I found myself recently reaching for a firstLeft function. The idea was that using
Either to model failure/success of a function, you either want to get the first error back (
Left) or all the results (
firstLeft :: [Either a b] -> Either a [b] Prelude> firstLeft [Right 1, Right 2, Right 3] Right [1,2,3] Prelude> firstLeft [Right 1, Right 2, Right 3, Left "gah!"] Left "gah!" Prelude> firstLeft [Right 1, Left "ooh!", Right 2, Right 3, Left "gah!"] Left "ooh!"
Benl immediately pointed out to me, this is in fact
sequence :: (Monad m, Traversable t) => t (m a) -> m (t a) Prelude> sequence [Right 1, Right 2, Right 3] Right [1,2,3] Prelude> sequence [Right 1, Right 2, Right 3, Left "gah!"] Left "gah!" Prelude> sequence [Right 1, Left "ooh!", Right 2, Right 3, Left "gah!"] Left "ooh!"
If this is not immediately obvious as to why these two are equivalent, see below. Type variables can effectively be curried:
sequence :: (Monad m, Traversable t) => t (m a) -> m (t a) -- m = Either a: sequence' :: (Traversable t) => t ((Either a) b) -> (Either a) (t b) -- t = : sequence'' :: [Either a b] -> Either a [b] firstLeft :: [Either a b] -> Either a [b]
sequence seems a bit confusing, I generally thought of this as a kind of
collect operation. Interestingly, the documentation for
sequence refers to this term too in describing it:
Evaluate each [..] action in the structure [..] and collect the results.
Over time the equities marketplace has evolved from phone-based trade execution, to using computerised channels such as Direct Market Access.
This is where a trader sends an order electronically to their broker who, assuming everything is OK, simply passes it straight through to the market. Unlike in the past, where the trader may have placed an order with the broker, granting them some discretion over when and how to execute it; DMA is simply a pipe giving you immediate access to the market. Think of it as a no-frills self-service offering.
As latency (speed) became more of an issue in recent years, firms investigated ways to speed up their DMA access times. One of the central tenants of DMA is that the broker still performs checks on each order to ensure that the firm has the capital required to settle the orders, and that the orders do not create a disorderly market. Somewhere along the line DMA evolved into something more aggressive: naked access. This is where the broker effectively allowed the firm to send orders directly to the exchange, completely bypassing the broker. In this model the broker trusted the firm to “do the right thing” and not send orders for things they could not afford to settle, or create disorderly market. Now that’s real trust.
It has been reported in the US that 30% of turnover was conducted via naked access around 2010, with the SEC moving to ban the practice the following year.
Regulators globally stepped in and questioned the validity of brokers outsourcing their responsibilities to trading firms in this manner. This delivered a new concept: sponsored access. The intention of this model is to be able to achieve the extreme speed permitted by naked access, whilst still ensuring the broker stays in control of their client’s trading. The problem with sponsored access is, that the manner in which a broker “stays in control” can be a little ambiguous.
Naked access has never been permitted in Australia, with a fairly strong regime of regulation around the checks that must be performed before electronic orders can flow through to the market. However, there is only a grey line between sponsored access and naked access.
A broker granting sponsored access to a trading firm would give them some software to run on their servers (or FPGA chip), so that when the firm’s algorithm decides to trade it calls this software to do a check as to whether the order is permitted according to the brokers’ requirements. If successful, the firm sends the order directly to the exchange. The broker never sees the order, only observing it in in the market just like everyone else. In the US there is a range of products which provide direct access to the firm, but perform risk checks in a hardware chip that is completely controlled by the broker. I don't see such products being pitched in Australia, leading me to believe that firms are being given software by their brokers to perform the risk checks. In this model (which I'd argue is exactly what firms were doing under naked access) there is a trust by the broker to the firm that they will do the right thing and make the necessary calls. Therein lies the moral hazard.
When you have a marketplace that rewards speed, as the electronic equity market does, and then you allow traders to have any kind of control over the risk checking of their orders – you have the ingredients for trouble. By definition the most profitable ultra-low latency trader is going to need the minimum possible risk checking, since every risk check costs time and therefore makes them more uncompetitive.
Many talk about the Knight Capital incident in the US in terms of it being a kind of "worst case scenario" reference point for algorithms run amok. I think it was far from it. Importantly, they did not lose more money than the capital base of their firm. In other words, while the shareholders lost a bundle, they did not need a bale-out. No third-party funds needed to be drawn to cover the losses.
During the malfunctions at Knight Capital, one thing other firms noticed that they were effectively buying massive amounts of liquidity – and paying away the bid/ask spread. The other firms started exploiting this fact (I mean, we all try to buy low and sell high, right?) and ramped up selling liquidity to them – and making near risk free profits. What if this had happened a lot faster, since it took a while for them to catch on that this was really happening? You can be sure next time there is a wounded trader in the market, they will be targeted - since they've seen this before. There is the potential that you could have massive losses (say trillions), inside a microscopic time period (say a few seconds).
If this were to happen you have a case where a firm has lost its capital base, and more - and who has trillions? The liability then goes to the broker, with as low as $10M in capital - that wont help much. Then it moves to the National Guarantee Fund with $100M. No stopping there either. Naturally many will take the view that this is "obviously" errant trading, and the trades should be busted. There would be a good case for this - but it would be a dangerous precedent since the trading would not in any way be "disorderly". Having someone (in error) buy massive liquidity from the market (buying at the ask and selling at the bid) does not in itself move prices - especially when a whole bunch of high speed peers are sitting there happily providing it and banking the spread. Whilst the root cause of Knight Capital's woes was someone releasing the wrong software into production, the system needs to be robust (antifragile?) enough to cope with these events.
There will be other causes to this kind of runaway event: for example it is quite possible that two ultra-high speed trading systems could interact with each-other in such a way that they behave in a completely unexpected way. This kind of bug could never be found in testing - since the algorithms would only ever "see" one another in the live market. Testing for every possible permutation of peer algorithm is, impossible.
The on-ramps for ultra-fast algorithms needs to be a lot more robust than it currently is, and ensure comprehensive risk checking is consistently performed - and in genuine real-time. That means the risk checks need to be ultra-low latency themselves, and certainly not performed on firm-controlled infrastructure. Put another way, making the trade-off between risk and speed needs to be done by someone who is not profiting from the speed and (as all firms/brokers are) passing the "tail risk" (read: really nasty unexpected events) to someone else.
Yes there is. The problem in all this is that the party in the transaction that stands to make the profits from trading the quickest, is the one who in practice enforces the risk checks. This needs to change – and I don’t mean by making the broker do it. That too is a compromise, as brokers will market their services over the speed (read: less risk checks) in their trading interface. There is skewed motivation for a broker to do sophisticated counter-party risk validation on each order (required to prevent a runaway algo) because of the latency cost involved. That is to say if they did, they would be considered uncompetitive with other brokers who did not have this requirement.
The solution to all this is: do the risk checks at the exchange. This has the dual benefit of ensuring a level playing field of comprehensive risk checks (everyone gets the same checks and corresponding speed penalty), plus the exchange is in a position to assess the fact that the firm has not run out of capital in a run-away malfunction. The exchange can vett each order, ensuring that the order flow is consistent with the capital base of the firm, not allowing each order until that validation is performed.
I should note that some of the more basic vetting is done by ASX on futures, but these need to be widened and applied to equities as well. The vetting rules on futures are fairly basic, but they seem to be enough for ASIC to think allowing naked access to clients is OK since that routinely happens. I personally think this is an accident waiting to happen.
Of course, a natural hedge to the runaway algo effect is to slow the market down and allow the market to set it's own pace – but that’s another story..
I have found the term High Frequency Trading to be somewhat overloaded, and now unusable term. Instead of referring to high-tech traders, the terms is now referring to more traditional traders who are dragged into the technology war in order to protect themselves. To throw both of these groups under the one "HFT" label is to completely misunderstand what is going on. I would like to propose a model for HFT that splits the term into 4 subcategories:
Of course the only remaining category on this list to describe the entire market is: humans. It is unfortunate that the press, regulators and folklore have type-cast these four communities under a single heading. Just like trying to describe a place, or a race of people under a single stereotype – so to thinking about HFT in it’s entirely is somewhat meaningless. It is uninteresting in the extreme to talk about the commonalities of a broad populations of people, which is exactly what the recent ASIC report did for HFT. No wonder they didn't find much.
Looking inside this very broad community is far more interesting. Under the HFT banner there is both predator and prey. Each of the categories outlined above has a very different mix of motivations, techniques; and therefore implications for the market. Some categories (like Algorithmic Execution) are often only employed because they’re forced to, in order to try to protect them against other more informed HFT’ers. They’re not the ones to look at in order to find causality.
Let’s focus on the most misunderstood, perhaps unknown, of these categories: ultra-low latency. These are not the “HFT” firms like Optiver who provide quotes to the media, that we are supposedly meant to take as somewhat representative of the whole HFT population. Instead, ultra-low latency specialists with one capability: speed. Nowadays, everyone under the HFT umbrella thinks they are "low-latency", since they think they are fast. I’m not talking about them, but rather the ones that dedicate themselves to being fast. The difference between low-latency and "ultra low-latency" is that the latter are the fastest.
When I say fast, I mean incomprehensibly fast. To be the fastest firm in the market to make a trading decision and send an order, you would need to be able to act within 5 micro-seconds. To put that in perspective it takes a 747 jet travelling at top speed (say 1,000 km/h) this long to travel 1.5 millimetres. Your LCD monitor will probably not be able to update itself and show this decision for another 2 meters. The technology used to employ this speed has gone beyond computer “servers” in any sense you would recognise. In contrast to what you may think: immense servers with banks of flashing lights, they instead use specialised FPGA chips which are essentially tiny hard-wired computers that are only capable of performing one task: trading. There is no hard disk or operating system, screen, mouse or keyboard – just a network port. Anything not directly relating to making a trading decision is stripped out - who needs it. Forget Windows or Linux there is no operating system to slow things down – the program is hard-wired into the chip.
Whilst being really fast is important – it is just a means to the real end objective: to be faster than the other guy. And being the fastest is not just about having great technology - there are another aspects that dictates the pace: such as compromise. If you are competing with another trader (and no doubt you are) who also has the latest technology you do, but they are just a bit more motivated than you to being the fastest – they still have an angle. Here is where the risks checks come in. Two traders with identical technology then inevitably get into a reverse auction on risk checking, as each risk check (or even sanity check) comes with a cost.
What we have now is a market that rewards the fastest traders with profit, since they are able to beat their competition to the punch. In order to be the fastest these traders have the perverse incentive to reduce the number and quality of their risk checks. Since every “risk check is slowing you down” – it is akin to a car race where helmets are not standardised. It would not be a suprise to see them to get lighter, and smaller. If you are in this game you have no choice to play along – join them or go out of business. This is why car racing has safety standards that cannot be compromised. Someone would notice if drivers got into a car without a helmet - in electronic markets we have no such visibility of the software employed by trading firms. We only find out what was going on after a crash.
As an aside: the idea that “kill switches” (as proposed by ASIC) can prevent any issues created at this scale of time simply does not make any sense. These mechanisms will always arrive too late.
I have developed a "high frequency TCA" methodology/technology that assists buy-side firms (and by extension sell-side) to see deeper inside their execution processes. While traditional TCA takes an aggregate approach to evaluating the performance of long-running trading intentions (parent orders), HF-TCA aims to focus on the quality of execution of each order that hits the market (child orders).
But liquidity is liquidity, right? If you look at child orders, some will win (meaning the price moves favorably after the fill) and some lose. On the whole it will balance out?
Well, that is the point - it doesn't. Trade data shows that uninformed traders (at high-frequency timescales) lose more than they win. This is due to a number of factors such as information leakage and adverse selection. The key to HF-TCA is measuring this effect to help you decide whether you have a problem big enough to solve.
Traditional TCA doesn't cut it on these problems since all buy-side brokers are potentially impacted by predatory strategies. Therefore comparisons among your broker panel will yield no discernible impact; they are all leaking.
In order to perform this type of analysis - more data is required. This data needs to come from the buy-side, in order to do the usual parent-child order analysis, but also to measure the increasing impact of information leakage in interacting with high-speed predatory traders.
Keeping up the momentum, the AFR posted this article on "Call for tax to rein in high-speed trading". Unfortunately for the title, it's already in effect in Australia. This article is based on a pretty good report](http://www.fsc.org.au/downloads/uploaded/Changing%20Technology%20in%20Capital%20Markets_2182.pdf) from Baseline Capital.
FSC chief executive John Brogden rejected any suggestion of an outright ban on high-speed trading of shares, but said regulators needed to keep a tight rein on the sector to prevent market manipulation and negative consequences for investors. “Both dark pools and high-frequency trading have a legitimate role to play in the market and as such we reject any calls for a moratorium or a ban on high-frequency trading in particular,” Mr Brogden said.
Well that is a step forward - a buy-side representative not calling for HFT to be put to the sword!
However, the FSC warned that a small minority of high-speed trading – estimated at about 5 per cent – could be deemed as “toxic or harmful”. This included “front running” of fund managers, where traders see the orders of other investors and use their high-speed computers to place their own trades ahead of these.
The strategy being referred to here is not "front running" (which is the use of non-public client order information to prompt another trade). The (predatory) strategy being referred to here is where an HFT will take public information and work out what someone else is probably going to do. It is employed by many trading firms, just on different (larger) time scales. Why is it suddenly wrong when the hackers do it?
There were also concerns about “quote stuffing”, where high-speed traders flood the market with a large number of orders for trades that are never executed, frustrating genuine investors.
There should be no concerns about quote stuffing in Australia. It has never existed, does not currently exist, and wont - due to the message tax already in place. What is there to be concerned about?
“The activities that really concern us are the predatory type activities where people are trying to observe what is going on in terms of large orders and get ahead of those orders,” Schroder Investment Management Australia chief executive Greg Cooper said.
The problem with this is two fold: (a) you don't need to be a HFT to exploit this strategy, and (b) all sorts of traders have always exploited this possibility. As is currently the case there has always been a variety of traders all trying to second-guess what the guy "one up" from them in terms of time-scales is going to do next. HFT have just created a new category - doing it on even smaller time-scales. The fact that there's a new kid in town, shouldn't switch our mind from the fact that everyone tries to exploit this kind of strategy.
Just like every other type of trader that does not exploit private client information (ie. does not front-run), they can only do it based on public information - as that is all they have access to. There is, however, a need for the buy-side to not be so predictable in the way they trade on the markets - to generate the public signals for HFT to feed from.
The large brokers successfully campaigned to have broker ID's removed in 2005, I think there is a limit to what the market micro-structure should be doing for them. At this point I think the onus is on the buy-side to trade smarter, not try to use taxes to make up for their shortcomings.
“That is the behaviour frankly that as a ‘buy-side’ person we are most actively trying to get away from. Certainly our view of the overseas experience is very negative. And people often look to the US more broadly in financial services as the country at the forefront of technology. In the high-frequency trading world, that is not a place we want to go.
Certainly there is situations overseas in the past where HFT's did front-run (eg. flash orders), but remember these were unintended consequences of reforms designed to protect buy-side. Careful what you wish for.
Under the FSC’s proposal, high-speed traders that make excessive orders for traders – also known as “messages” – that are not executed would be penalised with a new tax.
This might be playing with terminology - but such a tax already exists, as it applies to all messages. If the whole point is to leverage greater penalty on those who send a lot of messages, and don't do a lot of trades - the current system does that. An HFT placing 10 orders and canceling them will pay a tax of 20 messages, whereas a buy-side placing an order that trades will pay for 1 message. That looks to me like a tax that penalises excessive un-executed orders.
The report then goes on to say:
There isn’t really any true evidence we can find in Australia of genuine predatory high-frequency trading.
Ouch. This does seem inconsistent with the FSC's call for an additional tax, but in any case I'd argue I've got plenty of examples of predatory HFT. Having said that the scale of the activity is still very modest, and is very targeted at certain traders. Unfortunately for the FSC, it is their members that are the ones being targeted.
“There are activities that can be identified that look potentially like quote stuffing behaviours, which occur in high frequency trading in the US. “But those things are not really noticed in Australia.”
I'd be interested in knowing more, but nothing I've seen comes close to genuine quote stuffing, since the message rates are nowhere near high enough. I find the ones who come closest are usually two (buy-side) algorithms fighting it out, not HFT's trying to obfuscate the market.
Mr Brogden said any tax on orders should be levied only on those traders that put excessive “messages” through the market, so that it does not affect genuine investors.
The problem with this approach, is that message rates for genuine investors is increasing and will continue to as they employ more algorithmic trading techniques and seek to avoid adverse selection. While the market rules operate on the basis of "the first to respond gets to take the liquidity" - this will always be the case, and increasingly so. It is only a matter of time until real HFT strategies start to be employed by buy-side, just as self-defence against predatory HFT. Over taxing that can be detremental to on-HFT. And before you think: but what will they protect themselves against if the tax gets rid of HFT? Remember that predatory HFT does not require high message rates - it is just one signal that some of its strategies produce. And in Australia they currently don't.
Of course the alternative I propose for all this is to fix the market rules, then the whole argument around latency becomes moot: TimeMatch.
The AFR ran an article on HFT, purporting to report on an increased desire to push more supervision costs onto HFT. The central tenant of the article does not make sense, and perhaps by extension the Treasury report it's based on.
A substantial increase in algorithmic trading including HFT [high-frequency trading] – incorporating not only increased trading volume but increased cancelled orders and messages for each transaction executed – has led to a general increase in ASIC’s workload,
The terms are a bit muddled here - there are two items that ASIC charges the supervision levy. Messages: place, amend and cancel instructions; and Transactions: actual trades on the market. Since "messages" include cancellations, it really should read ".. but increased messages for each transaction executed"
The current cost recovery model was designed in an environment where trades rather than orders were driving most of the staff costs for ASIC’s market supervision operational work,” the regulator says in its report. At that time, detecting and prosecuting market abuse caused by trades was ASIC’s highest priority, although some work did arise from orders – for instance disruptions caused by HFT.
The strange thing is the message-to-transaction ratio has actually decreased (as you'd expect) since the introduction of the supervision levy. I don't think it's a fair representation to say the market has become more onerous to supervise, more-so that you got the model wrong in the first place, and you need to adjust it. I'd just prefer people state the latter rather than invent reasons :)
Aside from the reasoning, I think this is a sensible observation. Personally I'd look to levy on Messages entirely. After all, 2 place Messages are still required (at minimum) to cause a Transaction. Removing the latter from the formula would still "capture" them.
After promoting the merits of increasing the levy costs on high "message-to-transaction" ratio traders, a loophole emerges in terms of market maker discounts:
One option that could be considered would be to offer a reduction in the cost recovery levy, or perhaps even exempt continuous, two way market making activity from message based charges, subject to strict eligibility criteria and ongoing evaluation and measurement that the market maker is meeting its market making obligations.
HFT and market-making are very closely aligned - the former regularly representing themselves as the latter in order to gain cheaper market access from the exchanges. The obligations are not onerous, and there's little downside (other than a bit more market disclosure). This seems to go counter to the prior argument of the paper that high order-to-trade ratio participants need to pay more.
It would seem to me we have a difficulty here that needs to be addressed. ASIC and Treasury: how do you define HFT? The AFR article seems to keep up the inertia on this issue, in a blind-following-the-blind manner:
High-frequency traders use sophisticated computer systems to trade large volumes of stocks in short spaces of time. They rarely hold their positions for more than a few seconds and often flood the market with large numbers of trade orders, or “messages”, that they never execute.
There is little "flooding" in Australia, precisely because of the cost recovery scheme in place (and the topic of the article). The above is an extract of a sensible statement on HFT in the United States but does not at all describe what is happening in Australia.
ASIC gives their view on what HFT is:
HFT now accounts for 25 to 30 per cent of all lit market transactions and the high order-to-trade nature of HFT means it accounts for a substantially higher share of orders and messages than more traditional trading strategies do,
Personally I think this 25-30 percent figure is way over the top, but it all comes down to how you define HFT. It seems to me we are all spinning our wheels until we come to a definition on what HFT actually means. I think ASIC are including many forms of algorithmic trading (and not particularly fast forms) in this figure - and actually including the very people who have automated their trading in order to protect themselves against genuine HFT. It really does miss the point when you label them all "HFT" and place them in the one mental bucket with a note "to be charged more".
For the record, my definition of HFT has been styled in making the term fit the persona. Public opinion has already settled on "HFT" being the "guys with an advantage I can't match". Rather than trying to redefine public opinion, I define HFT as being the fastest form of algorithmic trading, the strategies that only work if you are the fastest. And by fastest, I mean the very fastest - the guys who actually win the race consistently - no matter how fast that is. Thes are the one with the advantage, the one everyone else is trying to protect themselves from, and the guys who sacrifices everything (intelligence, strategy and common-sense) for speed. By definition, 25-30% of the market cannot be the fastest.
The main difference between our definitions of HFT is: high message ratios. I don't think these are required for HFT, and can actually be a useful countermeasure to protect one's self from it. It is precisely the reason that Treasury has talked about providing discounts on the cost recovery to market makers - because fast and regular re-quoting helps you minimise bid-ask spreads in the face of adverse selection costs. I would purport the main reason many HFT commentators (and Treasury's paper) consider high quote-to-trade ratios as being a strong indicator of HFT is actually a misinterpretation. In the US, in order to be the fastest trader in the market, HFT firms have taken on the practice of "quote stuffing". This is where massive message rates of garbage (read: spam) orders are sent into the market in order to slow it down. This allows them to effectively "win" the speed race, by slowing down the competition. These high message rates do not define the activity however, they are just a side-effect in the US market where the regulator does not seem too concerned about noise levels on the electronic markets - not the case in Australia. So if you are looking for high message rates in Australia, you won't find HFT - you'll find other strategies (like market making) that employ it, but in this case for a useful purpose.
(NOTE: Granted the above paragraph is very much an oversimplification of the real-world, and that "high message rates" for HFT are orders of magnitude higher than for any genuine market maker. But here lies the potential for unintended consequences)
[http://adeleferguson.com/ Adele Ferguson] wrote a [http://www.afr.com/Page/Uuid/8167883c-4533-11e2-9c73-817ba4ddc986 not-bad piece] on the removal of broker ID's from the Australian equities market back in 2005.
One aspect of the story she didn't mention was the fact that pre-2005 only brokers could see the ID's (at least in theory). This created information asymmetry, which the brokers did not mind, since their clients would need to constantly ring them to find out the identity of orders they were interested in. Fixing this gap was certainly a positive.
At the time I felt against the "large investment bank" push to ban real-time ID's. But, I appreciated the fact that they were more of a "sitting duck" of "showing their hand" in allowing competitors to see what they were doing in real-time. Personally, I think their self-interest should lead them to act less predictably, but that's a story for another post.
Instead, I advocated (and still do) for choice. Let each broker decide whether their ID is displayed ("leaked") in real-time or not. Not only on a per-broker basis, but on a per-order basis. Giving the decision on a "all-in" basis to the regulator, I thinks makes no sense. ASIC steers well clear of making other less sensitive decisions on behalf of participants, why make this one?
Some brokers want disclosure - for example small cap stock traders want everyone to know they are behind a certain order with the view of getting other clients to contact them directly in order to assemble deals. I guess one wrinkle (and perhaps a reason the ASX is happy blanket with non-disclosure), is that it could use the lit market to advertise for business in dark pools. Not such a raw nerve in 2005.
I am currently doing some work on trade analytics to measure the effect of HFT and smart algos in figuring out what the investment banks are doing. There is a significant amount of leakage here, and that is without real-time broker numbers. Re-introducing them (this time for everyone presumably?) would make it a whole lot worse for them, as the algos would more precisely be able to track their most probable actions. Allowing the broker to choose his disclosure I think balances the dual interests of privacy (we don't want another incentive to move to dark/fragmented venues) and disclosure (some people want their ID broadcast).
I would say, for example, if [http://www.iress.com.au IRESS] wanted to distribute broker ID's for those clients who wanted them broadcast - why should (could?) ASIC stop it? (I suggested this possibility in 2005 and it was met with a "we would not be amused" rebuke). But, disclosure should belong to the order-owner, not to the regulator.
Much of this issue comes back to the question: who owns the data? Do exchanges, or even ASIC have the right to mandate what you //can't// say about your order? I've consistently argued that the "liquidity provider is king": and more control should be given to these participants a-la my TimeMatch proposal. Allow them to control their orders, and they'll be less likely to want to leave and trade in the dark, thereby harming the public benefit via fragmentation. Many of the inefficiencies in the market today are, I would argue, as a result of the lack of control that liquidity providers have over their orders.
Finally, some amusing contradictions:
The argument at the time to support the removal of broker IDs was that the big broking firms felt they were at a disadvantage when trading because it encouraged front running.
Front running is acting on privileged client order information. You can't front-run on public information. Exploiting someone's predictability in the way they trade, is something else entirely.
But since it removed broker numbers from the trading screen, it took away what many traders and investors considered an important piece of information in trying to predict where share prices might go. It can be argued that it spawned the quest for less and less transparency
One man's front running, is another's important piece of predictive information? Two sides of the one coin.
Previously I have proposed TimeMatch, a system where every order is auctioned in a timeframe that the order's owner chooses. This would mitigate the speed advantages enjoyed by the technology firms. It basically means that investors get to choose, when they place an order, how much they value speed of execution verses improved prices. They are not forced to value speed of execution over improved prices, as the current price-time priority rules say. Let's look at another thought experiment: a market modeled on the NBN.
To those not in Australia, the NBN is a new government owned monopoly provider of data services. It will give 90%+ of households fibre, and the rest a mix of fixed wireless and satellite delivery. NBN is wholesale only, so preserves the appearance of competition via value added retailers. It is in the early days of construction, but some towns are already live and operating.
(Just for the record, while I would love to get the NBN (like most, I'm not in the 3 year horizon), I'm not all that happy about the amount of money being spent on it. I think there should be a customer contribution to connectivity, rather than it being "free". I like the way they have done the network extension program, and wish the whole network could be built that way! There should have been some acknowledgement that the NBN really baled out Telstra from the consequences of its crumbling copper network which needed to be replaced. I would have connected the early adopters, but kept the late adopters on copper for a while yet.)
The NBN encapsulates the idea of: do it once, and do it properly. It was devised to avoid the situation where a network builder naturally wants to extract the maximum price from consumers, but the ability for them to slash the price acts as deterrent for anyone else to overbuild and provide direct competition.
Lets think of the equivalent in electronic markets: the NLN. The central problem it could solve is: liquidity fragmentation. By spreading out the liquidity in multiple liquidity venues, you pay higher prices since you need to try harder to get (and often can't get) the best price. Of course technology helps fill this gap (smart order routers), but they often create predictability that predatory HFT firms prey on. It comes with cost.
What this thought experiment is about is the formation of a government-run matching engine, but not an exchange. Instead of private operators all (ASX, Chi-X, Liquidnet etc.) all running their own matching engines, they would use one: the NLN.
Sounds crazy? Probably, but so did the NBN. Never thought that would see the light of day!
The NLN would be:
To facilitate this, liquidity would need to be mandated to go though the NLN (much like in the same way telcos were pushed at gunpoint to deal with the NBN). Broker crossings would be eliminated, and dark pools unable to run their own matching technology. Gone is the need to mandate meaningful price improvement, since in this system a trader can't miss out on a better price. The published NBBO (national best bid & offer) prices would be real - unlike the US where they are for "display purposes only".
It's always been an issue in the industry: when will the ASX bypass the brokers (or vice versa)? At this point I think there is a valid need for the two to remain, since you are only outsourcing the most commoditised aspect of exchanges to the NLN: running a computerised matching engine. Having said that, it is a fact of life that we all need to re-evaluate our contribution to the value chain. Ask Sony, Kodak or Microsoft.
Lets imagine you can only place orders on the NLN with a markets license. Market operators also have the ability to add value (like Telstra on the NBN) and differentiate themselves though:
OK, the ASX would feel it has a lot to lose. But will Telstra actually (despite their initial reactions) end up a stronger organisation post-NBN?
Brokers would no-longer be required to connect to more than one market operator. They would only do so to access the value-add services that they provide. As far as liquidity access is concerned, every market operator now has access to all the liquidity. Hello competition. Now trading via a "dark pool" will now be as safe, as you can't get dudded on price!
Of course the matching engine would be quite different to the ones we have in operation today. Instead of just accepting vanilla orders to buy or sell, market operators would contribute software algorithms - and run those algorithms on the NLN. This would allow innovation, without the latency arms race - since these algorithms would in effect be running at "zero latency". Now everyone has equal access, and everyone is a HFT.
Naturally proposals such as TimeMatch are still relevant in this imaginary world - that would be just one type of smart order that the market operators could design and provide to their broker clients.
The concept of a CLOB (Central Limit Order Book) has been discussed for many years in the US, and is essentially the same idea as the NLN. Unfortunately the US opted for Reg NMS and the rest is, well, history..
It's quite interesting to look at the regulations in play or on the table that would be unnecessary under a NLN scenario:
There is also a whole host of issues that are increasingly costly for market participants that would dissapear:
[http://www.theaustralian.com.au/business/opinion/andrew-main Andrew Main] wrote [http://www.theaustralian.com.au/business/wealth/why-hft-is-exploiting-an-unlevel-playing-field/story-e6frgac6-1226505646823 a piece] in The Aus on HFT, interviewing [http://au.linkedin.com/pub/zachary-may/3/69a/299 Zac May] the author of the [http://www.industrysupernetwork.com/wp-content/uploads/2012/09/140912-ISN-submission-to-ASIC-re-draft-market-integrity-rules-and-guidance-on-automated-trading.pdf Industry Super Network ASIC consultation paper response] I responded to earlier. Zac May says:
High Frequency Trading is the creation and then the exploitation of an unlevel playing field
HFT certainly didn't create the market we currently have. It is human intuition that said that the guy that matches the order first gets the trade (price-time priority). HFT's have just become the most efficient at adapting to the market that already existed. Maybe Zac is talking about quote stuffing (in the US - it doesn't happen here) where HFT's can slow everyone else down with noise to give themselves an advantage. Fair enough - but not relevant to this domestic debate?
HFT certainly uses it's strengths in the marketplace, just like any other market participant - including the people Zac May represents. They are managers of superannuation funds, who naturally use their strength of asset management scale to achieve other relative advantages. Things such as cheap brokerage, not able to be achieved by retail investors; but not something being hailed as an "unfair playing field".
To be clear, "exploitation" is an emotive term - but can be applied to HFT in the same sense it can be applied to anyone who trades for profit. The fact someone exploits an opportunity is hardly a big deal - assuming it is not fraudulent or illegal. The real query comes on the "unlevel playing field". What particular type of unfairness is he talking about here?
Some people are paying ASX to allow them to be as close as possible to where the bread drops, [...]
The direct way to correct this issue would be to refuse to provide market data in any way other than one: the fastest. The ASX would rip out a whole lot of older/slower technology, and say: the only way to get ASX data and trade is to be co-located at the ALC with an ITCH feed.
I can hear the outcry. Brokers (the suppliers to Zac May) would be up in arms that they are forced to change their technology and use super-fast feeds //that they do not want//. No doubt the costs would be passed through to the people Zac May represents. The reason they wouldn't want it, is that they know that the additional speed will not help them. They know that nothing will improve for them in the extra millisecond they gain, since they just don't have the capability to do anything with it.
How is it then, that the ASX is being unfair by providing a variety of services at different price points, using different technology for the customers to choose? The real issue is not that the ASX provide a variety of access mechanisms, but that some customers can move faster than others. Hardly ASX's fault?
You have to ask yourself why ASX provides privileged access to information for some participants. It's not just so they can get the information fast, it's so they can get it ahead of other people.
Exactly what privileged information does the ASX provide to some participants? None. They get the information ahead of other people because they can be bothered going to the pond to get the bread.
Besides, I would say the whole point of getting information fast, is so you can get it ahead of other people?
And that causes other investors to start asking themselves: 'What do they know that I don't?'"
Other investors need to acknowledge that there is a lot that HFT traders know that they don't. Remember HFT's only congregate in the pond to beat each-other: not other investors. They already have the other investors beaten who are waiting for the bread to be delivered to them via courier.
The issue is not that HFT's get a sneak peak at information, it is the arms race that is taking place in the pond. It creates a very fragile market, and one that is prone to chaos - as clusters of fast and stupid algorithms get up to mischief. All this in the knowledge that the ASX will have to bust the trades if things get really out of hand.
But, he said, it should be possible for regulators to eliminate that advantage. Some jurisdictions had introduced a half-second delay on orders, others were proposing random delays, and there was also talk of "sealed bid double auctions" at random intervals to remove the time advantage for HFTs.
One solution that means avoids all the above pitfalls, let the market decide how fast it wants to operate: TimeMatch. That way you don't force the genuine HFT liquidity providers to slow down (extra risk, wider spreads, more cost to Zac's members), you just allow everyone to trade at their own individual pace.
Also in Australia, a small ASIC levy on all orders, not just executed orders, hits the economics of HFT where only about one order in eight is actually executed.
This is already in place - it came in at the beginning of 2012.
Greg Yanco, head of markets at the Australia Securities and Investments Commission, copped criticism for calling for calm over the whole HFT debate but says he doesn't want to call for legislation against the few that might disadvantage the many.
Bravo! Greg continues:
"People see trades going through at best bid or offer, what's called the priority price," Mr Yanco said, "and assume they are HFT when they are actually dark pool crossings." He said that would cease to happen with the probable introduction next year of the Meaningful Price Improvement rule, under which you can only do smaller trades in dark pools at prices that are better than those available in the "lit" market.
Be careful of unintended consequences here. By bringing a rule like this in, you've now got a real concept of a NBBO (national best bid/offer price). Is this NBBO centrally aggregated? Where? By whom? How precise does the measurement of time need to be? How long after the NBBO changes, do you have to report a unimproved dark trade? The speed of light may seem fast but it's not infinite. This could be one step down the US road..
Others include having a mandatory "kill switch" in the offices of market participants doing algorithmic trading of any sort, including HFT.
I don't think this style of kill switch will achieve anything. Any algorithmic trader with any sense will already have one (for self preservation). It's their competence that's in question here - would you trust Knight Capital (and it's competitors) to build their kill switch properly? Kill switches need to be where you can trust them: at the exchange.
[http://www.smh.com.au/business/by/Gareth-Hutchens Gareth Hutchens] posted [http://www.smh.com.au/business/fears-mount-over-ultrafast-trading-20121029-28dn9.html this followup] to the weekend [http://www.smh.com.au/business/sell-and-buy-in-the-blink-of-an-eye-20121026-28b55.html article]. Lets look at the weekend article first:
Not everyone thinks so. Some sections of the financial industry are privately seething that Funke Kupper and others have been whipping up hysteria about high-speed trading. In the bars of Sydney's financial district, the topic is waved away with scorn. ''It's just become a scapegoat,'' is the refrain.
Funke Kupper is both responding to concerns in the public about HFT, and at the same time helping to further those concerns. He is in a difficult position. On the one hand you have people reading about HFT in the US, and incorrectly translating that to Australia. This creates a heightened sense of concern, which fed with the appropriate (low) level of media intelligence on the topic leading him to the conclusion: beat them or join them.
Beating them (read: educating them) is a difficult thing to do. There is certainly an argument that HFT is now being defined as all-things-evil, and the scaremongering is way overboard. If you try this angle you are only going to be painted as a sympathiser who, being paid good revenue by HFT firms, is trying to protect one's own interest. Joining them (read: laying the boot into HFT) is really the only choice you have available, and trying to position the ASX as the advocate of the Aussie Battler is the only way to go. Sorry HFT.
The other reason is, of course, it's the right thing for the ASX's long term business. ASX relies on being a diversified marketplace - that is, having all different types of liquidity - not just the stuff the computers generate. In fact, the market is a bit like any ecosystem - the predators will go wherever the prey are. The ASX would become a "ghost town" if the non-HFT left, as there would be nothing left for predatory HFT to feed off. ASX needs to be very careful the non-HFT liquidity does not go elsewhere. What if participants started feeling that Chi-X (or dark pools?) was a safer place to trade? You can afford to beat-up on HFT (computers don't get offended) as they wont go anywhere. The non-HFT is different: they now have a choice, and much more feelings. If they get skittish and leave - the HFT goes with them. Sorry ASX.
In fact, the ASX tried an experiment on creating a venue with no-one other than HFT on it: it's called [http://www.asx.com.au/tradingservices/asx-purematch.htm PureMatch] and was hailed as a high-performance market that trades faster than [http://www.asx.com.au/tradingservices/newmarketservices.htm TradeMatch]. I've only seen a few tumbleweeds over there.
One issue facing academics is the difficulty in identifying HFT. There are many firms specialising in this kind of work ([www.nanex.net Nanex], [www.hftalerts.com HFTAlerts]) but they are US-centric and focus on certain kinds of HFT activity. The kind of activity they focus on is the kind that happens in the US, but not here in Australia. There are a lot of technical reasons as to why it is difficult to do, but one I am planning to solve (in Australia) shortly. It's not impossible - but currently no-one (other than the HFT firms) have the raw data to do it.
How then, can a chart like this represent numbers of HFT activity on the various markets? Quite simply, they can't - and I think they are way overestimated. In Australia this 30% figure includes a lot of what I would call AT (algorithmic trading) activity that is not high speed. In terms of predatory HFT, I'm thinking sub-5%. There's isn't many papers or clicks in a story telling you to to be //less// afraid.
Unfortunately this image does not acknowledge the source of the top ten "stocks at risk". Evidence please - I'd love to know! I doubt there is real a "top 10" of stocks that predatory HFT targets: they'll go wherever the opportunity exists, and that will change quite quickly!
Funke Kupper believes that this fragmentation is a bad thing: the more the market splinters, he says, the more liquidity will get sucked from the main exchange into smaller off-market trading areas, making it harder and more costly for companies to raise capital. And in an emergency, that could be particularly worrisome.
For an organisation campaigning against market fragmentation, they've sure tried to create a lot of it! Off the back of the monopoly-like TradeMatch market they created PureMatch and VolumeMatch ([https://www.asxonline.com/intradoc-cgi/groups/participantservices/documents/communications/asx035012.pdf RIP]). Separate venues, splintering liquidity - at least Chi-X only has one!
If your mantra is "fragmentation is bad" then we should be looking to achieve competition in ways that don't create it, right? Perhaps we can take a lesson from the NBN - where fragmentation is also bad, and competition (of a kind) achieved.
As they [computerised traders] see it, high-frequency trading has become a scapegoat, an excuse used by brokers to explain why the sharemarket is still a disappointment more than four years after the financial crisis.
So true - the discussion around HFT is confused in terms of who it is talking about (confusion between HFT and AT), plus it is certainly a scapegoat for many things it is not responsible for.
That said, that does not mean that //all// concerns about it are invalid. But perhaps the fact they are being raised is more to do with the need for brokers and traders to blame someone for their problems, rather than the real magnitude of the issues.
''While some say high-frequency trading provides liquidity, I know some very senior bankers that privately describe it as providing only 'phantom liquidity','' he [Greg Medcraft] told the FINSIA conference this month.
While this may be partly true, I would counter with three points:
What we need to do is change the economics of the market so it does not reward the generation of phantom liquidity. (see TimeMatch)
The article then focuses on some ASIC submissions by some buy-side views like Zac May from [http://www.industrysupernetwork.com/ ISN]:
As he explains it, big firms with the most money are able to see market information before anyone else, given the speeds with which they operate. They then act upon that information, buying and selling shares before the normal retail investor. He says the fact that the ASX facilitates this means the market is inherently unfair.
The ASX facilitates a number of different channels for market data: ranging from co-location (the fastest), broker gateways (still fast), consumer websites (fast), and the newspaper. The fact the ASX facilitates these range of speeds is not unfair. The speed is more to do with the recipient of the information, and how fast they process it, since the fastest information is available to everyone on an equal access basis.
What is unfair, is the fact that the rules of the market hand an advantage to those who get the data first. If slower traders prefer better prices over speed - why not let them? Why force them into a speed race that they don't want? (see TimeMatch)
The new survey of more than 1750 market participants - including retail investors, brokers, fund managers, and analysts - took place from September 5 to October 7. It found nearly all respondents (96.6 per cent) believed algorithm trading and HFT were having a negative effect on the ASX, with nearly 90 per cent of respondents believing they were hurting the ability of the ASX to conduct a fair and transparent market.
HFT has a PR problem, for sure. And by extension (association?) ASX does too - the only difference now is we now know the sheer scale of it: 97%!
My only surprise is where they found the other 3%!
It also found that 78 per cent of respondents either agreed or strongly agreed that algorithm trading and HFT had made them reluctant to trade on the ASX.
The only doubt I have about these figures is the motivational gaming involved: I do sense there is a inclination for traders to 'blame' AT/HFT for their woes. I have not found that people in general have any appreciation for the actual risks they run with HFT, or any knowledge about how HFT operates. Therefore, I would say while this does confirm there is a PR problem, it does not confirm that the actual technical problems are anywhere near this magnitude.
While I think TimeMatch and other reforms could solve the technical problems, and possibly the PR - I am concerned that solutions more optimised for PR (eg. taxes, minimum resting times, cutting the pipes, burning HFT at the stake) could achieve the PR outcomes more effectively. Hmm..
When asked about the best way to control the impact on the ASX of algorithmic trading and HFT, most respondents (57 per cent) said the best thing to do was to ban them. Placing a minimum-order limit was the next most-cited response.
Oops - I spoke to soon. These punters do not yet realise that the a minimum-resting time would do nothing to curb HFT in Australia. As I've blogged about previously, it would most probably make things better for the really fast HFT to pick-off the not-so-fast algorithmic traders. This could even enhance the arms race, following in the footsteps of the US SEC who have successfully "reformed" the market into a more profitable environment for HFT on many rounds.
However, retail investors were much more likely to feel reluctant to trade on the ASX, with 85 per cent of retail investors either agreeing or strongly agreeing that they were now reluctant to trade.
It would be an interesting experiment (along the lines of a drug trial) to implement TimeMatch on a couple of securities and test the PR and technical outcomes. No taxes, no restrictions, no intervention. Any takers?
I think the resulting data would give more insight into the real issues than a survey.
When talking about HFT, it’s important we get this problem in perspective: predatory HFT is not to blame for taking that stock you bought for $2, ending up at 20c a few months later. I was reminded of this by a mate (who knows retail traders very well) who said HFT is a long way from the biggest problem faced by retail traders. Lack of knowledge and information that comes through established private networks would be higher on this list. Predatory HFT is akin to a “rounding” loss for retail traders, and only directly exposes you when you trade, not hold, securities.
Nonetheless, a rounding loss for an individual, when aggregated, can be substantial enough to worry about when it comes to the entire market. We just need to keep it in perspective and not blame all manner of things on HFT. My current concern is that the term “high frequency trading” is being defined as “everything about the market I don’t like”. Not a definition that’s going to help in identifying and solving the issues. I do feel there is a predatory class of computerised traders that are not improving liquidity or providing any value to the marketplace. But this is only a subset of the traders that are currently being scrutinised. It therefore makes sense that we look to understand it, before making wild assertions about what the regulators “should do about it”.
I’ve been on holidays for the last month touring around country NSW and southern Victoria, and reading the flow of press articles about HFT en route. What started out as a dribble has now turned into a torrent. There is a confusion, as most of the content published on HFT is not based on any knowledge – but rather prejudice and fear. It is deeply polarised, though the ratio of pro v anti-HFT rhetoric has shifted heavily from the former to the latter. HFT now has a PR problem, being found guilty in the court of public opinion; we are just not sure what the charges are.
On the ASX today we have had an explosion of algorithmic traders (“AT”). The fastest set of these AT’s we’ll refer to as high frequency traders (“HFT”). Everyone is getting these two confused. They are quite different animals and we need to understand the difference in their objectives, and operation. Optiver is often quoted in the press as the canonical instance of an HFT firm; however I would argue it is not an HFT firm at all. They are an AT for sure, being in the business of highly automated market-making and using sophisticated mathematical algorithms. They also seem very fast to a human (I’d guess a reaction speed of 0.0001 seconds), but that is actually not fast enough (in my view) to be defined as HFT. For mine, the essence of HFT is all about being at the front of the speed rate – whatever speed that requires. Quite simply, Optiver are not the fastest guys in the room – their strategies rely on sophisticated intelligent algorithms that are just too slow to win the outright speed race. I would argue the current benchmark for an HFT firm in Australia is to be able to trade faster than 0.00001 seconds. That may not seem like much difference to you – but in human terms think of Usain Bolt racing someone who can only manage to run the 100M in a minute or two; not much of a race!
The real risk in the current debate on “HFT” is that we confuse HFT with AT. What makes HFT dangerous is exactly the thing that separates it from other AT’s: speed. To achieve speed, you need to sacrifice smarts. Winning the speed race (remember there is no silver medal in this financial race) involves giving up everything to achieve it. That includes flexibility, sanity checks, and verification steps. The kinds of things that help prevent “run-away algorithms” such as Knight Capital. Speed is a cruel master, and requires complete devotion. What you end up with is an algorithm that is the fastest in the market, but is also very, very, fragile. By rewarding those in the market who are the fastest (through the price-time priority market rules) you are essentially encouraging behaviour that leads to volatility.
Think of a group of HFT firms – who are all in the running to be the fastest on any given trade: each has only the very minimal intelligence programmed into their algorithm; otherwise they wouldn’t be in the race in the first place. When these algorithms start triggering one another, they can create massive price volatility; the kind of volatility that is now a daily occurrence on the US markets. These “flash spikes” form a nuisance for retail traders in triggering stop losses, trading halts, and the risks associated with partial cancellation by the exchange. But I think the larger issue is that of perception integrity: events such do not promote the perception of a safe place to invest.
In short, AT is not the enemy when it comes to this kind of fast volatility. We need to make sure that non-HFT’s don’t get wiped out as collateral damage because they seem to match an uninformed picture of “HFT”.
The root of the problem is not the existence of HFT firms: they have merely adapted to the environment to maximise their profits. This is to be expected – in any other area we would describe this as “success”. The problem is the environment: a marketplace that rewards this behaviour in the first place. Fix the environment, and the marketplace will take care of HFT itself.
Price-time priority says that the “first in wins” when matching against others in the order book. It is where the advantage of speed is derived. This doctrine has outlived its usefulness, and it’s now time to re-balance the market rules since there is now a massive variation in the speed of traders. This is where I propose [[[blog:timematch-a-slower-order-book |TimeMatch]]]: an alternative set of market rules where every order is auctioned to the trader that is willing to give the best price. This way, technology works for the benefit of the marketplace – rather than against it. Without getting into a too much detail, retail traders would not notice anything other than they occasionally get price improvement on their orders. That is, they get a better price than they asked for – when their orders are caught up in fast price movements.
I feel this is a more natural solution to the issues surrounding the increased (now ridiculous) speed of trading technology – without harming all AT’s. I argue that the other proposals put forth in the public debate are ill-conceived knee-jerk reactions, without proper consideration of the unintended consequences:
There has been some interesting discussion on HFT from significant players in the ASX broking community.
Firstly, UBS got into the debate by calling on HFT to be curbed:
“There are some things that we support which aren’t necessarily great for our business, but we think they are actually right for the market structure,”
UBS is implying that HFT is good for their business, but they'll play the honesty card and admit it's not great for the market as a whole. I'd say their share of HFT business is somewhat underweight, and therefore see a competitive advantage in laying into HFT in the hope that it will be curbed.
They generously offered to offload the ASIC supervision levy onto HFT's:
Under the UBS plan, certain trades that do follow ASIC-approved guidelines would be charged a 1.7¢ fee on every message sent to market. The money raised would replace a $27 million levy that ASIC charges stockbrokers and the stock exchange to pay for supervision of the market. High-frequency traders that act as market makers and agree to ASIC’s rules would be exempt from the fees. Rules might include forcing traders to leave orders in the market for a minimum of between half and one second, and placing limits on the minimum size and value of trades.
Nice try, but it's not that simple. If you were to target firms that do not leave orders in the depth for long, you're not going to find much on the ASX. The lifetime of orders in the depth has not changed markedly in recent years because, well, if you were a HFT firm you would realise while these strategies (quote stuffing) are used in the US - they aren't here.
What you would find is that predatory HFT in Australia, being predominantly liquidity taking, would not fit the profile of what is being described - and therefore not pay anything extra. Here is a common problem with the "torch the bastards" reform approach: HFT is hard to define, and often does not even match your stereotype.
What if no-one sent these orders - who would pay the supervision levy then?
Similarly CBA got into the discussion with a number of pitches to similarly push their corporate preference: leave HFT alone.
There is a contradiction here:
“If HFT is actually providing liquidity and not distorting the market, why does it matter?”
If that is true, then it does not. But that "if" is not given for all of HFT.
“Our view is that it’s somewhat disingenuous to say that liquidity pools are bad because of HFTs, as some liquidity pools don’t have any HFTs.
I think the case against dark pools (lead by the ASX) is in relation to lack of regulation/licensing and liquidity fragmentation. HFT participation is only a potential additional concern, not a core reason for being opposed to it.
HFTs and liquidity pools are different concepts. CBA won’t allow any HFTs into its liquidity pool,” MacGregor says.
So if the implication is true that HFT is good (liquidity, not distorting, all that) then why does CBA not allow them into its liquidity pool?
“This enables execution away from predatory HFT funds...
Right - so predatory HFT is bad, but you think that is OK in the public market, but won't let it into your private pool?
Sounds like you're happy that the lit market has become more dangerous (real or perceived), bolstering the benefit of your own internal matching that doesn't let the bad guys in?
It does seem your views on all this are very proprietary and, like UBS, designed to bolster your own interests. We should view them in that context.
yet also provides the potential for better prices and quicker execution for retail investors.”
I am hoping this is a mis-(non?)quote. Better prices yes, quicker execution than HFT? Hmm.
This 3-on-1 interview was too good to refuse responding too!
RG: Elmer, this is far more serious than what you’re saying. What we’ve got looking at here is a totally immoral practice where a series of inside traders learn what’s happening in the market... As an ordinary retail investor, either for a fund or for themselves, as their orders come on, these guys slip in and get a better price – either the in or the out. It’s totally immoral, it’s wrong and it does the stock market no credit to Australia and you can see when you have immoral practices like that what can happen overseas.
What is 'immoral practice' on electronic markets usually depends on which side of a trade you sit. The large long-term institutional investor sees the medium-term trader as immoral, who sees the daytrader scalper as immoral. Each is trying to get in/out of stocks in advance of the other, and ride the created price change. Each is trying to eat the other's lunch.
Now, everyone sees insider traders as immoral - and for good reason. But are you really saying that HFT is insider trading? Are you serious? What 'inside information' are they trading on? If you are referring to the 'advance notice' of market data they receive, well the only reason that all firms are not receiving and acting on this data at the same time is they are not capable of it technologically. Should the investor that only gets prices from the newspaper complain that there are "insiders" out there using intra-day prices with newfangled gadgets?
You seem to be confusing this scenario with the US practice of flash-orders, which I would argue could be a form of inside/front running (since the poor retail client has no control of the disclosure and the fact they are about to being re-routed).
RG: Now, I think there’s only [one] solution, there’s only one moral solution and that is that the high frequency pipes should be cut and we have proper trading. And unless you do cut them, you will have dark pools.
How would you cut the pipes? Presumably you'd determine some arbitrary ratio or indicator that distinguishes the good guys from the bad, and put the bad guys out of business? As I've said previously:
The idea of regulating HFT out of existence disturbs me: how do you decide what is HFT? It is kind of like trying to get rid of criminal bikie gangs: do you define them by the fact they ride bikes, the number of tattoos they have, or the kind of beard? Perhaps you could invent some ratios, and research the correlation between these ratios and criminal activity. What you end up doing is making certain combinations of bikes, tattoos and beards illegal - not the activity itself.
Also, to think that the HFT's will just lie down and die? They will adapt to any prescribed ratios or rules - and if overseas experience is anything to go by they will then turn these new regulations to their advantage. The regulations requiring re-routing where a better price for an order is known on another exchange, or the ability for exchanges to provide price improvements from non-lot pools are two examples of this.
EFK: Well, that’s an extreme and almost medieval solution..
RG: No, no, no. It’s the correct solution. It’s the moral solution. It’s the right solution.
It is one potential solution - I think there are other, and much better, solutions.
RG: I think that we now have in Australia a situation where people are losing faith in the equity market, as they’re doing in the States too, [...]
I actually think the scaremongering about HFT in the press is doing a lot more to destroy the faith in the equity market, than predatory HFT itself is. I don't dispute for a second the fact that predatory HFT exists, but we need to get the scale of the problem right - otherwise the medicine may be worse than the disease.
and the first thing that you have to do is to make sure that they get a free and fair access to the market as they used to have and they no longer have
We have free and fair access - it's just that HFT's are doing more with this access than others are. The root problem is that the market rules (price-time priority) favours those who can move the fastest. We should start there, not trying to invent some new rules to specifically kill them off.
and the reason we’ve got dark pools is because of high frequency trading. The way to stop dark pools, which I can see the problem with them, is to stop high frequency trading.
I agree that dark pools are the 'canary in the cage' in that they indicate there is a problem. I would argue that HFT is not the 'problem' people are trying to avoid by going dark, however, they are just a symptom. The problem is the market rules favour those with speed - and dictate that 'first in, wins'. This works OK on human time scales (we are happy to compete with other humans on speed) but now everything has gone electronic, it confers an unfair advantage to those with ultra high speed. This advantage computer-against-computer is even more amplified than computer-against-human, as programmers do some dangerous things in order to squeeze the last microsecond of speed out. These compromises create some pretty stupid algorithms, that create what the boffins call some dangerous non-linearity. That's geek speak for wild market weather.
RG: Should we change the regulator so we get a tougher regulation to stop it, if you can't do it?
Is another set of bureaucrats going to do a better job?
EFK: I think we’re blessed with very strong regulators here who actually have a very deep understanding of this. But one of the things you have to recognise here is that we’ve changed the market structure and I think if we’d thought about it a little bit longer, two or three years ago, we may not have done that. But we’ve done it; it’s not going to go away. So, we now have to deal with the consequences of that and it’s a very important point right now.
The market microstructure appears to have changed as a result of change in the participant activity - not the market rules themselves. HFT has logically worked out that the market rules reward speed. These issues could all be solved without ASIC even getting involved. If the ASX provided more control to liquidity providers (minimum execution size, TimeMatch, exchange-side algorithms for starters) a lot of these issues would disappear as soon as the buy-side reacted to the new found tools.
These firms getting 'picked off' by HFT are big boys, and should be able to look after themselves. I mean, seriously: even retail traders have to go via a broker - why don't the retail online brokers in Australia advocate for their clients and provide tools to assist them? Instead of innovating in the market, why are we instead getting the lazy response of complaining that the regulator should restore their profitability on their behalf!
RG: And how did we change the market structure? Where did we go wrong? EFK: Well, we allowed fragmentation to proliferate, both with the introduction of another exchange...
Well, this may serve the corporate script of ASX to pin the blame on Chi-X for this - but the issue is more to do with the participants getting faster and being rewarded by the previously benign price-time priority rules with an advantage over the rest of the market. ASX and Chi-X have both (quote reasonably) assisted these firms, with providing ever-faster market engines. What I see is a lack of response from other side of the market; where are the non-HFT's asking for more control over their liquidity? A market that serves their purpose. They are too busy running to the regulator and complaining about the situation - I think an action that demonstrates their collective lack of understanding of what is actually the root cause of the problem. It is as if the non-HFT's believe that the 'HFT ripoff' is an inevitable part of electronic markets - and the only option is to regulate them out of existence.
and I’m not saying competition is bad by the way, what I’m saying is the consequences of fragmentation and the trading models that that allows to create can be damaging unless you manage it well. So, unfortunately whenever you change these structures, there are unintended consequences.
The inevitable increase in market fragmentation caused by competition (read: Chi-X) is only a small part of the 'nutrition' provided to HFT's. The far larger benefit is conferred to them by ASX's massive increases in matching engine speed. It is very hipocritical of the ASX to argue against competition in the name of fragmentation, given that the ASX have themselves a number of (fragmented) markets. Chi-X only have one.
In my mind, the fragmentation consequences of having two order books in Australia is an acceptable compromise in order to achieve competition. The ASX should practice what they preach and rationalise back to one order book; end their own fragmentation experiment.
While I think two order books are justifiable, a dozen or more isn't: having a plethora of dark pools is a major negative - and I'll be watching to see if their market share grows as an indicator on the perceived health of the overall market.
AK: But Elmer, the fragmentation you talk about here, but more especially in the US, allows high frequency arbitrage, right? I would have thought there’s nothing wrong with that? [...]
At the end of the day it increases the cost to the non-arbitraging traders.
[...] The problem of front running, which Bob is on about, surely is not caused, not brought about, by the introduction of another exchange?
Exactly right Alan. But then again there is no HFT front-running going on in Australia anyway.. :)
Perhaps the concept of 'front-running' has come from the practice of algos that 'sniff out' orders that are being executed by large institutions? After identifying what the institution is doing (by analysing public data - it is not front-running) they then jump in and add the to the cost paid by the institution. This kind of not-really-front-running needs to be recognised:
Of course dark pools cater to this type of problem, but I think not enough work has been put into solving this problem in the lit market. The exchanges could help with allowing them more control over their liquidity.
EFK: Yeah. So, I don’t confuse... There’s a connection between the two because fund managers tell us they’re going 'dark' because they don’t want to be in a place where they run into high frequency traders that front run.
Fund managers would say that; what I think they really mean is that they only know how to behave in a predictable enough manner, that the HFT's can always work out, in advance, what they're about to do. Sorry, that's not front running.
EFK: [...] One of the reasons there’s fragmentation is something we call internalisation. So, this is retail brokers matching trades within their own engines as opposed to passing them through the exchange;
Yes, this is a huge issue - and a 'market-within-every-broker' that has always existed. I can't help thinking, however, one of things making internalisation so attractive is that the broker has complete control over their own orders that get internalised (can't meet a HFT in there). Perhaps if ASX provided more control to liquidity providers, there would be less incentive to internalise everything they can. Of course the other reason is that ASX fees are lower on internalised trades, so there really is an economic incentive to do so.
EFK: Cutting the pipes is not the answer. SB: Pricing might be. EFK: Pricing is. RG: Or pricing the pipes?
RG: pricing the pipes higher will only embed the perception (and reality?) of conspiracy between HFT firms and the ASX. And who can blame them - they are there for shareholder return! If the ASX were to drive up the price of HFT data feeds, I can see the media crying: "ASX sells front-running feed to the even more select few!".
EFK: Well, you could argue that... Well, why don’t we just say if a retail broker internalises trades ‘in the dark’, just oversee them the way you oversee an exchange, you know.
I think it's a great idea to increase obligations on a dark pool / internalisation. Add some obligations that the exchanges face - that would be fairer.
EFK: So, the best market place, the most efficient market place, is the place where everybody gets together because that’s what the finance market tells us and we believe that’s true. AK: You just want your monopoly back.
Of course he does AK, but that's not going to happen!
As a wild thought experiment: imagine an NBN-style solution to this. This would give the ultimate solution from a 'minimise the spread by concentrating the liquidity in one place'-perspective.
ASIC would (presumably) run a single order book, on which Chi-X and ASX (and other newly regulated dark pools?) would operate. If EFK were serious about his desire for liquidity concentration (and not the self-interest of the ASX), then he would possibly advocate for this? Unfortunately for ASX, though, it puts the ASX competitors on equal footing with the ASX itself - can't have that!
Maybe it's not such a crazy idea (well, no more crazy than the NBN). I guess the exchanges would become super-brokers, and in a sense you end up making ASIC the new ASX. Just like the NBN, exchanges could no longer compete on technology (since the one order book would be used for all), however I think that ship has sailed. No-one other than predatory traders want markets to get faster from here. My view is that a matching engine is a commodity utility now: just like power or telecommunications.
[note: this NBN experiment has been explored further]
RG: You’re the one that’s doing the wrong thing. You have got the wrong... You’ve put these things in.
The market rules have (globally) become unfair to non-HFT traders, but this happened slowly. The ASX (or Chi-X for that matter) did not create HFT or encourage it in any way. HFT's responded to the price-time priority rules in Australia, the same way they have over the world: they tuned themselves to work the most profitable way within them. They literally are just hackers in the market, or in another language, unencumbered innovators.
What we actually need is for the ASX/ASIC/Chi-X to look at this situation and say: technology has changed the markets. The markets now need to fight back and adapt to those changes thrust upon it. Pushing the debate back onto ASX as to why they created this problem misses the point - they are reactionary in this regard (and I'd argue in general) not proactive (that's what I want them to be).
AK: Oh, are you saying that CHI-X can’t exist without high frequency trading? EFK: I suspect that model is more dependent on it than our model. In fact multiple exchanges really to some degree depend on that kind of model.
I'm not so sure about that. What I am sure about, however, is that if Chi-X provides a venue that allows liquidity providers more control over their orders, than ASX does, then (regardless of current reliance on HFT revenue) business will head their way. Investors will see that Chi-X provide a safer environment to transact, with less exposure to HFT predators. Of course, that is up to the competitive tension - hopefully this will now lead to some innovation.
AK: So, that’s an interesting statement. I mean so are you saying that the stock exchange competition actually relies on high frequency trading in order to have competition?
I think this is kind of true, if you make the assumption that the market rules can never change - ie. we are stuck with the doctrine of price-time priority forever. If you are able to question that doctrine however, I think there is a whole world of other possibilities..
In reviewing a number of the recent articles, I've outlined some proposed solutions to some problems created by ultra-high-speed HFT. Here is a summary of them all in one place. Unfortunately we cannot solve all problems at once - the ones I am trying to address here are:
The solutions have a simple theme: execution venues should allow liquidity providers to have more control over their orders. In this vein, the exchanges would implement:
The execution venues would use a TimeMatch-based order book (instead of the current price-time priority), allowing the liquidity provider to choose the time-frame over which a short auction auction will operate every time a trade is matched.
This allows the liquidity provider can specify the minimum size that each parcel of their order can execute in. eg. this would prevent 1-share orders from executing against them. Whilst this is beneficial to retail traders who then get stuck with a (expensive) single share, it also prevents algorithms from gaining insight into what others are doing.
Note that this allows 'all-or-nothing' orders, by specifying the minimum execution size as the size of the entire order.
A lot of the asymmetry in algorithmic tools lies in the fact that achieving low latency is an capital intensive exercise. Constant refinement, upgrades and improvements are required to keep an algorithm running at top speed - since all your competitors are constantly improving as well. There is always a select few in the market that have the best, at any given time, implementation of a particular type of algorithm. By 'best' I guess I am really saying, fastest - so we are assuming these algorithms are fairly simple and stupid beasts - and largely compete on their speed. Note that this competition does not benefit anyone (but themselves), so there is no commensurate improvement in market quality as a result of this battle.
Allow lit and dark liquidity to co-exist, don't banish the dark stuff to the dark pools! This is very controversial, and I need to do some more writing on it - however the liquidity provider is king. If he doesn't get what he wants, he goes somewhere else (read: fragmentation). He can get darkness somewhere, just at further cost to the market, so give it to him in the central market. More on this later..
One way to 'level this playing field' is to provide a perfect (from a speed point of view) implementation to everyone, on an equal access basis. This can be achieved by implementing algorithms within the exchange, and allowing trading firms to place these order types - which then get run within the exchange computers. This is not, and does not require co-location by the firm. It is simply a matter of a firm, instead of specifying a simple limit or market order, sending an algorithmic order. The exchange's computer then runs the algorithm, creating whatever limit and/or market orders result from the particular strategy.
This capability would have three key impacts:
The sorts of basic algorithms that could be useful:
Of course all these algorithms would need to come with a suite of features to prevent themselves from being 'sniffed out' by other algorithms, since the capabilities of these algorithms would be widely known. Things such as: time/volume randomisations and thresholds would allow them to avoid detection - by ensuring they behave slightly differently in every instance. All algorithms would be dark - obviously you don't want information being leaked on stop orders, or what underlying strategies are being used during execution.
Extra effort would be needed (over and above what exchanges usually provide) to also make these algorithms very safe. Given that these will never need to compete on speed (operating within the exchange's computers), it makes it a very good environment to implement a lot of intelligence to prevent the kind of non-linearity that currently causes massive nonsensical spikes in the market.
Would it be totally crazy to suggest the idea of firms being able to create their own algorithms, to run on the exchange's servers? Now, there are many areas of concern here, but stick with me!
One impact this would have (if implemented correctly) would be full diversity of algorithmic activity, ensuring you have a good spread of behaviour in the market - which leads to lower volatility. This would allow equal-access to algorithmic tools to all participants, meaning no technology war in order to be the fastest. Every firm would now be 'the fastest' and would now compete on the intelligence of algorithms rather than the outright speed. Speed is dangerous because it leads to poor and fragile behaviour - as safety and good strategy costs in performance.
For those worrying about algos seriously screwing with the markets in this scenario, well what can I say but that's a valid concern! But it may be possible to manage the risks. The algorithms would obviously be heavily 'sand boxed' and be prevented from impacting on the performance of the exchange itself beyond reasonable limits. I don't think it's impossible, though, to implement this in a workable fashion: but your quality control on this would be extremely high. The up-side is that because it's fully visible to the exchange you actually have the ability to control the process, rather than trust all the "Knight Capital"-esq firms: that they can do their software deployments correctly.
[to be continued..]
Business Spectator ran an article outlining the latest fear instalment on HFT in Australia.
These traders have been allowed to have a pipe into the major markets to give them an advantage over legitimate stock buyers and sellers. Business Spectator has led the world in highlighting the insider advantages these traders have secured (Getting the jump on high-frequency trading, July 18).
The referenced article makes some fundamentally misleading statements on HFT in Australia. It talks about an "insider trading network" operating in Australia by HFT. What nonsense.
Insider trading is trading for your own benefit based on non-public information. Whilst this has arguably been true in the US, with flash orders and a complex market structure, it has never been so in Australia. To imply that it has been in Australia is either running shot-gun over the facts, or deliberate scaremongering.
Put simply, there are no avenues for obtaining non-public information from the ASX by HFT firms. While some will argue that the price feeds they obtain (technically known as "ITCH") are faster than other feeds, you should remember that these technologies are available on a equal access basis at very reasonable cost. The advantage that HFT firms have, is in their technology to process the data - not obtain it before everyone else. The fact that HFT firms colocate is not, in fact, to beat non-HFT traders to the punch (they do this without raising a sweat); it's to beat other HFT firms.
HFT firms just do things quicker than retail traders, but this does not make them 'insider traders'. It just is a fact that the market rules (price-time priority) reward this kind of behaviour. Maybe we should direct our attention there, rather than conjuring up visions of queue-jumping cheaters? TimeMatch is a proposed solution that addresses the market rules, rather than just making illegal the most successful traders under the current ones.
The New York Times reports last night that global regulators are trying to stamp out the practice with new regulations. And of course because the large genuine institutions do not like being ripped off by the high frequency traders, and their pipes into the market, legitimate people have left the large exchanges. They have started what are called dark pools, where effectively legitimate institutions trade among themselves. It’s not ideal but better than the stock exchange system.
'Genuine institutions' (non-HFT) is right to be concerned about the impact of HFT in the US, and to a lesser extent, Australia.
Contrary to popular opinion, there is not an exodus to dark pools in Australia. The share of the lit market recent years has actually increased (marginally). What we've seen is a shift from broker crossings to dark pools. This actually is quite a positive indicator for the health of the ASX, but one that seems to be missed on the doomsayers. Unfortunately news telling you the sky is not falling in, does not attract as many readers. Any 'exodus' to dark pools has been from the broker crossings only.
One part of this story that gets neglected is that in the lit market there has been a shift (exodus may be too strong a word) from the continuous market to the opening and closing auctions. In my mind this reflects a very rational response to the continuous market. During the auctions there is less HFT gaming going on (not zero, but less) and you are less exposed to latency arbitrage since HFT's do not control the timing of execution - the exchange does.
Saying dark pools are not ideal is an understatement. They have negligable regulatory oversight, and don't have anywhere near the obligations that markets operators have to be fair to all participants. As I've previously said, the fact that someone is willing to use them indicates there is something wrong with the lit market. My hypothesis is that liquidity creators do not have enough control of their orders on the ASX, leading some to opt for other off-market possibilities such as broker crossings or dark pools. Broker crossings have no exposure to HFT, since the one broker has both sides of the trade, and dark pools often have buy-side friendly HFT countermeasures in them to prevent predatory behaviour.
The CEO of the ASX Elmer Funke Kupper in in his KGB Interview agreed that if the regulators failed in these current efforts, then cutting the pipes were the only solution.
He hardly agreed - this suggestion was proposed and EFK rejected it:
EFK: And so, what we need to do is rather than cut the pipes, which I think is not going to happen...
EFK: ...is make sure that – I understand your opinion, but I don’t think that that’s where we’ll go – is to make sure that the economic incentives of the behaviour are more aligned with the market place...
The above reading that EFK was proposing the 'pipes be cut' looks pretty deceptive..
My view is that the high frequency traders are making so much money and the regulators are so sleepy that the new regulations will fail.
Well, sleepy is pretty emotive: but yes keeping up with the impact technological innovation is difficult for a regulator. There is a lot of money in HFT so no doubt they attract a lot of talent to keep one step ahead.
I agree with the overall sentiment in that the regulators will find it tough: I actually think regulation is not necessary. The regulators have a lot of influence (with the threat of regulation) that I think can steer things in the right direction, but I think legislation is the last resort. A scheme like TimeMatch is a good market-based response to HFT to help, and a number of other innovations at the exchange level could level the playing field for all.
The idea of regulating HFT out of existence disturbs me: how do you decide what is HFT? It is kind of like trying to get rid of criminal bikie gangs: do you define them by the fact they ride bikes, the number of tattoos they have, or the kind of beard? Perhaps you could invent some ratios, and research the correlation between these ratios and criminal activity. What you end up doing is making certain combinations of bikes, tattoos and beards illegal - not the activity itself. There's bound to be some collateral damage there :)
The same thing happens with regulators trying to expunge predatory HFT: except you look at things like quote-to-trade ratios as the target (as Germany has done). It's a strategy of last resort - something you do to look busy, not something you actually expect to work.
Allowing the market to regulate itself (by exchanges providing greater control to liquidity providers) is a far more flexible solution, as it allows the market to adapt to new threats. As HFT change their strategy, the market can counterbalance the pressure far better than regulators can.
I hope I am wrong. There is only one way around the problem created by the pipes that the legalised insider traders have into the market – the pipes must be severed and the markets restored to fair places to do business.
How exactly would you 'sever the pipes'? Presumably the ASX would be prevented from providing trading services to HFT firms - so how would you define those HFT firms? Invent a whole lot of ratios, and pretend that the regulator knows best in picking the good guys (who get to remain in business) from the bad guys (who are out of luck). This is the root of the problem - you don't understand what is going on. You see symptoms, and you want to attack them directly. This is futile, as the root cause of the issue goes unresolved.
Removing all high speed (in human terms) trading firms would be a net loss for the health of the market. What we need is a way to help the good HFT, and curb (read: put of out business) the bad ones.
But in the US regulators are so bad that they seem corrupt. Next week the US Securities and Exchange Commission is hosting a round table on the topic but the agency has not proposed any major new rules this year. The conference is simply window dressing.
Agree. The regulation in the US has added to the advantage that fast HFT holds over the wider market.
In contrast, the German government this week advanced legislation that would, among other things, force high speed trading firms to register with the government and limit their ability to rapidly place and cancel orders – one of the central strategies used by the firms to take advantage of small changes in the price of stocks.
This reform would make a world of difference in the US, but none in Australia. There is no one-size-fits-all rule here. The answer lies in: control. At the moment the market is controlled (in the short-term) by those who are the fastest, since that is what the current market rules dictate (price-time priority). Change those rules (TimeMatch) and the control shifts: to the liquidity providers - of both HFT and non-HFT flavour.
The New York Times reports that in Australia, the top securities regulator recently stated its intention of bringing computer-driven trading firms under stricter supervision and forcing them to conduct stress testing, to protect “against the type of disruption we have seen recently in other markets.”
ASIC's proposals I think are fairly unimaginative and unless they can operate at the microsecond level (read: impossible) wont prevent the next market break. ASIC need to stop asking everyone to "do a better job" (stress test this, please put in kill switches etc.) and influence some real reform at the exchange level. At the end of the day you can trust the operation of the exchanges, you cant trust the firms behind it.
From another perspective it always seems that regulators are perpetually preparing themselves for the previous catastrophe, not looking forward to the next evolving one.
However the broadest and fastest changes have come out of Canada, where this spring regulators began increasing the fees charged to firms that flood the market with orders. The research and trading firm ITG found that the change had already made trading more efficient by reducing the crush of data burdening the market’s computer systems.
ASIC have already acted in this direction with the 'message tax' - and has reduced the quote-to-trade ratio on it's own. Not that the burden of data was high in global terms prior in any case.
It's good to see that the regulators acting but I just don’t think they are smart enough.
Their job is to create an environment where the smartest people can act for themselves, not create an environment where they potentially can't.
A friend sent me a link to the Industy Super Network's response to ASIC's Consultation Paper No. 184: Australian market structure: Draft market integrity rules and guidance on automated trading. Long title, but not a bad document. Here's my thoughts:
ASIC’s attention to High Frequency Trading (HFT), in particular, is appreciated. As discussed in more detail in our comments below, we encourage ASIC to carefully explore the potentially significant risks that HFT presents to market fairness and allocative efficiency, in addition to ASIC’s current emphasis on market integrity risks. A thorough and public assessment of these issues would be an important part of restoring confidence in the lit markets, which has been undermined by HFT.
There's a few points of confusion here. Firstly, it is almost assumed in this paper that HFT is bad for the market. I think it side-steps a few questions:
Now the paper does state some kind of definition:
HFT is, in our view, a subcategory of automated trading whose strategies rely on the capability to obtain and respond to market information at high speed with low latency. Typically, for these strategies to be successful, they must not only be fast, but they must obtain, process and react at higher speeds and with less latency than other market participants.
No arguments there, but how fast is fast?
My definition of HFT is currently (and yes it is a moving target): firms with a tick-to-trade (ie. being able to respond to market actions) of single digit microseconds (ie. less than 10µs). This means you are a specialist HFT firm, at the forefront of the speed race, and your strategies are pretty simple/stupid as they are no doubt implemented on FPGA technology. The many firms in the 3-digit microseconds latency (ie. 100µs+) I would argue are quick in human terms, but not HFT. Lots of prop trading and market making firms operate in this timeframe, and have very sophisticated automated strategies operating on servers running Linux.
To summarise my view of these questions on HFT:
Under these definitions, I would agree that HFT is generally bad for the market. By "bad" I mean it extracts a cost from the market, without providing a corresponding benefit (eg. liquidity). Predatory HFT is characterised by parties forcing themselves in-between two natural traders that would have otherwise traded, and extracting a 1c spread in the process.
I am all for traders protecting themselves (or being protected) against predatory HFT, but that is based on limited scope of HFT above. I would not want regulation to curtail what I consider to be the legitimate non-HFT activity of market makers and prop firms, using strategies which provide valuable liquidity. eg. Optiver, SIG, Tibra etal.
(i) Market integrity: HFT, particularly under stressed market conditions, can erode liquidity and contribute to a breakdown in orderly markets. The capacity of high frequency traders to flood the market with orders magnifies the risks to market integrity arising from automated trading in general.
Whilst I can understand the authors were thinking of specific instances of HFT activity, I think it misses the mark in terms of characterising the dangers of HFT.
HFT certainly does erode liquidity, but it doesn't do it under stressed market conditions. It does it all day, every day. Not only does it provide a disincentive for others to provide liquidity (via the threat of adverse selection), but also creates the appearance of liquidity that disappears when it is needed. Since they can move so fast they are able to provide what some people call "false orders" - that is, real orders that they can remove before anybody is able to trade against them. Maybe "shadow orders" would be more accurate? Thinking of HFT in stressed market conditions has been very popular since the flash crash: because you noticed it. What you don't notice is the constant drone of strategies being employed when the market is ticking along with no headline grabbing crashes.
HFT can contribute to a breakdown in orderly markets, but they certainly don't have a monopoly on that. One area that HFT excels in causing chaos is feedback loops - clusters of very fast algorithms start reacting to one-another, and being programmed the same way with a very stupid strategy (because they need to be so fast), can cause a quick spike or dip in prices. Several occurrences a day in the US is not uncommon, and they are starting to take hold in Australia.
The US equities market has shown how HFT can 'flood the market with orders', but this approach has yet to take off in Australia. My view is that that due to the message tax, predatory HFT in Australia will be a liquidity taker, not a liquidity (noise) creator. As such the whole notion of HFT's creating a flood of orders I think is a misunderstanding of the Australian market. Targeting regulatory action at such a non-problem will be of no benefit.
(ii) Fairness: Structural advantages (largely derived through technology) can unfairly redistribute profits from traditional long term investors to HFT traders. Many HFT offers to buy and sell are not bona fide and their objective can be to confuse other investors.
As long term investors, super funds are actually less exposed to HFT, as the exposure is proportional to the portfolio turnover. The consummate "day trader" is the one most exposed.
Whilst technically buy and sell orders on an exchange are always bone fide, the above interpretation is understandable given that if the fastest firms in the market have the ability to quote with little risk of being matched. These firm can withdraw orders should there be any mood shift in the market (eg. the orders ahead of them in the queue start to get executed). In this situation, these firms are able to withdraw their liquidity exactly at the time it would normally benefit the market. As such these orders can be thought of as like 'shadows' - constantly moving beyond your reach. This kind of liquidity is just an illusion as it cannot benefit you - disappearing just as you go to grab it!
While the technology-poor participant will argue: 'advantages derived through technology are unfair', are you really suggesting that advantages of all kinds should be removed? If so, why not regulate that brokerage is the same for retail as buy-side: that is an advantage. At the end of the day the marketplace is has a mix of participants with different advantages - that is not a bad thing in itself. We should look at the market rules before trying to consider some lowest-common-denominator approach to ensure that the fastest firms get slowed down. The free market can resolve these problems, it just needs the ability to innovate. Dark pools are an attempt to provide trading platforms more amenable to non-HFT, but I think the more regulated (lit) markets are yet to show their potential for innovation.
(iii) Efficiency: HFT, insofar as it is chartist trading, would appear to risk undermining the efficiency of the market, particularly their capacity to facilitate the allocation of society’s scarce resources at low social costs.
Not sure what's going on there, but in my experience human traders are chartists; and HFT focus more on sub-tick opportunities.
ISN recommends that ASIC impose a moratorium on HFT at the exchange level requiring orders that are received to be maintained until exposed to matching,
What is meant by this - how would you define HFT for the purposes of this short-term shutdown? I'm assuming that your definition of HFT is not as tight as mine, meaning all market makers would be offline..?
What you are advocating here is elimination of amend and cancellation instructions, so a placed order stays until someone matches it. This would actually create a disincentive to create liquidity, giving the HFT firms a leg up since they would have a captive market of targets that cannot withdraw their orders.
which matching would occur in pulses separated by meaningful periods of time.
This would have the effect of slowing the market down - in principle a good idea, however I prefer to allow the market to control the speed rather than exchanges to lock everyone onto the same speed. This idea is explored in TimeMatch.
This would be combined with complimentary rules against internalisation prior to such pulsing and the opportunity for potentially better prices.
Ah.. attacking the self-interest of broker crossings, and the foundation of dark pools; yes the brokers have held onto this one (against their collective interest I feel) since the market begain. Yes I agree that broker crossings are 'queue jumpers'; but not sure if it is possible to force a broker to always join the public queue on each occasion? A fine-tuning reform for another day perhaps.
The moratorium would allow technological and market developments to proceed only after the risks have been carefully studied by ASIC. Further, the benefits of technology could be brought to bear for all investors in the public interest, not just a small subset of traders.
Personally I think it is too late to de-automate the market. Perhaps the cleanest way to do it would be to bring the chalkies back, but then again they weren't perfect either (being tall and loud on the exchange floor had advantages)! Almost everyone is now using execution algorithms, and it's a very slippery slope between there and HFT. Having regulators choose arbitrary measures like quote-to-trade ratios, or randomised response times and the like is very dangerous. We are better off fixing the rules that created the game play in the first place; it is possible but we need to question some of the fundamentals.
Having technological progress limited by ASIC's study would be problematic to say the least, why not just allow the market participants to defend themselves - then you have a more adaptive response as the market inevitably evolves.
The problem with predatory HFT is that it only benefits themselves, there is no commensurate benefit (of competition) to the other participants in the market. Once we fix this anomaly, the technology effect will flip - and become a benefit, not a threat.
HFT firms' often rely on, and pay for, certain structural advantages: Co-location. By paying fees to the market centre or exchange, an HFT firm can receive a privileged location for the HFT firm’s trading computers, which allows the computers to receive data from the market centre more quickly, and send responses back more quickly.
In Australia, co-location is available to everyone on an equal access basis. Costs are fairly reasonable (sub $10k/month at the ALC). If I were building any trading business, I would be asking why not position all infrastructure at the ALC/Equinix, regardless of latency. Co-location at these centres is certainly not a ultra-premium service only affordable by a few.
Customised and time-advantaged data feeds received directly from the exchange or market centre. For a fee, HFT firms can receive market data faster than the public tape and the data might include information that is not included on the public tape.
In Australia these are available to everyone, and there is no special feeds available just to HFT firms with extra data than on the public tape. Contrary to much of the media reporting there are no flash orders or similar in Australia (on the lit markets at least, who could say about the dark pools?). Both exchanges use ITCH which is actually technically incapable of customising the feed to specific recipients (as a result of the fact it is so optimised for performance).
While there are a bunch of delivery technologies (ITCH, FIX, OM API, ITC) - they have different performance and usability characteristics. One is not designed to provide a faster feed at a higher price point. When considering the 'advantage' of having a technology that is 1 millisecond faster, you need to ask: what would you do with that millisecond?
Having a standard broking firm co-located at the ALC with that extra millisecond will mean: nothing. They do not have the technology or business model in place to make the most of it. We therefore shouldn't be singling out cases where there is faster technology, and say that that is a case of exchanges playing favourites to HFT firms.
Access to special telecommunications infrastructure. An example from the US: to facilitate trading strategies across securities markets (located in New York) and commodity derivatives 3 markets (located in Chicago), boutique telecom companies have laid special fibre optic cable that is more direct than the pre-existing cable between these cities (saving about 1.4 milliseconds in message travel time versus the original cable) and a microwave beam system (saving perhaps 4.5 milliseconds versus the original cable), with another microwave beam system under construction (saving perhaps 6 milliseconds versus the original cable).
Yeah, this stuff is pretty insane (and profitable - under the current market micro-structure). I'm not aware of any microwave links Equinix-ALC but I guess someone is at the very least looking at it.
While this is all pretty disconcerting from a competitive standpoint, just remember that these types of speed enhancements aren't really a concern for you. People using these kinds of telecommunications services are already faster than you, they are jumping onto even more insane services only to beat other HFT firms. If these really expensive, exclusive services did not exist - they would only be giving up profits to someone else in the speed club - you're unfortunately not a member.
For some HFT firms, affiliation with trading systems operated within the same corporate group (such as internalised trading platforms, including dark pools).
This is a major issue. I think it is fair to make dark pool operators get a license (a new kind - something more than an AFSL!); that provides oversight in the myriad of moral hazards that they face. I think the following obligations should be considered:
Whether dark pools should get a markets license, or a new type of license depends on who you work for. My view is that they should at least be better than an AFSL.
Some markets will provide participants, for a fee, with a “flash” of an order before routing it to another market for execution. [...] ASX has publicly confirmed it does not issue “flash orders” [...] Whether Chi-X issues flash orders is not readily disclosed to the public [...]
It is impossible for the ASX to have ever offered flash orders without it being widely known. However, the absence of flash orders is more to do with regulated order routing obligations (or lack thereof) than whether exchanges want to help HFT firms.
Similarly, Chi-X Australia certainly do not employ flash orders. I am sure they would verify this fact if asked.
We have been advised, but have not independently verified, that most Australian dark pools do not permit flash orders and in fact prohibit participation in the pool by high frequency traders (or offer an option for investors to prevent their order from interacting with HFT)
Yes talking to dark pool and HFT firms is a little like dealing with MI-5, you wont get much confirmation from them even about the obvious, because they are actors in a war. Depending on how strict your definition of 'flash order' is, one might be able to argue that polling an internal order book (or dark pool) could be similar to an 'internal flash order'? That's about as close an association as I could make based on my knowledge.
Yes dark pools are generally designed for non-HFT firms so they let you specify all sorts of execution limits like who you do (and don't) want to trade with. However as their name suggests, they are not as transparent and don't have the same obligations to their clients (at this point) as the markets operators have.
[...] regulators [...] have allowed a stratification of market data dissemination and trading messages to occur. Regulators have already determined that selective disclosure is unlawful -- if a listed company proposed to sell its annual report to a select group of hedge funds a day before public release so that those funds could trade on the information to the disadvantage of long-term investors, this would be enjoined and punished. This is true even though market prices would presumably move closer to fundamental value sooner as a result of the trading. Put another way, the regulatory interest in market fairness and rewarding judgement and analysis in investing outweighed potential efficiency gains arising from unfair structural advantages. The evolution of disclosure obligations in respect of market information has been allowed to evolve outside of the view of fundamental investors, without significant buy side consultation and participation, and an unlevel playing field has come about. We believe this difference between disclosure of fundamental information by listed companies and market information by exchanges should be carefully considered, studied in light of first principles of securities regulation, and remedied as appropriate
I guess what you are saying here is: treat market data (prices) the same way that you company announcements. The suspension of a stock during a price sensitive announcement would be analogous to the gaps in-between the 'pulses' of the market. There are then large swathes of time for everyone to become equally informed.
I can see what is being said: but I feel it is a little lazy for the buy-side to complain at this late stage of the technological advance of electronic trading. Firstly, I think the buy-side should have invested prior to this in technology (or at least funded the sell-side to), to counter-balance the situation. The buy-side have enjoyed a massive reduction in market access (ie. brokerage) as they have screwed the sell-side down and down. What the sell-side have done is sell the same simple execution services, in a reverse auction, while all the time having it cost the buy-side more through adverse execution. What should of happened is that execution services brokerage did not plummet to it's "almost zero" level - but the product kept up with the technology race.
In terms of an unfair playing field being created due to market data stratification, I would simply return to my previous point: no-one gets a really fast data feed unless you can actually do something with it. There is no point the buy-side getting the best data feed possible, if they don't have the right tools to go behind it. It is not the exchange providing different products (as long as on an equal access basis) that causes stratification, but the business models of the firms themselves. HFT's are fast because they have great data feeds AND really fast trading systems - buy-side would not waste the money on the really fast kit as they have nothing they can do with it with the extra millisecond - making no difference to their business. As such I would argue it is in fact not 'unfair' that the buy-side effectively has a slower price feed than HFT. This disparity is a result of the buy-side business model: it is not the fault of the exchange's market data policy to sell data that participants will consume at different speeds.
This is not to say that there is no unfairness here, but simply to point out it is not the exchange's market data products that create the unfairness. The unfairness is the market matching rules that cause it to award liquidity on a 'first in' basis.
Concerns about fairness break down into at least three sets of potential problems. First, that high frequency trading benefits may essentially be transfers from traditional positional investors, on the one hand, to HFT firms, on the other hand, arising from structural advantages, which transfers redistribute profits from long-term investors to HFT participants (and market infrastructure providers who receive a fee for stratifying data and market access underlying the structural advantages).
Having one type of firm make profit out of another firm, does not make it unfair (well, it feels like it if you're the one losing ;). For it to be unfair it needs to be established that what they are taking out of the intra-day trading 'pool' is more than what they are contributing: eg. liquidity, convenience, risk reduction etc.
In this regard, an example of note is the claim by Nanex Research that position-neutral HFT (i.e., quasi market making) appears to generate significant net losses to long-term investors. According to this research, HFT results in immaterial price improvement and a net cost to long-term investors measured in billions over the period studied.
You really can't use [http://Nanex Research (who I have the utmost respect for) to frame an argument in Australian equities, other than looking at what Australia may look like if we adopt the same market structure. These guys are great at exposing the crazy things that HFT are getting away with over there, but the markets are completely different as Australia does not have:
All the above features create opportunities for HFT, and is largely what Nanex are analysing. Nanex is very US-centric, and doesn't necessarily mean anything for the Australian market. HFT is extremely adaptive - and behaves differently in each market.
We are not aware of any other segment of society that tolerates the ordinary course making of offers that are not bona fide, to seek to confuse others, or to undermine the infrastructure of commerce.
Whilst I agree with the general thrust of this statement, I do hesitate on the distinction between seeking 'to confuse others', and seeking 'to protect one-self'. There are many techniques in trading used by buy and sell-side that are specifically intended to prevent others from figuring out what you are doing. To avoid being 'front-run', someone executing a large order will often introduce a lot of random behaviour to avoid others from figuring out what they are doing - and trade in front of them. Is this behaviour seeking to confuse others? Yes. This is one example of a legitimate reason to create confusion.
Lastly, when it comes to fairness, HFT expands the opportunities to engage in manipulative behaviour that is difficult to detect. [...] the capacity to police for HFT conduct has, however, been questioned.
Absolutely. ASIC have a hard enough time keeping up with humans, let alone algorithms. It is frustrating to see algos get away with the same actions that humans wouldn't. My view is you want to (1) remove the need for this type of enforcement as possible, by allowing the market to police itself; and (2) have very smart and efficient surveillance for the rest. I think the former can be achieved, in part, by allowing liquidity providers to control their orders better (as already described) plus having exchange-level algorithms made available.
Exchange-implemented algorithms provide a perfect level playing field for all participants, and increase the risk of unwanted execution by those implementing deceptive techniques such as layering. These are essentially 'zero latency' algorithms that cannot be out-paced by HFT firms, and give equal access to all participants.
Yet that is precisely what is happening when market centres provide selective advance disclosure through co-location, privileged data feeds, and other advantages paid for by high frequency traders
I think the way to refute this claim is via a thought experiment: if you took the feeds of the non-HFT firms and made get their data a full millisecond faster than those taking the current 'high speed' ITCH feeds; there would be no impact. Whilst highly automated, technology-focused firms are very sensitive to latency, firms that are driven by human response-times, are not.
It is very appealing to think that there is a conspiracy against the non-HFT firms, in the way that exchanges are dealing with them. Maybe this is the case in the US, but certainly not in Australia. If anything there is a conspiracy of denial by the non-HFT's as to who should do something about the advantage that the HFT's have obtained. Much of the buy side is blaming everyone else (except themselves) - whereas I say much of the blame lie closer to home.
The rise of HFT has the potential to shift the character of markets, making them more technical and less based on fundamentals.
Yes, I would argue it is the market rules (combined with fast technology) that creates an environment of advantage for technology-driven trading strategies. HFT is just the strategy that is most profitable in this environment. ie. HFT is the symptom - not the problem.
With respect to the costs of markets as they perform their capital allocation function, the infrastructure costs underlying HFT are significant. HFT has produced explosive growth in trades, but even greater growth in quotes or orders (most of which are cancelled).
HFT cannot be blamed for higher numbers of, and smaller, trades. These are driven by algorithmic execution tools (buy-side driven), which are not HFT. There has been no respective explosion in quote numbers in Australia (unlike the US).
We are not aware of any estimates on a financial system-wide basis of the costs to continually expand capacity, and for market participants to receive and wade through the data. On top of this would be regulatory costs to surveil this behaviour. We understand that ASIC has increased its budget allocation to electronic trading matters significantly. Similarly, the opportunity costs of talent and R&D to produce ever-faster market interaction have not been quantified, but likely are significant.
Absolutely true. At the big picture level -this technology war is just not adding value, and I think it will largely disappear (on the scale we are now seeing) if we take the market rules back to ones that support the natural market (what ISN calls fundamental traders).
ISN supports ASIC’s efforts to reduce the risk of disruptions to market integrity arising from poorly designed algorithms, including the requirement that market participants using automated trading have appropriate filters and a kill switch.
I feel the 'kill switch' is flawed policy, for many reasons. For ultra-fast HFT, the switch will be so lightweight (in order to be fast) as to be useless. For the rest, the policy will only be as good as the quality of the kill switch implementation, which of course will vary. A kill switch would not have helped Knight.
My view is that these safety switches need to be implemented at a more trusted level: the executions venues (exchanges). This provides the benefit of no-disadvantage (in terms of speed), and not needing to trust that every participant is doing the right thing. These are tasks best performed at scale, not by every firm individually. It is always the weakest link that will break.
I usually try to start with the parts of a story I agree with, but lets just start with a part that I'm confused about:
[...] Australian Securities Exchange chief executive Elmer Funke Kupper this week said he wanted to "tighten the economics of the practice" to make it less profitable.
I've seen this quote a bit since the interview it came from, but I'm really interested: how will you make it less profitable? By charging them more (and become a de-facto partner), or by enabling the traders they make money from, to avoid giving it away. I'd argue that TimeMatch achieves the latter.
[...] market operator Chi-X has helped to enable the trend towards HFT, Australian Securities Exchange chief executive Elmer Funke Kupper [...]
Chi-X has not helped HFT any more than the ASX, who actually have trading platform dedicated to it (PureDeal) plus extensive infrastructure to support their operation (ALC). Not sure you can point fingers on that one towards Chi-X, I think the more accurate message is: market fragmentation (caused by competition) has helped to enable the trend towards HFT.
The ACTU will join the debate today by seeking a moratorium on all high-frequency trading, so that the regulator can do a full study of the costs and benefits of the practice. Failing that, the peak union body wants to stop the very fastest trades by imposing a "minimum resting time" of one second for all transactions on Australian markets.
Minimum resting times (stopping traders from canceling orders until one second after the order was created) wont really do much to curb the ill-effects of HFT in Australia. This would make a massive difference in the US (though one second would create a lot of unintended consequences) to eliminate quote stuffing, but we don't have those issues in Australia.
The detrimental HFT activity in Australia is more characterised by the speed of an HFT trader responding to an order that appears from a non-HFT trader: called 'taking liquidity'. As I've blogged about with my TimeMatch scheme, this race creates no value for anyone except the HFT trader. It does not create the traditional benefit of greater market participation: increased competition (read: better prices for the non-HFT trader).
My personal view, a one second minimum resting time would be a disaster and make the HFT situation worse. This scheme actually exposes traders even more to other HFT predators who know that other traders are unable to remove their orders when they want to, within that second. In other words, the good HFT would be penalised, and the bad HFT would be rewarded. Go figure!
Another recommendation in the ACTU's formal submission to ASIC, obtained by The Australian, is a ban on "flash orders", in which traders gain advance notice of incoming orders, buy up stock ahead of the imminent purchase and sell them to the buyer.
Flash orders are bad things, I agree - but they don't exist in Australia.
My somewhat cynical view of them (in the US) is that they really are a by-product of badly designed market regulation, not the pure invention of HFT's. One example of how rules intended to help you, can end up harming once you realise that light does not travel at infinity.
It says additional fees should also be imposed on HFT operators who cancel orders regularly, seen as a common tactic by traders moving in and out of the market very quickly.
Traders moving quickly is not necessarily a bad thing, it can just mean they are being more efficient and able to (in the case of a market maker) able to quote tighter buy/sell spreads. This kind of crude mechanism should be at the end of a very long list of things we should try first, the simplest being allow the market itself to decide how fast it wants to operate.
Last Sunday the ABC broadcast a reasonable review on HFT: Attack of the algorithms. A little sensationalist and populist in broad assertions (we've got used to far worse!), but there was a lot of good content. However, what it lacks I think is the ability to communicate the magnitude of the problem. In other words: how much is predatory HFT "costing" other investors.
Some great points were made:
And then you’ve got to ask yourself, ‘Well, who funds this?’ And at the end of the day, some of the high-frequency trading gains have got to be made from the more traditional participants in the market such as the superannuation funds.
Great point - not that we should automatically think that HFT is 100% 'drag' on the market. You should expect to pay something for liquidity, ie. for the ability to trade what you want when you want. I think we are paying for liquidity, which is fine; but in other cases we are also just paying - for no benefit in return.
High-frequency trading, HFT, makes the public uncertain of the integrity of the Australian Securities Exchange. Professional stockbrokers such as myself have reduced confidence in the ASX as a result of observing HFT. If an operator manually entered HFT trade types, we would be penalised for manipulative trading. There should not be one rule for man and another for machines programmed by man.
Exactly! One major inconsistency has been the fact that humans seem to be able to be accused of things that machines can't. I remember discussing with the ASX some time back that central to the definition of market manipulation is: what is in the mind of the person placing the order? What if there is no person, and therefore no mind.. just an algorithm trained to look for profit generating feedback loops. There you have it - a technology driven pump-and-dump module!
Human traders now have a perfect excuse when getting a 'please explain' on some orders (perhaps too aggressive in posting/deleting orders): 'oh that's an algorithm'.
Stan Correy: The most common criticism of high-frequency traders is that with state-of-the-art technology plugged into the stock exchange data centre they can manipulate the market; by getting information about share orders milliseconds before anyone else they can manipulate prices. In the market, it’s called front running. Paul Hilgers (Optiver): Front running means that I have an information before anyone else has an information. Front running means I have access to order flow other people don’t have. Well, I can only speak for my firm here—but I know a lot of my competitors—we don’t have any clients. We use our own money, our own capital, to trade and provide liquidity. So how can I front run order flow I don’t actually have? The information we use is available to everyone.
Exactly - doing things very fast is not front-running; go Paul!
I any case, I would not even categorise Optiver as an HFT. Depending your level of cynicism, they are either too slow (specialist HFT's are 10x faster), or too useful to the market to be HFT. Optiver are a market maker, and as such add valuable liquidity to the market as a whole. Optiver are more prey than predator in the speed game, as the sophistication of their models are too complex to be implemented in the super-fast FPGA technology employed at specialist houses.
It's a common misconception (to think Optiver is HFT), but it goes to heart of the issue: the mantle of HFT goes to the fastest in the market. Get consistently overtaken by others, and you are no longer HFT.
There is a way that algorithmic traders ('slow' and HFT alike) can, in Australia, carry out a sort of front-running in Australia. It is where algorithms specifically sniff-out those looking to execute large trades in the market, by identifying the patterns. I think this is what many people are really talking about, rather than the technical definition of front-running. This is obviously pretty grey, but my view would be that traders (usually other algorithms!) shouldn't be so predictable if they don't want to be read.
"Alan McGrath was watching high-frequency traders at work. Each transaction of 30,000 shares was a blink and you’ll miss it moment, generated by algorithmic computer programs, also known as robot traders."
These emerging items on FKP are orders, not transactions (trades). They were not arriving in rapid succession (a small batch once per second), and were easy for a human to comprehend and respond to. This example cannot be categorised as HFT - but instead a badly behaved (buggy) algorithmic trading program.
Remember: all HFT is algorithmic trading (usually really simple algos), but not all algorithmic trading is HFT.
"The share market is a place where businesses are valued. It should be a place where people have confidence that the businesses are fairly valued, based on events—macro and micro. That’s all that should be affecting markets. It should not be affected by computer programs and mathematicians. When you turn on the stock market and make it a game, a casino, you are creating havoc."
Share markets are more than a place where businesses are valued - they are also a place where liquidity is valued. That is the price you pay for being able to trade at the time you want to. Remove computer software and mathematics from the market and you'll find it a very lonely place waiting for natural counter-parties to arrive. Whilst HFT may not always improve the process of valuing of businesses, it can help in improving liquidity. Just remember that HFT is like bacteria: it's not all bad. Overdose on antibiotics and you may regret it.
Before Usain Bolt has even lifted his feet off the blocks, the trading algorithm has already bought and sold millions of shares, or flooded a market with false offers to buy and sell. Then they’re gone.
The orders that HFT (or algorithms) generate are never 'false'. They are just generated by people who can move very swiftly in the market, and probably withdraw the orders before anyone can execute against them. They are the equivalent of jockeying in football, where players shift their body to create an impression of movement, only to do something else. The movements are real, just like the orders.
Given that the market rules in the US (unlike Australia) allow cost-free creation and deletion of orders is it any surprise that, like with email, there is a lot of spam? These 'spam' orders are certainly a problem, but not because they are 'false'. They are a problem because everyone else in the market (actually, it's mainly directed at other HFT firms trying to keep up) must sift through this information to figure out what is really going on.
In the US, on the New York stock exchange, about 70 per cent of all trading is high-frequency trading by these computers and algorithms and these are not going away.
This is a widely quoted statistic, as it on the face of it, quite shocking. While it is technically true, it is hiding what is really going on. The reason this number is so high, is not because the proportion of HFT is actually that big - but because almost all the non-HFT orders have been stripped away by the time it gets to the NYSE. It demonstrates NYSE's lack of non-HFT market share (or [[[blog:order-flow |toxic order-flow]]]), not the fact that the overall market is dominated to this degree by HFT.
So their business model relies on using better computer technology, which is better predictive algorithms, faster line speeds, effectively trading in nanoseconds to take small amounts of profit on what is billions of dollars of trade. And then you’ve got to ask yourself, ‘Well, who funds this?’.
Great question, but nobody trades in nanoseconds. Sorry, it takes a nanosecond for light/electricity to travel about 30cm - and takes the Australian exchanges (who are very fast) about 100,000 nanoseconds just to match a trade.
The impact of HFT, high-frequency trading, on Australian super funds is an emerging issue, with the potential to make our retirement savings more vulnerable.
HFT does not make your savings more vulnerable, it makes you more share trading more vulnerable. Whilst you could argue that concerns over HFT impact the perceived integrity of the asset class, it is really only going to be an issue when you buy and sell. Hopefully your superannuation is not being day traded, if so I hope you know what you're doing.
Some of them are very good for the market and play an important economic role. They provide liquidity; they provide what we call price discovery, or price improvement.
A small point, but HFT certainly does not provide price improvement. Price improvement is where you give a trader a better price than he was expecting, or prepared to trade at. A successful HFT strategy is one that gets the trade before anyone else does, and giving the dubious benefit to the counter-party of executing his order 1 microsecond faster than your competitors (who all gave it a shot) - not paying any price improvement.
But regulators are increasingly aware of the risks posed by these so-called [HFT] bad guys, which I’ll call parasitic traders. In the regulator community they call them ‘robots gone mad’.*
'Robots gone mad' is a good title to give to a buggy trading algorithm; but it's not necessarily anything to do with HFT. A dodgy algorithm doesn't have to be fast to cause damage.
'Parasitic traders' are HFT traders that use their speed to leech out profit by sitting in-between two counter-parties that would have otherwise traded together, and extracting a profit in the process. They don't add any value to the market (since the counter-parties would have traded anyway) but forced their way into the value chain.
Whilst I understand why you might associate parasitic traders and HFT, they are distinct from 'robots gone mad'. The idea that anyone such as a regulator could confuse these two is of concern.
Now, we had a flash crash and a flash rise in a stock called David Jones in Australia on 30 June. And no one thought on that day to turn the stock off. It went up on speculation which was later denied, yet the stock market let it trade: money was made, money was lost. No one thought about turning that little stock’s market off—just one stock, not the market—no one thought about it.
What happened to David Jones is not new. Remember the AMP float in 1998? What happened was an sharp irrational rally (read: flash rise) and it was driven by humans. The price spiked up to $45 just after listing in a buying panic, only for the price to just as quickly crash and close the day at $23. Markets are sometimes chaotic beasts, with so many participants responding to the actions of other participants in tight feedback loops. We need to be careful to keep things in perspective and not blame algos/HFT for every market aberration. Prices can be unstable in times of uncertain supply/demand.
Alan McGrath: "I don’t not what [colocating in ASX's Liquidity Centre] costs, but I’m sure it costs, you know, in the tens of thousands of dollars to have your server there. And I know there’s articles that they’re all on exactly the same length’s cables, so that one doesn’t get an advantage from the other. And, yeah, it’s a level playing field for those that want to spend the money to go that way, but certainly from a private trader that’s not an option.
It costs under $10k/month to have access to the top-tier price feeds in the ALC, and locate some of your own servers here. ASX has done a great job at providing genuine equal-access to this key infrastructure.
The question is: what would the private trader do with a price feed that was able to process prices a few milliseconds faster at the ALC, than the private trader does at their desk? The answer is of course: nothing. The reason the algorithms colocated in the ALC can respond to price action faster than the private trader is that they are computer programs that can act faster than humans. In fact the reason the algos move to the ALC is not to beat other private traders (they achieve that in any case), but it is simply to beat other algos that they compete with.
Put the private trader in the ALC, and it won't make any difference. However, give the private trader some algo tools to trade on their behalf, that are co-located in the ALC etc. and you might be onto something. But is it the HFT firms 'fault' that the private trader is not doing this? There is a need for the brokers to provide tools to private traders like Alan McGrath, allowing them to compete. Is this not a shortcoming of private traders and their brokers initiative?
Here are some examples of how the TimeMatch public order book would work. You may want review the premise of this design in the previous posts, as well as the terminology. The following key is used to interpret the status of various orders in the examples:
|Potential||Yes||Yes||Yes||Potentially matching against another order, if the market is stationary.|
|Hopeful (taker)||No||Yes||Yes||Taker is committed to the auction, could be removed if another taker offers price improvement.|
|Auction (maker)||No||No||Yes||Order will match against the opposing taker (Hopeful order) or at another improved price.|
|Trade||No||No||No||Order is executed and complete on both sides|
We'll start with a simple example that shows the way the public order book works today - ie. instant execution with no auction period:
The notation for bids:
Party: the counter-party owning the order
Time: the length of time that the party wishes to participate in auction, after price overlapping
Price: limit price of the order
|0.000||10 (0) Z|
|1.000||X (0) 10||10 (0) Z||Z X @ 10|
The above table shows an order book a seller "Z" showing an order which is matched instantly by "X". Both parties are asking for instant execution - denoted by "(0)" in both cases which means the auction time period is zero. The "10" is the price in both cases, so we have the simplest of trade matches.
|0.000||10 (1) Z|
|0.001||X (0) 10||10 (1) Z|
|1.001||X (0) 10||10 (1) Z||Z X @ 10|
Here we have a mismatch of time periods: the seller wants to trade on a 1 second auction, whereas the buyer wants to trade instantly. In this case, since the market is stationary, the order book will match them up as nothing has occurred in the intervening second. There is no 'auction' per-se in this situation, and someone could have jumped over the top of buyer "Z" at any time up to the match. This type of "stationary non-auction" scenario is present to merely provide natural liquidity between investor types when nothing is moving and there is no additional time period risk to the liquidity creator.
|0.000||Z (0) 10|
|0.001||Z (0) 10||10 (1) X||X Z @ 10|
The liquidity creator Z has requested instant execution. Liquidity taker X then enters with an exposure period of up to 1 second. This match is done instantly since the taker's time period is greater than or equal to the liquidity creator's. Since there is no adverse selection of the slower order (since it arrived later), this is ruled to not expose them.
Of course there is some potential for latency arbitrage here if you know that X is about to hit the order book. This is an issue that can be addressed - and an area for further refinement.
|0.000||10 (1) Z|
|0.001||X (0) 10||10 (1) Z|
|0.002||Y (1) 10
X (0) 10
|10 (1) Z|
|1.002||Y (1) 10
X (0) 10
|10 (1) Z||Z Y @ 10|
|1.002||X (0) 10|
In this example buyer X has come in to take Z's liquidity, but is not willing to bid in an auction - only willing to trade immediately. At this point Z does not know whether X is more informed. Immediately after this, Y jumps in front of X on the queue and is willing to submit to an auction of 1 second, which is good enough for Z. A 1-second auction is started, and since no-one out-bids Y with a higher price the trade is matched. X is left unfilled.
|0.000||10 (1) Z|
|0.001||X (0) 10||10 (1) Z|
|0.002||Y (1) 10
X (0) 10
|10 (1) Z|
|0.003||X (0) 11
Y (1) 10
|10 (1) Z|
|1.002||X (0) 11
Y (1) 10
|10 (1) Z||Z Y @ 10|
|1.002||X (0) 11|
In this example there is a strange situation where the price divergence on the basis of latency. While a fast trader outbids a slower one on a price basis, they are not willing to commit to an auction. Price improvement cannot be passed on to Z in this case because doing so would encourage late and informed orders to arrive at the end of auctions - handing them the advantage of adversely selecting their opponents. In this scenario Y was granted the match at a lower price due to the fact they committed to the auction first, thereby rewarding them for their willingness to expose themselves to the auction.
|0.000||10 (1) Z|
|0.001||X (0) 10||10 (1) Z|
|0.002||Y (1) 10
X (0) 10
|10 (1) Z|
|0.500||Y (1) 10
X (0) 08
|10 (1) Z|
|0.501||Y (1) 10
X (0) 08
|09 (0) W
10 (1) Z
|0.600||Y (1) 10
X (0) 08
|08 (0) W
10 (1) Z
|W X @ 8|
|0.600||Y (1) 10||10 (1) Z|
|1.002||Y (1) 10||10 (1) Z||Z Y @ 10|
The scenario shows where fast prices can move throughout auctions for slow trades. Here the price drops and executes while an auction is still in progress - with different prices. The trades come out in a different order to the order instructions that generated them. We can see here that traders are not held-up at all by slower traders, they all progress concurrently.
|0.000||10 (1) Z|
|0.001||X (0) 10||10 (1) Z|
|0.500||Y (0) 11
X (0) 10
|10 (1) Z|
|1.100||W (0) 12
Y (0) 11
X (0) 10
|10 (1) Z|
|2.000||V (0) 14
W (0) 12
Y (0) 11
X (0) 10
|10 (1) Z|
|2.500||W (0) 12
Y (0) 11
X (0) 10
|10 (1) Z|
|3.000||U (1) 13
W (0) 12
Y (0) 11
X (0) 10
|10 (1) Z|
|3.500||W (0) 14
U (1) 13
Y (0) 11
X (0) 10
|10 (1) Z|
|3.600||W (0) 15
Y (1) 14
U (1) 13
X (0) 10
|10 (1) Z|
|4.000||W (1) 15
Y (1) 14
U (1) 13
X (0) 10
|10 (1) Z|
|4.500||W (1) 15
Y (1) 11
U (1) 10
X (0) 10
|10 (1) Z|
|5.000||W (1) 15
U (1) 10
X (0) 10
Y (1) 09
|10 (1) Z||Z W @ 15|
|5.000||U (1) 10
X (0) 10
Y (1) 09
This is a fairly typical latency arbitrage scenario, where a liquidity creator Z is exposed by news creating bidding amongst liquidity takers. Instead of handing the order to the fastest trader, the order book requires takers to commit to a time period before allowing them to lock an auction in. Prior to the auction being called there are no executions, despite overlapping price since the market is not stationary.
What we've ended up with is a market where the informed investors efficiently pick-up profits from the uninformed investors. Given this environment it is no surprise that the market has responded with a complex array of execution alternatives. The explosion in the choice of execution venues has been swift - both additional exchanges and dark pools, MTF, ECN. These options create complexity, additional cost and fragment the liquidity that is available - so there must be some perceived threat that is causing participants to leave the transparent public order books for the uncertain world of alternatives. Of course the objective of avoiding more informed investors is only one reason why people opt for dark pools - information leakage probably being the primary reason. But as the available technology increases (both in speed and smarts), the disparity between you and the Usain Bolt of traders widens, thereby making you increasingly uninformed on a relative basis.
The structure of the market today is a result of the battle between uninformed and informed investors. Uninformed traders have the temptation to leave the exchanges for the safety of a dark pool in order to achieve protection - but at what cost? There is certainly a cost paid by the market as a whole, so is it worth it?
Dark pools can be like a canary in the cage - if they're significant, it indicates there is a problem. The problem is not that dark pools exist, it is the fact that anyone wants to use them. It should be noted that there is not a trend away from the public order book towards dark pools in Australia. Contrary to popular commentary, traders are not moving from the public order book into dark pools. They are moving from other off-market trading schemes (like broker crossings).
Solving the problem of the informed-imbalance is one I want to tackle (next), but prior I want to make the point that it is one that should be solved in the public order book. Everyone should have equal access to the order book, and be able to give and take liquidity.
Some exchanges, like NYSE, have suffered from toxic liquidity because the uninformed orders that everyone is looking to trade against, have been removed by the time they get there.
NYSE's retail liquidity program is looking to bring back some price competition for uninformed (retail) orders, and thereby encourage uninformed liquidity back. Is it any wonder - they have already lost all the retail order flow, so now they have nothing to lose.
The ASX is a good example of an execution venue that provides and has always provided good equal access - eg. non-market-makers are not prevented from providing quotes to the orders book. The Australian retail brokers don't sell their order flow to other firms. So the ASX is one place where you do get a healthy mix of liquidity: informed and uninformed alike. Going forward, however, I see an issue as dark pools try to take liquidity away providing the uninformed investor with the promise of protection against information leakage and some HFT techniques such as latency arbitrage.
Australia is at a critical juncture. One path leads to "choice", complexity, non-transparency and market fragmentation - requiring the help of HFT to smooth the prices via arbitrage (at a cost). The other leads to maintaining liquid, transparent public order book market that has a healthy mix of investor types dealing with one another. As traders consider the impact of ever increasing technology 'arms race', they are encouraged to choose: better to hide or safety in numbers?
My hope is that we can reforming the public order book such that uninformed investors are looked after, without having to resort to liquidity fragmentation in order to pursue price improvement.
TimeMatchno-TM is my current thought experiment on designing a modern public order book. Flowing on from my previous posts, it contains special features to discourage uninformed investors from fragmenting the liquidity of the market via disparate venues in the hope of achieving better prices. The idea behind TimeMatch is that we improve the public order book to the point that uninformed investors experience no speed/technology disadvantage, but continue to benefit from the larger liquidity pool.
You could think of TimeMatch is a public order-book alternative to NYSE's retail liquidity program.
One of the relatively unquestioned aspects of order book design is price-time priority. It says that you should get priority of execution if you post the best price. If the price is equal to someone else's - then it goes to the one who was first. If you think in terms of human processes, this intuitively seems fair. Give me the best price available, but give it to the guy that's been in the queue for the longest amount of time. No queue-jumpers please?
Over time the dynamics of the market have changed fundamentally with speed of technology. We now operate in a market where the speed of the market is at least in part working against the interests of uninformed traders.
For uninformed investors, execution speeds of a split-second are undoubtedly better than that of several seconds - we can work more efficiently and have a shorter period of uncertainty particularly when working with several orders at once. I would argue, however, there is no benefit to the uninformed investor if their order is executed in a few microseconds, as opposed to 1 millisecond. (it probably takes your LCD screen ~7 milliseconds to update, so no - you won't notice :)
When response times get very low, a technology race emerges and the benefit of an ever-faster market platform shifts to the trader that can move the fastest. When trading against an uninformed order, the high-speed algorithms all race to trade against it first. This means your order was awarded to the firm that was able to move the fastest, getting you only a benefit of faster execution - not a better price.
So yes, speed is good - but it's not everything. At some point I don't value it at all. I have no problem with Usain Bolt being faster than me, but I just choose not to race him!
So how then do we meet the needs of a variety of traders in a single venue, given that some prefer speed over all else (HFT) and others would prefer a better price (buy-side/retail)? Instead of complex and inefficient schemes such as mandating minimum order resting times, slow moving markets or imposing targeted HFT taxes - how about we let investors decide?
An investor should decide, when submitting an order, the time period over which the execution (or auctioning) of the order will take place. This could be thought of as "price-time-time priority". Orders would only be able to match against one another if the time period (supplied by the buyer and seller) and the price are overlapping.
Once a match has taken place, an auction would be conducted for the time period specified. This would allow any other trader to out-bid the liquidity taker, in supplying price improvement to the liquidity creator. Thus the liquidity creator has the potential of price improvement from other bidders who may arrive at the market after the initial bidder.
This market-based solution allows for the participants themselves to decide how quickly they want their orders to be executed. A scheme such as this has the potential to address the issues of adverse selection and latency arbitrage where faster traders monetise their speed advantage from the market.
The effect of this kind of order book would be that each trader in the market would analyse what their execution timeframe is. HFT firms would choose "time 0" for instant execution of their orders (no auction), while a human trader may chose to run a 200 millisecond auction for "fast enough" execution with the potential for price improvement amongst the algos.
You would then see a reversing of the trend that occurred in development of high speed matching venues. Instead of uninformed investors 'cheap' orders being disproportionally picked up by HFT, they would go back to matching more naturally against other uninformed investors. This is exactly what off-market trading mechanisms such as broker crossings, and dark pools try to achieve, but creating a lot of other question marks at the same time. This would result in improved prices for both uninformed sides, with no cost. Of course HFT would still be allowed to compete on the same basis as humans, but their speed would not be able to be used to obtain access to the cheap orders of those who don't value it.
The best thing about this method is that the market regulates it's own speed. If the market decides it wants to keep trading 'instantly' with no auction - then all traders will send their orders with "time 0". I think once the buy-side do their trade cost analysis, they will quickly find the answer as to whether this is of benefit to them.
This would help achieve my stated aim of building better liquidity - helping prevent the departure of the most uninformed orders to fragmented venues.
Next we look at some practical examples.
(Note: the context of this discussion is short time intervals, and by that I mean shorter than human response times)
There are many different ways to categorise different types of orders, but for our purposes lets put them in buckets of: informed and uninformed. This is the actual terminology used in the firms who deal with order flow.
Uninformed order flow are orders created by people/machines who aren't "in the know". They have no unique analysis or advantage over others in the market. For all intents and purposes these orders are pretty random, and not time critical. A human deciding and manually keying in an order to buy a stock in their break at work is a good example of this kind of order flow.
Informed order flow are orders that are made by people (read: computers) who are the more informed part of the market. They are more informed for two reasons: they have smarter strategies/methods to predict future prices or they are just faster at processing the publicly available information that everyone else has.
In general the objective of short-term electronic trading is to trade against uninformed order flow as much as possible. This type of order flow is not as smart, and on average will deliver you a profit. Trade against informed order flow, and the odds are seriously stacked against you. These are the orders that all the smart guys avoid, and leave for someone else to pickup.
How do you tell the difference between uninformed and informed orders? Well, there's a number of factors here, but mainly: who placed the order, and is the order a good deal for me. One of the benefits of trading via an exchange is anonymity. We are prevented from knowing who placed an order, so we are then left to take a guess. based on the available information.
Here are a few techniques for finding uninformed orders:
The US and Australian equity markets are quite different to one another when it comes to how firms in general exploit order flow:
Historically US investors have been used to sending orders to market makers, and not having direct access to the order book on an exchange. This allowed retail clients to only take the quotes offered by market makers/specialists, but not to provide quotes themselves. Therefore, a market makers' ability to make profits was linked to the amount of order flow coming from the uninformed sources (over which they have no control), and price pressures from their competitors (which they can). The market maker with the better price and tighter spread would get the business.
What happened next was that firms got aggressive in wanting more exclusive access to this order flow. They did not want to compete on price (which transfers profit to the uninformed investor), so instead they engaged in buying order flow.
This involves paying an uninformed order source (like a retail brokerage) money to send them orders instead of going direct to the specialists/market makers - potentially subsidising the brokerage rates of the retail brokers as they sought out more 'uninformed investors'. This allows them to have first right of refusal of any orders, before it is exposed to the wider market via some kind of exchange. They are aware, for example, if there has been a slight dip in an overseas market or a similar stock, the instant an order comes in from a retail client for a security. This allows them the exclusive option to take or leave the order depending on whether it is a good deal for them. This is called adverse selection.
In Australia, we have always been more used to having direct access to the central limit book in that the market was not dominated by market makers. In order to obtain access to the uninformed order flow, firms had to compete in an open, transparent and competitive market on the ASX.
In order to gain exclusive access to retail order flow over-the-counter CFD providers emerged. They were able to achieve the same thing - get access to the unformed order flow, without having to compete for the business on a price basis. They can choose which trades to take on, and which to effectively pass through to the exchanges. Knowing the client (thanks to regulations requiring them to establish investor "suitability") allows them to drill further into analytics on clients, and look at other clues to determine patterns of which clients may or be profitable to bet against.
What we have today is a market that is very efficient at removing uninformed orders. Unfortunately this creates order flow residual that only contains informed order flow, or 'toxic order flow'. It are the orders that you don't want to trade against (if your objective is to make a profit).
Order flow toxicity is one postulated contributing factor to the 2010 flash crash, since the levels in the hour prior to the crash were quite high. High toxicity means market makers are losing money, and withdraw from the market to protect themselves from further losses. This loss of liquidity then leads to increased volatility and larger price movements.
There are measures for order flow toxicity such as VPIN; which calculates the probability that orders are informed. You can think of retail order flow being near 0, and orders from a know-all computer with really smart predictive capability being near 1. Some have proposed monitoring VPIN in the US in real-time, with the SEC estimating it would cost several billion dollars per year just to do that.