HFT in my backyard | VI

When you think about it,  it really is one tangled, small cottage industry. Everybody knows everybody.
Everybody knows what everybody is doing, and nobody talks.

One of my informants.

This is the final episode of the “HFT in my backard” series, season 1. I have been too busy to post it before (I work on HFT in my spare time). What’s more, I realize that I accumulated so much data that I really can’t write about all the stuff I have now on my desk about those microwave networks used by high-frequency trading firms. The first part of my book was titled 6, then I added the second part 5, and I think that given all the data I got, I could easily write a new sequel titled 4 about the networks but I’m afraid that I don’t have the time for that. This final episode is more a like a medley (“Miscellanea”) about various topics than a conclusion. With an interesting cliffhanger.

FACTS AND INACCURACIES

This investigation on microwave was possible mainly because a lot of public data is available online. Aside from the towers I visited (mainly in Belgium) and some tips I had from people working in the industry, public documents are cool but using them doesn’t mean that you have found “the truth” – here the truth means: knowing exactly where the competitors have installed dishes, and which are the exact paths they designed to go from a tower/data centre to another. By checking the Ofcom website (UK) or the ANFR website (France), you can learn about the frequencies  reserved by competitors here and there, but that doesn’t mean they really have put dishes here and there, so you have to compare the licenses with planning applications (if they are online), and so on. Some countries are quite transparent (UK), but others are dark pools (Belgium). Besides, even if a competitor has both a frequency license and a granted permit for a tower, that doesn’t mean he’s really using the location – perhaps he moved on elsewhere, etc. Sometimes you have to speculate, and trading is all about speculating. Sometimes you are right but sometimes you don’t, sometimes you make money but sometimes you lost money…  and the same is true with the microwave networks. What’s more, some of them are still in progress, and others are trying to get better paths than they had before. They move, so it is quite difficult to have a definitive map showing the real paths – by the way, I have found that the first meaning of the word trade, back in the 14th century, was path.

What’s more, when investigating the networks you have to deal with a lottttt of technical details and I know that may have been mistaken on some of technical facts. All this to say that even with  good public data and fair discussions with the people from the HFT/microwave industry who gave me some tips, it’s very difficult to be sure about the true and definitive routes the competitors designed to get data from London to Frankfurt, and vice versa. The most interesting here (from an anthropological point of view) is the way HFT shapes geography and how firms have to deal with nature to achieve the best paths, being faster than competitors.

THE CHANNEL AGAIN

I talked a lot about the channel in the previous episodes (the difficulties you may have to cross it with microwave; the speculation on frequencies; etc.), so let’s start with the sea. As I wrote in Part II, the old military Houtem tower purchased by Jump early 2013 (which was the very beginning of the series) enables the firm to have dishes far higher than competitors who don’t have the chance to own such a tall tower. Thanks to this mast, Jump can go straight to Basildon by going through Ramsgate and bypassing the old RAF Swingate tower (they save approximatively more or less 6 kilometers, i.e. more or less 20 microseconds).

Channel_2015

I realized later that when I started to work on microwave, at the same time the BBC released a short movie about Basildon and the microwave networks, with some pictures of the HFT dishes lost in the fog of Swingate :

Capture d’écran 2015-04-09 à 09.16.42

Capture d’écran 2015-04-09 à 09.17.12

Capture d’écran 2015-04-09 à 09.16.55

The last photography is interesting : this is a boat crossing the channel. I have been told that big container ships may block the microwave signal between France/Belgium and the UK. A lot of boats navigate the channel daily, coming from (or going to) the Rotterdam port. Here is a snapshot from marinetraffic.com, showing some ships around the channel (on Tuesday 9, 2015, 8:30 CET):

Capture d’écran 2015-04-09 à 09.29.38

With the tall Houtem tower, there is little chance that the Jump signal could be blocked. By contrast, some competitors may have problems with these boats. It appears that Jump isn’t/will not be the only firm to cross the channel by bypassing Swingate. Other competitors are in the ranks –Optiver and McKay Brothers, at least:

Capture d’écran 2015-04-09 à 09.24.09

Capture d’écran 2015-04-09 à 09.37.26

Both Optiver and McKay have Ofcom licenses in Dunkerque (France) and around Ramsgate (UK) (Vigilant  also has a license in Ramsgate), so I bet all the firms will use this route one day to cross the channel (if they don’t, they will probably have difficulties to beat Jump). In Dunkerque, Optiver seems to want to install dishes on one of these “chimneys” owned by the Polimeri Europa France (a petrochemical industry firm):

26696664

Two kilometers away, McKay acquired licenses at an elevator owned by Arcelor Mittal (a steel company):

26696940

This other photography shows the bottom of the elevator, with a boat waiting for commodities.

60158506

This boat is not really tall, but what would happen if a very tall boat comes to rest in front of the McKay dishes – which seem to be between 70 and 80 meters high? That said, boats are not the only bothersome obstacle – a swarm of birds may also disrupt the signal. Here is a snapshot (given to me by a HFT firm) showing that noise may take precedence over signal. If you have more noise than signal, your microwave network is dead (that’s why HFT firms always need fiber as a backbone).

degredation

INFORMATION TRANSMISSION AND HISTORY

The fact McKay put dishes on an elevator is quite amazing (if you know a little the history of the Chicago commodities markets, then you are aware that the industrial history of the grain elevators around the Windy City is related to the history of telegraph). Much could be written about the microwave networks in the US and about those used by HFT firms between New Jersey dans Chicago. Thanks to this map (from Laughlin et al., 2012), it’s possible to see how HFT is reshaping the line-of-sight microwave networks. The New Jersey-Chicago-Washington corridors are clearly visible:

Capture d’écran 2015-04-09 à 10.34.31

I didn’t investigate these networks but on this blog  there is an good story about the first microwave network built between New York and Chicago, it was in 1949:

34Jumps01

Technologies have changed: in 1949 you needed 34 towers between two cities, but in 2015 the best HFT providers/firms use 22 hops only:

Capture d’écran 2015-04-09 à 10.44.18

© McKay Brothers

That’s interesting to see that a firm like Auburn (Data Systems) needs more towers than the first network build in 1949 (the firm was one of the first applicants to license frequencies, but one of the last networks to turn up). There are other amazing stories about the US networks: here you can read about the new communication mast the city of Mahwah, New Jersey, erected in 2014 to allow both McKay and Auburn (among others) to install dishes. In this official document we learn that, in 2013, the township received several “bids for placement of communication equipment on a 30 foot pole extension”. Auburn proposed $64,000, Jump offered $35,000, Anova $35,000 but the winner was “a data company” firm named Wireless Holdings Network, who offered to pay a $246,024 (!) annual fee to put dishes there. Before that, the pole only made $40,000 a year, so we have the definitive proof here that HFT has some real social and public values. I don’t know who is behind Wireless Holdings Network, but it seems the firm is affiliated to SMG Holdings, and SMG Holdings has a subsidiary named Anova-Tech, and Anova provides laser networks between some of the exchanges in New Jersey. How interesting is that. Even if laser is as fast as microwave, laser networks have higher bandwidth but they are more affected by weather (if a dense fog  laser beams won’t travel more than a few hundreds of meters). If Wireless Holdings Network/Anova agreed to pay a huge annual fee to the township of Mahwah, is it because they put laser antennas on this tower? Or only because they wanted to be sure to be on this tower, so they paid the price? – pure speculations again.

In Randolph (New Jersey), McKay had dishes on a tower where Auburn wanted to be too, but the township of Randolph refused the Auburn planning application, so Auburn was jealous and sued the township, arguing that if the city of Randolph approved the McKay installation there were no reason to refuse Auburn’s one. The Battle of the Dishes. There are other stories, but the United States of America are not my backyard, so I didn’t really work on. That said, there is this interesting story about HFTs exploiting a loophole at the CME data center (Aurora, Illinois). Wall Street Journal‘s Scott Patterson wrote about the case in May 2013, and a lawsuit was filed against the CME in 2014. In short, “some customers experience a latency of a few milliseconds between the time they receive their trade confirmations and when the information is accessible on the public feeds”, told a CME spokeswoman. The fact HFTs receive information before other participants is much debated – markets have different polices about that: Eurex (Frankfurt), for instance, sends trade confirmations publicly to all the participants, not only (i.e. before) to the firm who sent the order. Here, the CME loophole gave a 1 to 10-millisecond advantage to HFTs, and with microwave data can travel between Aurora and New Jersey in more or less 4 milliseconds, meaning that an algorithm could decide to trade on Nasdaq thanks to information coming from Chicago but not even public in Chicago. This chart that shows the evolution of the response time between the CME and the Nasdaq:

LatencyEvolution

© Greg Laughlin

According to one of my informants, “the sharp current-day response is largely a consequence of the fact that the CME seems to have substantially reduced the unpredictable delay between the timestamp at the match and the print at the exchange gateway” (meaning that the CME probably – but partially – succeeded in resolving the loophole issue). “This likely puts pressure on the economic viability of second-tier microwave networks, while increasing the value of the fastest networks”.

FREQUENTLY ASKED QUESTIONS

Before “concluding” this first season with the recent microwave moves around my backyard, a quick Q&A session – as I received a lot of emails asking about various issues about the networks. Obviously, I don’t have all the answers (this goes without saying) but sometimes I have leads.

  • Why don’t we find the name of big HFT firms like Virtu on your map?
    I don’t know. Probably because they don’t build their own network (unlike Optiver or Flow Traders). They probably use microwave network providers. Call McKay Brothers, or even Custom Connect.
  • Are you on the McKay Brothers payroll?
    Well… no. Some readers (mainly people from the HFT industry, I think) were wondering why I was so kind with McKay Brothers. I don’t think I particularly praised McKay Brothers but 1) as a microwave network provider, McKay is the most transparent firm: they publicly publish their “real” latencies, and above all their parent sister company, Quincy Data, also makes public the products they sell (for instance: data about the Bond Futures traded at the Eurex exchange, in Frankfurt); other companies (such as network provider Perseus, who resells bandwidth of the Jump network) also publish latencies, but not with the required accuracy – what’s more, there is no details about the data/products involved. 2) Although the firm has offices in Oakland, McKay is also a French company, with offices in Paris, and as a French I defend the French industry;) 3) That being said, I am fully transparent: I knew Stéphane Tyc (the co-founder of McKay Brothers) before investigating the microwave area (I was a kind of “go-between” between him and Nanex’s Eric Hunsader when Stéphane published a white paper about the Fed Robbery in 2013) but that doesn’t mean I advertise McKay Brothers’ network. 4) Stéphane Tyc has interesting ideas, about best execution and transparency in reporting trades, even if its “libertarian-style” (my words) does not always please everyone.
  • Is it possible to hack the microwave signal?
    I am doubtful.
  • What about the pirates?
    This investigation was mainly about geography, nature and speed. But I know French regulators Arcep/ANFR checked some sites in September 2014, they wanted to be sure that the ones who asked for frequencies really installed dishes here or there (they have legally 18 months to do it before the license expires). It seems that for instance CIC has no dishes in France. In Monts des Cats (where Custom Connect and Flow Traders are), 14 links/licenses are not used (meaning operators asked for frequencies but did not put devices – a kind of tower/frequency squatting?). This field inspection allowed the French regulator to “clean up” a little the licenses (they will inspect other HFT dishes next June) and confirmed that Custom Connect has a direct path between Mont des Cats and Dover (the longest path crossing the channel, i.e. 106 kilometers).
    Capture d’écran 2015-04-10 à 10.55.34
  • Why some HFTs need to get information a few milliseconds before other traders?
    This is the $0,001 question. First answer: it’s all about data, as price is an aggregate of informations coming from various locations, and some products seem to be more sensitive to latency than others. Other answers: I’m working on that.
  • What about the different London locations?
    I did not detail the London metro area before. In and around London (the Basildon-Slough route), the networks use millimeter wave (whose frequency range is different than microwave). There are different locations in London: Basildon, Slough, The London Stock Exchange and Thomson Reuters LHC (add Interxion). The London metro networks are quite dense, and my map is far from being accurate (I just picked up some paths on the Ofcom website), more work would be required to have a full map of those networks (and I won’t do it).
    Capture d’écran 2015-04-10 à 08.51.24
  • How much Getco paid Latent to cross the channel?
    In Part IV I wrote about the fact Getco (under its Global Colocation Services subsidiary) purchased some paths to Latent Networks, including the Dover/Dunkerque route. Given the speculation on frequencies around the channel, I was very curious to know the amount, so I wrote Ofcom and I got a kind answer, saying that they don’t know the prices of the trade – it’s  private matter between the buyer and the seller only. We’ll never know.
  • Are there other existing, or future, microwave networks in Europe?
    Yes. A company named 12 Horizons said in June 2014 to have build a microwave network between Frankfurt (Eurex) and Zurich (SIX). The network seems real (sometimes companies make announcements but it’s only for advertising – for instance, Anova announced to have built in 2013 a laser network between London and Frankfurt but this is not true). I vaguely searched for data about the towers in Germany and Swiss but nothing relevant. What I know is there are old military microwave towers in this corridor, so I would’t be surprised to learn that they are now used by high-frequency traders. In this McKay Brothers presentation, you can see some possible future microwave routes (in dashed lines).Capture d’écran 2015-04-09 à 13.46.18
BOATS, PLANES AND PROJECT EXPRESS

In Part V I speculated about the future of transatlantic transmissions. Building a wireless network between Europe and the US is not a simple task but some are really working on that – floating islands? tropospheric ballons? (Amazing fact, by the way: some years ago, Island’s designer, Josh Levine, had imagined a system with a couple dozen Cessnas flying in circles with microwave repeaters on their bellies; the Cessnas would be so high [~ 5,000 feet[ that they would have line-of-sight of more than 100 miles between them. It would have been a challenge to keep the planes in the air but, according to Levine, “so, so much faster and cheaper than all these microwave towers”). I bet that we’ll learn about some wireless transatlantic networks sooner than expected (check for example this article about SpaceX). The deployment of wireless networks all around the globe is a fascinating issue. “As the planet wires itself and its computers ever more tightly together in an ever-lower latency web of radio links and optical fiber, it no longer seems like a particular stretch to float an Electra hypothesis in which computational nodes and their interconnections assume a global role comparable to that now filled by the biological organisms”, Greg Laughlin rightly writes by extending the “Gaia hypothesis” invented in the 1970s by James Lovelock (“organisms interact with their inorganic surroundings on Earth to form a self-regulating, complex system that contributes to maintaining the conditions for life on the planet”). This is a major point being from anthropology.

There is no wireless transatlantic network for now. In Part V I speculated about the new Hibernia Project Express cable, supposed to save ~6 millisecond between Europe and the US. This cable has been long announced and it’s now in progress. We know that the landing station in the UK is in Brean, but I have made a mistake about the location of the station. The photography I previously published is not the Hibernia landing station, but the one of Vodafone:

Capture d’écran 2015-04-10 à 09.44.17

The Vodafone landing station in Brean

I have been duped by the fact  Jump, Vigilant and now New Line Networks reserved microwave frequencies around the house. I don’t know where is the exact location of the Hibernia station but  that is not important as the company won’t allow customers to pick up circuits at the landing station and re-reroute them via microwave. As far as I know, the Express cable will be first offered between the NY4 data centre in Secaucus (New Jersey) and the LD4 data centre in Slough (UK) – that’s all. Customers won’t be able to use microwave between the landing station in Brean and the Slough facility. That means all the competitors who booked frequencies between the two locations will have to wait a little bit:

Capture d’écran 2015-04-10 à 10.00.28

THE NEW LINE NETWORKS CLIFFHANGER

It is a shame that I can’t share all the stuff I have accumulated during this 6-month investigation on the HFT microwave networks in Europe. I’ll keep a close eye on those networks (mainly because it’s quite amazing to map the HFT world). Before writing this last episode, I had never thought that a new competitor would show up: New Line Networks (NLN). I talked about it in a previous post and I just want to add a few commentaries. New Line Networks is a joint-venture between Jump Trading (under the name World Class Wireless) and KCG (via their Geodesic Networks subsidiary). “KCG Holdings, Inc. (NYSE: KCG) announced today that its subsidiary, Geodesic Networks, LLC, and World Class Wireless, LLC have created a joint venture, New Line Networks LLC. New Line Networks will explore opportunities to leverage infrastructure investment, including re-selling network bandwidth to industry participants and third party vendors. Through this venture, New Line Networks will bring together complementary network and communication infrastructure while simultaneously providing additional data transmission bandwidth to the marketplace” said the announcement on February 12, 2015. The fact two major HFT firms from Chicago come together to build a new network is an unexpected event.

NLN first move was to buy all the UK licenses of  Communication Infrastructure UK in February 2015. That’s intriguing because Communication Infrastructure was one of the first microwave network here (at least they reserved a lot of licenses) and later the company was purchased by Perseus, in March 2014. Have in mind that Perseus resells the Jump network bandwidth to customers, so I was betting that if Perseus took possession of Communication Infrastructure, that was because they wanted to have their own network, so that they wouldn’t have to use the Jump one anymore. Are we to understand that Perseus abandoned this project? If not, why they sold all the Communication Infrastructure licenses to New Line Networks? The NLN second move was in March 2015: they bought 104 Ofcom licenses to Jump. With both the Communication Infrastructure and Jump licenses, New Line Networks has potentially a full network once they put dishes on all the towers – at least in the UK, I didn’t find the name “New Line Networks” on the French public websites (yet?).

Capture d’écran 2015-04-10 à 11.33.37

As shown in the map above, the NLN will use the now famous Houtem tower in Belgium (with a path going to Ramsgate and then to Basildon); NLN purchased some paths going to Brean, even if the Hibernia Project Express cable won’t allow microwave there, and that’s the reason why NLN also has paths going to Sennen, Cornwall (see Part V). The fact NLN will use the Houtem tower may explain why Jump purchased it for $5,000,000 – this was not cheap, but now it’s time to monetize the tower via New Line network? Pure speculation, again.

I have been told that the New Line Networks sales representative are touring the small world of HFT, hunting for customers. NLN is now a direct competitor of Perseus and McKay Brothers (and Custom Connect). What is most surprising here is that two HFT trading firms are now selling microwave communications to their HFT competitors, whereas McKay claims to be “private and independent. Not affiliated with any trading firm or exchange”, as said on their website (bold is mine). I don’t know what to think about that – are they trying to “nick” McKay’s clients? A veteran in the HFT microwave world told me that “any kind of alliance between HFT firms is very surprising… as this is not how industry works”; another admitted that “the New Line Networks move to create a cartel is very disconcerting”. “Cartel” may be a strong word, but perhaps not so far from the truth. I don’t know, but this is an unexpected cliffhanger.

The most interesting is that all the Jump and Getco founders started electronic trading on a platform/tower erected above the pits of the CME in 1998-1999 (Part I), where for the first time there computers and algorithms could really compete with human traders, screen to face, with the help of Globex. If (for some people) the high-frequency area started in 2007, we may say that the prehistory probably was in Chicago at the end of the 1990s, when some the first HFT firms were arbitraging computerized prices (on the tower) and human prices (in the pits). Some of them did it by using radio headsets: a trader working for the firm was dealing in the pit and transmitted the pit prices to another “trader” who was dealing on Globex. Without headsets, without radio waves, arbitrage was impossible. The Chicago futures pits will close on July 2, 2015, but some former pit traders are now behind state-of-the-art microwave networks, here in my bakyard, searching for tall towers to get data as fast as possible with the help of waves, trying to “arb” prices between London and Frankfurt. That’s how an anthropologist roamed the countryside, listening to the sound produced by the wind caressing the guys of the towers, a continuous sound not so far from the one you can hear in the data centers where electronic pits are now established – the breath of the computers being cooled. Trust me or not, but there is poetry in HFT. Now it’s time to move on. The next series will be about monks, because they erected towers long before the US Army built a phallic mast in Houtem. This is called steeples.

Advertisements

A quick note

A new academic paper about HFT titled High Frequency Trading and Extreme Price Movements came out on November 26, 2014. Led by Jonathan Brogaard (University of Washington), Ryan Riordan (Queen’s School of Business), Andriy Shkilko and Konstantin Sokolov (Wilfrid Laurier University), the study focus on “the relation between high-frequency trading and extreme price movements” and conclude that HFTs perform a “stabilizing function in markets during periods of stress”, when prices move more than usual – in short, the paper claims that in volatile times HF traders don’t vanish and still provide liquidity. As usual, as soon as a new academic paper is published about high-frequency trading, the tension escalates and the findings are debated. Yesterday Themis Trading (not a pro-HFT) published a response to this paper, and others replied to the response, and so on (by sheer coincidence,  Apple’s stock went wrong for a minute yesterday, allowing Matt Levine to write this post both about Apple and this new study).

I won’t dispute the conclusions here as the way some HFTs disappear (or not) in volatile times is a very debated issue – even among academics. But in my (humble) opinion and once again, this new study (like others) raise questions about 1) the data set used and 2) the definition of “a HF trader”. The first thing I noticed is that the data set used study is the same Jonathan Brogaard, Terrence Hendershott and Ryan Riordan used for another article titled “High Frequency Trading and Price Discovery” published earlier this year (but the first version of of the paper was available from April 2013). This data set, provided by Nasdaq, may be one of “the few U.S. datasets that identifies the trader types” but span 2008 and 2009 – in other words, the data is quite old. What’s more, it only includes 26 “firms that act as independent HFT proprietary trading firms”. That means some big banks having HFT desks are not included. Since these banks may be both non-HFTs and HFTs, it’s therefore normal to exclude them in order to have a more accurate data set by excluding non-HFTs activities, but that also means some data is missing.

Furthermore, the definition of a HF trader is rather ambiguous since it’s Nasdaq who “determines whether a participant is an HFT firm based on its knowledge of the firm’s activity”; so we have no information about the Nasdaq definition of a “high-frequency trader”. The new study states that “a firm is more likely to be identified as HFT if it trades frequently, holds small intraday inventory positions, and ends the day with nearly zero inventory [my bold]”. The words come from the authors, not from Nasdaq. Fair enough, but by discussing recently with the CEO of a big market maker (operating both in Europe and the US), I learnt the firm has huge overnight positions: several billion dollars. Even if these positions are not “directional” (in the sense that long/short positions are hedged with short/long positions – i.e. the positions are not liquidated, only the risk is liquidated), this major market maker doesn’t finish a day with a flat inventory. The question, then, arises: electronic market makers being HFTs (but not all HFTs are market makers), is this big market maker included in the Nasdaq data set? The answer is “yes”, but that means the definition of a “HF trader” given by the authors is probably not the one of Nasdaq (“nearly zero inventory” versus huge overnight positions). The way people define HFTs versus non HFTs is a crucial issue and I think one would expect more precise definitions.

So, the data set is old, includes 26 (prop) firms only and the paper studies 40 of the major stocks (by market capitalization). One of the author, Andriy Shkilko, admitted yesterday on Twitter that “there are no other large nonproprietary data on HFT available to” the academics, adding that “newer more extensive datasets have been hard to come by”, and that’s true. Data is missing (besides, I have been told that some data – from the CME for instance – is quite expensive), and that’s problematic. Just a few words about the Brogaard, Hendershott and Riordan study released last year, “High Frequency Trading and Price Discovery”. Again, I don’t dispute the conclusions of the study (HFTs improve price discovery) but this paper had a strange career. The first draft was posted in April 2013, and in November 2013 you could find it on the European Central bank website as a “working paper”.

It was strange because the ECB states that “this Working Paper should not be reported as representing the views of the European Central Bank (ECB). The views expressed are those of the authors and do not necessarily reflect those of the ECB.” It was strange because the data set used by the authors (the same as the one used for the new study published last week) includes “120 randomly selected stocks listed on NASDAQ and the NYSE”… nothing to do with stocks listed on European markets/MFTs/etc. Then the ECB version of the study was echoed by the Financial Times: “The ECB said the views expressed in the working paper were those of its authors and not necessarily those of the ECB. But the study’s conclusions could influence the thinking of the Frankfurt-based central bank, which is taking greater responsibility for bank supervision.” That’s interesting because the FT titled the article “High-speed trading ‘aids efficiency’, ECB says” [my bold]. Look at how an academic article studying HF traders and 120 stocks listed in the US suddenly became relevant for the EU market microstructure, whereas US and EU market structures are quite different. It inot the fault of the authors here, but the fact ECB put the study on their website conducted some journalists to write that “HFTs help price discovery” in general, and that’s a little bit problematic – again, no matter the truth, I only talk about the way a 2008-2009 US small data set become an absolute truth for some people in 2013.

Last week I was in Copenhagen for a workshop titled ”Investigating High-Frequency Trading. Theoretical, social and anthropological perspectives“. Most of the speakers were not from the HFT industry (they were mainly sociologists) but one representative of Bank of England was there. His presentation covered “material from two research papers on high frequency trading currently under production at the Bank of England. Both papers use unique transaction-level data from the UK equity market”. The talk was very interesting in the sense that these two studies “show how the informativeness of trades varies in the cross-section of HFT firms and also whether and how HFT activity is correlated within stocks and what this means for price efficiency”. I can’t seriously make a report with the notes I took (the studies compare passive versus agressive HFTs, intraday activities, etc.), but what was surprising is the fact the data set used only include the activities of 10 HFT firms – not even 26. The two papers should be released in the public domain in the next couple of weeks, and I bet these studies will be much discussed.

HFT in my backyard – III

INTRO

Het nieuwsblad, July 11, 2013

Last Friday evening, I received a tip via telephone, “You should go to the Houtem tower tomorrow morning. Jump may work on the network.” Unfortunately I couldn’t make it back to Houtem. But it would’ve been amazing to see contractors adding or realigning dishes, or working on equipment. Who knows what they got up to on Saturday! While it appears the €5,000,000 Houtem tower was the best deal the Belgium government made in 2013, the question of whether the purchase was a good deal for Jump still remains. The military history of these towers prior to their use for HFT is fascinating.

To learn more about Houtem tower, take a look at these photos, check out this Google Group, and read an account of the removal of all equipment, antennas and associated hardware by the 1st Communications Maintenance Squadron: “The site in Houtem Veurne, Belgium, has a 799 foot tall communications tower. Not only did Airmen have to climb this enormous tower, but if the support lines gave way and it fell in the right direction, a portion of the tower would have landed in France. While many of the technicians won’t miss the painstaking climb to the top of that tower, most are sad to see it go since it was such a challenge”. A “Cable and Antenna Maintenance Craftsman” who worked for the United States Air Force based in Ramstein, Germany, inadvertently reveals in his LinkedIn profile that long before HFT, Houtem tower was all about high frequency: “ Team project lead of microwave antenna shroud replacement at the 800ft level, Houtem, Belgium radio relay site” and “[I] installed and maintained microwave, Ultra High Frequency, Very High Frequency, and High Frequency antenna systems, cabling, connectors and their structures.” The “U.S. Air Force Wire and Cable Dawgs” Facebook page also includes anecdotes and the only photograph I found which was taken from the top of the tower:

10527323_10152148558085951_8005352295835710795_n

From the HFT microwaves industry I have also heard many stories. Two competitors visiting the same tower at the same time studiously avoided eye contact. A contractor working for various competitors complained about “all of these American guys asking for high towers… they just don’t exist!” Of course, some exist and are now used by high-frequency traders. There is this anecdote about Jump: “They were at a structure in Belgium waiting on an authority who controls this structure to give them permission. It was taking awhile. So they put a tower on wheels at the site. That pissed some people off. But then the tower was moving a lot. So they started pouring concrete around the base of the vehicle. The police were called out and there were forced to stop.” I don’t know if it’s true, but it’s fun – se non è vero è bene trovato. I was also told about a contractor who was supposed to install a dish for a firm on a tower already home to another HFT firm’s dishes. The workers made a mistake and pointed the new dish toward a tower used by a competitor!

Most invaluable to researching these microwave networks was the book The Death of a Pirate. British Radio and the Making of the Information Age (available in French here) by one of the greatest historians I know, Adrian Johns. This must-read tome focuses mainly on the famous UK pirate radios located on boats in the 1960s, but also cover the radios installed on the incredible Shivering Sands Army Forts, which were part of the Maunsell army Forts used for anti-aircraft defence during WWII. Thanks to these standing structures, the dishes were both more stable and taller than those on boats. By squatting in the old forts, broadcasters including Screaming Lord Sutch’s Radio City could use their height to reach more listeners and thus more revenue via advertising. It was all about height.

Shivering Sands Here are the precise locations of the Shivering Sands and Red Sands forts:

Thames Estuary See how the Shivering Sands forts are exactly on the straight line between NYSE’s facility in Basildon and Frankfurt’s Deutsche Börse. Optiver and McKay Brothers may each have a path very close to Red Sands as well (but I’m doubtful). I wonder if any HFT players thought to use these old forts. It would be amazing to replace the old pirate radio dishes with the big Andrew dishes Jump Trading favors. Torrent provider The Pirate Bay has already tried to install servers on another fort further north, the famous  Principality of Sealand to convert it to a “data haven”).

The most interesting discussion in The Death of a Pirate, however, is the history of the first radio pirates. They were not pop broadcasters earning cash using waves in international waters, but rather people who fiddled with the British Broadcasting Corporation (BBC) receivers to listen to non-BBC programmes like Radio Normandy. Waves are fruits of nature, to me a common good. When the first pirates emerged, states quickly monopolized these fruits and then were obliged to sort them. The International Telecommunication Union (ITU), originally founded in 1865 as the International Telegraph Convention, was charged with allocating waves. In the present, high-frequency traders have to pay state regulators for permission to use microwaves. If a firm wants to cross the channel from France to England, it has to ask state regulators for frequencies. Since most frequencies in the area were already booked or squatted, crossing the channel wasn’t an easy task – one competitor had some difficulties to cross the Channel. HFT is hard, and pirates may not be dead.

CUSTOM CONNECT

According to Bloomberg, “‘Custom Connect’s network was the first microwave link connecting the main financial centers in Europe that multiple trading firms could pay to use […] About 25 companies use Custom Connect’s network, including trading units at investment banks and high-frequency trading houses,’ said CC co-owner Jan Willem Meijer. ‘The network, which cost about 5 million euros ($6.8 million) to build, relies on 13 towers across Europe’”. Jump purchased a single tower at Houtem for the price of a whole network. My investigation of Custom Connect (CC) commenced in early July. I was visiting the CC website and at one point a chat window popped up and someone asked if he could help me. It was Jan Willem Meijer. When I told him I want to know where the CC towers are, he answered, “I can’t share. Confidential”. What a pity. On the CC website, however, there’s an  article from De Volkstrand (English version here), containing a map.

© De Volkskrant | April 30, 2014

© De Volkskrant | April 30, 2014

While not a truly accurate map, it’s interesting. Belgium is the least transparent country regarding data, but I was able to use various dossiers to find some towers housing CC dishes. Others I located thanks to Cartoradio, the official map of the French Agence National des Fréquences (ANFR), one of the French state regulators with Arcep. If you click on the right tower, the names “Custom Connect”, “Global Connect”, “Latent Networks”, etc., will appear and there you have all the details about the dishes. Azimuths are of particular interest, as they tell us where the signal goes. This allowed me to draw my first paths on Google Earth:

Custom Connect

Comparing the map from the Volkstrand article with mine revealed a problem with the Genappe tower (depicted as a white rectangle in the image above). Custom Connect should have two dishes on this tower, but the tower is north of the paths I drew. I have no data to link this tower to another, so my bet is on CC setting up a new network using the Genappe tower to replace the one in the south of Belgium. This is an educated guess, of course, but it is within the realm of possibility. Now let’s take a look at how the CC network crosses the channel:

Custom Connect - France-England

Custom Connect has dishes on a 107.2 meter guyed tower at Mont Lambert, near Boulogne-sur-mer (where Vigilant also likely have dishes), and the Cartoradio data tells us the two dishes have a 327.2 degree azimuth. This means the CC signal goes to another guyed tower to the south of Dover. CC also has dishes in Wrotham and Dunkirk, England, but reviewing my papers before writing this post, something disturbing jumped out at me.

When I started my map by drawing CC paths in July, I was unaware of the precious data one could find on the Ofcom website. Ofcom is the English regulator which allocates radio waves to customers such as radio and television broadcasters, mobile operators… and high-frequency traders. The Ofcom map allowed me to check where operators have asked to use frequencies, which in turn allowed me to find towers in England that may be used by these operators. One competitor gave me the tip to use Ofcom’s map to search for licenses used to cross the channel. Click here for a list of the English-licensed operators using the 2-26 gigahertz range (which is used by high-frequency traders) in the north of France and Belgium:

Capture d’écran 2014-09-30 à 14.00.51

You find Vigilant Global (aka Smartable LLC), McKay Brothers (aka Decyben), Jump Trading (aka World Class Wireless), Flow Traders (aka Global Connect), Optiver and Latent Networks but not Custom Connect, and that’s strange. I decided to compare the different data coming from the different regulators. I downloaded all the available public data from the French ANFR. The files list all the towers and all the dishes used (or not) by all operators. I dug a little and realized that all (without exception) the Custom Connect dishes in France are listed with this tag: “Dish authorized but not activated”. Very strange.

I dug more. I realized that the planning application for the CC dishes in Wrotham was very recent as the firm received an approval in July 2014. But the network is said to be operational since March 2013; Wrotham being in the middle of the Dover-Slough pathway, does that mean Custom Connect didn’t have a Dover-Slough link prior to July or August 2014? That would mean CC was only providing a network between Frankfurt and Basildon. There is more: the planning permit for the CC dishes in Dunkirk was granted in January 2014 – that’s problematic because Dunkirk is in the middle of the Dover-Basildon way, and this network is supposed to be operational since March 2013. What’s more, there is no Ofcom license with the name “Custom Connect” in Dunkirk. How they send a signal to Basildon? The most interesting is the planning application for the Dover tower:

© Dover District Council

First of all, the permit was submitted on July 25, 2014, and it’s not approved yet – files say CC wants to remove the dishes pointing to Boulogne, add a dish to go to Wrotham and keep the dish which points to Dunkirk. Secondly, the files also say two new dishes will have a 109° azimuth, so they won’t point to Boulogne but somewhere further north. Last but not least: the Ofcom database tells us CC has two licenses in Dover – 0996747 and 0956492. The license number 0996747 is needed to send the signal from Dover to Wrotham, and you can locate the license 0956492 in… Basildon. All of this sounds strange. I doubt it’s possible to build a path from Dover directly up to Basildon – if it’s possible, why all the competitors need at least one more tower between Dover and Basildon? What’s more, if CC has a license for a Dover-Basildon path, what about the planning application for the Dunkirk tower? That makes no sense (I can’t go more into technical details, but those who know will understand: it’s impossible to get a license on Dunkirk using the frequency assignment CC have at Dover and Basildon as the Dunkirk tower is in the same parity as Basildon and Dover.) So the situation is:

CC

I talked with people from the microwaves industry. I have been told it would be possible to cross the channel without a license but that would have been tough – and not very legal. I also heard rumors about a competitor having installed dishes without permission on the Dunkirk tower (some say that could be Custom Connect). But rumors are rumors, and I prefer facts. Having said that, it would not be the first time a HFT player would sneakily install dishes. Once Jump installed illegal dishes on the famous Swingate tower and they humbly apologized:  “The two largest dishes proposed for World Class Wireless were actually installed in advance of the submission of the planning applications. Both WCW and Arqiva [the owner of the tower] wish to offer their sincere apologies for this oversight and no disrespect was meant towards the Council”. LOL. Both the Dunkirk and Swingate towers were a part of the famous Chain Home system built in 1937 (I’ll detail on Part V); they have been very important for the Royal Air Force during WWII and now they are a part of UK’s history. Installing dishes without permission on these historical monuments is a “felony”. These monuments are the crown jewels of the United Kingdom – be respectful, guys!

I don’t know about the illegal dishes in Dunkirk, but I know for sure CC doesn’t have radio licenses there. As for the channel, I have found all the licenses needed by the different competitors, except the Custom Connect one – perhaps they are hiding very well, using other names, etc. I visited again the Custom Connect website, hoping the chat window would open but nobody was there. But I discovered a very amazing detail. The motto of the firm seems to be a statement once made by Steve Jobs. Check the LinkedIn profile of Olav van Doorn (the other co-owner of Custom Connect). The description of the company starts with these two sentences: “‘It’s better to be a pirate than to join the Navy’ (Steve Jobs). Nothing describes better what Custom Connect is all about”!

What? Custom Connect runs a piracy business? No way… I thought radio pirates no longer existed! During my three-month investigation I have heard about some guys trying to pirate frequencies but these tales may by myths. I can’t believe one of the fastest state-of-the-art network in the world would pirate radio frequencies. That can’t be. But that would be amazing: the old Maunsell army Forts, built in 1937, have been used by the pirate radios in the 1960s; so I would not be surprised to find pirates in Dunkirk and Swingate, as the tower was built in 1937 too, for the same military purpose. There is much mystery about Custom Connect – data is probably missing – but what cannot be disputed is the fact Jump Trading once installed illegal dishes on the Swingate tower. Pirates are not dead.

OPTIVER

I didn’t know who I would find when I was drawing my first paths on the map. I was pretty sure to meet McKay Brothers (I knew they were working on a network) but first I met firms I had never heard of before. One is Communication Infrastructure Ldt. It’s probably a network provider à la Custom Connect, not a prop trading firm. I have found them everywhere, from Cornwall to France. One of the anonymous e-mails I have received when investigating (yeah, I received anonymous emails!) said: “Communications Infrastructure is a shell with no customers and only licenses to sell to a sucker like…” – stop. Wait a minute. I won’t publish the name of the “sucker” yet, as the firm will be a guest star of Part IV. But I confirm: Communication Infrastructure asked for a lot of licenses in England but I have found no planning application. I even doubt they have installed dishes (in France, some of them are “not activated”.) [UPDATE, 10/06/2014: they have at least one planning application granted, here | sorry for the mistake]. An empty shell for now?

I also got to know NexxCom, which built a microwave network between New Jersey and Chicago. I discovered them in Cornwall, at the very end of the microwaves paths. There are two kind of competitors in Europe: some are only interested by linking London and Frankfurt (Custom Connect, McKay Brothers); others decided to spread their networks to Chicago (Vigilant, Optiver, Jump Trading, Flow Traders, Comm. Infra and NexxCom), so they have to go to South Cornwall where they can meet the Atlantic Crossing 1 (AC1) landing station – from there, data is sent in the US thanks to optic fibers. But in two years Cornwall will no longer be important, so I don’t really understand why NexxCom submitted a planning permit in Four Lanes last February. (I have heard NexxCom may be supported by a HFT firm but I have no name.)

Capture d’écran 2014-09-30 à 15.56.52

One big competitor – far more serious than Communication Infrastructure or NexxCom – is Dutch prop firm Optiver. Optiver has some elegant paths and a good reputation in the industry. Some say they may be the fastest – being the fastest is cool, but the most fundamental task is to remain the fastest, and that’s not easy. Optiver has two licenses to cross the channel and they co-locate with Custom Connect and McKay Brothers in Dunkirk. I visited the Hannut tower in Belgium and shot their cabinet:

P1010252

Optiver has interesting paths in Cornwall (they have offices in Chicago):

Capture d’écran 2014-10-01 à 11.00.46

FLOW TRADERS

You won’t find the name “Flow Traders” (another Dutch prop firm) in any legal document, you’ll only read the name “Global Connect”. I had never heard of “Global Connect” before, but the postal address is Jacob Bontiusplaats 9, 1018 LL Amsterdam, where Flow Traders is located. I have nothing special to tell about Flow. Some say they were the first European firm to build a microwave network between London and Frankfurt (and they have licenses to cross the channel). Like Optiver, Flow Traders has paths until Cornwall. There is a third big Dutch firm in Amsterdam, IMC, but I couldn’t find evidence IMC has a microwaves network, even if in this LinkedIn page published on January 31, 2014, we learn the firm is “seeking a technical Project Manager or Lead Engineer to improve our efforts in building out the fastest network paths used for transmitting financial market data between major financial locations around the globe”. A new competitor may appear on my map soon.

VIGILANT GLOBAL

Here is Vigilant Global, a Montreal-based prop trading firm. Vigilant is now a subsidiary of DRW.  One people from the industry once confessed: “Vigilant is my best competitor.” Another said: “I actually admire Vigilant/DRW. They were first. They seem to be pretty class acts about it all.” Yes, Vigilant were the first to build a network in Europe. They probably initiated their network 2011, perhaps even in 2010 (there is a funny anecdote about an old dish on the Equinix data center in Frankfurt). By working on the map I understood why some people in the industry admire Vigilant: they are really good.

Ann Brocklehurst wrote about the difficulties Vigilant encountered when they tried to install dishes in Benfleet, but I don’t think this was a big problem. Why Vigilant is good? First, they don’t ask for a lot of licenses before installing their dishes (Jump ask for a lot of frequencies but often they don’t use them at all, even if they have to pay the UK regulator). I think the Vigilant engineers checked all the possibilities very carefully before asking for licenses. Secondly, they have some very elegant paths. I would be very curious to know if Vigilant really built these paths from Boulogne to Cornwall:

Above the sea

These are long and complex paths (microwaves don’t like dampness nor fog). But their best move is probably this path:

Vigilant

To understand it you must have in mind that a new optic fiber cable is in progress between England and the US: the notorious Hibernia cable (the plan has a long and fun history I will detail in Part V). It has been said the Hibernia cable should be faster than the AC1 cable by saving six milliseconds – that’s why the title of my book is 6. But the landing station of Hibernia (in Brean, Somerset) is far from the AC1 landing station (Whitesand Bay, Cornwall):

Networks before Hibernia

The Hibernia cable is due to open in two years. If the high-frequency traders can save six milliseconds with the help of new optic fibers, they won’t use the AC1 cable anymore. That would be stupid to have these fast microwaves networks without exploiting a new faster cable. That means allllllll the paths going to Cornwall will be dead in two years. And the map will look like that:

Networks after Hibernia

So Jump Trading, Vigilant and Flow Traders already booked licenses/frequencies to build new paths between Slough and Brean (Optiver may be around too), and Vigilant has anticipated the Hibernia cable with this long and classy path:

Vigilant

I would love to have a chat with the Vigilant engineers who designed the paths. These people definitively have good neural networks.

MCKAY BROTHERS

Let’s finish with McKay Brothers. Even if the first Chicago-New Jersey microwaves network was built by Alex Pilosov’s Windy Apple Technologies, everyone thinks McKay are the fastest in the US. In this document released by Aviat Networks (Aviat provides technology to most of – if not all – the HFT players, both in the US and in EU) you can read this: “The McKay Brothers LLC low latency network between Chicago and New York—the world’s busiest route for low latency financial networks…”. McKay is now a fierce competitor in Europe. Their network is still in progress but I have found towers where they have put dishes in place. One of their best path is between Liège and Sint Pieters Leeuw (in Belgium). Here is the only Belgian document I have found with all the details about the dishes put by a private operator (I parsed more than 18,000 documents to find it):

McKay | Ougrée-Sint Pieters Leeuw

That’s how I could draw this 96,3 kilometers path between two of the tallest towers in Belgium:

Ougrée-Sint Pieters Leuw ± McKay Brothers

TSPL2he Sint Pieters Leeuw tower is the tallest standing structure in Belgium (300 meters). McKay having the fastest microwaves network in the US, I am not surprised to find their dishes on this tallest structure – a new “cathedral of capitalism”. I went to visit the building and shoot McKay’s dishes. Again, the tower is surrounded by potato fields, and right at the basement of the tower there is a small pen where a calf was quietly watching me shooting the tower. This beautiful building is owned by Norkring. Norkring is a is a network provider managing 24 transmitter masts in Flanders and BrusselsThey even offer colocation: “On eight locations, Norkring België offers all facilities for hosting telecommunication operators, local broadcasters, emergency services, television operators, the police, or any other party that wishes to set up a wireless network. Several heights are possible for installing antennas on our transmission towers, ranging up to a level of 300 meters.” I don’t know where are these eight towers (you can’t find this information on the Norkring website) but I have been told Vigilant could use one (or more) Norkring tower. Other HFT players may also use the Norkring network.

© Norkring

But the most interesting part of the Sint Pieters Leeuw tower are these incredible stairs. I have been asked to appear in a French documentary about HFT and market microstructure, and I think the best place to shoot is there:

Sint Pieters Leuw | Norkring

One last thing about McKay Brothers (and the parent company Quincy Data): they are transparent. They are the only providers to publish both prices and latencies “client port to client port”. “Client port to client port” means that latencies (e.g. 2.195 milliseconds between Basildon and Frankufrt, but McKay will be faster soon) are calculated for a trip between a co-located server (inside NYSE’s facility in Basildon) to another colocated server (inside Equinix data center in Frankfurt). I think they are the only ones to publish these latencies. What’s more, McKay Brothers co-CEO Stéphane Tyc advocates for more transparency in the market industry (check the last white paper he released two weeks ago about better execution). I think McKay Brothers goal is to “level the playing field” in the small world of microwaves networks. They are on the tallest structure in Belgium, but will they be the fastest? All bets are off.

OUTRO

One network is missing: Latent Networks. I had never heard of them it I found some of “their” towers. I tried to get information about them but the people behind the company did everything possible to hide. So I decided to dig. Then the first anonymous email arrived in my mailbox. Then the name of a big HFT firm from Chicago showed up – not Jump Trading. Then I started a strange relationship with a kangaroo. I’ll devote Part IV to Latent Networks. Stay tuned.