Time is money

If traders are located 100 miles away from an exchange, they face a
delay of one millisecond whenever they seek to trade a price via their
computer screen. Few serious investors can afford to be that late to
prices that flash so quickly. The blink of a human eye takes 300
milliseconds; many traders now operate in the smaller realm of

Here is the longer story.  I liked this bit:

Mr Greifeld said there might have to be measures to ensure speeds
within the co-location facilities were the same. “We might have to give
everybody the same length cable, believe it or not,” he said.

For the pointer I thank Christian Meyer.


Maybe someone can explain to me what advantage microsecond-timing gives you. What's the sequence of events that actually leads to a few milliseconds making a difference? I mean, sure...I can understand the difference between real-time quotes and the free 15 minute delayed stuff on Yahoo! Finance, but this isn't about being the first to submit a trade when Steve Jobs steps on stage and announces Apple's going to start making cars or whatever.

Requires registration, didn't read.

However, I am extremely skeptical. One error: the blink of an eye is much faster than one-third of a second. Another error, more relevant: delays from cable length are going to be dwarfed by delays from processing - including both the time for the processors in network routers and switches to route and switch the packets, and the time for the processors in the computers executing the trades to execute the trades.

Furthermore, anyone seeking to trade a price "via their computer screen" is going to have delays of many seconds while their brain makes a decision and then causes their fingers to twitch. Further delays of milliseconds or microseconds caused by network transmission would be inconsequential, and trying to reduce those delays by co-locating or using shorter cables is an exercise in imagination and futility, akin to audiophiles marking up their CDs with green pens.

Eddie, there are traders in South Africa and Singapore who sweat blood over the delays on tiny, tiny amounts of time.

I work for a high-frequency exchange in a technology role.

Traders absolutely believe that a millisecond or 2 makes a huge difference. But in reality, it doesn't. Here is how it essentially works.

A maker sends out a price for an instrument. A taker decides that they want to hit that price and put an offer in. Oftentimes several people do. The maker will then HOLD the order for several milliseconds as it tries to see where the market is going. If the price improves, from their perspective, they reject the order. If it deteriorates they accept the original offer. This "timer" varies by exchange and the relationship they have with the maker. Could be as little as 30 ms or as high 100+ ms. So regardless of whether you see a 1 or 2 ms delay in the market data the maker still gets last say.

The fear that the traders have when it comes to latency is that they are receiving out of book prices. Those are very bad because if the market has moved considerably then they will have the order rejected if it is beneficial to them and accepted if it hurts them. Thus they want to see deltas between send time of the market data and the received time of the market data by the algo box.

BTW, this stuff is only relevant for programmatic trading. If humans are involved in the trading transaction none of this is relevant as the human traders add delays magnitudes higher than a few miliseconds.

There is more of this in humanity's future, not less. Historically, the financial industry has often pioneered the adoption of high-end computing and networking technology and infrastructure.

In the future, the growing importance of low-latency connectivity (in online multiplayer gaming and other areas) will ensure that the human race will never wish to disperse farther than a fraction of a light second from one another. Even the Moon will be too far, Mars impossibly out of the question. (Even by today's standards, Mars is much too far: imagine never being able to have a conversation with your loved ones, with all communication limited to asynchronous exchange of e-mails and YouTube videos).

And if we ever manage the trick of leaving flesh and blood behind and uploading ourselves into a computing environment, thereby leaving behind the slow reaction times of organic neuron-controlled thoughts and actions, then we will have the same speed requirements as today's flash-trading computer programs. The entire human race (or what it will have evolved into) will probably exist within a geographic area much smaller than one of today's major cities, with the wealthy living closest to the core and running on the fastest multicore processors ("buying time" will be a literal transaction, not a metaphor). Recolonizing the rest of the planet, let alone the Moon or the rest of the solar system, would be strictly the province of radical cultists and hermits.

These guys are full of crap. In terms of network latency, one millisecond barely qualifies as rounding error. Traders may believe that microseconds count but that level of service is literally impossible to guarantee.

I can assure you that millisecond timing is very important to traders and that 1 millisecond is worth several tens of millions of dollars to the right firm. I know a project manager at a very good firm and they are printing money, simply because they have lots of former NASA engineers doing networking and programming.

Citadel supposedly made 40% on 1.2billion last year from high frequency trading.

BTW, I should say the SEC rule I proposed would consist of offers within a minimum amount of time, say half a second or one second. The Taker would HAVE to take, reducing the reward for information based on miliseconds. Don't take the rule too literally. I'm just sincerely interested to see what high frequency's defenders have to say.

The "series of auctions" approach would be much improved over the current FIFO system.

It should probably be only once every 5 seconds too.

Probably don't need to tell this crowd, but there are plenty of "serious investors" that are more concerned about what a stock price will do in the next 300 days than what it will do in the next 300 microseconds.

As a retired systems engineer with a lot of experience in distributed systems, testing and performance qualification, system architecture, distributed transactions (such a widely spread joat that I'm master of none of them), this is a "been there, done that" de ja vu all over again "discovery of the limits of the speed of light" drive off into the weeds of misunderstanding.

In the early 90s, I was involved in work on distributed computing projects testing multi-site data centers separated by a hundred miles as a conservative first step. Yep, the speed of light did add a little to the equation, but just two or three completely replicated systems in the same room able to detect the failure of another while working at full speed dwarfed the distance delays.

When the first WTC bombing occurred, we and our competitors spent a lot of time in NYC working with banks and traders, and while the speed of light came up early in the discussions of technical issues, that was quickly replaced with lots of other issues, with money being the biggest concern, and that is with people willing to spend millions to avoid having the bank fined multiple millions for failure to meet a deadline which would mean they lose their job or at least their bonus.

To say a hundred miles is a millisecond says a dedicated point to point link, presumably dark fiber. That is still extremely expensive, (and there should be two independent fiber pairs to prevent backhoe failure), so the link becomes shared with lots of different customers sharing the link which means lots of switches/routers and unpredictable traffic and thus unpredictable delays, and soon the round trip delay is reliably 5 to 10 ms.

Time is money for sure. Are you willing to pay a million dollars a months for dark fiber to get 1ms or will you settle for fifty thousand to get 5-10ms?

I don't know what the pricing is today, but a decade ago, the telcos certainly had lots of dark fiber sitting around, but they weren't going to lease it cheap but instead price it as if you were going to compete with them on voice and data over the dark fiber. And unless you have end to end dark fiber or at least T3/E3 or SONET or site to site microwave or some other similar end to end TDM, the speed of light isn't the primary cause of the link delay.

And I've heard of lots of people claim "money is no object because billions depends on being fastest" who failed to get all the signatures on the POs to buy those milliseconds.

Sub second response times?

Fooled by randomness.


In theory short term arbitrage means that longer term investors see more reliable price signals than they would without it. In the case of high frequency trading, I assume that means that day traders who intend to hold stocks for whole hours will get less noisy price data. If you discretized the process you'd end up amplifying the noise, since there's no price information between time-steps, you'd end up with restrospecitively wrong prices/volumes at each time step and thus amplified noise. Intuitively it seems there must be some point at which it becomes pointless to get any faster, but I think that point is only defined by the speed of everyone else's trading systems ....

Of course high frequency trading is actually just very fancy and expensive chartism, which means it can't ever reliably make money, because every pattern in prices is public and be equally exploited by all traders, which means ultimately no-one gains from it. From their point of view, though, this doesn't help - if they refrain from buying the latest and fastest kit they guanrantee they won't make any money, whereas if they do its just almost certain that they won't!

Yep, it makes a difference. I attended a conference where they listed off a number of such things that would be affected by variable latency: the person with the fastest video link always "wins" TV debates, gambling, and computer-aided arbitrage. Maybe you don't make a killing on any one transaction, but fractions of a penny on thousands of 'em ... why does this sound familiar, Michael Bolton?

"the human race will never wish to disperse farther than a fraction of a light second from one another"

Perhaps it would be more accurate to say "nobody who wants to remain fully connected will ever disperse farther ..." Just as people signed up to go to the New World in the 17th century, I'm sure you will find a few that would prefer to be with their own than to remain here with the herd.

Isn't this just another way in which Wall Street has an unfair advantage over the average Joe? It would be better for all trades to take place over five-minute slots, allowing enough time for buyers and sellers to negotiate. This would narrow the bid/offer spread, especially for thinly-traded securities, and also reduce the current unfair dependence on market-makers.

Financial markets should serve businesses and society at large; they shouldn't serve Wall St. There is no social need for trades to take place faster than every five minutes.

HFT really is of great use to abritrageurs who can and do make money by exploiting very small, temporary inefficiencies in pricing. As more and more arbitrageurs come on the scene, obviously the opportunities dry up in a given market.

In terms of index ETF arbitrage, I have a question: my former employer, an institutional asset manager, would sometimes wind up buying or selling 5-10% of all the shares of a given ETF. That's not because we were a huge asset manager but rather because some ETFs (particularly the non-U.S. ones) simply don't have a large market. We would never trade these ETFs on the exchange because we might move the price significantly but rather the trading desk would call up our institutional broker who would then go out into the market (whether it was block trading or some other technique, I don't know), buy up the underlying shares of the ETF, and then "package" them into an ETF for the customer's accounts.

Given this, I would imagine the arbitrage opportunities in many ETFs are very small. If you can't buy or sell a significant amount of ETFs without moving the market, the profits from index arbitrage for many instruments would be relatively low, right? Too low to interest a large hedge fund or institutional investor, it seems to me. The trades we made were always for medium-term buy-and-hold purposes, not for arbitrage.

I read with dismay the various comments made above.

Decreasing network latency is always accretive to value.

Amazon has internal studies showing that immaterial milliseconds change in the median rendering times of web pages increase sales.

Poundstone's _Fortune's Formula_ talks about the origin of the Kelly Criterion being based on a study of WHY so much of Ma Bell's revenue came from dedicated circuits to saloons. The answer - the bookie wanted to know the results before people knowing the results could rush to place bets.

In the simplest case, beating your competitor to be at the top of the order queue gives you the advantage to take the profits before your competitor.

I have spent 15 years doing optical and advanced networking. I recently earned an obscene amount of money trying to help a company trim 3milliseconds off of their network topology (patent filed).

When governmental organizations like Labor, Fed, Treasury release information to their news agencies they do this in a completely fair manner. As the digital exchanges terminate circuits and local loops to their customers they ensure that they are all exactly the same length.

Temporal latency matters both to allow one to the front of the queue, and separately to enable folks to have faster cycles in decision making.

The article and thesis holds tremendous water, the arguments against it porously flow.


This brings to mind an interesting thing: the social benefit and the importance of a product or idea can be measured by how much money you can reap from it. Think about it: Who got the richest over the past 20 years? The person who help make the most useful thing, the computer, and then the internet. ie. Bill Gates, who had one of the best ideas ever. Who Else Made Money? Michael Milken, Junk bond king. He profited off of this computer boom by using them in his high speed junk bond trading schemes.

Money is nothing in and of itself, but (alluding to the theory or should I say law of subjective value) it represents costs, tradeoffs, and therefore, value. It means that you have created value because someone (a whole lot of people) gave up something to buy your product, they gave up something else for you. If you're making a lot of money, a SHIT load of people are giving things up for you, which means you're socially benefitting a lot of people.

** this works in most situations. Some legal situations of making a lot of money can be attributed to herd behavior, ie. investors making millions off of CDOs and other ABSs. But my thesis still holds true for the majority.

BayAreaAlan "I have spent 15 years doing optical and advanced networking. I recently earned an obscene amount of money trying to help a company trim 3milliseconds off of their network topology (patent filed)."

No confirmation bias at play here, then? I'd feel the need to self justify if I'd spent fifteen years helping trading firms get to the front of the queue. Perhaps you could elaborate on the huge benefits to the rest of us from millisecond trading versus, say, 1 second trading? Your rather lofty and assured answer tells me nothing. Also, what's the opportunity cost of having so many hugely bright people engaged in helping trading firms queue hop? Seems more than a little tragic to me.

It's absurd to think that the millisecond doesn't "cost" anyone anything. It absolutely does. The firms which book the trades and manage the process (Goldman Sachs, one of the largest, for example) have a massive advantage in seeing the book just as it hits the wire. They can do a form of frontrunning by engaging in flash trades. Technically, these firms say they don't "share" information between systems or allow the information to overlap in order to gain an advantage, but several people have begun to speak up and talk about this, and it's not looking like this is precisely true.

For more precise coverage of this, and better explanations, you should visit www.zerohedge.com. While some of the posters and writers there are a bit nutty and conspiratorial in nature, the vast majority are legitimate traders and they have massive amounts of insight about how the system is rigged. Anyone who ignores zerohedge is doing so at their own peril. The site is supportive of markets and capitalism, but NOT CRONY Capitalism, which is what we have right now.

For FT articles, google a quoted line and click on the google result.

Not to burden the discussion with actual numbers, but the RTT on an InfiniBand link inside a datacenter is 8 microseconds, while ethernet clocks in at about 50 microseconds. Routing at layer-3 would add latency in the milliseconds, but the whole point of these setups is to avoid that.

On those time scales the speed of light delay becomes important over distances of kilometers and tens of kilometers, respectively. This suggests that Greifeld's concern about cable lengths is either misplaced or driven by perception amongst his customers, rather than based on real physical concerns. Nevertheless, the length of the fiber run between the datacenter and the exchange would be a real concern on these time scales.

The dedicated connections required to make all of this work are pricey, but not unobtainable. I got a quote for a 10Gig DWDM wave in the DC metro area for $15K per month. I could believe that prices in NYC are higher, but not the "million dollars a month" someone quoted above.

So, trades on millisecond time scales are easily technically plausible. I can't speak to the desirability of such rapid trading, but I think that the cures are likely to be worse than the disease. Unless someone can point to concrete harm caused by the practice, it seems better just to let people who want to compete for millisecond arbitrage do so. If nothing else, the finance industry's spending in this area helps make high-end computing and networking gear affordable to researchers in the physical, life, and earth sciences.

This article is about trading that is automated with computers. (Humans would not be able to distinguish the difference between a few microseconds.) Since computers are involved, it seems like there is more that can be done to minimize delay than decrease to the physical distance between computers. A lot of things affect computer speeds. It may be possible to speed up a transaction by using a different program or algorithm on the computer. One can also increase the processing speed or the number of processors so that calculations can be done in parallel. I was surprised that the article didn't mention any methods other than co-location. -Stephen N WCU 1257

Comments for this post are closed