeds any other participants’, need the speed to jump out of the way, the super low latency lines are only worth the cost to them. Everyone else who would like lower latency lines (for a variety of good reasons other than chasing HFT) than they currently have at a price they can justify will now have a chance to buy the lines the HFTs abandon. So, your calculation of social cost is way too high.
Also, even your earlier claim that what you calculated is the effect of slower price discovery is just not correct. Faster price discovery is important not because nobody can breathe until the trade goes through but for purposes of efficient resource allocation. The moment information that changes prices appears in the universe, all prices become incorrect. The more quickly prices can reflect reality, the less misallocation there will be in the economy. How much less at each time increment? I don’t know. But, technology allows us to move only in small increments at a time and this is only the latest increment.
Which brings me to another point: Regulation NMS necessitated HFT by market intermediaries in 2007. Since then we’ve had several incremental speed improvements. Setting aside concerns about risk appetite, lack of available capital and reduced pricing ability, by your logic one winner should have emerged to take all already. Since this hasn’t happened, I think there’s ample reason for you to re-examine your winner-takes-all theory.
One of us is right. One of us is wrong. I will come back here later to propose a way to settle this.
I sent you a link to the latter type of economist through email messaging
Correction: facebook messaging. It’s in your “other folder” and there are two academic papers referenced on the subject of information gathering and its social benefit and cost.
In which case 50% of the customers buy my oranges and 50% buy his, and compared to the situation before I opened my orange grove, most of my profits are still a transfer from the other guy, except I only transferred about 50% of them instead of all of them.
I just (in the case where there’s only one other seller) got 50% of the market by making my product infinitesimally better. How is that not a transfer in the same way that most of the gain from HFT is a transfer?