Saturday, 9 March 2019

Speedbumpiness: the yin and yang

Speedbumps are good and bad. It's a yin and yang equation.

Here is a short list of my personal evaluation of some bumps or proposed bumps:

ICE - gold and silver futures - bad
IEX - terrible
TSX Alpha - bad
Eurex limited options - bad
FX last looks - very bad
FX arseymmetric bumps - bad
NYSE American - bad but funny

William Poundstone opens Fortune's Formula with an enjoyable example of a speed-bump:

THE STORY STARTS with a corrupt telegraph operator. His name was John Payne, and he worked for Western Union’s Cincinnati office in the early 1900s. At the urging of one of its largest stockholders, Western Union took a moral stand against the evils of gambling. It adopted a policy of refusing to transmit messages reporting horse race results. Payne quit his job and started his own Payne Telegraph Service of Cincinnati. The new service’s sole purpose was to report racetrack results to bookies.  
Payne stationed an employee at the local racetrack. The instant a horse crossed the finish line, the employee used a hand mirror to flash the winner, in code, to another employee in a nearby tall building. This employee telegraphed the results to pool halls all over Cincinnati, on leased wires. 
In our age of omnipresent live sports coverage, the value of Payne’s service may not be apparent. Without the telegraphed results, it could take minutes for news of winning horses to reach bookies. All sorts of shifty practices exploited this delay. A customer who learned the winner before the bookmakers did could place bets on a horse that had already won. 
Payne’s service ensured that the bookies had the advantage. When a customer tried to place a bet on a horse that had already won, the bookie would know it and refuse the bet. When a bettor unknowingly tried to place a bet on a horse that had already lost…naturally, the bookie accepted that bet. 
It is the American dream to invent a useful new product or service that makes a fortune. Within a few years, the Payne wire service was reporting results for tracks from Saratoga to the Midwest. Local crackdowns on gambling only boosted business.
The speedbump delay to the bookies allowed them to not disadvantaged by someone having better information. They used a last-look to only take profitable bets.

The scale of it shouldn't be underestimated, Fortune's Formula continues describing the land before RICO,
The growth of General News Bureau paralleled that of the American Telephone and Telegraph Company. In 1894 Alexander Graham Bell’s telephone patents expired. Within a few years, over 6,000 local telephone companies were competing for the U.S. market. AT&T acquired or drove most of them out of business. Though AT&T’s techniques were more gentlemanly than Annenberg’s, the result was about the same. The government stepped in with an antitrust suit. The legal action was settled in 1913 with an agreement that AT&T permit competing phone companies to connect to its long-distance network. In 1915 the first coast-to-coast telephone line went into operation. The following year, AT&T was added to the Dow Jones average. With its now-legal monopoly and reliable dividend, AT&T was reputed to be a favorite stock of widows and orphans. 
Few of those widows and orphans realized how closely the phone company’s business was connected to bookmaking. General News Bureau did not own the wires connecting every racetrack and bookie joint. It leased lines and equipment from AT&T, much as today’s Internet services lease cables and routers. Both telegraph and voice lines were used. As the system grew more sophisticated, voice lines provided live track commentary.AT&T’s attorneys worried about this side of the business. An in-house legal opinion from 1924 read: “These applicants [the racing wire services] must know that a majority of their customers are bound to be owners of poolrooms and bookmakers. They cannot willfully blind themselves to these facts and, in fact, set up their ignorance of what everybody knows in order to cooperate with lawbreakers.” 
On legal advice, AT&T put an escape clause in its contract with the wire lessees. The clause gave the phone company the right to cancel service should authorities judge the lessee’s business illegal. AT&T continued to do business with bookies—while officially it could claim to be shocked that gambling was going on in its network. By the mid-1930s, Moe Annenberg was AT&T’s fifth largest customer.
The Sting had a nice dramatisation of a dramatisation of such a shop,



So, in my leading "leading" examples, I proposed all of those existing speedbumps mentioned are bad. Yet I suggest there is a yin and yang to them. Where is the yang? Let's meander on.

Good speedbumps

Speedbumps are rational. They may offer a benefit: one type of fairness. Think of an extreme speedbump, such as waiting a day and getting a fair price by formula. Daily VWAP trading is an example of this. VWAP may no longer be the best benchmark to consider, but, in its day, this kind of approach was considered a fair way to play Mr Market's random walk. It is a kind of speedbump that captures much of the essence of fairness.

Back in the old daze of Jefferies, their high touch business noted many of their clients bought stocks other clients sold. There was an opportunity to match these clients and give them a benchmark price. Barra was contracted to write the complex and sophisticated code of comparing two lists of stock to buy and sell. That was a mistake. Barra earnt a silly amount of fees until finally being bought out more than a decade later.

That business for Jefferies did very well, especially as the fees came down. The idea of a lower price for electronically matched orders did not sit well with Jefferies high touch business, so three employees rolled out and started ITG with POSIT in tow. ITG was quite the innovator until they went off the rails, or perhaps stayed on them would be a better analogy.

This kind of dark trading is parasitic, like an index fund, as it relies on price discovery outside the tent but it can be considered fair enough with the benefit of not impacting prices with sizeable blocks.

It's a balance.

Such dark mechanisms have several drawbacks. There is a perverse incentive to maximise information toward the deadline for placing a buy or sell. If you want to buy at tomorrow's clearing price, you may think differently in the afternoon after a huge rally as compared to how you evaluated the scenario in the morning. There has been a war in gaming and anti-gaming techniques with a particular focus on this deadline issue.

Consider if everybody went this route. There would be no reference price to price off. This doesn't work obviously. It seems of a silly theoretic worry until you consider thinly traded assets. That is the nature of a parasite, it may be good to a host until the host is overwhelmed.

Another major drawback is the delay in setting the price. We don't live in an isolated bubble. There is a utility to speed. Faster trades not only allow better information transmission, but they also allow improved risk management due to the lessening of uncertainty and extra trade opportunities. Faster trades also allow more frequent replenishment of liquidity being priced at best.

Consider an auction every second versus one that is hourly. The auction every second is going to be the better resource for pricing and risk. There is more opportunity to hedge, spread, arb, and replenish. The shorter time frame will be the price leader. Do you think an option would be priced off the second hand or the hour hand?

That's a difference of 3,600 to one. It seems kind of obvious but then we lose objectivity when we enter into the mind of the machine with the ascent of the algorithm. A microsecond to a millisecond is over anthropomorphised into eye-blinks and the like. The lack of human scale causes our rationality to wander.

A ten-microsecond exchange can do twenty transactions in the time it takes a two hundred microsecond exchange to do one. Which one do you think may provide better price discovery, risk management and replenishment opportunities?

CODA and Budish et al push elements of slowness akin to a speedbump for continual batch auctions. If they are fair there may be a utility to such speed-bumped thoughts. Prof Eric Budish from Chicago Booth pushes his hobby horse idea of rapid batch auctions as a way of circumventing the need for speed. There is merit to the argument but it similarly suffers from parasitic tendencies, inefficiency, and some gameability via a deadline focus.

I feel such dark activity has some merit and place in modern markets. It was kind of an upstairs thing that migrated to the ATS dark pool. Arguably it should be excluded from public markets and remain in the land of the ATS. My simple view is prices should be visible, tradeable, and striving for efficiency. That is, I am suggesting parasitic markets and their darkness should be excluded from privileged public pools. Hidden orders, icebergs, pegs, and their kin confuse this model with integrated darkness and seem unsuitable for efficient public price discovery.

Good speedbumps are possible but they should have a limited role in public markets.

FX


Foreign exchange is far from simple. There is circling randomness and all sorts of going-ons to confuse the punter in various pools. However, there is also a pretty clear and worrying narrative.

Many banks and institutions had fx pools that they ran for their benefit. The so-called single name pools. They didn't like being picked off in their pools by faster players. Their main way of curbing you in the old days was to give you very limited credit so you couldn't really trade and make any sizeable bank.

The pools both got better at tech and were also threatened by the slow realisation their behaviour was anti-competitive and challengeable. So the credit squeeze eased. One way to give themselves an edge was to add a delay so the bank could decide if they would accept your trade or bin it. This was last-look. They held your order for a while and if it was a loser for the bank, they'd frown and disown it.

That doesn't seem the worst thing in the world when your alternative may be paying a 5% spread at retail, but it violates the principle of visible quotes being tradeable. Unfortunately, it is worse than it seems. Many house traders ran their books more aggressively. With the beneficial hindsight of last-look, there was a tendency to avoid even fair orders and to only accept orders that ran in favour of the house and against the client. The strive for profit maximisation led to the occasional thread of perverse client exploitation.

Remember the other features though, even if last-look was fair it still introduces some inefficiency by limiting the ability to hedge, replenish, and otherwise manage your risk efficiently. This matters more to some than others but no one likes being taken advantage of.

Last-look has a deserved bad name.


IEX


I've covered IEX to death. I despise their speed-bump as it is not symmetric, fair, nor simple. It is just a dumb construct, not all of which was IEX's intention but largely so. You can read about it here in Speed-bump 101. The most troubling thing is how IEX misrepresent their flawed system and poor execution outcomes. The SEC has to take some heat for this as it allows IEX to exist as a public exchange with less than 30% of their handled trading being lit. Bad SEC. Bad regulator. Bad. Bad. Bad.

However, IEX had a different intention at the outset that wasn't so bad.

What IEX isn't(click to enlarge)



It was kind of meant to look something like the above picture. It doesn't as it leaks information and has an uneven and unfair lack of colocation, but let's ignore that for now.

A closer approximation of IEX
(click to enlarge)


The intention from IEX was that the speed-bump would be symmetric and allow them to set a fair price, e.g. a reasonable, for some definition of reasonable, midpoint price, for buyers and sellers to match. A kind of dark order. The speed-bump delay allows them to consume all the incoming market data information and look into the future, albeit a short lookahead, and devise an algorithm to decide on the matching price.

There are a few problems I find with this. Firstly, all US NMS public exchanges have dark orders and I don't like it. I like IEX less as it is darker. It is mainly dark. This parasitic behaviour detracts from the price discovery process that should be the role of such privileged markets.

Secondly, the delayed lit quotes you see in the market data feed are like a hall of mirrors. They might not really be there by the time your order winds the long and windy cable, literally long and windy from the spool of wound fibre. Them there fake quotes are disturbing the force. Quotes should be actionable.

The other thing IEX prevents is fair innovation. Its lame dark order types and bogus logistic regression based crumbling quote indicator prevent brokers or other participants rolling their own algorithms. I have written about how IEX's innovation kills innovation. Prevention of innovation is something you'd hope a regulator would raise an eyebrow to. The speed-bump protects IEX as it is the only one that can see into the future in its internal world.

IEX gets worse the closer you look at it. It has separate issues as a speed-bumped lit exchange beyond being a slow and inefficient exchange with high fees, such as auto fading, but meanderers know that already. Let's meander on.


The asymmetric speed-bump: the last-look-a-like demon


TSX Alpha is quite a big turkey. It offers an asymmetric speed-bump. I think of it as arse-emmetric. It smells that bad.

It is kind of a last look where a delay, randomised to 1-3 milliseconds, exists on the aggressive order allowing the price posting order, the resting or passive order, time to get out of the way.

This is not quite last-look but it has some similar features. On such an asymmetric speed-bump a resting order can get out of the way if its master thinks the market is now adverse. It's not the same as last-look as you don't actually know if an order could be headed your way, but it is similar in that you may just decide to always fade if the market is not in your favour near the end of your bump time. That's quite an abuse.

You get the same house of mirrors effect as a symmetric bump where visible quote may not be tradeable. The asymmetry has the benefit where the posting price process, market making, is derisked so participants may quote larger and tighter as they can duck and weave without committing. It looks better but it is not. It's smoke and mirrors.

It's not as simple as that as losing queue priority is still a thing, but you get the idea: fake prices are bad.

Regulators have been hoodwinked into allowing asymmetric speed bumps.

One consolation for such asymmetry is that it isn't as dumb as the Aquis Exchange's post only if you are a prop biz model. That is an amusing but dangerous piece of stupidity. The Aquis model is seemingly "F$%^*&# it!", let's just make it unfair by segmentation. There is a curious reliance on external players in such a place. At least the SEC realised the danger of that model in the US and is unlikely to countenance such a beast.


Is this so new?

Not so new, but a hundred years ago buckets shops were subtler than relying on "The Sting" last looks. They had other ways and means.

In Reminiscences of a Stock Operator by Edwin LeFevre (1923) it was noted that such last look games didn't really need to be played by the unscrupulous bucket shops as there were plenty of other ways to skin a cat,
"Bucket shops in those days seldom lay down on their customers. They didn't have to. There were other ways of parting customers from their money, even when they guessed right. The business was tremendously profitable. When it was conducted legitimately I mean straight, as far as the bucket shop went the fluctuations took care of the shoestring"
The modern equivalent would be the highly levered account at a spread betting or CFD shop.

Nevertheless, our protagonist found his own problems when his visible tape and true market were out of synchronisation by delay,
"The ticker beat me by lagging so far behind the market. I was accustomed to regarding the tape as the best little friend I had because I bet according to what it told me. But this time the tape double-crossed me. The divergence between the printed and the actual prices undid me. It was the sublimation of my previous unsuccess, the selfsame thing that had beaten me before. It seems so obvious now that tape reading is not enough, irrespective of the brokers' execution, that I wonder why I didn't then see both my trouble and the remedy for it."
I had similar problems in the early nineties trying to trade S&P futures from Sydney via a direct voice box to the Chicago floor. Delays are a killer for many a seemingly reasonable strategy.

I suspect a fresher finance community may be arguing about similar features in another hundred years.

Are speed bumps really all bad? What about a 1-microsecond speed bump?


Reiterating: the idea of delayed darkness is not altogether bad. I just don't like the concept of parasitic darkness being encouraged in public exchanges. Let them go hang out at an ATS or broker's shop. Keep public markets, open, fair, and efficient. We should be encouraging open and fair public price discovery.

I used to worry more about overly complex order types rather than the pegs, hiddens, and icebergs. That was until IEX came along with its perverse, leaky, innovation killing, speed-bump with high fees and misleading snake oil salesmen. To think it all came to widespread attention thanks to the conflicted Lewis helping out his old Netscape "The New New Thing" buddies with their investments in Spread Networks and IEX. The true nature of the dark side of this dark fader makes IEX's flawed model a priority concern when considering the health of public markets.

One of the great levellers of the playing field was the widespread adoption of colocation facilities along with the equalisation of cable lengths. This ensured fairness in the exchange data centre. It is also a kind of speed-bump as the cable lengths are typically not de minimus. A couple of microsecond asymmetric speed-bump may be seen as a fair technology field as typical variation in 10GbE hardware stacks is 0-2 microseconds. That is a slippery slope though and quite the contextual evaluation as competitiveness with FPGAs is now down to single-digit nanoseconds. It is a bit of an easier argument than hundreds of microseconds. Hundreds of microseconds just make an exchange a poor inefficient clunker.

So, I'm arguing speed bumps may have a place. Really big ones for tomorrow's parasitic match seem fair fare for a dark place. Tiny ones, say a couple of microseconds with today's tech, for stimulating fairness with a minimal contextual impact so that efficiency is barely impacted look somewhat benign and arguable. Keep the ones in between out of the public markets, please.

There are other ways exchanges could approach making venues fair. Inverting IEX's approach by bringing the outside in, instead of keeping it out, via customer provisioning of consolidated external feeds may be a better way to spend an IEX sized investment and even things up. Another interesting way may be making an entire region, or planet, a fair co-lo equivalent with timed protected gateways, perhaps even in a customers premise. Also, distributing news, such as SEC Edgar reports, by co-ordinated in co-lo release would be an admirable and inexpensive fairness measure that would make a few RF networks redundant. The SEC should get on that.

We are far from exhausting the exhausting fairness debate.


Happy trading,


--Matt.

1 comment:

  1. What would be gameable about the budish frequent auctions?

    ReplyDelete